Remote Sensing of Wet Forests Proefschrift · Remote Sensing of Wet Forests Proefschrift ter...

RIJKSUNIVERSITEIT GRONINGEN

Remote Sensing of Wet Forests

Proefschrift

ter verkrijging van het doctoraat in deWiskunde en Natuurwetenschappenaan de Rijksuniversiteit Groningen

op gezag van deRector Magnificus, dr. D. F. J. Bosscher,

in het openbaar te verdedigen opdinsdag 18 september 2001

om 13.15 uur

door

Joost Johannes Martinus de Jonggeboren op 3 april 1969

te Groningen

Promotor : prof. dr. ir. P. J. C. KuiperReferent : dr. W. Klaassen

Beoordelingscommissie : prof. dr. J. T. M. Elzenga prof. dr. R. A. Feddes prof. ir. P. Hoogeboom

ISBN : 90-367-1492-3

Remote Sensing of Wet Forests

Joost de Jong

This study was carried out at the Centre for Ecological and Evolutionary Studies (CEES)from the University of Groningen, Faculty of Mathematics and Natural Sciences, withfinancial support from the Space Research Organisation Netherlands (EO97/048). ERStandem images were provided by the European Space Agency (AOT.NL302).

Map ref. 41° N 93° W(Lewis/Newman/Gilbert)

An unseen ruler defines with geometryAn unruleable expanse of geographyAn aerial photographer over exposedTo the cartologist’s 2D images knowsThe areas where the waters flowedSo petrified the landscape growsStraining eyes, try to understandThe works, incessantly in handThe carving and paring of the landThe quarter square, the graph dividesBeneath the rule, a country hides

Interrupting my train of thoughtLines of longitude and latitudeDefine and refine my altitude

Contents

Abstract

General Introduction1.1 Global water resources 91.2 Rainfall interception by forest 101.3 Rainfall interception models 151.4 Rainwater storage measurement techniques 181.5 Identification of potential radar application 191.6 Radar sensing of forest 201.7 Thesis outline 22

Rain Storage in Forests Detected with ERS Tandem Mission SARAbstract 232.1 Introduction 232.2 Models 252.3 Study site and data 282.4 Data processing 302.5 Results 312.6 Discussion 352.7 Conclusion and recommendations 37

Radar Measurement of Rain Storage in a Deciduous TreeAbstract 393.1 Introduction 393.2 Radar 403.3 Site and instrumentation 433.4 Results 453.5 Discussion 513.6 Conclusion 52

Monitoring of Rain Water Storage in Forests with Satellite RadarAbstract 534.1 Introduction 534.2 Model 554.3 Validation 584.4 Sensitivity analysis 634.5 Discussion 66

Estimations of Rainwater Storage in a Deciduous Forest Canopy by SatelliteRadarAbstract 715.1 Introduction 715.2 Site description 725.3 Measurements and methods 735.4 Results 775.5 Discussion 845.6 Conclusion 86

Potential of Radar for Measuring Forest Wetness6.1 Introduction 876.2 Radar observations of wet forests 876.3 Sensitivity of radar backscatter to forest wetness 896.4 Minimising the influence of soil backscatter 906.5 The potential of radar for monitoring rainwater storage in forest canopy 91

Literature 95List of symbols and abbreviations 103Samenvatting 105Nawoord 109Curriculum vitae 111

Abstract

This thesis deals with the determination of the amount of rainwater that is retained inthe canopies of forests during and after a rainstorm. Satellite radar is used as a tooltowards this purpose. The relevance of this research topic arises from the increasingscarcity of fresh water for human society. A quarter of the global land surface iscovered with forests. A dense forest canopy intercepts most raindrops and stores themtemporarily on the surface of leaves, needles and branches. The canopy of a foresttherefore acts as a water reservoir at the boundary of the solid earth and atmosphere.Forest rainfall interception models can be improved, when temporal and spatial data ofthe amount of rainwater storage in forest canopies become available. Whether radarcould fill this observational gap, is investigated as follows.

A model that simulates the radar backscatter from forest was adapted to includestorage of rainwater in the canopy of the simulated forest. The first simulationsindicated that radar backscatter is more sensitive to rainwater storage in deciduousforest than to storage in coniferous forests, due to the different forest structure. Anexperiment with a ground-based radar was carried out to validate the model results.Radar backscatter of a deciduous tree (ash) was monitored simultaneously withmeteorological parameters. The measurements demonstrated the ability of ground-based radar to continuously monitor the wetting and drying of leaves. Furthersimulations indicated that the most promising radar for retrieval of the amount ofrainwater storage in deciduous canopies is a co-polarised C-or X-band SAR, like forexample carried by the ERS and RADARSAT satellites. For suchlike radar, themodelled maximum sensitivity of radar backscatter to rainwater storage was in theorder of 2 dB. The application of rainwater storage retrieval was demonstrated byanalysing 1 year of ERS-SAR data of a 14-year-old poplar forest. The radar-estimatedamount of rainwater storage in the canopy agreed with in-situ measured storage, aftercorrection for the contribution of soil backscatter. This correction was possible whensoil moisture content was known.

The general conclusion is that the amount of rainwater storage in deciduous forestcanopies can be estimated by satellite radar, on condition that soil moisture contentand forest structure are known. Radar measurements of both a solitary ash and 14-year-old poplar forest indicated that during rain a 0.1-mm-thick waterfilm retains onthe surface of leaves. Satellite radar appears to be sensitive to rainwater storage onleaves in the upper parts of the canopy. This region is important for surface-atmosphere exchange processes. Ground-based radar appears to be suitable tocontinuously monitor the wetting and drying of leaves. A hilltop-based radar thatobserves the surrounding landscape may combine the excellent qualities of satelliteand ground-based radar; it can continuously monitor the wetting and drying of theupper parts of a forest canopy in a landscape.

9

Chapter 1

General Introduction

1.1 Global Water Resources

This thesis deals with the determination of the amount of rainwater that is retained inthe canopies of forests during and after a rainstorm. Satellite radar is used as a tooltowards this purpose. The relevance of this research topic arises from the increasingscarcity of fresh water for human society. The water withdrawal currently approxi-mates 1/10th of the total fresh water resources. About 2/3 of the withdrawn water isused for irrigating 15% of the cultivated lands, which supply the food for half theglobal population (Shiklomanov, 1997). The food demand is expected to increase thecoming 50 years with 50%, due to a larger and richer human population(Alexandratos, 1999). The fraction irrigated land will probably double (Tilman et al.,2001). Climate change is expected to interfere with the water resources by intensifyingthe hydrological cycle: both evaporation and precipitation above land will increase.The increase in precipitation will probably occur in the form of extreme events, whichimplicates an increased risk of flooding (Bengtsson, 1997; Jones, 1999; Easterling etal., 2000). After earthquakes, floodings were the type of natural disaster that causedmost economic losses during the past 50 years (Munich Re, 1999). It is estimated thatin 2025 more than half the global population will reside in countries that suffer fromfresh water stress, taking into account the increased antropogenic water demand andclimate change (Arnell, 1999; Vörösmarty et al., 2000). Therefore, the protection,maintenance and, where necessary, restoration of water resources is a high priority forthe human society. The study of waterflow after rainfall through various catchmenttypes is essential (Andersson et al., 2000; UNESCO, 2000).

A quarter of the global land surface is covered with forests (FAO, 2000). Anundisturbed forest reduces the risk of floods. It regulates streamflow by acting as abuffer that accumulates precipitation and releases it slowly. On the other hand, a forestuses water (e.g. Brooks, 1928; Bosch and Hewlett, 1983; Whitehead and Robinson,1993; Swank et al., 2001). Trees intercept precipitation, which may evaporate duringand after a rainstorm before it reaches the ground. This part of the precipitation thatnever reaches the ground is denoted with the term interception. In addition, trees loosewater by transpiration. Interception is high for forest, up to 50% of the total precipita-tion (e.g. Schellekens et al., 1999; Ataroff and Rada, 2000), due to the tight couplingof the forest canopy with the atmosphere. In spite of the fact that research of forestinterception has been conducted since the 1860’s (Ebermayer, 1873), a recent reviewreaches the following conclusion: “It is difficult to draw general conclusions aboutinterception losses by particular forest types because they always depend on the typeof rainfall and other meteorological conditions… Characteristics of a forest that

Remote Sensing of Wet Forests10

affect interception are not always easy to identify and quantify” (Crockford andRichardson, 2000).

In this chapter, the objective of the thesis is formulated. The chapter is organised asfollows: in the first three sections, the interception process, models and measurementmethods for the amount of rainwater storage in forest canopies are reviewed. Thefollowing section highlights the gaps in knowledge, and identifies a potential applica-tion of satellite radar. Some main principles of satellite radar remote sensing arebriefly described. The last section deals with the thesis objective and outline.

1.2 Rainfall Interception by Forest

Horton’s classical physical description of the interception process is printed in textbox1. The mathematical description and terminology used in this thesis are printed intextbox 2. As will be pointed out in this paragraph, the translation of Horton’sdescription into the bulk quantities interception, storage and throughfall is difficult,because partitioning of rainfall is determined by both atmospheric and biologicalfactors, which may interact in a complex mode.

Textbox 1

It is a matter of common observation that the percentage of precipitation reachingthe ground in forest or on fields with growing crops is very small in the earlieststages of rain, increasing as the duration of the storm increases, the total amountreaching the ground being small for short light showers, and increasing for severeprolonged storms. General observations also lead, to the following conclusions: When rain begins, drops striking leaves are mostly retained, spreading over the leafsurfaces in a thin layer or collecting in drops or blotches at points, edges, or onridges or in depressions of the leaf surface. Only a meagre spattered fall reaches theground, until the leaf surfaces have retained a certain volume of water, dependenton the position of the leaf surface, whether horizontal or inclined, on the form of theleaf, and on the surface tension relations between the water and the leaf surface, onthe wind velocity, the intensity of the rainfall, and the size and impact of the fallingdrops. When the maximum surface storage capacity for a given leaf is reached,added water striking the leaf causes one after another of the drops to accumulate onthe leaf edges at the lower points. Each drop grows in size (the air being still) untilthe weight of the drop overbalances the surface tension between the drop and theleaf fi1m, when it falls, perhaps to the ground, perhaps to a lower leaf hitherto moresheltered. These drops may also be shaken off by wind or by impact of rain. The leafsystem temporarily stores the precipitation, transforming the original raindropsusually into larger drops. In the meantime the films and drops on the leaves arefreely exposed to evaporation (Horton, 1919).

General Introduction 11

The dynamics of storage and the factors controlling storage will be described onbasis of the results of a laboratory experiment (Grah and Wilson, 1944). This experi-ment was conducted as follows: a number of small plants, up to ~50 cm height, weresprayed with water. The species were a coniferous Monterey pine (Pinus radiata), anda coyote bush (Baccharis pilularis), an evergreen deciduous species from the familyof compositae. The weight increase was measured every minute, and the falling ofwaterdrops from the plant was monitored by sound. The weight difference betweenthe dry and wet plant was attributed to water storage, because transpiration losses werevery low, and water uptake by leaves was considered to be insignificant; liquid watercan not enter stomata because of the high surface tension (Dybing and Currier, 1961;Schönherr and Bukovac, 1972), and the resistance of the cuticula to water transport ishigh (Kertiens, 1996).

Three consecutive phases are generally distinguished in rainfall interception. In thefirst phase, the plants wetted up gradually (Figure 1.1). A number of reasons havebeen put forward for the gradual wetting of the canopy. Upper leaves may shelterlower leaves (Horton, 1919), sheltered undersides of leaves and bark saturate slowlyby rainsplash (Herwitz, 1985), and drops falling on wet parts do not moisten dry parts(Calder, 1986). Theoretically, the wetting-up rate depends on the droplet size of therain: storage increases slower by rain with large-sized droplets than by rain withsmall-sized droplets, because every part of the canopy has an equal change to get hitby a droplet (Calder, 1986). The droplet size depends on the precipitation type.

Textbox 2

The waterbalance of a vegetated area may be expressed by the equation(Leonard, 1967):

t

SEFDTP

∂∂

++++=

where S = moisture stored on the surface of vegetation, P = precipitation rate, T= rate of precipitation passing directly through the canopy, D = water drainagerate from the leaves, F = stemflow, E = evaporation rate from vegetative surface.The unit of S is mm, that of the other fluxes mm.h-1. Rearranging results into thewater input of the soil:

t

SEPFDT

∂∂

−−=++

In this thesis are the following terms used for the partitioning of cumulativeprecipitation:

storage: S

interception : SEdt +∫ throughfall : ( )∫ ++ dtFDT


Figure 1.1. The rise and fall of storage for two small trees due to spraying with water. The sprayingstopped when the weight increase stabilised. The plants were shaken when drainage stopped. This figurewas published by Grah and Wilson (1944).

Convective precipitation, occurring frequently in the tropics and known as tropicaldownpour, contains large-sized droplets. Clouds contain small-sized droplets. Thecapture of cloud water by vegetation, and subsequent drain of excess water to theground may contribute significant to ecosystem water input on mountain slopes(Marloth, 1903; Phillips, 1926; Ataroff and Rada, 2000). Wind influences the wetting-up rate, because a tree canopy intercepts rainfall more efficient, when rainfall has alarger horizontal velocity (Herwitz and Slye, 1996).

The water-exposed plant weight stabilised after a certain time of spraying (Figure1.1). This is the second phase of the interception process. All excessive water will runoff to the ground. The plants were saturated with water, and therefore the maximumweight increase will be denoted as the saturation storage. The saturation storage is,just as the wetting–up rate of the canopy, influenced by the rainfall intensity: it islarger for fine rain than for downpour (Phillips, 1926). The saturation storage betweentrees, wetted by large-sized droplets, and trees wetted with a fine spray differs theo-retically and experimentally with a factor 2, because large-sized droplets have a higherkinetic energy than small size droplets, and thus shed more retained water from thesurface of leaves (Calder, 1996; Calder et al., 1996).


Evaporation after rain has stopped, is the third phase in the interception process.After spraying stopped, the wet-tree weight decreased rather quickly, till drain ceasedat 2/3th of the saturation storage. That level is defined as the maximum storagecapacity, which is in fact the maximum amount of storage that may evaporate afterdrain and rain stopped under calm conditions. A similar fast decrease in storage wasobserved by Aston (1979), but to a lesser extent by Theklehaimanot and Jarvis (1991).The plants were shaken for 10 s to simulate wind disturbance, and about 50% of theremaining stored water drained to the ground. The same reduction in storage has beensuggested by Horton (1919) on basis of limited field data, and by Hörmann et al.(1996) on basis of the waterbalance of a 97-year-old beech stand. On the other hand,Bouten et al. (1996) observed in a very dense Douglas fir stand that wind only influ-enced the amount of storage in the treetops.

The storage capacity differs between species (Figure 1.1). The range in maximumstorage capacity of Dutch forest stands varied between 0.2 mm in a young poplarstand without leaves till 2.6 mm in a dense Douglas fir stand (Elbers et al., 1996;Bouten et al., 1996, see also Figure 1.3). In general, most storage is retained on thesurface of leaves or needles. Therefore, the total surface area of the leaves and theamount of storage that a unit of leaf area can retain are the most important factorsdetermining the storage capacity (e.g. Rutter, 1963, Aston, 1979). However, in sometropical forests storage provided by bark is estimated to be 30-50% of total storage ina large canopy tree that has been thoroughly wetted, since the surface area of bark islarge compared to that in temperate forests (Herwitz, 1985).

Retained rainwater resides on the surface of leaves as a waterfilm, or as drops ofvarying size, depending on the species (Figure 1.2, Horton, 1919). The form of storage(droplet, waterfilm) is related to the contact angle of the water with the leaf (Fogg,1944, 1947). The contact angle is a function of the surface tension. The apparentcontact angle is also influenced by the surface roughness, because air entrapped at sur-face irregularities enlarges the contact angle (Cassie and Baxter, 1945). Interceptedrain on a smooth and hydrophilic leaf has a small contact angle, and forms a thinwaterfilm. When such a wet leaf receives additional rain, this drains easily. Inter-cepted rain on a rough and hydrophobic leaf forms small-sized droplets with a largecontact angle (>90º). These droplets roll easily over irregularities on the leaf surfacebecause their gravitational centre is located relatively high above the leaf surface.Leaves on which the droplets form intermediate contact angles theoretically retainmost rainwater, as these droplets may become large and yet do not roll off easily. In astatistical analysis of the storage capacity of sub-alpine species, Monson et al. (1992)found that leaf pubescence is the most important morphological factor determining ahigh storage per unit leaf area. Similarly, a rough and flaky bark can store more rain-water than a smooth bark (Voigt and Zwolinsky, 1964; Herwitz, 1985). The wettabil-ity of leaves, and consequently the storage capacity of the tree, may change over time,because the cuticular waxes on the surface of leaves weather. For example, the contactangle of droplets of Scots pine (Pinus sylvestris) decreased in 30 months from 110° to90° on unpolluted sites and to 75° on air-polluted sites (mainly SO2, Cape, 1982).Fungi may also increase the wettability of leaves by secreting hydrophobins, a protein


Figure 1.2. Sketches of typical leaves, showing the mode of water storage. Retained water may formfilms on leaves of many other species. This sketch was published by Horton (1919).

that radically changes the nature of the leaf surface: hydrophobic surfaces turn intohydrophilic surfaces, and hydrophilic surfaces turn into hydrophobic surfaces(Wessels, 1997).


The laboratory experiment of Figure 1.1 was executed under conditions of negligi-ble evaporative demand. Evaporation depends on three properties: the amount ofenergy available for evaporation, the vapour pressure deficit, and the exchange rate ofthe air near the leaf surface to the upper-atmosphere (Penman, 1948). The last is deter-mined by factors as windspeed, turbulence, and forest canopy structure (e.g.Lankreijer et al., 1993; Gash et al., 1995; Roberts, 2000). Measured evaporation ratesduring rain generally were well below 0.4 mm.h-1 (e.g. Gash et al., 1999; Lankreijer etal., 1999). Highest averaged evaporation rates, in the order of 1 mm.h-1, were indi-rectly determined for some tropical locations (Bruinzeel and Wiersum, 1987; Dykes,1997; Schellekens et al., 1999). After rain ceases, remaining storage in the canopyevaporates. Estimates of the contribution of this part to total interception variedbetween ~15% (Bruinzeel and Wiersum, 1987; Schellekens et al., 1999) till more than50% (Carlyle-Moses and Price, 1999; Aobal et al., 1999). Llorens et al. (1997)provides an example of the influence of evaporation on interception. He applied aprincipal component analysis on data recorded in a 33-year-old pine stand (Pinussylvestris) in the Spanish mountains. Highest interception (49% of incoming precipi-tation) was observed for medium events with a low rainfall intensity and a very dryatmospheric condition. Interception for comparable rainfall events with a wet atmos-pheric condition was low (15% of incoming precipitation), due to a lower evaporationrates during rainfall. Lowest interception (13% of incoming precipitation) wasobserved for storms with a high rainfall intensity and a dry atmospheric condition. Therainfall-intensity related slower wetting up and lower storage capacity of the canopycould explain this low interception.

1.3 Rainfall Interception Models

Knowledge about the interception process is formalised in models. The major modelapproaches will be reviewed in this paragraph. The emphasis will be on the processestaken into account.

The first semi-empirical model of rainfall interception was constructed by plottingprecipitation against throughfall (Figure 1.3, Horton, 1919). The interception lossduring the wetting-up phase was taken equal to the amount of precipitation. When theprecipitation rate exceeded the storage capacity, the interception loss was calculatedwith a regression equation as a function of the amount of precipitation. The slope ofthis regression equation was related to the averaged evaporation rate during all storms,and the intercept was set equal to the averaged storage capacity.

Most current research models are based on simulating the waterbalance during andafter individual rainstorms. In the classical model of Rutter et al. (1971), the canopywas considered to be a single compartment that was filled by rain, and emptied byevaporation and drainage. The canopy waterbalance was calculated with a timestep of


Figure 1.3. Recorded throughfall and precipitation per rainstorm in a young poplar stand (left), and adense douglas fir stand (right). The poplar forest measurements are described in detail in chapter 5, andwere conducted by the Alterra institute. The douglas forest measurements were conducted by theUniversity of Amsterdam, and described in Bouten et al. (1996). The line is the regression line after thecanopy is saturated with rainwater. Throughfall already starts before the canopy is saturated.

1 hour. Part of the rain fell directly through the canopy to the ground. This directthroughfall fraction was assumed to be proportional to the fraction canopy cover(Aston, 1979). The drainage rate was an exponential function of the amount of storagein the canopy. The evaporation rate was calculated from atmospheric data according toPenman (1948). The so-called Penman equation was formulated for a flat water body.However, a partly wet canopy cannot be considered as a flat water body, see forexample Shuttleworth (1975). The evaporation of stored rainwater was thereforereduced proportional to the storage fraction, or S/Smax, where S is the amount ofstorage and Smax the maximum storage capacity. Here starts some confusion of thedefinition of the maximum storage capacity: is it the maximum amount of storageafter a rainstorm, when drainage stopped under calm conditions (Horton, 1919), or theminimum quantity of water required to wet all surfaces (Rutter et al., 1971)? Rutterstates: “It is comparable to the field capacity of the soil; as useful in practice and asdifficult to define precisely”. In this thesis is Horton’s definition followed.

Research models of the Rutter type became more complex in the course of time.The atmospheric transport within the canopy has been simulated by representing thecanopy with several horizontal layers, and calculating the atmospheric transportbetween these layers (e.g. Butler, 1986; Watanabe and Mizutani, 1996; Klaassen,2001). The drainage rate has been set proportional to the rainfall intensity (e.g.Massman, 1983; Calder, 1986; Liu, 1997). A recent model considered individualleaves in the canopy as tipping buckets: when the total force, acting on the leaf surfaceas a result of both the water stored and the raindrop hitting the leaf, was greater than acertain factor, the leaf tipped, and the water storage on the leaf was reduced to aminimum storage. The leaves were ordered in horizontal layers, that could be hit byrain with a horizontal component (Xiao et al., 2000).


Estimates of the long-term waterbalance of forests are sufficient for most practicalapplications (Putuhena and Cordery, 2000; Murakami et al., 2000; Vertessy et al.,2001). The Rutter model was therefore simplified to run on less detailed data (e.g.Gash, 1979; Mulder, 1985; Gash et al., 1995; Liu, 1997; Zeng et al., 2000). The Gashmodel has been widely tested. This model is an analytical solution of the Rutter modelunder the assumption that the time span between the successive rain events issufficiently long to allow the canopy to dry completely, and the rainfall andevaporation rate of individual storms can be replaced by an average rate during allstorms. After calibration of the model, and on the condition that the number of stormsis large enough to average out uncertainties in single storms, the Gash model predictsthe interception loss mostly within 15% (e.g. Bruijnzeel and Wiersum, 1987; Dolman,1987; Carlyle-Moses and Price, 1999; Jackson, 2000). In a tropical lowland rainforest,Dykes (1999) found after 20 storms a difference of 50% between measured andmodelled interception loss. This difference was reduced till only 2% after 46 storms.The performance of the Gash model and the more detailed Rutter model haveregularly been compared. Aobal et al. (1999) found that both the Rutter and Gashmodels underestimated interception during high rainfall events in the Canary Islands.He supported Lankreijers et al. (1993) assertion that the greater complexity of theRutter model does not make it better than the Gash model. In the observational recordof a Puerto Rican rainforest, Schellekens et al. (1999) had a big storm (~200 mm rain).The Rutter model underestimated interception during that storm with more than 50%,while the Gash model performed surprisingly well, because it averaged out variationsin precipitation rate during the storm.

Another user community, the meteorologists and climatologists, simulates the waterand energy balance on regional, continental or global scales. Their models have largegridcells, hundreds till thousands of square kilometres. Rainfall interception is com-monly simulated in these models by incorporating a simple bucket model, as describedin chapter 2, in the so-called Soil Vegetation Atmosphere Transfer (SVAT) schemes(e.g. Deardorff, 1978; Sellers et al., 1986; Noilhan and Planton, 1989). Shuttleworth(1988) noticed that a major uncertainty in these models was the spatial heterogeneityof rainfall within the gridcell. The canopy was simulated to saturate quicker when therain fell on a fraction of the gridcell, than when the rain was assumed to spread overthe whole gridcell. As a consequence, simulated throughfall increased and interceptiondecreased. For example, modelled Amazonian climate changed from an evaporationdominated till a run-off dominated system by restricting rainfall to an arbitrarily frac-tion, 1/10th, of the gridcell (Pitman et al., 1990). Additionally, the assumption of con-stant rainfall over the whole gridcell resulted in a model sensitivity to the gridsize. Thecalculated interception loss for a 300 x 300 km American agricultural/prairie area,simulated as a single gridcell, or on a 6.25 x 6.25 km grid, was 27% and 7% of totalprecipitation, respectively (Ghan et al., 1997). Noilhan et al. (1997) found similardifferences for a French area. Current SVAT-schemes use an arbitrarily correction forthe subgrid heterogeneity in rainfall (Sellers et al., 1996; Wang and Eltahir, 2000).The gridcell storage capacity is generally set equal to the spatial-averaged storage


capacity of the vegetation present within the gridcell. The storage capacity of thevegetation is assigned on basis of the landuse, observed by satellites.

1.4 Rainwater Storage Measurement Techniques

A reason for the poor model performance in simulating single events is the scarcity ofproper measurements that can be used for model calibration. Storage during rain ismostly indirectly estimated from the waterbalance. The reliability of this approach islimited. Precipitation and throughfall are generally measured at temporal resolutionscoarser than the resolution of the interception process itself, which is in the order ofminutes, because of the technical problems in accurately monitoring small amounts ofthroughfall (Calder and Hall, 1997; Lundberg et al., 1997). Accurate measurements ofthe bulk evaporation rate during rain have been claimed since recently (e.g. Stewart,1977; Hörmann et al., 1996; Mizutani et al., 1997; Klaassen et al., 1998; Gash et al.,1999; Lankreijer et al., 1999). This bulk evaporation rate includes evaporation fromthe soil and falling droplets. During daytime, the transpiration of the tree is alsoincluded, as not all stomata will be covered with liquid water. The interception processcan therefore only be modelled accurately when direct measurements of the storageare available (Dekker, 2000).

Storage during rain has been measured directly by monitoring the weight increaseof a tree when wetting it, see for example Figure 1.1. Cut trees have been wetted artifi-cially by spraying (Grah and Wilson, 1944; Theklehaimanot and Jarvis, 1991) or usinga rainfall simulator (Aston, 1979; Calder et al., 1996). A major disadvantage of thecut-tree method is the destructive character; storage can only be monitored during afew experiments before the tree leaves are wilted. To overcome this problem, alysimeter, which is a container filled with soil and one or more trees planted in it, hasbeen used (Dunin et al., 1988). Another non-destructive approach to measure storagewas based on the attenuation of electro-magnetic radiation through a forest canopy(Calder and Wright, 1986; Bouten et al., 1991). The transmitter and receiver wereoutlined at opposite sides of a forest stand, and moved along towers to acquire verticalprofiles. γ- and X-band (~10 GHz) radiation have been applied. The attenuation of thetransmitted signal depended, among others, on the total amount of water between atransmitter and receiver. Both techniques, lysimeter and attenuation of electro-magnetic radiation, have only been applied sporadically due to the complexity of themethod and the necessary infrastructure.

The feasibility of two other non-destructive methods for measuring storage duringrain has been investigated for non-forest types of vegetation. Wigneron et al. (1996)analysed the brightness temperature of a wheat field and the underlying soil with adownward-looking microwave receiver at 1.4 and 5 GHz. The brightness temperatureof the soil and the wheat could be separated by using a radiative transfer model. Thebrightness temperature of the wheat increased due to storage after spraying it withwater. Despite this encouraging result, Wigneron et al. (1996) argues that thistechnique is not practical for satellite application, because the resolution of passive


microwave satellite sensors is coarse, and the received signal is highly sensitive to rainin the atmosphere. Calvet et al. (1994) deduced for example a moisture index for anAmazonian forest. The spatial resolution of the gridcells was 148 x 95 km, and thecorrection for rain was performed by using surface measurements of rainfall rate andradiosoundings of the atmospheric water content. In another domain of the electro-magnetic spectrum, at 1165 nm, the reflectance was found to correlate with thewetness of grasses (Madeira et al., 2001). A disadvantage of this method is thesensitivity to cloud cover, vegetation structure, and observation angle.

Storage after rain has ceased is traditionally determined from plots of throughfallagainst precipitation (Horton, 1919). This indirect, graphical method to determine the(maximum) interception storage is, with a few adaptations concerning the (subjective)data selection, still common practice (Wilm and Niederhof, 1941; Leyton et al., 1967).These methods underestimate the maximum storage capacity, because most rainstormsdo not saturate the canopy fully due to the slow wetting of the canopy (Klaassen et al.,1998). Another approach to determine the storage capacity of a tree is based on theweight difference between wet and dry parts (leaves, branches and stems) of the tree,and to scale the storage on individual parts up to the whole canopy storage, on basis ofthe leaf and branch area of the canopy (e.g. Rutter, 1963; Herwitz, 1985; Hutchins,1988; Monson, 1992; Liu, 1998; Llorens, 2000). The wet weight increase may bedetermined before and after shaking the samples to simulate wind. This scaling-upmethod theoretically results into an overestimation of the canopy storage capacity,because generally not all surfaces get saturated with water during a rainstorm: shel-tered parts wet slowly, and intercepted water drains following preferred pathways, likethe undersides of branches.

1.5 Identification of Potential Radar Application

First a summary of the previous paragraphs is given: environmental topics as freshwater resources and climate change ask for models that predict the large-scale wateruse of forests. This water use has two components: interception and transpiration.Interception significantly reduces the water availability in forests, and has therefore tobe explicitly predicted. Interception depends on atmospheric and site-specific factors.The influence of these factors is only partly quantitatively known by lack of propermeasurement techniques. Few experiments indicationed that the storage capacity offorest may vary by a factor 2, due to the influence of windspeed and droplet size. Mostplot-scale models predict interception for single events poorly, even after extensivecalibration, because these models only account for precipitation amount, averagedevaporation rate and maximum storage capacity. The uncertainties average out for alarge number of events. Large-scale models are modified plot models, that take thespatial heterogeneity in rainfall into account by applying an arbitrarily correctionfactor. Simulated large-scale interception is more uncertain, even when the spatialdistribution of rainfall is taken into account, because of the variability of rainfallintensity, windspeed, evaporation rate and vegetation storage capacity.


The uncertainty in predicted large-scale interception, and the influence of thevarious factors upon interception, can be assessed when data of large-scale intercep-tion are available. Large-scale hydrological monitoring commonly uses the catchmentwater budget (e.g. Brooks, 1928; Murakami et al., 2000; Putuhena and Cordery, 2000;Swank et al., 2001). The temporal resolution of this method is poor, because it takestime for the precipitated rainwater to flow from the tree canopy to the stream. There-fore, this method is not useful for investigating the processes driving rainfallinterception by tree canopies, as intercepted rain evaporates fast. Upscaling of plotmeasurements is not reliable due to the mentioned spatial variability of the factorsdetermining interception. An example of the reliability of upscaling plot measure-ments to a catchment is the assessment of the yearly water yield of a small catchment(100 ha) by four methods (sap-flow, soil water budget, eddy covariance and catchmentwater balance), operating at observational scales from a single tree to the completecatchment (Wilson et al., 2001). The estimated water yield varied by a factor 2, whichwas partly attributed to rainfall interception. This uncertainty is expected to increasefor a larger catchment, or for shorter temporal scales. A key factor to unravel theinfluence of various factors upon interception is the amount of rainwater storage in thecanopy of trees. This variable could hardly be measured at a research plot, let alone ata large catchment. As will be described in the next paragraph, radar remote sensingmay be used to overcome this problem.

1.6 Radar Sensing of Forest

A photograph taken by the crew of one of the first spaceflights demonstrated thepotential of synoptic views from space for studying precipitation events (Hope, 1966).Passive and active (radar) microwave sensors, with typical frequencies between 1-10GHz, are very sensitive to the water content of the observed terrain (Ulaby et al.,1981). The spatial resolution of spaceborne microwave systems is in the order of 103

km, which is too coarse for studying a highly spatial heterogeneous process as rainfallinterception. This spatial resolution is determined by the physical size of the antennae.The spatial resolution of radar can be improved till 10 m by using so-called SyntheticAperture Radar (SAR). Satellites equipped with SAR monitor the earth continuouslysince the early 1990’s.

Radar backscatter of forests originates from the canopy and the underlying soil.The fraction of soil backscatter depends on the properties of the radar, and those of theforest and the underlying soil (Ulaby et al., 1981). Well known forestry applications ofSAR are the classification of forest type, estimation of woody biomass and monitoringof the extent and timing of inudations (Kasischke et al., 1997). Soil moisture contentcould be retrieved from SAR-backscatter of open boreal forest (Pulliainen et al.,1996), while SAR-backscatter of dense forest is so stable that rainforest has been usedfor calibrating radar (e.g. Bernard and Vidal-Madjar, 1989; Zink and Bamler, 1995).SAR-backscatter of forests was nevertheless repeatedly found to change after rainevents over forests, see for example Figure 1.4, and Way et al. (1990), Dobson et al.


Figure 1.4. Colour composite of ERS-1 SAR images. The extent of this image is 80 × 80 km. The hori-zontal line in the south of the image is the river Rio Negro, the light spot the city Manaus, and most otherparts forest. The images were acquired at 15 May, 24 July and 11 December 1992. Heavy rainstormsmoved westward over the area at 11 December. The magenta colour in the forest results from abackscatter decrease of 2.5 dB. This decrease was attributed to attenuation of the radar backscatter by thetropical downpour. The wet forest appears as a greenish colour, caused by a backscatter increase of 0.6dB. The backscatter of the dry areas changed ample, within 0.15 dB. This image was published byLichtenegger (1996).

(1991), Ahern et al. (1993), Pulliainen et al. (1994), Rignot et al. (1994), and Proisyet al. (2000). The mechanism causing the SAR-backscatter change of forest after rainwas unknown at the start of this project.


1.7 Thesis Outline

The objective of this thesis is limited to the determination of the potential of SAR tomeasure the amount of rainwater storage in the canopy of forests. The underlyinghypotheses are:

1: radar backscatter is quantitatively related to the amount of rainwater storage on thesurface of vegetation;

2: the influence of soil backscatter can be removed from the total recorded backscatter.

These hypotheses were tested using the following research approach: a radar back-scatter model that took the amount of rainwater storage into account was first devel-oped. A feasibility study was performed with this model, and compared with satelliteobservations of a large mixed forest (Chapter 2: “Rain storage in forest detected withERS tandem mission SAR”). Whether radar backscatter from a tree in the field wasquantitatively related to the amount of rainwater storage, was investigated during anexperiment with ground-based radar. The slightly upward-looking radar monitored thebackscatter from the canopy of a tree without influence of the soil (Chapter 3, “Radarmeasurement of rain storage in a deciduous tree”). For satellite applications, soilbackscatter can be removed by two fundamental different approaches. The firstapproach is to make a clever choice of the radar configuration, in such a way that thecontribution of soil backscatter is minimised. This has been studied in Chapter 4,“Monitoring of rainwater storage in forests with satellite radar”. The second approachis to separate the backscatter contribution of vegetation and soil by using additionalinformation. Two different separation methods have been tested in Chapter 5 “Estima-tions of rainwater storage in a deciduous forest canopy with satellite radar”. Chapter 2-5 are written as articles that have or will be submitted to various scientific journals. As aresult of this approach, some overlap exist in the text of the chapters. The thesis endswith a synthesis of the results.

23

Chapter 2

Rain Storage in Forests Detected with ERS Tandem Mission SAR

Joost de Jong, Wim Klaassen, and Albert Ballast1

Abstract. Rainfall interception by vegetation is a major component in the hydrological balance at theland surface. Small-scale variations in rainfall interception occur when both rainfall and land surface arehighly variable. A key parameter of interception is the amount of rain stored on vegetation. As radarbackscatter is strongly influenced by the free-water content of vegetation, SAR remote sensing might beapplied to analyse large-scale rainfall interception. We concentrated in this study on C-band radar andrainfall storage in forests. The backscatter sensitivity to wetness is simulated with a radiative transfermodel, which has been modified to describe the changes in dimension and dielectric constant of leavesand needles due to wetting. The simulations indicate that backscatter may decrease when a sparse conif-erous forest is wetted, while the backscatter of a closed forest is found to increase with 1-4 dB due to rainstorage. Thus, the sensitivity to storage strongly depends on the type of forest. The simulations areempirically tested by analysing two sets of successive SAR image pairs from the ERS tandem mission.Given the short time between these measurements, it is argued that backscatter changes are mainly causedby changes in rain storage. The observed backscatter change is compared with wetness change estimatedby a standard hydrological model using ground-based rain radar observations as input. The observedbackscatter change between a wet and a dry forest varied between 0.7 and 2.5 dB, in the range of thesimulations. It is concluded that C-band SAR is sensitive to forest wetness, although for a quantitativeassessment of water storage on forest additional information on at least forest structure is needed.

2.1 Introduction

Rainfall interception is defined as rain that is intercepted by vegetation and evaporatedwithout reaching the ground (Horton, 1919; Rutter et al., 1971; Gash, 1979). Intercep-tion accounts for approximately 20% of net evaporation from the earth land surface(Choudhury et al., 1998), and for 25-40% from temperate forests (Linacre and Geerts,1997). It is therefore a major component in the hydrological balance at the land sur-face. As water vapour is the most important greenhouse gas, interception influencesclimate (Gates et al., 1996; Sellers et al., 1997a). 1 Department of Physics, University of Groningen

Reprinted from Remote Sensing of Environment, volume 72, J. de Jong, W. Klaassen and A. Ballast, RainStorage in Forest Detected with ERS Tandem Mission SAR, Pages No. 170-180, Copyright (2000), withpermission form Elsevier Science.


Precipitation, evaporation and interception are processes occurring at a spatial scaleof 5 m to 50 km (Blöschl and Sivapalan, 1995). Large-scale climatological and hydro-logical models have grid cells of ~1000 km2. As a consequence, interception within agrid cell is heterogeneous. This subgrid heterogeneity strongly influences the mod-elled hydrological balance and climate (e.g. Shuttleworth, 1988; Pitman et al., 1990;Dolman and Gregory, 1992; Eltahir and Bras, 1993; Noilhan et al., 1997). Current ap-proaches to account for subgrid heterogeneity are poorly validated, as only local inter-ception measurement techniques are available. A large-scale measurement techniquewould therefore be a useful tool in the validation of large-scale interception models.

A key parameter in the interception process is storage, here defined as the amountof rain stored on vegetation. Most storage occurs on the foliage, where rain forms awater layer on hydrophilic leaves, or drops on hydrophobic leaves (Horton, 1919). Thebackscatter measured by a Synthetic Aperture Radar (SAR) is positively related withthe water content of the vegetation cover, and therefore a SAR might be able to quan-tify large-scale rain storage. C-band backscatter of forests increases up to 3 dB shortlyafter rainfall (e.g. Bernard and Vidal-Madjar, 1989; Way et al., 1990; Dobson et al.,1991; Ahern et al., 1993; Pulliainen et al., 1994; Rignot et al., 1994). On the otherhand, a slight backscatter decrease is also observed for a wet coniferous forest(Schowengerdt, 1983), and densely vegetated agricultural areas (Bergen et al., 1997).The mechanism causing the observed backscatter change after rainfall is still poorunderstood (Hobbs et al., 1998; Saich and Borgeaud, 2000). Uncertainty arises, as therecorded total backscatter is not only influenced by the water stored on top of thevegetation, but also by moisture and structure of both soil and vegetation.

A promising approach to single out the factor rain storage on vegetation is back-scatter change detection between successive days with a wet and a dry vegetationcover. The vegetation and soil structure can be assumed to be unchanged. Moreover,this method may distinguish between the quick drying of the vegetation and the slowdrying of the underlying soil. Preliminary model calculations indicate that the back-scatter change of a forest with a wet and a dry canopy varies between 0.5 and 1.8 dBfor C-band (Klaassen et al., 1997).

The feasibility of measuring large-scale rain storage by SAR with successive daysof change detection is investigated by radiative transfer model simulations and obser-vations. Successive SAR observations with the same incidence angle from wet and drydays are available from the ERS-1 and ERS-2 tandem mission. We therefore concen-trate on the ERS C-band SAR. Furthermore, this study will focus on forest, becauseinterception and backscatter of forests are well studied, and thus data are available forvalidation. A problem arises, as direct observations of rain storage are not availabledue to the almost complete lack of measurement techniques. The validation is there-fore carried out indirectly by estimating forest wetness with a standard hydrologicalmodel. These simulations are used to relate the forest backscatter change with theforest wetness change and thus to estimate the potential of SAR to measure large-scalevegetation wetness and rainfall interception.

Rain Storage in Forests Detected 25

2.2 Models

InterceptionThree phases are generally distinguished in the interception process. The first phasestarts at the beginning of rainfall. A dense canopy intercepts most raindrops, relativelyfew fall through the canopy without contact. The canopy can retain only a certainamount of rainfall, the so-called maximum storage capacity Smax. The second phasestarts when stored rain exceeds Smax and excessive water runs off to the ground. Thethird, drying phase starts when rainfall stops.

Following Deardorff (1978), interception is modelled by considering the canopy asa bucket, which is filled by precipitation and emptied by evaporation. The bucketflows over if Smax is reached. Assuming a closed canopy that intercepts all raindrops,variations in total storage S (in mm water) are calculated by:

∂∂S

tP R E= − − (2.1)

where P the precipitation flux, E the evaporation flux from the wet canopy and R therunoff flux from the canopy, all in mm.h-1. Run off only occurs if S exceeds Smax. Thevalue of Smax is calculated by assuming that each leaf or needle is covered with a 0.2-mm-thick water film on one side (Dickinson, 1984), so:

S LAImax .= 0 2 (2.2)

where LAI is the one-sided leaf area index and Smax is in mm rain. The evaporationflux is (Deardorff, 1978):

ES

SE p=

max

23

(2.3)

where Ep is the potential evaporation, which is calculated with the Penman-Monteithformula with zero canopy resistance (Monteith, 1965). This formula is based on theenergy budget, and is a function of the aerodynamic resistance, humidity, temperatureand available energy. The aerodynamic resistance is calculated from the windspeedand the aerodynamic roughness. The latter is estimated from the aerodynamic rough-ness for heat (Lankreijer et al., 1993).

Radar backscatterRadar backscatter of forest is simulated by the three-layer UTA Radiative TransferCanopy Model (UTARTCan, Karam et al., 1992; Karam et al., 1995). The inputparameters of the model are the density, size, orientation, and dielectric constant of


trunks, branches, leaves or needles, grouped in horizontal layers. The forest is in ourcase schematised as two separate, continuous layers, a canopy layer and a trunk layer.The canopy layer consists of leaves or needles, and branches. The trunk layer islocated under the canopy layer and consists of only trunks. As second-order scatteringonly contributes significantly on cross-polarised backscattering (vertical to horizontalpolarisation and vice versa) (Karam et al., 1992), and the ERS-SAR is only verticalpolarised, the model is restricted to simulate first order scattering effects. In otherwords, the incoming radar wave is reflected and transmitted by the vegetation cover.Backscattering from the soil occurs either directly or via double bounce via trunks orother canopy elements; in these cases the vegetation cover attenuates the radar wavetwo times.

The scattering and attenuation in the canopy depends on the dielectric constant,size, orientation and density of each vegetation cover element (leaf, needle, branch ortrunk). The dielectric constant of a single element is mainly determined by its watercontent. This water is free or bound to organic molecules, with both forms of waterhaving different dielectric constants. A single, effective dielectric constant of eachelement is calculated with the Debye-Cole dual-dispersion model (Ulaby and El-Rayes, 1987), which is based on the assumption that the inhomogenities inside anelement are smaller than common radar wavelengths (X, C, L, P-band; i.e., 3-67 cm).As a consequence, each element can be considered as a homogeneous medium, andthus the effective dielectric constant is obtained by volumetric averaging of thedielectric constants of both forms of water. So, the basic form of the Debye-Cole dual-dispersion model is:

ε ε ε ε= + +r f f b bv v (2.4)

where ε is the dielectric constant and v the volume fraction. The subscripts r, f and bdenotes a relatively unimportant residual component due to the solid matter of thecanopy element, the free water inside the element, and the bound water inside theelement respectively. The volume fractions and the residual dielectric constant arecalculated as a function of the gravimetric water content Mg with the followingempirical equations (Ulaby and El-Rayes, 1987):

( )v M Mf g g= −055 0 076. . (2.5a)

( )v M Mb g g= +4 64 1 7 362 2. . (2.5b)

εr g gM M= − +17 0 74 616 2. . . (2.5c)


The dielectric constants of the free water and bound water are calculated with theDebye and Cole-Cole equations (Ulaby and El-Rayes, 1987) and resulted in respec-tively 73.2-j28.8 and 9.8-j5.5. As mentioned in the introduction, most of the stored rain is retained on the leaf orneedle surface. The influence of this water is estimated by adapting the effectivedielectric constant and the thickness of the leaves or radius of the needles in the radia-tive transfer model. This adaptation will be described in the following paragraph.

Equation (2.2) denotes that a leaf or needle can retain the equivalent of a 0.2-mm-thick water film. This water is stored as droplets, or as a film. The film thickness orthe size and distance between the droplets will be small compared to common radarwavelengths. Following the Debye-Cole dual-dispersion model, stored rain is consid-ered part of the homogeneous leaf. We assume that all stored rain is free water. Con-sequently, the volume fraction of stored rain vs can be added to the free-water term ofEquation (2.4):

( )ε ε ν ε ε= + + +r s f f b bv v (2.6)

As well as changing the dielectric constant, storage also changes the dimensions of theleaf or needle. The thickness of a wet leaf is calculated by assuming that stored rainforms a homogeneous film on the leaf. Additionally, stored water is assumed to bedistributed homogeneous over all leaves. Deciduous leaves are schematised as discs.The thickness of a wet deciduous leaf is now related to storage by:

d d dlf s= + (2.7a)

with

dS

LAIs = (2.7b)

where d is the thickness, lf denotes the leaf, and s the water layer. A needle ismodelled as a cylinder. The radius of a wet needle is obtained by adding the volume ofthe stored water (Vs) to the volume of the needle:

π πr l r l Vw d S2 2= + (2.8a)

with

V r lS

LAIS d= ⋅05 2. π (2.8b)


which results in:

r r rS

LAIw d d= +2 (2.8c)

where rw and rd are the radius of the wet and the dry needle respectively, and l thelength of the needle. The factor 2πrdl in Equation (2.8b) is the surface area of a needleand 0.5 is brought into this equation as the storage is a function of the one-sided LAI,which is in case of needles half the surface area.

2.3 Study Site and Data

Figure 2.1a shows the area of interest (AOI), a parallelogram of 20 km × 85 km nearthe centre of the Netherlands. The large forest in the south is the Veluwe, ± 350 km2

mixed temperate forest on sandy soil. The function of this forest is partly productionforest and partly nature reserve. The most recent forest stands of the Veluwe wereplanted around 1900, and the oldest already existed in the Middle Ages. Most foreststands are dominated by coniferous species (spruce, larch and pine: Picea, Larix andPinus spp.) and have a small fraction (<10%) large (>15 m) deciduous trees like oak,birch and beech (Quercus, Betula, and Fagus spp.) (Ten Houte de Lange, 1977). LAImaps of the Veluwe do not exist. Published plot measurements of the LAI at theVeluwe range from 2.8 (Dolman et al., 1998) up to 11 (Bouten et al., 1996). A fieldsurvey showed that these plots are extremes. The averaged LAI is estimated to be 5.

The SAR image pairs of the AOI were acquired by the ERS-1 (first day) and ERS-2 (second day) on 7 & 8 September 1995 and 25 & 26 May at 10h30 UMT (12h30local summer time). Both SAR’s operate at C band vv-polarisation, and the ERS-1 andERS-2 orbit and incidence angle were identical. A fractional landuse map on 1 × 1 kmgrid was taken from a GIS database (anonymous, 1997). Precipitation rates originatefrom the rain radar of the Royal Dutch Meteorological Organisation. Figure 2.1b is anexample of a (resampled) rain radar image. With a time resolution of 15 minutes and aspatial resolution of 2.5 km, the rain radar recorded a number of precipitation rateclasses: 0.1-0.3, 0.3-1, 1-3, 3-10, and 10-30 mm h-1. The whole AOI is located within100 km distance of the rain radar.

TABLE 2.1

Relevant characteristics of the observed area at the moment of satellite overpass.

Date 7 September 8 September 25 May 26 MayShower top (km) - 5.7 3.9 4.5Surface windspeed (m.s-1) 3.3 4.5 2.8 4.1Surface wind direction (°) 146 108 254 244Soil moisture content at 3 cm (vol. %) 14.9 15.0 20.5 21.1


Figure 2.1. The location of the area of interest near the centre of the Netherlands, showing: (2.1a)landuse; (2.1b) the precipitation rate at 8 September during the moment of ERS overpass; and (2.1c) themodelled storage in the pixels with a subgrid fraction forest > 75% at the same time. Note that all pixelswith rain and 2 km around the shower are excluded from further analysis.

Detailed meteorological and hydrological observations were carried out in twoforest stands. One location was situated in a pine stand in the centre of the Veluwe, theother in a poplar (Populus spp.) forest 20 km west from the centre of the AOI. Themeasurement sites and equipment used are described in Elbers et al. (1996), andDolman et al. (1998). The measured parameters include precipitation, run off, avail-able energy, humidity, and windspeed. Soil moisture at 3 cm depth was also measuredat the site within the Veluwe. Table 2.1 sums some parameter values at the moment ofsatellite overpass, to which we will refer to afterward.

To simulate the backscatter sensitivity to forest wetness with the radiative transfermodel, suitable input data must be chosen. An exact parameterisation of the forest inthe radiative transfer simulations was not tried because the forest structure varieswithin the measurement grid of 1 × 1 km. Instead, backscatter sensitivity to wetnesswas simulated for a few characteristic forest stands. As most of the Veluwe is coveredwith coniferous species, a dense and a sparse coniferous forest are simulated. To showthe difference between a coniferous and a deciduous forest, the backscatter sensitivityfrom an average deciduous stand is also simulated. The forest types are: a low coniferous


TABLE 2.2

Forest structure parameters used in the radiative transfer model simulations. The forest is schematised asa canopy layer above a trunk layer. These parameters are based on Rignot et al. (1994).

black spruce white spruce balsam poplarCanopy layer Canopy-layer height 5.1 14.7 10.1 Leaves/needles LAI 1.5 9.1 3.6 Mean length (cm) 0.8 1.6 6.8 Mean thickness/diameter (cm) 0.1 0.1 0.03 Orientation 0.5sin(ß) 0.5sin(ß) 0.5sin(ß) Maximum water layer (mm) 0.2 0.2 0.2 Primary branches Mean length (m) 0.62 1.13 2.0 Mean diameter (cm) 1.81 2.24 1.50 Density (branches/m3) 1.31 2.37 6.69 Orientation sin9(ß-30°) sin4(ß) sin9(ß+60°) Secondary branches Mean length (m) 0.39 0.57 1.0 Mean diameter (cm) 0.81 1.04 0.75 Density (branches/m3) 1.31 2.37 6.69 Orientation sin9(ß) sin9(ß) sin9(ß+60°)Trunk layer Trunk-layer height 5.1 16.7 20.1 Trunks Mean height (m) 5.1 16.7 20.1 Mean diameter (cm) 6.5 21.3 22.5 Stem density (stems/m2) 0.137 0.0654 0.106

black spruce stand with LAI 1.5, a coniferous white spruce stand with LAI 9.1,and adeciduous poplar stand with LAI 3.6. The structural parameters of these stands aregiven in Table 2.2. The gravimetric water content of the leaves/needles was set at60%, and the water content of the wooden elements at 50%. These are normal valuesfor healthy trees in the Netherlands (Hoekman et al., 1995). The maximum storage isassumed to be the equivalent of a 0.2-mm-thick water layer on each leaf or needle.The structural parameters of the soil were in all simulations set equal. The surfaceroughness had a rms height of 1 cm, and a correlation length of 4 cm. The volumetricmoisture content of the soil was set to 20%, a value close to the maximum soilmoisture measured during the satellite overpasses (Table 2.1).

2.4 Data Processing

Data were processed on a 1 × 1 km grid. Pixels with a subgrid fraction forest largerthan 75% were selected with the GIS database. It was assumed that these pixels werefully covered with forest. Spatial variation in precipitation input was obtained from


rain radar measurements. These rain radar images were resampled to the model gridwith nearest neighbour sampling, and the precipitation rate classes were replaced bythe logarithmic mean of the class concerning. Other input data were assumed to bespatially constant. The potential evaporation rate was estimated by first calculating thelocal potential evaporation rate at the two locations with known meteorological data,and were then averaged. If measurements were available from one location only, thesewere applied. This occurred at 7 September before 13h30 UMT due to device failure.For the remainder of the time, the correlation between the evaporation rates at the twolocations is 0.86 and the mean difference 0.02 mm.h-1. Spatial averaging of evapo-ration rate therefore introduces only minor deviations in the resulting storage, which isin agreement with Ghan et al. (1997). The spatial distribution of Smax does have on thecontrary a large influence on modelled storage (Eltahir and Bras, 1993). Measure-ments of Smax within the test area vary between 0.5 (Lankreijer et al., 1993) and 2.5mm (Bouten et al., 1996). The influence of Smax was therefore analysed by using thevalues 0.5, 1.5 and 2.5 mm.

The backscatter change of the ERS-1 and ERS-2 image pairs were derived by firstcalculating the linear backscatter coefficient σ0 from ERS PRI-products with the ERS-SAR Toolbox, based on the calibration procedure of Laur et al. (1997). The calibratedimages were next georeferenced, averaged to the 1 × 1 km grid and converted to dB.The calibration includes compensation of ADC saturation effects, as the ERS-2 ADCsaturation is significant decreased compared with the ERS-1 ADC saturation.Compensation of ADC saturation increased the backscatter of the ERS-1 on averagewith 0.2 dB over the selected forest pixels. Another source of false backscatterchanges might be rainfall, as it attenuates C-band backscatter by a few percent (Ulabyet al., 1981). The rain radar detected rainfall at a maximum of 5.7 km above thesurface (Table 2.1). Assuming an incidence angle of 23° and a shower height of 5 km,this rain could distort the radar signal up to 2 km behind the shower. Pixels with rainas well as pixels up to 2 km around the shower were therefore excluded from theanalyses. After this data selection, 117 and 119 forest pixels were analysed from theMay and September image pairs. The resulting accuracy is claimed to be within ± 0.4dB (Laur et al., 1997).

2.5 Results

Simulated sensitivity of C-band radar to vegetation wetnessThe simulated sensitivity to intercepted rain for three different forest types is shown inFigures 2.2a-c. The backscatter from the primary and secondary branches are addedand only the backscatter terms that contribute more than -30 dB are shown. Figure2.2a shows the backscatter sensitivity from the sparse black spruce stand to storage.The most important terms contributing to the backscatter are the soil and the branches.Dry needles are relatively unimportant. The backscatter from the trunks and via thesoil-trunks, soil-needles and soil-branches are small. When needles become wet, thebackscatter of needles increases with 7.8 dB at the maximum storage. The backscatter


Figure 2.2. Modelled backscatter of: (a) sparse black spruce stand; (b) dense white spruce stand; and (c)poplar stand as a function of the rain storage on the needles/leaves. Shown are all backscatter contribu-tions that exceed -30 dB. With increasing storage the backscatter from needles/leaves increases while thebackscatter from the soil and branches decreases.

of wet needles is still an unimportant fraction of the total backscatter. However,wetness not only causes a backscatter increase from needles, it also enhancesattenuation by needles. As a result, the backscatter from the soil and branchesdecreases with -3.0 dB and -1.3 dB, respectively, and the total backscatter of the blackspruce forest stand decreases with -2.2 dB.

The same processes, increased backscatter of leaves and decreased backscatterfrom branches and soil due to attenuation, occur in the other forest stands, but, as willbe seen, the significance of these processes depends on the forest stand parameters.Figure 2.2b shows the simulated sensitivity of the dense white spruce stand. The back-scatter from the branches is the most important term in the dry state. The backscatterfrom the soil and the trunks is negligible. The backscatter from needles increases dueto wetness with 6.0 dB, and the backscatter from branches decreases with -3.1 dB. Asa consequence, the needles become the most important scatters in the wet white spruce


Figure 2.3. Areally averaged cumulative precipitation (mm), potential evaporation rate (mm.h-1) and rain-fall storage (mm) modelled assuming Smax is 1.5 mm (a) in September, and (b) in May.

stand, and the total backscatter increases with +1.1 dB. Figure 2.2c shows the back-scatter sensitivity of the poplar stand to wetness. The total backscatter is dominated bythe backscatter of leaves. This increases with 4.7 dB at the maximum storage. Thetotal backscatter increases with 4.4 dB when leaves get wet. This is slightly less thanthe increased backscatter from the leaves due to enhanced attenuation of the otherbackscatter terms. These simulations show that the backscatter sensitivity of a forest towetness depends on the relative contribution of the leaves or needles (i.e., on the foreststructure). The difference in backscatter between a completely dry and wet forest mayvary between -2 dB and 4 dB.

Rain storage at the moment of satellite overpassRain storage is modelled as a spatial and temporal variable. An example of the spatialheterogeneity of the modelled storage assuming an Smax of 1.5 mm is shown in Figure2.1c. Note that Figure 2.1b shows the rainfall at the same time. It rains in the centreand the North of the AOI. Because the rain passed the AOI from South to North, theSouth of the AOI is already partly dry. The storage varies between 0.0 and 0.5 mm.The spatial differences result from the rainfall amount and the drying time since thelast rainfall.

Figure 2.3a shows the spatially averaged storage modelled with an Smax of 1.5 mmin September as a function of time. The figure also contains the main drivingparameters: rainfall and potential evaporation flux. The potential evaporation fluxshows a clear day-night rhythm: low evaporation fluxes in night time and high evapo-ration fluxes in day time. Moderate rainfall saturated the canopy in the first afternoon.Some drizzle fell in the evening. Although evaporation rate was low during the night,still a significant part of the canopy dried during this night. The stored water starts toevaporate rapidly after dawn. It rained intensely the following afternoon, but as thenight was dry a large fraction of the storage evaporated before dawn. That same morning,


TABLE 2.3

Rain storage at the time of satellite overpass from the pixels used in the analysis as a function of themodel parameter Smax. Shown are the parameters describing the spatial variability of rain storage.

Smax 7 September 8 September 25 May 26 Maymean min max mean min max mean min max mean min max

0.5 0.0 0.0 0.0 0.1 0.0 0.3 0.2 0.1 0.3 0.0 0.0 0.01.5 0.3 0.0 0.5 0.3 0.1 0.6 1.2 1.0 1.3 0.2 0.0 0.42.5 0.7 0.1 1.4 0.7 0.3 1.0 2.1 1.2 2.3 0.8 0.1 1.3

drizzle with bright clearings passed the Veluwe, so vegetation was partly wetted again.The figure shows that the ERS-1 and 2 overpasses were both in the drying phase of thecanopy. The situation in May was quite different; see Figure 2.3b. It rained almostcontinuously the first night. Combined with a low evaporation rate this resulted in ahigh storage in the morning. It rained again the following afternoon and evening, andadditionally dew occurred (indicated by a negative potential evaporation). Due to thehigh evaporation rate most of the stored water evaporated quickly in the morning. TheERS-1 and ERS-2 overpasses were again both in the drying phase, but the amount ofstorage was much higher during the first overpass.

Table 2.3 shows the range of storage modelled with the different values of Smax atthe moments of satellite overpass in the AOI. Smax = 0.5 mm is regarded as the lowerlimit, and Smax = 2.5 mm as the upper limit of expected storage capacity. A large Smax

causes more rain accumulation, and the canopy stays wet longer. For all values of Smax,this table shows that the canopy was dry (or at the most half wet) on both days inSeptember, while the canopy was almost saturated during the first overpass and sig-nificant drier during the second overpass in May.

Observed backscatter changeFigure 2.4 shows the relation between backscatter change and storage change calcu-lated with Smax = 1.5 mm. A positive value means more backscatter, or a wetter canopyon the first day, and a negative value the inverse. The scatter on the x-axis (storage) iscaused by variations in precipitation rate and in drying time after the rain events. TheSeptember data show distinct spatial storage change, but the variations are always lessthan 1/3 of the maximum storage capacity. The averaged storage change is insignifi-cant. The averaged backscatter change in September is also insignificant, and 93% ofthe pixels have a backscatter change less than ±0.4 dB, the expected measurementnoise. The September observations therefore agree with the simulations when changesin rain storage are small. In May, the storage was large on the first day and small onthe second. The backscatter change varied between +0.7 dB and +2.5 dB, with anaverage of +1.3 dB. As most of the Veluwe consist of rather dense mixed forests, thisobservation is within the range of simulated sensitivity of dense forests.


Figure 2.4. Backscatter change versus canopy storage change modelled with Smax is 1.5 mm. A positivevalue means a higher storage or a stronger backscatter on the first day. The canopy was almost dry duringboth September overpasses. The first day was wet and the second day was dry during the May overpasses

2.6 Discussion

The feasibility of measuring large-scale forest wetness by using C-band SAR back-scatter changes between successive days was empirically tested. The observed sensi-tivity to rainfall storage agrees with modelled sensitivity. The significance of thisresult is analysed, based on a discussion of the underlying models.

The maximum modelled backscatter change due to wetness (+4 dB) is slightlyhigher than the maximum observed backscatter increase, mentioned in the introduc-tion (+3 dB) and observed in this study (+2.5 dB). It will be qualitatively discussedwhether this might be caused by a systematic deviation of the model. The firstassumptions concern the rain storage at the scale of the leaf. Equation (2.2) assumesthat all rain is stored on leaves and needles. However, rain may also be stored onbranches and trunks. In case of coniferous species, one could even imagine that rela-tively much rain is stored as small drops at the transition of the twig to the needle. Thestorage on leaves/needles and the resulting backscatter from these is therefore over-estimated. As the sensitivity to wetness of the backscatter from trunks and brancheswill be small due to the small relative volume change and consequently smalldielectric constant change, this overestimation of the backscatter from the leaves willonly be partly compensated by increased backscatter from trunks and branches. Thestored rain is next considered as free water, while a small part of the stored rain might


be bound to the surface of the leaf or needle. The water adheres to this surface due tothe physical processes of surface tension differences between the water and the leafsurface and friction (Horton, 1919). As these bonds are weak compared to chemicalbonds, this assumption causes only a slight overestimation of the modelled sensitivity.At a slightly larger scale, modelling the needles and the small branches in case ofconiferous forest as independent entities is a rough approach. Needles may be regu-larly ordered at distances smaller than the radar wavelength (5.7 cm), and back-scattering interaction between the needles might therefore occur. The next set ofassumptions discussed apply to the scale of the canopy. The storage is assumed to behomogeneously distributed in the canopy. Model calculations indicated that the uppercanopy dries faster than the lower canopy (Watanabe and Mizutani, 1996), thussuppose that the upper canopy is dry while the lower canopy is still wet. The back-scatter change of the upper canopy will be zero while the lower canopy backscatterwill partly be attenuated by the upper parts. This assumption would therefore result inan overestimation of the sensitivity. The canopy is also assumed to be continuous. Inreality, there will be gaps in the canopy, even in dense forests. As a result, part of thebackscatter will arise from the soil and the backscatter sensitivity to leaf wetness willfurther decrease. Thus, most model assumptions probably result in an overestimationof the sensitivity of radar backscatter to water storage. The simulated backscattersensitivity is therefore regarded as the maximum sensitivity.

The SAR observations are indirectly validated on ground truth via modelledstorage. The modelled storage depends strongly on the precipitation rate, the maxi-mum storage capacity, and the evaporation rate between the end of the rain and themoment of observation. The accuracy of precipitation rate derived from rain radar islow (Stewart et al., 1998), just like the accuracy in the value of Smax and the evapo-ration rate from a wet canopy (Klaassen et al., 1998). If, as before the first overpass inMay, it rained intensely for several hours, then the canopy is saturated with rain, andthe uncertainty in precipitation rate is not important. In that case, the uncertainty inactual storage is mainly determined by the uncertainty in Smax. On the other hand, ifsmall showers occur and the canopy does not saturate, like before both Septemberoverpasses, then the uncertainty in Smax hardly influences the actual storage. In thatcase, the uncertainty in storage is determined by the precipitation rate and evaporationrate uncertainty. Given the uncertainty in ground truth, it is concluded that theSeptember data are not useful to test the SAR observations and the May data are onlyuseful for a qualitative comparison of the SAR observations.

Although the observed backscatter changes in May can be explained by wetnesschanges, the backscatter may have been changed by other causes. The significance ofsome often mentioned causes for temporal changes in backscatter (Hobbs et al., 1998;Saich and Borgeaud, 2000) are therefore evaluated for the present situation. The back-scatter depends on the forest structure, which might change due to altering wind. Onthe scale of Beaufort, the wind changed between the May image pair from windforce 2to windforce 3 (Table 2.1). The wind direction stayed almost constant. According tothe description of the Royal Dutch Meteorological Organisation, small branches startto move at Beaufort windforce 4. It is therefore assumed that the wind did not cause


any relevant backscatter changes. Furthermore, backscatter depends on soil moistureand soil moisture might change after rainfall. Soil moisture was measured at 3 cmdepth at the measurement site within the Veluwe. The measurements in May tookplace in a rainy period. The soil was wetter the second day (+0.6%, see Table 2.1) thanthe first day, due to rainfall between the moments of ERS overpass. This higher soilmoisture would result in a higher backscatter on the second day. The vegetation waswet the first day and dry the second day. This would cause an increased backscatter onthe first day. We observed the latter, and this is therefore attributed to rain storage inthe canopy.

2.7 Conclusion and Recommendations

We examined the feasibility of detecting forest canopy rain storage with changedetection between successive C-band SAR images. According to simulations, thesensitivity of radar backscatter to forest wetness is strongly dependent on the type offorest. The modelled sensitivity ranged from -2 to + 4 dB. Due to model assumptions,this sensitivity is the maximum sensitivity. The relative contribution of the leaves orneedles to the total backscatter is the parameter that governs whether backscatter willincrease or decrease with increasing storage of rain. Backscatter change observationswere made by the ERS tandem mission and compared with modelled wetness changeas ground truth. The accuracy of ground truth is only qualitatively fair due to uncer-tainties in input parameters. The observed backscatter change between a wet and a dryforest varied from +0.7 dB to 2.5 dB, which agrees with the simulations. It is thereforeconcluded that C-band short-time change detection is a feasible approach to discrimi-nate between wet and dry forest.

A disadvantage of satellite observations appeared in the September dataset, whenboth observations dealt with a partially wet forest and measurements could not be usedto detect changes in rain storage. The September example shows that al least one dryobservation is desired to monitor rain storage. This condition reduces the number ofsuitable observations. Another disappointing result is that the backscatter sensitivity tocanopy wetness strongly depends on forest structure. ERS-SAR observations alone aretherefore not sufficient to quantify rain storage in forests. Additional information onforest structure is needed for that purpose.

The forest structure might be retrieved by multiple wavelength radar remotesensing. An additional advantage of multiple wavelength observations is that thesensitivity to storage is wavelength dependent, and therefore a combination of wave-lengths could perform more accurate. Finally, using modelled wetness as ground truthintroduces uncertainties. It is therefore recommended to use direct local measurementof rain storage as ground truth.

39

Chapter 3

Radar Measurement of Rain Storage in a Deciduous Tree

Joost de Jong, Henk de Groot, Wim Klaassen, and Piet Kuiper

Abstract. The potential of radar to estimate the amount of rain, stored in a deciduous tree canopy, isexperimentally investigated. A ground-based X-band radar was pointed at the canopy of a mature ash(Fraxinus excelsior). Radar backscatter increased during a shower until 2 mm rain had fallen. Withfurther rain, radar backscatter maintained a constant level. Radar backscatter decreased exponentiallyafter rain stopped. Given the correlation between radar backscatter variation and the variation in waterstorage due to rainfall and evaporation, it is argued that X-band radar can be used to monitor the amountof rain, stored in the canopy of a forest. This new direct measurement technique therefore promises greatimprovements to forest hydrological research.

3.1 Introduction

Interception is defined as precipitation that is temporarily stored on vegetation andevaporates without reaching the ground. Interception accounts for approximately 25 to40% of the net evaporation of temperate forests (Linacre and Geerts, 1997, p. 98). Akey parameter in interception is the amount of precipitation temporarily stored onvegetation, which is called storage (Horton, 1919). The interception process is stillsubject of investigations (e.g. Calder and Wright, 1986; Lankreijer et al., 1993;Klaassen et al., 1998; Gash et al., 1999). Uncertainties in the interception process arethe evaporation rate, the maximum storage, and to which degree the precipitation rate(Aston, 1979; Calder, 1986), and windspeed (Hutchins et al., 1986; Hörmann et al.,1996) influence maximum storage.

Few non-destructive measurement techniques have been developed to monitorstorage at canopy scale (Olszyczka and Crowther, 1981; Calder and Wright, 1986;Bouten et al., 1991). These techniques are based on attenuation of gamma or micro-wave radiation passing through a tree canopy or forest stand. A disadvantage of thesemethods is the complexity of the experimental set-up: to obtain a profile of storage thetransmitter and receiver have to be moved vertically along two towers at oppositesides of the forest stand.


A large part of storage in the canopy of a deciduous tree is located at the surface ofleaves (Horton, 1919; Helvey and Patric, 1965; Aston, 1979). Theoretically, radar issensitive to storage at the surface of leaves (De Jong et al., 2000a). Radar has theadvantage of an integrated transmitter and receiver. If radar could quantify storage, aflexible, simple and relatively cheap measurement technique would become availableto monitor storage directly.

The aim of this study is to investigate the potential of ground-based radar to moni-tor storage in a deciduous tree. The study was executed using an existing portableradar, of which wavelength, range and Doppler-velocity resolution were optimised forthe experiment. The costs were kept low by keeping the radar, radar use, and radarsignal processing as simple as possible. Weather parameters were monitored simulta-neously to estimate the amount of rain storage in the canopy. The data from the firstweek of October 1999 were analysed to relate the radar signal to weather parameters.

3.2 Radar

GeneralRadar transmits an electromagnetic wave that might be reflected by a target, andsubsequently received by the radar. The amount of reflected power or backscatterdepends on the target parameters and the wavelength of the radar. An important targetparameter is the dielectric constant. In case of vegetation, the dielectric constant ismainly determined by the water content of the plant tissues (e.g. Tan, 1981; El-Rayesand Ulaby, 1987). Intercepted rain is stored at the surface of leaves and branches as afilm of water or as droplets. As the thickness of the water film or droplets is smallcompared to common radar wavelengths (~3-60 cm), the dielectric constant of a leafor branch with stored water at its surface theoretically equals the volumetric averageof the dielectric constants of the leaf or branch and the stored water (De Jong et al.,2000a). Therefore, radar backscatter is proportional to storage.

Signal processingRadar essentially receives a reflected radiowave with a certain power, N. Reflection orbackscatter takes place at a certain distance or range, r. The range is retrieved from thedelay between transmission and reception of the signal. A measure of radar back-scatter is the radar cross section, σ0. The radar cross section in decibel (dB) is relatedto N and r by the radar equation:

= N

C

rlog

40 10σ (3.1)

where C denotes the radar constant, which accounts for the system parameters of theradar. A precise measurement of the radar constant C is complicated. Under the

Radar Measurements of a Wet Tree 41

assumption of a constant range, C is easily eliminated by redefining σ0 as the back-scatter difference relative to a reference measurement, Nref :

=∆

refN

Nlog100σ (3.2)

By using the averaged received power as Nref, the resulting backscatter change, ∆σ0,becomes the relative deviation from the mean backscatter. Moreover, N may now berecorded in any arbitrary unit.

Some additional signal processing is executed to correct for crosstalk, a systemeffect. Despite the isolation between the transmitter and receiver, part of the trans-mitted power leaks to the receiver. This leaked power is denoted as the backgroundspectrum. The background spectrum is quantified by pointing the radar beam into theair. The backscatter change from the target is calculated by subtracting the backgroundspectrum, Nb, from the received power. This results in:

−−

=∆bref

b

NN

NNlog100σ (3.3)

Equation (3.3) is based on the assumption that the radar constant C does not changewith time. This has to be verified by regular calibration. A sufficient calibrationmethod of radar is feeding a part of the transmitted power through the receiver and theelectronics (Ulaby et al., 1981, H10). The background spectrum satisfies this criterion,and the radar is therefore calibrated on the background spectrum at ranges where notarget is present.

The signal is also sensitive to fading. Fading is caused by the interference of waveswith a different phase, arising from reflection of the radar wave from different parts ofthe canopy. The consequence of fading is a stochastic variation in the radar signal. Toestimate ∆σ0 accurately, a sufficient number of independent samples have to beaveraged. Two approaches to averaging are (i) spatial and (ii) temporal (Ulaby et al.,1981, H7). Temporal averaging is only effective when the target changes in time.When the tree moves in the wind, each sample is expected to be independent, andtemporal averaging is feasible (McDonald et al., 1991). Advantages of temporalaveraging are: (i) it suppresses noise in the electronics, as noise also has a stochasticnature, and (ii) it keeps the system simple and cheap compared to the spatial averagingapproach, as a directional scanning utility is not needed. We therefore opt for temporalaveraging to suppress fading. The reliability of the measurements will be secured byusing the Doppler shift of the received signal, as this additional characteristic of thereceived signal is a direct measure of the motion of the tree in the direction of theradar beam. The standard deviation of the Doppler velocity, σv, is calculated as ameasure of the Doppler shift.


Figure 3.1. The 20-m-tall ash photographed from the location of the radar. The centreof the radar beam was a few meters below the top of the tree. At that range, thediameter of the radar beam was 3 m. The weather station is visible in front of the ash.


3.3 Site and Instrumentation

SiteThe experiment was conducted on an experimental field of the University ofGroningen, the Netherlands (6° 40’ E, 53° 10’ N). The field is a tree bordered squarewith sides of 60 m. The radar was pointed at an ash tree (Fraxinus excelsior) in thecentre of a tree line, clearly separated from the other trees (Figure 3.1). The height ofthe ash was 20 m, the crown projection 250 m2.

RadarThe Micro Tree-Radar (MTR, Metek GmbH) is an FM-CW Doppler radar, adaptedfrom the Micro Rain-Radar MRR-1 (Peters, 1995; Klugman et al., 1996). The radarhas an integrated transmitter and receiver based on the homodyne principle. This tech-nique results in a strong coupling between the transmitter and the receiver. The radarprocessed the reflected power from raindrops with fast-Fourier transformations on-lineinto arbitrary digital units for 16 range cells and 32 Doppler-velocity cells. The treeradar on-line processing was optimised compared to the rain radar processing by: (i)decreasing the resolution of the Doppler-velocity classes because branches and leavesmove slower than raindrops, and (ii) decreasing the range-cell resolution because theradar is used at a relative short distance from the tree. Additionally, the operation fre-quency was set at X-band (10.4 GHz). X-band backscatter theoretically arises mainlyfrom leaves (McDonald et al., 1991; Karam et al., 1992). The radar was put on arobust aluminium stand, which was equipped with a small roof to avoid the radar andthe parabolic reflector from getting wet. The transmitter and receiver were kept at aconstant temperature of 60 °C to prevent temperature induced sensitivity shifts in theelectronics. The reflector was heated to prevent dew formation on its surface. Theradar characteristics are presented in Table 3.1.

TABLE 3.1

Characteristics of the Micro-Tree Radar.

Type frequency modulated, continuous wave (FM-CW)Transmit power 10 mWBeamwidth 3 degreesPolarisation verticalFrequency 10.4 GHz (X-band)Averaging time 5 minNo. range cells 16Range cell resolution 5 mNo. Doppler cells 32Doppler cells resolution 0.029 m.s-1

Weight (incl. Frame) 6 kg (25 kg)


Figure 3.2. Mean background spectrum in every range cell and Doppler-velocity cell in digital numberson a logarithmic scale. The background spectrum is the transmitted power that leaks to the receiver. Thefirst range cell did not contain data. The background spectrum was highest in the second range cell withzero Doppler velocity.

The radar was installed at 60 m distance from the centre of the tree. The radar waspointed at the upper canopy because even small showers wet the upper canopy.Moreover, the largest tree motions occur in the wind-exposed upper canopy, reducingthe influence of fading. The shape of the radar beam passing through the canopyresembled a cylinder with a length of 16 m and a diameter of 3 m. The on-line proc-essed radar data were logged on a HP palmtop computer.

Weather stationWeather parameters were monitored by a weather station at 2.5 m height located in thecentre of the field (Figure 3.1). Essential instruments were: a cup anemometer forwindspeed measurement (Vector Instruments AR100), wet and dry bulb thermometersfor water vapour pressure measurements (Vector Instruments H301 Psychrometer), awetness sensor (Campbell 237) and tipping bucket rain gauge for rain measurements(Campbell ARG100). When the windspeed was ≤ 0.2 m.s-1, the cup anemometerrecorded a windspeed of 0.2 m.s-1 to compensate for errors due to stalling. Theresolution of the rain gauge was 0.2 mm. The 5-min-averaged weather parameterswere logged on a Campbell XR10 datalogger.


Figure 3.3. The variation of the background spectrum relative to the mean background spectrum in rangecell 9, recorded during 9 days and plotted versus time of the day. The background spectrum had asystematic and a noise component.

3.4 Results

Signal processingThe background spectrum was obtained by pointing the radar beam into the air from29 November till 6 December 1999. The temporal averaged background spectrum washighest at shorter ranges and low velocities due to the homodyne principle of the radar(Figure 3.2). A number of unexpected isolated peaks were visible in higher Doppler-velocity cells. The total background spectrum varied less than 0.4 dB with time. Themaximum deviation from the mean was found in range cell 9, on top of the peakvisible in Figure 3.2. The background spectrum in range cell 9 was correlated with thevariation in the other isolated peaks, visible in Figure 3.2. The variation in backgroundspectrum in range cell 9 had a systematic component with a period of 24 hours, and arandom component of smaller amplitude (Figure 3.3). The latter component is attrib-uted to noise. Weather conditions were variable during this observation period (frost,rain and storm). It was checked that the variations in background spectrum were notcorrelated with variations in weather variables. Even falling raindrops did not influ-ence radar backscatter.


Figure 3.4. The received power of the tree and the background spectrum. According to field measure-ments, the tree was present in range cell 11, 12, 13 and 14.

The reflection from the tree was measured from 1 till 7 October 1999. This resultedin a strong backscatter increase in range cells 11, 12, 13 and 14 (Figure 3.4). Theserange cells coincide with the location of the tree. The signal in the adjoining rangecells 9, 10, 15 and 16 was slightly increased. This increase was attributed to leakagedue to the method of on-line processing of the received signal. The difference in thereceived power between the October and the background spectrum was insignificantfor the other 8 range cells, which indicates that these range cells are useful for radarcalibration, and that the radar constant did not change in time.

The influence of background spectrum variations on the accuracy of the tree meas-urements was assessed. The derived ∆σ0 from the tree is strongest influenced by varia-tions in background spectrum (Nb), when the recorded signal from the tree (N) is rela-tively low compared to the background spectrum, see Equation (3.3). The accuracy istherefore determined by selecting the lowest received power from the tree, and calcu-lating ∆σ0 from the tree with the highest and the lowest recorded backgroundspectrum. The resulting difference in ∆σ0 for range cell 11, 12, 13 and 14 was 0.50,0.23, 0.13, and 4.66 dB, respectively. The poor accuracy of range cell 14 results fromthe low tree reflection in that particular range cell. Range cell 14 was thereforeexcluded from further processing. As the reflected powers in range cell 11, 12 and 13were strongly correlated with each other (correlation coefficient > 0.9), the receivedpowers of range cell 11, 12 and 13 were summed to determine the total treebackscatter change, ∆σ0. The influence of background spectrum variations on ∆σ0 was< 0.25 dB, even for the lowest recorded tree backscatter.


Figure 3.5. Hour-averaged backscatter change, Doppler velocity standard deviation, temperature,windspeed, and cumulative rainfall during the first week of October 1999. The marks on the x-axis are setat midnight. Some weather data were missing on 4 October. Weather was stormy in the first half of theweek, and calm at the end of the week. It rained regularly.


Tree radar reflection in relation to weatherThe tree backscatter change, ∆σ0, Doppler-velocity standard deviation, σv,temperature, windspeed and precipitation are shown in Figure 3.5. Weather data from4 October between 11 a.m. and midnight are missing. Due to the location of theweather station in a clearing, the measured windspeed was lower than the windspeednear the upper canopy of the tree. The reliability of our windspeed measurements wastherefore assessed by comparing it with windspeed measured at Groningen Airport byRoyal Dutch Meteorological Organisation (KNMI). The windspeed was measured at10 m height above grass at Groningen Airport, which is located at 8 km from ourexperimental site. It was found that the hour-averaged windspeed in the clearing wasstrongly correlated, a correlation coefficient of 0.9, with windspeed at Groningen air-port. The windspeed at the airport was 5.6 times higher than the windspeed in theclearing. We prefer to use the windspeed in the clearing as a measure of the windspeednear the top of the ash, as gusts of wind occur locally. The windspeed at the top of theash is assumed to be a constant multiple of the windspeed in the clearing. The correla-tion coefficients between radar and weather parameters are presented in Table 3.2.

FadingFading is only averaged out when successive samples are independent due to treemotion. Therefore, we focus on the relation between σv and windspeed, the drivingforce of tree motion. Figure 3.6 is based on hour-averaged data, as 5-min-averageddata would result in larger scatter in the relation between windspeed and σv. For wind-speed above 0.5 m.s-1, windspeed and σv were related. At low windspeed, σv was notsensitive to measured windspeed. This is attributed to coupling between adjacentDoppler cells, which results in an offset in σv, and motions smaller than the resolutionof the Doppler velocity cells. One might think that measurements influenced by fadingshould therefore be best excluded on the basis of windspeed and not on σv. On theother hand, low windspeed occurred at night (Figure 3.5), when the atmosphere wasstable. In a stable atmosphere, the windspeed at low heights in the clearing will becoupled badly with windspeed near the top of the canopy. For example, very lowwindspeeds (≤ 0.2 m.s-1) occurred in the clearing at night, while the windspeed meas-ured at the airport never dropped below 1 m.s-1. Consequently, we could not objec-tively identify measurements influenced by fading. We therefore preferred an arbitrarythreshold, based on the most direct measure of tree motion: the Doppler velocity.

TABLE 3.2

Pearson correlation coefficients between the radar signal and weather parameters.

temperature wind direction windspeed rain intensity humidity

Backscatter change 0.13 -0.15 0.20 0.71 0.29

Doppler-velocity standard deviation 0.70 -0.29 0.91 -0.05 -0.38


Figure 3.6. Hour-averaged Doppler standard deviation measured in the upper canopy of the tree versusthe windspeed measured at 2.5 m height. It is noted that 0.2 m.s-1 is the lower detection limit of the cupanemometer used for the windspeed measurement.

With the threshold σv < 0.05 m.s-1, 11 hours with data were excluded. Excluded weree.g. the lowest measured ∆σ0, which occurred simultaneously with the lowestmeasured σv and windspeed. From the resulting data, the ∆σ0 of just one observationwas below -1.1 dB. As this observation borders an excluded observation, fading mightoccur during a significant fraction of the time span of this observation.

Sensitivity of radar backscatter to storageRain intensity and ∆σ0 of the tree were found to be correlated (Table 3.2). Radarbackscatter theoretically depends on storage. As storage might be related with rainintensity, the sensitivity of ∆σ0 to storage is investigated with data of 5-6 October,when two small showers with respectively 0.8 and 0.6 mm of rain occurred (Figure3.7). The 5-min-averaged ∆σ0 was used to reveal the temporal behaviour most clearly.It should be noted that the small integration time results in a larger measurementuncertainty. The radar reflection increased immediately after the beginning of theshower, and was almost proportional with cumulative precipitation. When rainfallstopped, the backscatter decreased in a few hours to the same level as before theshower, -0.5 dB. The negative value of a dry canopy results from the averagedbackscatter being defined as 0 dB, while the wet canopy backscatter values weregenerally above 0 dB.


Figure 3.7. The 5-min-averaged and hour-averaged backscatter between 5 and 6 October 12h00. The x-axis hours are equal to the hours in Figure 3.5. After the beginning of the shower backscatter increasedstrongly and decreased exponentially. During the first shower the rain gauge measured 0.8 mm rainwaterand during the second 0.6 mm rainwater.

Evaporation of storage is relatively small during a shower, as the atmospherichumidity is high. Consequently, storage can be assumed to be a single function of theamount of precipitation. The increase in ∆σ0 after the beginning of all showers wastherefore studied to assess the relation between storage and ∆σ0. First, a data selectionwas applied. Each shower was assumed to start as soon as the rain gauge detected rain,after at least one hour of “no rain”. The “no rain” test was used as most storage causedby a previous shower was assumed to have been evaporated within one hour. Theshower was defined to end when the rain gauge did not record rain for the followinghalf-hour. Consequently, it could stop raining for almost half an hour during theshower, causing the canopy to dry. Showers with a total precipitation of 0.2 mm, thedetection limit of the rain gauge, were finally excluded due to the limited accuracy ofthe rain gauge measurement. 14 showers fit the selection criteria. The 5-min-averaged∆σ0 was plotted versus the cumulative rainfall since the start of the shower (Figure3.8). The backscatter level at the start of each shower was scattered. This scatter wasattributed to two processes: (i) the canopy might be partly wet due to a previousshower, dew, or rain with an amount less than the resolution of the rain gauge, 0.2mm, and (ii) fading and noise resulting from the high temporal resolution of 5minutes. The backscatter increased with precipitation until the precipitation amountreached 2 mm. After 2 mm of rain, the backscatter remained relatively stable around∆σ0 = 1.75 dB, indicating that storage reached its maximum. Only 2 showers exceeded3 mm rain. The decrease of ∆σ0 after 5.6 mm rain had fallen is caused by drying of thecanopy. The other scatter visible during the two largest showers could be noise andfading. It is however striking that during two sharp decreases (at 4 mm and 6.2 mm),


Figure 3.8. Backscatter as a function of cumulative precipitation for 14 showers. Each shower has its ownsymbol. During each shower the canopy could dry for maximal 30 minutes.

the windspeed had a temporary maximum, above 1.2 m.s-1, and on the third sharpdecrease, at 5 mm, the precipitation rate strongly increased till 10 mm.h-1. Part of thestored rain may thus have been shaken off by gusts of wind or driving rain, causingthe backscatter to decrease. The backscatter averaged over the whole week with a drywetness sensor was -0.25 dB. With a water saturated canopy ∆σ0 of +1.75 dB, the sen-sitivity of X-band radar to storage appears to be 2 dB.

3.5 Discussion

The radar measured the wetness of a tree in a reproducible manner. This result will bediscussed in relation to a possible future application of radar in forest hydrologicalresearch. Attention will be paid to the measurement precision and the correlation withother direct measurements of storage.

The measurement precision was influenced by radar system effects and by fading.A system effect, which influenced the measurement precision, was the variation ofsome peaks in the background spectrum with the 24-hour period (Figure 3.3). Thesource of these variations was unknown, but it was demonstrated that variances inbackground spectrum, which included noise, in the range cells where the tree was pre-sent could account for maximal 0.25 dB. The measurement precision on a time scaleof hours will be less influenced by the 24-hour period in background spectrum. Understable environmental conditions, the hour-averaged backscatter change indeeddeviated only a few tenths of dB, at least, as long as the tree moved, while the5-min-averaged ∆σ0 deviated ~0.5 dB from the hour-averaged ∆σ0 (e.g. Figure 3.7).


The deviations of the 5-min-averaged ∆σ0 were explained by the short interval oftemporal averaging, causing fading and noise to influence the measurement precision.The precision of the 5-min-averaged ∆σ0 is therefore assessed to be 0.5 dB, and thehour-averaged backscatter precision is assessed to be 0.25 dB.

Fading might seriously influence backscatter. Our approach to reduce fading wastemporal averaging of backscatter from the moving tree. As our measurements wereperformed in a windy season and on a wind-exposed tree, only a few measurementswere excluded. To our experience, it was very difficult to measure small motions onthis wind-exposed single tree. Tree motion will be reduced within a forest, and thetemporal averaging approach to reduce fading will be less suitable for application inforests. We recommend therefore to suppress fading by spatial averaging over inde-pendent samples in future radar measurements of single trees or forest stands.

The experimental results indicate that this radar is useful to quantify rain storage.The similarity in backscatter change during and after the showers in the night of 5 to 6October was in agreement with similar shaped exponential decrease observed byLarsson (1981), Calder and Wright (1986), Bouten et al. (1991), Theklehaimanot andJarvis (1991), and simulated by Rutter et al. (1971). A second result that points to theusefulness of radar to quantify water storage was the wetting of vegetation by 14showers. During the wetting, the 5-min-averaged ∆σ0 falls within a bandwidth of 1 dBaround the mean ∆σ0, which agrees with the estimated precision of 0.5 dB.Differences between the showers could therefore be attributed to fading or noise.Other causes of differences between the showers are the initial wetness of the canopy,and the difference in evaporation rate during the shower. The general trend, however,was a saturated canopy above 2 mm precipitation, in agreement with other rainfallinterception observations (Rutter et al., 1971; Hancock and Crowther, 1979; Aston,1979; Theklehaimanot and Jarvis, 1991).

3.6 Conclusion

The X-band radar backscatter of the deciduous canopy increased proportionally withcumulative rain. The backscatter stopped increasing after 2 mm of rain, indicating thatthe canopy was saturated. The difference in backscatter between a dry and a rain-satu-rated canopy was 2 dB. The radar cross-section decreased exponentially within hoursafter rain stopped, in agreement with other published measurements on evaporation ofstored water after rainfall. It is concluded that radar can monitor storage in forestcanopies. As additionally the radar apparatus used in this experiment had a lowweight, and an integrated receiver and transmitter, the general conclusion is that radaris a new, flexible and simple tool for forest hydrological research. It is recommendedto suppress fading by spatial averaging over independent samples, instead of temporalaveraging over moving tree samples.

53

Chapter 4

Monitoring of Rainwater Storage in Forests with Satellite Radar

Joost de Jong, Wim Klaassen, and Piet Kuiper

Abstract. The sensitivity of radar backscatter to the amount of intercepted rain in temperate deciduousforests is analysed to determine the feasibility of retrieval of this parameter from satellite radar data. Abackscatter model is validated with X-band radar measurements of a single tree exposed to rain. A goodagreement between simulation and measurements is observed and this demonstrates the ability of radar tomeasure the amount of intercepted rain. The backscatter model is next applied to simulate different satel-lite radar configurations. To account for forest variability, the backscatter difference between a wet anddry forest canopy is calculated for four deciduous tree species, above a wet and a dry soil. On average, thesimulated backscatter of a wet forest canopy is 1 dB higher than the backscatter of a dry forest canopy atco-polarised L-band, and 2 dB at co-polarised C and X-band. The simulated sensitivity is in agreementwith observations. It is argued that current satellites can retrieve the amount of intercepted rain at bestwith a reliability of 50%, due to the variability in soil moisture, species composition and system noise.We expect that the reliability will improve with the launch of the next generation radar satellites. Theresults of this analysis may also be used to assess the influence of rain, fog or dew upon other radar appli-cations for temperate deciduous forests.1

4.1 Introduction

Since the launch of the ERS-1 in 1991, imaging satellite radars continuously monitorforests. Examples of products generated from radar images are maps of forest bio-mass, forest type, tree moisture content, and even soil moisture content and floodingunder forest canopies (Hess et al., 1990; Dobson et al., 1995; Pulliainen et al., 1996;Moghaddam and Saatchi, 1999). Radar images acquired during or just after rain areunsuitable for most applications, because radar backscatter of a forest changes by wet-ting (Bernard and Vidal-Madjar, 1989; Way et al., 1990; Ahern et al., 1993; Rignot etal., 1994; Proisy et al., 2000; De Jong et al., 2000a). This sensitivity of backscatter toforest wetness may result in a new application of radar images: retrieval of the amountof intercepted rain in the canopies of forests.

2001 IEEE. Reprinted, with permission, from IEEE Transactions on Geoscience and Remote Sensing(under review). This material is posted here with permission of the IEEE. Such permission does not inany way imply endorsement of any of the University’s products or services. Internal or personal use ofthis material is permitted. However, permission to reprint/republish this material for advertising orpromotional purposes or for creating collective works for resale or redistribution must be obtained fromthe IEEE by sending a blank email message to [email protected].


A dense forest canopy intercepts most raindrops at the beginning of rainfall. Thiswater is retained as small droplets or as a thin waterfilm upon the surface of leaves andbranches (Horton, 1919). A canopy can retain up to a certain amount of water, themaximum storage capacity. After a rainstorm, stored water quickly evaporates, mostlywithin hours. A forest canopy is therefore a temporal water reservoir at the surface-atmosphere boundary. This reservoir can also be filled by interception of cloudwateror dew. Both precipitation and maximum storage capacity are distributed spatiallyheterogeneous. The dynamic and spatially heterogeneous behaviour results in largeuncertainties in the quantification of rainfall interception (Shuttleworth, 1988; Pitmanet al., 1990; Eltahir and Bras, 1993). It was for example demonstrated that themodelled Amazon basin surface-climatology would change from a run-off dominatedto an evaporation dominated regime by ignoring spatial heterogeneity of rainfall.Other theoretical studies have shown that the interception loss of precipitation calcu-lated with spatially averaged parameters deviates a factor 3 to 4 with the interceptionloss calculated with detailed models accounting for heterogeneity (Noilhan et al.,1997; Ghan et al., 1997). An adequate representation of the hydrological balance atthe land surface is one of the main uncertainties in global climate simulations (Gates etal., 1996). A snapshot of the spatial distribution of rainwater storage could providedata to determine spatially averaged interception, which in turn can be used to tunelarge-scale climatological, meteorological and hydrological models.

Whether the amount of rainwater storage can be measured by radar, depends uponthe sensitivity of radar backscatter to storage. Scarce reports give insight into the back-scatter processes when a forest becomes wet. Shortly after rain, the backscatter of amixed forest increased with 0 dB at P-band (0.44 GHz), with 1-2 dB at L-band (1.25GHz), and with 2-3 dB at C-band (5.26 GHz), for all polarisation directions (Dobsonet al., 1991), demonstrating that the radar sensitivity to storage of rainwater dependsupon the radar frequency. Forest type also influences the backscatter change after rain.The rain induced C-band backscatter increase of nearby deciduous and coniferousstands was 2 dB and 1 dB in Alaska, and 1.5 dB and 0.9 dB in France, respectively(Rignot et al., 1994; Proisy et al., 2000). For the same amount of rainwater storage inthe canopy, model simulations resulted in a stronger backscatter increase for decidu-ous forest than for coniferous forests (De Jong et al., 2000a). Therefore, the strongerbackscatter increase of wet deciduous stands was probably caused by the difference inforest structure. The maximum sensitivity of radar backscatter to storage and how thisdepends on the radar properties and forest structure, remains uncertain due to thelimited number of reported observations.

The objective of this study is to determine the potential of present and futureimaging satellite radars for storage retrieval, or in other words: is the maximum radarsensitivity to storage large enough to enable storage retrieval from radar data? Toanswer this question, the relation between radar backscatter and rainwater storage isanalysed for a variety of radar configurations. The analysis is performed with aphysical model, due to lack of detailed observations. To secure the soundness of thistheoretical approach, the model is first validated with small-scale radar measurementsof a single tree, and the final result is compared with large-scale observations. The

Monitoring of Storage with Satellite Radar 55

simulations are restricted to temperate deciduous forest, because this is a promisingforest type: observed backscatter changes after rain were largest for deciduous forest,and this biome occurs in regions with humid and semi-humid climates (Röhrig andUlrich, 1991). As forests differ from each other, the simulations are executed for fourdeciduous tree species, above a dry and wet soil, to account for differences in foreststructure and soil moisture. The results of this analysis may also be used to assess theinfluence of rain, fog or dew upon the radar applications that are mentioned in the firstparagraph of this introduction.

4.2 Model

The theoretical base for simulating backscatter from rain-wetted vegetation was laid inour previous work (De Jong et al., 2000a): the dielectric constant of the rain-wettedvegetation parts can be calculated from the amount of rainwater storage on the surfaceof the leaves or branches, and the amount of water inside the leaves or branches. Theradar backscatter from the forest as a whole is simulated next with a radiative transfermodel. The main models for these calculations will be described in this paragraph.This array of models is extended in the present study, because in-situ measurements ofthe amount of rainwater storage are very complicated (Calder and Wright, 1986;Bouten et al., 1991; Theklehaimanot and Jarvis, 1991). The total amount of rainwaterstorage in the tree is instead modelled from precipitation by (Aston, 1979):

[ ]satc SkPsat eSS −−= 1 (4.1)

where S is storage, or the total amount of rainwater that is retained on the surface ofthe plant, Ssat the saturation storage capacity and Pc cumulative precipitation, all in mmper unit ground area. The empirical factor k is set equal to 1-p, where p is the propor-tion of rain falling through the canopy. This equation is only valid when evaporation islow, thus during and shortly after rain. The saturation storage capacity is the maxi-mum amount of storage that a plant can retain during rain. It is approximated by themaximum storage capacity, Smax, which is defined as the maximum amount of storageafter a long rainstorm under calm weather conditions, when drainage from the canopystopped (Horton, 1919). The difference between these two parameters is the amount ofrainwater that after rain drains from the plant under calm conditions (no wind), whichis generally a small fraction of total storage (Aston, 1979). The specific storagecapacity, S0, is the maximum storage capacity per unit plant area. S0 is given in mm,and equals the waterfilm thickness on the surface of a leaf when the retained water isspread out evenly over one side of the leaf. Smax is calculated from S0 of leaves,branches and trunks, and their one-sided surface area, A, by (Herwitz, 1985):

S S Ai ii

max ,= ∑ 0(4.2)


ε ν ε ν εwet leaf dry leaf dry leaf water water=

+

where the subscript i stands for the entity: leaf, branch or trunk. S0,i may differ betweenspecies. For a given species, S0,leaf is generally lower than S0,branch and S0,trunk (Herwitz,1985; Liu, 1998). S from Equation (4.1) is scaled down to storage per unit plant areaby substitution of S instead of Smax in Equation (4.2). Under the assumption that eachunit plant area in the canopy has the same chance to get hit by a rain droplet (Calder,1986), storage per unit plant area on leaves, branches and trunks is even, till storage onleaves reaches S0,leaf. Storage on branches and trunks is still lower than S0,branch andS0,trunk, and additional storage is located on branches and trunks.

A second model relates the amount of storage per unit plant area on leaves andbranches with backscatter by volumetric averaging of the dielectric properties of theretained water and the tree. The equations for calculating the dielectric constant of wetleaves are given. The dielectric constant of wet branches and trunks is calculated withthe same approach.

The effective dielectric constant of a wet leaf is (De Jong et al., 2000a):

(4.3)

where v is the volume fraction and ε the complex dielectric constant. The effectivethickness of the “leaf with waterfilm” entity increases with the thickness of theretained waterfilm on the surface of the leaf. The volume fractions of retained watervwater and the leaf vdry leaf are respectively the thickness of the waterfilm and that of thedry leaf, divided through the total thickness of the leaf with waterfilm. The watercontent inside the leaf determines the dielectric constant of the dry leaf. It is calculatedwith the Cole-Debye dual-dispersion model (Ulaby and El-Rayes, 1987):

residuefreefreeboundbounddry leaf v εενεε ++= (4.4)

where the subscript bound stands for water inside the leaf that cannot oscillate freelyto the applied radarwave because it is bound to organic molecules, and free for the freewater inside the leaf. εresidue accounts for a residual term due to solid matter inside theleaf. This term and the volume fractions are calculated from the gravimetric fractionwater of the leaf, Mg, by (Ulaby and El-Rayes, 1987):

( )2

2

3671

644

g

g

bound M.

M.v

+= (4.5)

( )0760550 .M.Mv ggfree −= (4.6)

216674071 ggresidue M.M.. +−=ε (4.7)


Figure 4.1. Geometry of a forest canopy and the backscattering pathways in the first-order solution modeof the radiative transfer model: (1) crown scattering, (2) crown-ground interaction, (3) trunk scattering,(4) trunk-ground interaction, (5) ground scattering. This figure is drawn after Karam et al. (1992).

The dielectric constant of bound water is calculated as a function of the radarfrequency by (Ulaby and El-Rayes, 1987):

50

1801

5592

.bound

.jf

.

+

+=ε (4.8)

The dielectric constant of water at a temperature of 10 ºC (the averaged air tempera-ture during the validation measurements) is (Stogryn, 1971):

fj

jffree

σε

18

6.121

1.799.4 −

++= (4.9)

where f is the frequency in GHz, and σ the ionic conductivity of the free water insidethe leaf. The ionic conductance has a value of 1.27. The dielectric constant of inter-cepted rain, εwater, is also calculated with Equation (4.9). Retained rainwater isassumed to be pure, and therefore σ = 0. Forest backscatter is finally simulated with aradiative transfer model (Karam et al., 1992). The forest is schematised as anassemblage of dielectric disks and cylinders with a given orientation, organised inmaximal 3 horizontal layers. Disks represent leaves, and cylinder branches and trunks.The model accounts for the first-order backscatter pathways of Figure 4.1. Second-


order scattering calculations (e.g. soil-trunk-leaf) demands extensive computationalresources, and only contributes to the absolute value of cross-polarised scattering(Karam et al., 1992; McDonald et al., 1991). This study aims at the backscatterdifference for a large number of forest situations, and therefore second-orderscattering is ignored. The soil backscatter is calculated with the integral equationmodel (Fung et al., 1992), using a soil dielectric constant determined by soil moisturecontent and soil composition (Hallikainen et al., 1985).

4.3 Validation

The models are validated with radar measurements of a single tree exposed to rain.This procedure is described in three parts. First, the radar measurements. These meas-urements are only briefly described as details can be found in chapter 3. The nextparagraph is about the collection of input data in order to run the models. The finalpart of this section describes the model simulations.

Radar measurementThe measurements have been conducted at the experimental field of the University ofGroningen, the Netherlands (6° 40’ E, 53° 10’ N). Figure 4.2 is a sketch of the experi-mental set-up. The tree was a mature ash (Fraxinus excelsior), 20 m tall. The foliageformed a 3-4 m thick surface layer around the centre of the tree. The tree branched outjust above the ground, and at breast height 12 stems were present. The radar was aground-based FM-CW radar, built by METEK GmbH, Elsmhorn, Germany. Theoperation frequency was 10.4 GHz (X-band) and the polarisation vertical. The radarwas located at 60 m from the tree centre and pointed at the leafy upper-canopy. Thebeam width was 3°. A weather station was installed halfway the radar and the tree.Data of the first week of October 1999 were processed. A datalogger recorded the 5-minute-averaged radar backscatter in linear units. To secure that the recorded datawere the average of independent samples (and fading was excluded) measurementswith wind-driven tree motion were only processed. This motion was secured byapplying a threshold on the Doppler shift in the reflected signal. The backscatter inlinear units was converted to dB. The mean backscatter of the dry tree was set at 0 dB.The wetness of the tree was assessed with wetness sensors in the crown of the tree andat the weather station. A rainstorm was defined as being preceded by at least one hourof ‘no rain’, to allow the tree to dry from previous rainstorms. The recorded back-scatter of 14 storms passed the data selection. To reduce scatter in the data, backscatterwas averaged over all rainstorms. The number of storms that exceeded 1, 2, or 3 mmwere 10, 7 and 2, respectively. The backscatter data recorded after 3 mm rain wereexcluded because of the low number of storms that exceeded 3 mm.


Figure 4.2. Sketch of the experiment with the location of the X-band radar (X), weather station (W) andthe tree. The arrow points at the wind direction during the experiment. The radar observed the windexposed side of the tree.

Tree measurementThe model requires density, geometrical and dielectric data from the tree parts tosimulate radar backscatter. The large-scale structure of the ash was a thin surface layerwith a high density of leaves and branches, which surrounded an inner region withmainly trunks and major branches. The ground-based radar observed a cross-sectionwith dense foliage. In the model, downward-looking radar observes a closed canopy.Because an ash has only sun leaves and no shade leaves, we assumed that the sides ofthe trees are not foliated in a closed canopy forest. Therefore, the schematised ashforest was assigned a crown and trunk height representative for the centre of the soli-tary tree. The heights were 3.5 m and 16.5 m, respectively. This was measured onphotographs of the tree before and after leaffall. It was assumed that the throughfallfraction p equalled the gap fraction in the canopy for light transmission. The latterparameter was determined on photographs of the ash: it was 0.23. The value of k wasconsequently 0.77.

The small-scale dimensions of the tree parts were derived as follows. An ash haspinnate leaves with 7-13 separate leaflets, attached to a long, central nerve. The centralnerve and the attached leaflets were the smallest entities in the simulations; the leafitself was not regarded as an entity. The number density was estimated by counting theleaflets, nerves, and secondary branches present at the end of a primary branch anddividing the numbers through the sampled volume, ~0.94 m3. This part was represen-tative for the surface layer that contained the foliage. The dimensions from 543leaflets, 107 nerves and 17 secondary branches were measured. The leaflet length wasmeasured and related to leaflet area with a regression equation (area = 1.05 ×length2.72, R2 = 0.98, n = 40). The leaflets were divided into 5 classes to avoid


frequency-dependent oscillations in the simulations. The density and dimensions ofthe large branches and trunks were obtained from field measurements and photographstaken after leaffall. The orientation of trunks, branches and leaves is described in themodel by a probability density function (pdf), which is a sine or cosine function of theangle β with the vertical normal (Karam et al., 1992). The pdf’s of the trunks andbranches were derived from photographs taken after leaffall. The pdf’s were fitted tohistograms that described the orientation of the branches and trunks in classes of 10degrees. The visual impression was that the orientation of trunks and branches did notchange after the tree shed its leaves. Leaves were assigned a sin β orientation such thatthe direction in which the normal to the leaf surface is oriented is uniformly distrib-uted over a spherical surface (McDonald et al., 1991).

The gravimetric fraction water of leaflets, Mg, was determined by the difference inweight of fresh and oven-dried leaves. The leaves were picked from the lowerbranches. The Mg varied between 0.59 and 0.65. The slightly low value of Mg = 0.60was used to calculate the dielectric constant because the radar observed the top of theash, and measurements in another ash indicated that the water potential in the top of anash is lower than that at lower branches (Cochard et al., 1997). Branches and trunkswere assigned the same water content. This value was found to be representative forthe outer 0.5 cm of the trunks and branches. The specific storage capacity of leaflets,S0,leaflet, was determined for 40 leaflets according to Liu (1998). The method was asfollows: after the determination of the fresh weight, each leaflet was submersed inwater for 20 seconds. The droplet at the tip of the leaflet was removed with a blottingpaper to simulate wind driven shake-off, and the wet weight of the leaflet was deter-mined. S0,leaflet was finally calculated by dividing the wet and dry leaflet weight differ-ence through the one-sided area of the leaflet. The averaged value of S0,leaflet was 0.06mm, and the standard deviation 0.016 mm. These values are representative for the

TABLE 4.1

Model input for the ash simulations. The branches and leaves were present in a 3.5-m-thick foliage layer, thetrunks in a 16.5-m-thick trunk layer. Additional input data are Mg = 0.6, and k = 0.77. The specific storagecapacity of leaflets is determined in the laboratory (first value) and by adjusting the model (second value).

radius(m)

thickness(cm)/

length (m)

density(m-3)

orientation(pdf)

specificstorage

capacity (mm)Leaflet 1 1.9.10-3 0.005 65 sin β 0.06 / 0.09Leaflet 2 6.8.10-3 0.005 260 sin β 0.06 / 0.09Leaflet 3 1.3.10-2 0.010 553 sin β 0.06 / 0.09Leaflet 4 2.1.10-2 0.015 505 sin β 0.06 / 0.09Leaflet 5 3.0.10-2 0.020 260 sin β 0.06 / 0.09Nerve 5.5.10-4 0.13 220 sin β 0.09Branch 1 8.0.10-3 2.00 1.1 cos4 β 0.21Branch 2 3.6.10-3 0.38 18.1 sin2 2β 0.21Trunk 7.9.10-2 16.5 2.9.10-3 cos β, 0°>β>15° 0.21


Figure 4.3. The simulated X-band backscatter as a function of the amount of retained rain. The total back-scatter is the sum of the contributions from leaflets, nerves and branches in linear units. This sum is nextconverted to the logarithmic decibel scale. The leaflets cannot retain more water after 0.41 mm is stored,and additional storage is located on the surface of branches and nerves. This additional storage hardlyinfluenced total backscatter.

amount of storage when excess water drained. The S0, branch and S0, nerves were acquiredby submersion of 20 nerves and 20 small branches for 1 minute (Hutchins et al.,1988). The area of the branches and nerves was calculated from their length andradius. S0,trunk was taken equal to that of the branches. The combined results are sum-marised in Table 4.1.

SimulationsFirstly, the backscatter processes were simulated under conditions of a canopy thatbecomes wet during rain. Figure 4.3 shows the vertical polarised X-band backscatterand the contributions of leaflets, nerves and branches as a function of rainwaterstorage. Trunks and soil were not included in these simulations, because the ground-based radar only observed the densely foliated upper canopy, where the trunks hadapproximately the same radius as the branches. The contribution of leaflets dominatedtotal backscatter. With increasing storage, backscatter from leaflets increased, andbackscatter from branches and nerves simultaneously decreased, due to enhancedattenuation by wet leaves. Backscatter from leaflets increased till storage per unit plantarea reached the maximum storage on leaflets, at S0,leaflet = 0.06 mm. This happenedwhen total storage S was 0.41 mm. Nerves and branches were not yet saturated withrainwater and retained additional storage. Nerve backscatter increased more thanbranch backscatter by this additional storage, because the dielectric constant of wetnerves increased more than that of the branches. This is logical because the effective


Figure 4.4. Measured and simulated vertical polarised X-band backscatter change of the experimental treeas a function of cumulative precipitation. The measurements are the average of 14 rainstorms. The back-scatter change during the individual rainstorms is plotted in Figure 3.8. In that figure, 0 dB is set at theaveraged backscatter (wet and dry tree) during the whole measurement period, while in this figure 0 dB isset at the average backscatter of the dry tree. The difference is 0.25 dB. The original simulation is per-formed with sampled input parameters. The second simulation is executed by increasing the water storageon leaflets with 50% till 0.09 mm per unit one-sided leaf area.

dielectric constant is the volumetric average of the dielectric constant of the dry nerveor branch and that of the retained rainwater. Both nerves and branches had an equalthick waterfilm on their surface, while the volume of the nerves was smaller than thevolume of branches. On the other hand, the increased backscatter from nerves andbranches hardly influenced total backscatter, because this backscatter was dominatedby leaflet backscatter.

The simulations were compared with the measurements after transforming storageto precipitation with Equation (4.1), and setting the backscatter of the dry tree at 0 dB(Figure 4.4). The correlation between simulations and measurements was high, R2 =0.84, especially in the early stages of the rain storms (Pc < 1.8 mm, R2 = 0.93). Thesimulated backscatter increase was lower than the measured backscatter increase. Thecorrelation between simulations and measurements improved when the value of S0,leaflet

was increased from 0.06 mm till 0.09 mm (R2 = 0.90). We chose to increase S0,leaflet

because the previous simulations indicated that the total backscatter was dominated bybackscatter from leaflets. The original value of S0,leaflet was determined under labora-tory conditions. The radar observations were made during rain. The laboratory deter-minations of S0,leaflet were representative for the amount of storage after rain, whendrainage of excess water ended. The radar observations were made during rain. There-fore, the amount of storage on the surface of leaves during rain could be higher thanthe laboratory determined value.


Other processes may also change the modelled sensitivity of radar backscatter torainwater storage on the surface of leaves. The water content of leaves was assumed tobe relatively low, Mg = 60%. A simulation with Mg = 65% resulted in a 0.3 dB lowerbackscatter change due to wetting, increasing the discrepancy between the measuredand modelled backscatter. Alternatively, the water content inside the leaves couldincrease during wetting of the tree, because liquid water practically blocks stomata,and transpiration of leaves stops (e.g. Ishibashi and Terashima, 1995). When thegravimetric water content of the leaves was increased from 60% to 65% duringwetting of leaves, the modelled radar sensitivity to precipitation appeared to be only0.1 dB higher than the previous simulations. The reason for this low sensitivity to theinternal water content of the leaves was that the dielectric constant of the wet leaveswas dominated by the waterfilm on surface of the leaves, and not by the internal watercontent of the leaf. Rainwater on the surface of leaves was assumed to be pure, but itmight be little saline. Elements like Na+ and K+ leach for example from the interior ofleaves to the waterfilm on the surface of leaves (e.g. Gordon et al., 2000). Because ofthis leaching, rainwater is expected to have a lower salinity than the water insideleaves, or σ <1.27. According to Equation (4.9), a suchlike salinity of rainwater hardlyinfluences the dielectric constant of water at high frequencies. Finally, the orientationof wet leaves could change due to the larger weight of wet leaves. This was notexpected to be important because only measurements with wind-driven tree motionwere used. Therefore, the most likely explanation for the discrepancy between themeasured and simulated backscatter change is the underestimation of the amount ofrainwater storage on the surface of leaves. However, in the discussion we will comeback whether the large increase (50%) in the amount of rainwater storage on leavesduring rain is realistic.

4.4 Sensitivity Analysis

After the model validation with small-scale radar measurements, the model wasapplied in a theoretical analysis to relate the sensitivity of satellite radar to rainwaterstorage in an arbitrarily deciduous forest. Forest structure and soil moisture varieswithin a forest. The influence of forest structure was taken into account by calculationof the backscatter change between a wet and dry forest for four different single speciesforests. The species were: a beech (Fagus sylvatica), two poplar species (theeuramerican clone Populus robusta and Populus balsamifera), and an ash (Fraxinusexselsior). These species were chosen from the most abundant species in a temperatedeciduous forest (Röhrig and Ulrich, 1991). Undergrowth was ignored because thiswas assumed to be insignificant under a closed canopy. Storage in each tree wascalculated by assignment of an equal thick waterfilm on leaves (0.09 mm), branchesand trunks (0.21 mm), based on the ash schematisations. As will be argued in thediscussion, the maximum waterfilm thickness on the surface of leaves, and hence themaximum backscatter change, is in the same order of magnitude for the simulated treespecies. The dielectric constant of all vegetated parts was calculated from Mg = 0.6.


Soil moisture variations were taken into account by execution of the simulationsfor three different states of soil wetness: a soil that remains dry or wet, and a soil inwhich wetness increased when the canopy intercepted rain. The dry and the wet soilwere assigned a volumetric water content of 10 and 20%. This difference in soilmoisture content was higher than observed in the top 5 cm of bare soils during and inthe first 5 days after several rain events in the Netherlands (Borgeaud and Floury,2000). We applied this ‘worst-case scenario’ to compensate for uncertainties in soilreflection modelling, because the influence of surface roughness upon radar back-scatter is very complicated. Describing the surface roughness by root-mean-square(rms) height and correlation length is only a simple approximation (e.g. Zzribi et al.,2000). It was for example found that the rms height and correlation length increasedwhen sampled over longer transects (Davidson et al., 1998). The values we used (rmsheight 1 cm, correlation length 4 cm) were based on an Alaskan forest soil (Rignot etal., 1994). Such a soil is smooth at L-band and rough at C- and X-band (Fraunhofer-criterion, Ulaby et al., 1981). Comparable values (rms height 1.2, correlation length 5cm) have been used for analysing the backscatter of an Amazonian floodplain forest ina sensitivity analysis over the same range of frequencies and incidence angles as thisstudy (Wang et al., 1995). Aggregate properties of the forest stands are given in Table4.2.

Next to varying forest structure and soil moisture content, the simulations wereexecuted for different radar configurations. The radar configurations were simulatedby varying the frequency f (1-10 GHz), incidence angle θ (20-60°), and polarisation(vertical, vv, horizontal, hh, or cross-polarised, hv) of the radarbeam. The values of fand θ were chosen to include most present and proposed imaging satellite radars(Huneycutt and Zuzek, 2000).

TABLE 4.2

Tree and soil properties used for the sensitivity analysis simulations. All trees are deciduous and abundant intemperate deciduous forests. Note the variation in tree height, stem density and leaf area index.

Canopypoplar

(P. robusta)poplar

(P. balsamifera)beech

(F. sylvatica)ash

(F. excelsior)Total height (m) 17.1 30.2 14.5 20Height foliage layer (m) 11.4 10.1 3 3.5Stem density (stems.ha-1) 217 1060 500 478Leaf area index 2.8 3.6 6.7 6.2Stem diameter (cm) 26 22.5 8 15Tthickness leaves (cm) 0.022 0.030 0.013 0.016S (mm) 0.33 0.64 0.73 0.73S leaves only (mm) 0.23 0.33 0.60 0.56Source Hoekman et al.,

1995Rignot et al.,

1994Proisy et al.,

2000This study

SoilWater content 10 vol. % (dry), 20 vol. % (wet)Composition 10% sand, 40% silt, 50% claySurface roughness rms height 1 cm, correlation length 4 cm (over a transect of 1 m)


Figure 4.5. The simulated backscatter change between a wet and a dry tree for four different tree speciesabove three different soils as a function of frequency for a vv-polarised radar with θ = 20°. All wet trees wereassigned a waterfilm thickness of 0.09 mm on one side of the leaves. Differences between the simulationsare therefore solely caused by differences in forest structure and soil moisture content.

Influence of tree species and soil moistureThe simulations for a radar with one incidence angle (θ = 20°) and polarisation (vv)are highlighted to demonstrate the variability caused by the tree species and soilmoisture (Figure 4.5). It is emphasised that the simulated backscatter differencebetween a wet and a dry canopy is the slope of the relation between storage per unitplant area and backscatter, because each species was assigned an equal thick waterfilmon leaves, branches and trunks. This waterfilm thickness may differ slightly from themaximum waterfilm thickness, which is unique for each species. The simulated back-scatter change differed largely between the forests at low frequencies. These largedifferences were caused by the variable contribution of soil backscatter to total back-scatter. Backscatter increased most when wetness of both soil and canopy increased.Backscatter change was least when the soil was already wet under the dry canopy,because the relatively large contribution of wet soil backscatter to total backscatterstrongly reduced the radar sensitivity to storage. The contribution of soil backscatter tototal backscatter reduced at frequencies above 6 GHz, and the differences in simulated


backscatter change became small, within 1 dB. These differences were caused by theforest stand structure. At 10 GHz, the highest backscatter change was simulated forthe ash and beech stand, which had the thinnest leaves, and consequently the highestdielectric constants of wet leaves (the dielectric constant of wet leaves was calculatedas the volumetric average of the dielectric constant of the dry leaf and the waterfilmon the surface, which had the same thickness in all simulations).

Influence of radar configurationFor each radar configuration, the backscatter change due to storage was calculated for4 tree species × 3 states of soil wetness = 12 unique forests. Per radar configuration,the results of these 12 simulations were averaged to estimate the sensitivity of back-scatter to rainwater storage in an arbitrarily forest (Figure 4.6). The averagedbackscatter increased when the forest became wet, except for simulated hv-polarisedbackscatter, which decreased when the forest became wet. According to the model, hv-backscatter was dominated by backscatter from branches. When leaves became wet,they strongly attenuated the backscatter from branches, and the simulated total back-scatter decreased. The maximal change was simulated for hh-, vv- and hv-polarisationat respectively 5, 6 and 8 GHz. This maximum appeared at steep incidence angles,except for hv-polarisation, where it appeared at θ = 40°. In all cases, the absolute valueof the maximal change was in the same order of magnitude, 3 dB. The averaged back-scatter change for a radar with hv-polarisation, θ = 20°, was close to 0 dB, and itstrongly depended upon the species. This strong dependency of backscatter upon thespecies resulted in a large standard deviation. The standard deviation of the other radarconfigurations was large at low frequencies, when soil backscatter was most influ-encing. The standard deviation decreased with increasing frequency, till a minimumwas reached at 8-9 GHz. The sign and order of magnitude of the simulated backscatterchange were in agreement with the observations mentioned in the introduction, maxi-mal +3 dB at C-band and +1 dB at L-band, except for hv-polarised backscatter, whereobservations and simulations had an opposite sign.

4.5 Discussion

The objective of this study was to assess whether rainwater storage in deciduousforests could be retrieved quantitatively with imaging satellite radar. The approachwas simulating the backscatter change due to rainwater storage in the canopy of treesfor a number of forest tree species, after a validation of the model. Quantitativeaspects of the validation and sensitivity analysis are first discussed. The results arenext evaluated with large-scale observations. The main research question is addressedat the end of this discussion.


Figure 4.6. The average and standard deviation of the simulated backscatter change due to rainwater stor-age of four different forest trees above three different soils as a function of the radar configuration. Thefigures are organised according to the polarisation direction of the radar, which are from top till bottom avertical, horizontal and cross polarisation. Again, each wet tree had a waterfilm thickness of 0.09 mm onthe surface of leaves.

ValidationThe model was validated with radar measurements of a single tree. The input data forthe waterfilm thickness on the surface of leaves was determined in the laboratory. Thecorrelation between simulations and measurements improved by increasing the water-


film thickness on leaves with 50%. Unfortunately, we did not find any measurementsof the waterfilm thickness on the surface of ash leaves during wetting. Instead, indirectevidence that justifies this increase will be discussed. The waterfilm was determinedin the laboratory by weighting a fresh leaflet, submersing it in water, removing thedrop at tip of the leaflet with a blotting paper, and reweighting the wet leaflet. Thisdetermination resulted in the amount of retained water when gravimetric and reten-tional forces were in static equilibrium, and excess water was removed. On the otherhand, the radar measurements were performed during the first stages of rainstorms,when the tree moved in the wind. Physical models suggest that in the first stage of arainstorm extra water is retained on the surface of leaves before equilibrium develops,and excess water drains to the ground (Rutter et al., 1971). The best available directmeasurements for a comparison of the amount of rain water storage on the surface ofleaves during and after rain (when drainage of excess water ended), were acquired bywetting plants or plant parts in the laboratory and measuring the plant weight before,during and after wetting. The amount of water storage on coyote bush (Baccharispilularis) during spraying with water was 47% higher than the amount of waterstorage after spraying and drainage of excess water ended (Grah and Wilson, 1944).For 5 eucalyptus species, the amount of water storage on the surface of leaves duringsimulated rainfall was 15-30% higher than the amount of water storage after drainagestopped (Aston, 1979). Exceptions were E. maculata, which had a 50% higher storageduring simulated rain, and Acacia longiflora, which had a 70% higher storage duringsimulated rain. Further evidence was provided by Monson et al. (1992). The waterfilmthickness on leaves was determined for a number of alpine species by two differentmethods. The first method was identical to our method. The second method was thatcomplete specimen were wetted with a rainfall simulator. All excess water was imme-diately after wetting stopped removed by shaking. This amount of excess water thatcould drain was scaled down to the waterfilm thickness on leaves. The second methodresulted in a larger S0,leaf than the first method. The difference was even up to an orderof magnitude. It is concluded that during rain the storage on the surface of leavescould be 50% higher than after rain, and underestimation of waterfilm thickness in thelaboratory is a plausible explanation for the difference between model and measure-ment. Therefore, radar appears to be suitable for quantifying storage in a deciduoustree canopy, on condition that detailed data on forest structure and weather are avail-able.

Sensitivity analysisThe sensitivity to storage in a temperate deciduous forest was next assessed by simula-tion of the backscatter of four forest stands. Storage in these stands was calculated byextrapolating the values of S0,leaf and S0,trunk from the ash to the other species. It will bediscussed whether this extrapolated storage approximated the maximal or saturationstorage for each species, and the simulated backscatter change is consequently themaximum backscatter change for all tree species. Measurements of S0 or Ssat were,again, not available for the modelled species, and therefore stand aggregate values of S


(see Table 4.2) were compared with published values of Smax. The maximum storagecapacity of a poplar stand was 0.40-0.66 mm in summer and 0.20-0.39 mm in winter(Elbers et al., 1996). The LAI of this stand was 3.5. Under the assumption that thedifference in maximum storage between summer and winter could be attributed torainwater retention on leaves, the maximum waterfilm thickness on the surface ofpoplar leaves had to be at least 0.06-0.08 mm, in agreement with the used value.Maximum storage after rain stopped of two mature beech stands was 0.5 and 0.8 mmin winter, and 1.2 and 1.3 mm in summer (Kändler, 1986; Hörmann et al., 1996). Bothobservations indicated that storage on leaves in a beech stand is in the order of 0.5-0.7mm, also in agreement with the simulations. On the other hand, measured wintertimestorage strongly deviated from modelled storage on branches and trunks, indicatingthat S0,branch and S0,trunk differ between species. Differences in S0,branch and S0,trunk

influenced simulated backscatter change hardly, because the volume of retained wateron branches and trunks was several orders of magnitude smaller than the branch ortrunk volume. The dielectric constant of wet branches consequently approximated thedielectric constant of dry branches. Recapturing, the maximum storage capacity perunit plant area of leaves, S0,leaf, was in the same order of magnitude for all simulatedspecies. The simulated rainwater storage on leaves approximated this maximal value.The simulated backscatter change between a wet and a dry tree therefore equals themaximum backscatter change due to storage for each tree species. The simulatedspecies are abundant in temperate deciduous forests. It is therefore concluded that thesimulated average backscatter change is indicative for the theoretical maximal back-scatter sensitivity to storage in an arbitrarily temperate deciduous forest.

ObservationsThe quality of the simulations was evaluated by going back to the observations thatwere mentioned in the introduction. Just after rain, the AIRSAR observed backscatterof a forest increased with +1-2 dB at L-band and with +2-3 dB at C-band for allpolarisation directions (Dobson et al., 1991). The modelled co-polarised backscatterchange was in agreement with these observations. Modelled cross-polarised back-scatter change contradicted the observations, because it decreased. According to themodel, hv-backscatter was dominated by backscatter from branches. The backscatterfrom the branches decreased due to enhanced attenuation of wet leaves. The branchbackscatter decrease was more important than the backscatter increase from wet leaves.We therefore expect that hv-backscatter from leaves was underestimated by the model.Experimental evidence was not found for this hypothesis. Without additional experi-ments, a statement on the sensitivity of cross-polarised backscatter to storage cannot bemade. The other observations of wet deciduous forest were made with the radar of theERS-satellite (Rignot et al., 1994; Proisy et al., 2000). The backscatter increased with1.5-2 dB when the forest became wet. This value was slightly lower than the simulatedbackscatter increase. Two reasons can explain the difference: (i) part of the stored rain-water was evaporated, or (ii) due to the steep incidence angle of the ERS-radar, θ = 23°,some direct soil backscatter was received through gaps in the canopy. A large


contribution of soil backscatter to total backscatter weakens the sensitivity to storage.Radar with a flatter incidence angle would receive less direct backscatter from the soil,because the radar wave has to pass a longer path of vegetation before being reflected atthe soil. It is therefore expected that for natural forests the sensitivity of a co-polarisedradar with θ = 20° is reduced till it approximates that of a co-polarised radar with θ =40°. The theoretical sensitivity to storage is therefore assessed to be 2 dB ± 0.75 dB for aco-polarised C- or X-band radar, and 1 dB ± 1.25 dB for an L-band radar.

Feasibility of storage retrieval from satellite radar dataGiven the theoretical sensitivity of radar backscatter to rainwater storage in temperatedeciduous forests, is storage retrieval feasible from data acquired by present or futureimaging satellite radar? Presently available data are recorded with the ERS or RADAR-SAT satellite, which have a C-band radar with one polarisation direction. The measure-ment precision of such a radar generally is within 1 dB (Laur et al., 1997). Taking thisadditional uncertainty into account, the sensitivity to rainwater storage will be in theorder of 2 dB ± 1 dB, which results in an error of 50% in retrieved wetness. This water-film thickness has to be extrapolated to total storage in a canopy. Therefore, a feasibleapplication of the current satellites might be to distinguish wet and dry parts in a largeforest. This spatial information could contain valuable information on the spatial distri-bution of rainwater storage in the canopy of trees for hydrologists and climatologists.

The next generation satellites may acquire vv, hh and hv-polarised C-band imagessimultaneously (RADARSAT-2, expected launch late 2003). It has been demonstratedthat the dielectric constant of a forest canopy can be estimated elegantly from multiplepolarised radar data (Moghaddam and Saatchi, 1999, 2000). The coverage access ofRADARSAT-2 at Equator is every 2-3 days and above 70° latitude daily. RADARSAT-2 is modified to support a proposed tandem mission with RADARSAT-3. With such ahigh temporal coverage, and on condition that additional rainfall data are available, it ispossible to distinguish observations of wet days from a baseline containing observationsof dry days. Changes in forest structure and soil moisture content will be visible in thedry day baseline. The backscatter change between the observations made on wet daysand on dry days can be interpreted in terms of forest wetness. We therefore expect thatthe retrieval of the amount of rainwater storage in deciduous forests will improve withthe launch of the next generation radar satellites.

71

Chapter 5

Estimations of Rainwater Storage in a Deciduous Forest Canopy

by Satellite Radar

Joost de Jong, Eddy Moors1, Wim Klaassen, Piet Kuiper,Paul Saich2, and Maurice Borgeaud3

Abstract. ERS satellite radar was used to estimate the amount of rainwater storage in the canopy of ayoung deciduous forest. The amount of rainwater storage was retrieved after correction for the contribu-tion of soil backscatter. This correction was executed with two different approaches. In the first approach,the soil backscatter contribution was retrieved from another radar image, which contained the backscatterof the dry canopy. It was acquired within 24 hours of the image with the wet canopy. Resulting storagewas in agreement with in-situ measured storage for 2 analysed events. In the second approach, soil back-scatter contribution was modelled from in-situ measured soil moisture content. The radar-retrievedamount of rainwater storage deviated maximal 50% with in-situ measured storage for 4 analysed events.The agreement of satellite retrieved storage and plot measurements implicates that radar can be used tofill an observational scale gap between plot and catchment measurements.

5.1 Introduction

Rain intercepted by a forest is temporarily stored on the surface of leaves, branchesand stems. The wettability of these parts depends on the tree species (e.g. Horton,1919; Fogg, 1947; Herwitz, 1985; Monson et al., 1992; Brewer and Smith, 1997).Evaporation of intercepted rain may reduce the amount of water that reaches the forestfloor up to 50% (Schellekens et al., 1999; Ataroff and Rada, 2000). The retention ofwater by leaves influences several biogeochemical cycles: CO2 exchange between treeand atmosphere reduces due to blocking of stomata by liquid water (e.g. Ishibashi andTerashima, 1995; Field et al., 1998), deposition rates of hygroscopic gasses and aero-sols to wet surfaces speeds up (e.g. Wesely and Hicks, 2000), and nutrients as K, Mgand Ca leach from the leaves interior into the retained waterfilm (e.g. Guentzel et al.;1998, Gordon et al., 2000). The occurrence of a waterfilm on leaves also influencesecosystem health, as wet leaves are an excellent culture for pathogens (Butler, 1996).

Large-scale processes such as deforestation, increased atmospheric CO2 concentra-tion, and changed fresh water supplies are a threat of the future. It is therefore desiredthat hydrological and geochemical processes are monitored at the appropriate temporaland spatial scale (Anderson et al., 2000; Schultze, 2000). The available observational

1 Alterra, Wageningen, 2 University College, London, 3 ESA-ESTEC, Noordwijk


scales are determined by the technical state of the art. Large-scale hydrological andbiogeochemical monitoring commonly uses the catchment budget (e.g. Brooks, 1928;Murakami et al., 2000; Putuhena and Cordery, 2000; Robinson, 2001; Swank et al.,2001). The poor temporal resolution of this method is not useful for investigating theinfluence of rainwater retention in tree canopies, as intercepted rain evaporates fast.On the other hand, upscaling of plot measurements may result in large errors, due toheterogeneous landuse, tree species, precipitation, and evaporation within the catch-ment. An example is the assessment of the yearly water yield of a small catchment(~100 ha) by four methods, operating at observational scales from a single tree to thecomplete catchment (Wilson et al., 2001). The estimated water yield varied by a factortwo. This deviation was partly attributed to rainfall interception, and is expected toincrease for a larger catchment, or for shorter temporal scales. We think that aparameter describing the spatial and temporal distribution of the wetness on leaves canbe used for more reliable upscaling of local hydrological, biological and geochemicalmeasurements to large-scale ecosystem processes.

From a feasibility study, it was concluded that satellite radar has the potential tomeasure the amount of rainwater storage in deciduous forests at large scales (De Jonget al., 2000a). The objective of this study is to evaluate the performance of satelliteradar in quantification of the amount of rainwater storage in the canopy of a deciduousforest. The radar data were acquired by the C-band SAR of the ERS-satellites. The testsite is a poplar stand. Simultaneous ground measurements were executed in theframework of a forest hydrology study (Elbers et al., 1996; Dolman et al., 1998). Soilbackscatter is expected to contribute to total backscatter due to the low leaf density ofthe poplar stand (one-sided leaf area index, or LAI = 3.5). The specific challenge ofthis study is therefore to derive an adequate method for a relatively precise correctionfor the soil backscatter contribution.

5.2 Site Description

The experimental site was situated within the Fledite forest, the Netherlands (52° 19’N, 5° 27’ W, Figure 5.1). The sandy clay loam soil was covered for 76% with poplars(Populus spp.), and for another 17.5% with mainly ash (Fraxinus spp.), oak (Quercusspp.), maple (Acer spp.), and beech (Fagus spp.). The trees were planted in the early1980’s in homogeneous stands for production. Undergrowth of stinging nettle (Urticaspp.), cleavers (Galium spp.) and grasses was present below the poplar stand. In 1995,the phenology in the forest was observed visually. It was as follows: the poplarsbloomed half April. At 27 April only buds were seen on the tree, and on 4 May allleaves emerged. The undergrowth was 90 cm high at 18 May, while caterpillars hadreduced the canopy leaf area by 10%. The undergrowth might go up to 1.5 m high,and had an estimated maximum LAI of 2.5. At 13 July the canopy was fully devel-oped, while the undergrowth started to perish. The poplars start shedding their leavesat the end of September. The growing season is defined for this study to extentbetween 4 May and 4 October.

Estimations of Rainwater Storage by Satellite Radar 73

Figure 5.1. Colour composite of the Fledite forest and surrounding agricultural fields. The extent of thearea is 8 × 8 km. The arrow points at the location of the experimental site. This composite containsimages of 25 May (red), 29 June (green) and 3 August (blue). The weak colour of the forest indicates thatthe forest backscatter did not change much compared to that of the surrounding agricultural fields. Theseasonal backscatter change of these fields was described in Saich and Borgeaud (2000).

5.3 Measurements and Methods

Hydrological measurements and methodsA scaffolding tower with an automated weather station was placed in the centre of theplot. Weather data were logged with a temporal resolution of 30 minutes. Volumetricsoil moisture content at 3 cm depth, mv,3 cm, was measured with a frequency domainsensor. Rainfall above the canopy was recorded with a tipping bucket rain gauge thathad a resolution of 0.20 mm. Rain falling through the canopy was recorded with athroughfall gutter with a length of 10 m and a tipping bucket rain gauge at the outlet.


The gutter rain gauge had a resolution of 0.075 mm. Hydrological data were loggedwith a temporal resolution of 5 minutes. Stemflow of 6 trees, and throughfall collectedby 36 rain gauges were weekly read.

The amount of rainwater storage in the canopy in mm depth, S, was deduced fromthe high temporal resolution precipitation and throughfall measurements according tothe water balance of the canopy:

ETPdt

dS−−= (5.1)

where P is the precipitation rate, T the throughfall rate and E the evaporation rate, allin mm.h-1. The throughfall, recorded by the gutter, was corrected for spatial heteroge-neity effects by the weekly read rain gauges under the canopy. Stemflow was ignoredas it contributed less than 1% of total throughfall. Evaporation rate of the completelywet canopy was calculated by using the Penman-Monteith equation with the surfaceresistance set to zero (Monteith, 1965):

( )γλ

δρ

+∆

−∆= ap

p

qgcAE (5.2)

where A is the available energy, ga the aerodynamic conductance, δq the specifichumidity deficit, ∆ the slope of the saturated specific humidity temperature curve, cp

the specific heat of air, ρ the density of air, λ the latent heat of vaporisation, and γ thepsychometric constant. Evaporation rate of retained rainwater in a partly dry canopywas reduced according to (Rutter et al., 1971):

pmax

ES

SE = (5.3)

where Smax is the maximum storage capacity, which is defined as the maximumamount of rainwater that is retained in the canopy shortly after a rainstorm, whendrainage stopped under calm conditions. Equation (5.3) was only applied when S <Smax. The value of Smax was 0.4 mm in summer, and 0.22 mm in winter (Elbers et al.,1996).

Radar measurements and methodsThe radar images were collected with the Synthetic Aperture Radar (SAR) of theERS-1 and ERS-2 satellites. Two sets of images were analysed. The first set isdenoted as the tandem mode set, and the second set as the single mode set. The tan-dem mode set was acquired during the tandem mission of the ERS-1 and ERS-2 satel-lites. The satellites followed the same orbit, and acquired images of the location 1 day


apart. Available were 2 tandem image pairs. Both pairs included a rain event justbefore or during the satellite overpass, and a dry overpass on subsequent days. Theimages were acquired with an identical viewing geometry at 22h40 local time. Thesingle mode set comprised all images collected by the ERS-1 satellite during the day-time overpass at 11h35 local time. The time lag between these overpasses was 16 or19 days. Again, the viewing geometry was identical for all images. An area ofapproximately 40 × 40 pixels (~0.25 km2) was defined around the location of the scaf-folding tower. The radar backscattering coefficient σ0 (in the logarithmic unit dB) ofthis area was derived from the radar images according to the procedure of Laur et al.(1994). The difference in Analog Digital Converter of the ERS-1 and ERS-2 SAR wastaken into account in the processing of the tandem mode set. The backscatter intensitywas not corrected for attenuation by rain, because the rainfall rate during the satelliteoverpasses was relatively low (2.4 mm.h-1), and according to Table 5.29 of Ulaby etal., (1981) such a rainfall hardly attenuates radar backscatter coefficient.

The objective of this study was to retrieve storage S from the recorded backscattercoefficient σ0. The relation between σ0 and S was ambiguous, because σ0 was alsodetermined by forest (canopy and soil) structure and moisture content. We thereforeused an inverse approach. The branches, trunks and soil structure changed less and atslower temporal scales than storage, soil moisture content, and LAI. The canopymoisture content was relatively stable in a nearby poplar stand (Hoekman et al., 1995).Based on these dynamics, radar backscatter was modelled as a function of the vari-ables storage, soil moisture content, and LAI. In these simulations, the branches,trunks, soil structure, and canopy moisture content were taken constant. The resultswere summed in a so-called Look-Up-Table (LUT), because of the non-linearity of therelations. Details of the generation of the LUT will be described in the followingsection. We continue with describing the approach for retrieval of storage fromrecorded backscatter. The LUT enabled S to be retrieved for any observed σ0, on thecondition that soil moisture content and LAI were known. The soil moisture contentwas determined in two ways: (i) it was retrieved from observations with a dry canopy,and applied for wet canopy observations. This approach used a-priori knowledge toestimate the wetness status of the canopy, and assumed unchanged soil moisturecontent; (ii) soil moisture content was measured in-situ. It was not tried to retrieve LAIfrom SAR-data, because other satellite sensors can provide fair estimates (e.g. Tian etal., 2000; Zhang et al., 2000). Instead, in-situ measured values were used. From Sep-tember 1995 onwards, the LAI of the poplars was systematically measured with anoptical device (LAI2000, LI-Cor, Inc. Lincoln, Nebraska). The measured values of1997 were used for the missing period May-August 1995, because of the agreementwith the visual observations of 1995.

The LUT was generated as follows: the modelled total radar backscatter was thesum from the direct backscatter from the soil, σ0

direct soil, and the backscatter from thecanopy and the underlying soil, σ0

closed canopy:

( )σ σ σ0 0 01= + −p pdirect soil closed canopy (5.4)


TABLE 5.1

Forest canopy and soil structure in the radar backscatter simulations. The leaves and branches were pres-ent in an 11.4-m-thick foliage layer, and the trunks in a 4.8-m-thick trunk layer.

Canopyradius(m)

thicknessheight (m)

moisturecontent (g.g-1)

numberdensity (m-3)

orientation(pdf)

Leaves 0.033 0.00022 0.60 71.8 sin(β+60)Branches1 0.00265 0.09 0.50 4.37 cos(β)Branches 2 0.0105 0.47 0.50 4.67E-1 cos(β)Branches3 0.042 1.04 0.50 1.89E-1 sin2(2β)Trunks 0.123 4.8* 0.48 9.17E-3* sin6(2β), 0°>β>4°

SoilComposition 50% sand, 30% silt , 20% clay (sandy clay loam)Surface roughness rms height 2 cm, correlation length 4 cm

*The original data of Hoekman et al. (1995) were adjusted to our experimental site by decreasing the stem height with1.4 m, and increasing the stem density with 20%.

where p is the gap fraction. It is noted that this summation was executed in linear units.σ0

closed canopy was simulated with the radiative transfer model of Karam et al. (1992). Inthis model, the canopy of the forest was schematised by two horizontal layers. Theupper layer contained leaves and branches, and the lower layer trunks. Both layersattenuated backscatter from the soil under the canopy. Radar is only sensitive to waterstorage on the surface of leaves. To account for storage, leaves were assigned on theirsurface a waterfilm with thickness ds (de Jong et al., 2000a). This adaptation was vali-dated on radar measurements of the canopy of an ash tree with a ground-based radar(chapter 4). The backscatter from the bare soil σ0

direct soil and the soil under the vege-tation cover was simulated with the model of Fung et al. (1992). The vegetation andsoil dielectric constants were calculated from their water content with the semi-empirical models of Ulaby and EL-Rayes (1987) and Hallikainen et al. (1985). Theinput parameters for the Karam model were taken from Hoekman et al., (1995), seeTable 5.1. These parameters were based on samples in a comparable poplar stand inthe same forest by Vissers and Van der Sanden (1993). The surface roughness of theforest soil was taken from Rignot et al., (1994), and the soil composition was based onElbers et al., (1996). The value of the gap fraction p was determined by fitting themodel output with known environmental variables (a dry canopy, in-situ measured mv

and LAI) to the radar measurements. Advantage of this semi-empirical approach is thatit compensates for uncertainties in the models and input parameters.

The radar retrieved thickness of the waterfilm on the surface of leaves, ds, wasscaled up to the amount of rainwater storage in the whole canopy, S, to allow a com-parison with the direct measurements of that parameter. The upscaling was performedunder the assumption that retained water was distributed vertically homogeneous inthe canopy. The total amount of water storage on leaves, Sleaves, was calculated fromthe waterfilm thickness on the surface of leaves, ds, by:


S LAI dleavess= ⋅ (5.5)

When a forests is thoroughly wet, total storage also includes storage on branches andtrunks, or Swood:

woodleaves SSS += (5.6)

Swood was written as a function of Sleaves under the following assumptions: the waterstorage on trunks and branches fill-up and empty in proportion to storage on leaves

( leavesmax

leaveswoodmax

wood SSSS = ). The maximum water storage on the wooden elements

was constant during the whole year, and equal to the maximum storage capacity in

winter ( wintermax

woodmax SS = ). The measured maximum storage capacity in summer is rep-

resentative for the maximum storage capacity of a fully developed canopy, and to

include storage upon wooden parts of the tree ( woodmax

summerleaves,max

leavesmax SSS += ). The

maximum storage on leaves in autumn and spring depends linearly on the leaves den-sity, which changes over time. It was therefore scaled to the averaged LAI in summer

( summerleaves,maxsummer

leavesmax SLAILAIS = ). These assumptions resulted into:

( )wintermax

summermaxsummer

wintermaxwood

SSLAILAI

SS

−= (5.7)

Substituting LAIsummer = 3.5, summermaxS = 0.4 mm and winter

maxS = 0.22 mm into Equations

(5.5), (5.6) and (5.7), gave:

( ) sdLAIS += 27.4 (5.8)

5.4 Results

Qualitative interpretation of measurementsRadar backscatter had a distinct seasonal pattern (Figure 5.2, Table 5.2). The first andlast winter time observations coincide with the occurrence of snow, resulting in a lowbackscatter, -10 dB. 30 January also was a day with snow, but a high backscatter valuewas observed, because snow may have been fallen after the satellite overpass. The soilmoisture content and related backscatter intensity was high for the rest of the wintertime observations, -6 dB. Backscatter dropped between 4 and 20 April with 1.3 dB,which was attributed to attenuation of soil backscatter by the undergrowth. The back-scatter during the growing season from May till October fluctuated between –6.8 dBand –8.1 dB. After the poplars shed their leaves, the backscatter increased in Octoberagain till snow occurred.


TABLE 5.2

Radar backscatter values and environmental conditions during the satellite overpass. Snow observationswere made by the KNMI in de Bilt, at 30 km distance. The value of the LAI in the period May-Septemberwas taken from 1997 measurements.

date temp.(°C)

wind-speed(m.s-1)

atmosph.humidity

(%)

soilmoisturecontent(vol.%)

potentialevapo-

ration rate(mm.h-1)

4 hourcum.

precip.(mm)

24 hourcum.

precip.(mm)

LAI(m2.m-2)

other

Tandem mode set

29 Aug. 11 2 94 22 0.1 2 1.8 3.330 Aug. 12 1 89 22 0.1 0 1.2 3.33 Oct. 16 3 91 39 0.3 0 1 0.74 Oct. 14 2 98 39 0.0 2.8 2.8 0.7 rain

Single mode set

10 Jan. 5 6 76 - 1.4 0 7.6 0 snow30 Jan. 5 7 72 - 2.0 0 5.4 0 snow16 Feb. 9 8 70 - 2.8 0 0 08 Mar. 5 5 77 52 1.3 1.2 5.8 04 Apr. 9 5 75 50 1.7 0 0 0

20 Apr. 7 5 73 50 1.9 0 1.4 0.19 May 11 5 80 32 1.5 0.8 1 0.625 May 18 4 66 47 1.7 0 6.2 1.313 Jun. 15 6 63 50 3.4 0 0 2.429 Jun. 22 7 74 44 3.5 0 0 3.218 Jul. 17 3 96 37 0.3 4 5 3.33 Aug. 26 5 53 29 5.0 0 0 3.422 Aug. 25 6 73 24 3.1 0 0 3.67 Sep. 20 4 69 22 1.3 0.2 1.4 2.826 Sep. 14 5 100 23 0.1 0.6 0.8 1.3 rain12 Oct. 16 2 96 39 0.2 0 0.2 0.131 Oct. 10 6 95 36 0.3 0 0.2 016 Nov. 12 6 90 37 0.4 0 3.8 05 Dec. -3 5 64 39 0.6 0 0 0 snow

21 Dec. -2 0 97 41 0.0 0 0.8 0 snow

Focusing on the growing season between May and 4 October, the strongest back-scatter was recorded during a rainstorm and the weakest after a period of drought,when soil moisture content reached a minimum. Figure 5.3 shows that σ0 was relatedwith mv,3 cm, after exclusion of wet days, which implies that the soil backscatter con-tributed to the total backscatter. Wet days were defined as days with rain during orwithin 4 hours before the overpass. The correlation σ0 with mv,3 cm was significant(Pearson’s correlation coefficient: 0.80, significance level 0.10). σ0 of wet days wasup to 1 dB higher than the neighbouring observations. The exception was 9 May,when it rained just before the satellite overpass, while the measured backscatter was


Figure 5.2. The seasonal variation in radar backscatter, forest development and hydrologic properties.

lower than the previous and following radar observations. The relatively low soilmoisture content was the result of a period of drought, and could explain thiscontradicting observation on 9 May (Figure 5.2). The 9 May observation demonstratedthat the soil might dry and refill within a month. Assessment of the soil moisturecontent at 9 May by interpolation between the neighbouring observation of 20 Apriland 25 May would introduce large errors.

Figure 5.3. The relation between radar backscatter and soil moisture content at 3 cm depth for dry days.


Figure 5.4. Measured precipitation, throughfall and resulting storage at days with a potentially wet can-opy. The arrows point at the satellite overpasses.

In-situ measured rainwater storageThe water balance was plotted for the days when rain occurred in the 4 hours preced-ing the satellite overpass, and the canopy was likely to be wet (Figure 5.4). It waschecked that on the other observational days the canopy was likely to be dry. On 18July, 26 September and 4 October the satellite overpass was during or just after a rain-storm, and the water storage in the canopy was almost equal to Smax. The canopy wasapproximately half wet on 9 May and 29 August, when the canopy had been dryingfor some time. At 7 September the simulated canopy was dry. The observation on 7September was questionable because other (half-hour averaged) weather parameterschanged abruptly just before the satellite overpass. The available energy reduced forexample from 130 till 70 W.m-2, before it increased till 290 W.m-2. This reduction inavailable energy may be due to evaporation of some rainwater that was not detectedby the rain gauge. To account for suchlike uncertainties, the accuracy of the storagedetermination was assessed from the precision of the precipitation and throughfallmeasurements. The detection limits of the rain gauge and throughfall gutter wererespectively 0.20 mm and 0.08 mm. The expected measurement uncertainty of these


Figure 5.5. Root-mean-square error between measured and modeled backscatter for the dry days as afunction of the gap function, p.

instruments was therefore 0.10 mm and 0.04 mm. The real amount of retained rain-water in the canopy during rain was expected to be between S + 0.14 mm and S - 0.14mm, due to these measurement uncertainties. The precision in the evaporation termwas ignored, as Equations (5.2) and (5.3) provide a fair estimate of evaporation(Klaassen, 2001), and a 20% error in cumulative evaporation rate resulted in a smallerdeviation than the deviation caused by the rain gauge, for this particular forest.

Rainwater storage estimated by satellite radarThe first step in retrieval of water storage from the satellite radar data was the genera-tion of the LUT. The backscatter of both soil and canopy were integrated in the LUT.The fraction direct backscatter from the soil, p, was estimated by fitting modelled σ0

for S = 0 mm, in-situ measured mv,3 cm and LAI, to observed σ0 of the dry days between9 May and 4 October. The fit was judged by calculating the root-mean-square errorbetween the modelled and observed σ0. A minimum root-mean-square error of 0.18dB was reached for p = 6% (Figure 5.5). The agreement of the model result for p = 6%with the observations when no leaves were present, and it was not snowing or freez-ing, was ambiguous. The difference between the model and the 8 March and 4 Aprilobservations was reasonable, 0.2 and 0.4 dB. The model overestimated σ0 on 20 April,12 and 31 October with 1.5 dB. In spring, this deviation was attributed to the presenceof undergrowt, which was ignored in the derivation of the LUT. In autumn, freshlyfallen leaves lying on the ground may attenuate the soil backscatter. The resulting LUT for p = 6% showed that the theoretical backscatter decreasedwhen the leaf area increased (Figure 5.6). The sensitivity of σ0 to soil moisture contentwas 2.6 dB when the trees shed their leaves. The sensitivity of σ0 to soil moisturecontent decreased till 1.4 dB for LAI = 2, and till 0.9 dB for LAI = 4, due to enhancedattenuation of soil backscatter by leaves. When the leaves became wet, the attenuationof soil backscatter increased further, and the sensitivity of σ0 to soil moisture content


Figure 5.6. Graphical representation of the look-up table that related leaf area index, soil moisture con-tent, and storage with radar backscatter. The minimum and maximum values were based on the measuredvalues. The displayed layers represent from bottom till top a waterfilm thickness of 0, 0.02, 0.04, 0.06,0.08 and 0.10 mm on the surface of the leaves.

decreased till 0.5 dB for LAI > 2. The sensitivity of σ0 to the waterfilm thickness ds

was for LAI = 1 smaller than 1 dB. It varied between 1.6 dB above a dry soil and 0.7dB above a very wet soil for LAI = 2, and between 1.8 dB above a dry soil and 1.4 dBabove a very wet soil for LAI = 4. This indicates that the sensitivity of σ0 to ds is rela-tively insensitive for variations in LAI when soil moisture content is low.

The waterfilm thickness ds of the tandem mode images was retrieved under theassumption that soil moisture content was not known, and did not change overnight.The first step was to determine mv from the observations of the dry canopy by usingthe LUT. This resulted into mv = 30% for 30 August and mv = 35% for 3 October.Storage on the wet day was looked-up for this retrieved soil moisture content andobserved σ0, resulting in a waterfilm thickness on the surface of leaves, ds, of 0.04 mm(29 August) and 0.11 mm (4 October). Storage of the single mode set of images wasretrieved by using in-situ measured mv,3cm. The qualitative interpretation of the radarmeasurements already showed that the time span between two single mode observations


Figure 5.7. In-situ and by radar measured storage. The range in storage was derived from the propagationof possible errors in the original hydrological and radar observations.

was too long to assume unchanged soil moisture content. The resulting water layerthickness on the surface of leaves was 0.00 (9 May), 0.03 (18 July), 0.04 (7September), and 0.11 mm (26 September). It is noted that both radar observationsduring rain resulted in the same value of ds, 0.11 mm. Analogous to the in-situ deter-mination of S, the precision of the radar-retrieved storage was assessed. The measure-ment accuracy of the radar backscatter measurements was claimed to be within 0.4 dB(Laur et al., 1992). The precision of the retrieved storage was estimated by retrievingS for σ0 ± 0.2 dB.

Comparison of radar and in-situ measured rainwater storageThe radar retrieved waterfilm thickness was transformed to S with Equation (5.8). Forthe tandem mode set, the radar retrieved S deviated ~0.13 mm from the direct meas-urements of S (Figure 5.7). This deviation was within the measurement precision ofboth methods. For the single mode set there is less agreement between the methods.The radar underestimated storage on 9 May and 18 July, and overestimated storage on7 and 26 September. The maximum difference between the methods was 0.28 mm. Insearch for the cause of the scatter in the single mode dataset, the correlation of the dif-ference between Sin-situ and Sradar with the main environmental parameters was calcu-lated (Table 5.4). The difference between Sin-situ and Sradar showed a remarkable highcorrelation with soil moisture content. Radar underestimated rainwater storage whensoil moisture content was high.


TABLE 5.3

Correlation coefficients of environmental variables with the difference between radar estimated storageand in–situ estimated storage.

temperature windspeed humidity soil moisturecontent

potentialevaporation

rate

4 hcumulative

precipitation

24 hcumulative

precipitationLAI

0.56 0.20 -0.21 -0.94 -0.13 -0.62 -0.48 -0.16

5.5 Discussion

The objective of this study was to evaluate the performance of satellite radar in esti-mation of the amount of rainwater storage in the canopy of deciduous forest. Thestudy site was a 14-year-old poplar forest with maximum LAI = 3.5. Radar backscatterof this forest appeared to be related with both soil moisture content and the amount ofrainwater storage. The observed sensitivity of σ0 to soil moisture and rainwater storageis compared with other observations of the ERS satellites. The sensitivity of σ0 to soilmoisture content, ~1 dB, was less than the observed sensitivity of σ0 to soil moisturecontent in bare fields, ~10 dB, (Weimann et al., 1998), or in agricultural fields withyoung crops, ~4 dB (Quesney et al., 2000). It was in line with observed sensitivity tosoil moisture content in boreal forests, < 2dB (Pulliainen et al., 1996). The observedand modelled sensitivity to rain storage, 1-1.5 dB, was slightly less than the observedbackscatter change, 2 dB, of a mature poplar stand in Alaska after rain (Rignot et al.,1994). The observed sensitivity to soil moisture content and rainwater storage wasthus in agreement with other observations.

The amount of rainfall storage was retrieved from the radar observations after cor-recting for the backscatter contribution arising from the soil. Sradar and Sin-situ agreedwith each other when the soil backscatter contribution was determined from a radarobservation, acquired within 24 hours. The agreement reduced when the soil back-scatter contribution was modelled from in-situ measured soil moisture content. Thelargest deviation occurred on 18 July and 7 September, when the difference betweenthe radar retrieved storage and in-situ measured storage was ~0.5Smax. The deviationwas correlated with soil moisture content. Soil moisture content was only one of theaxes of the LUT (Figure 5.6). The other axis were LAI and S. All axis represent awater reservoir in the forest. Assumptions concerning these water reservoirs will bediscussed in the next sections.

The first water reservoir was the soil. The measured soil moisture content at 3 cmdepth was used to model the soil backscatter contribution. The soil moisture contentnear the surface might be higher than measured mv,3 cm during and after a rainstorm dueto slow infiltration rate of throughfall. A simple calculation indicated that the amountof precipitation in the 24 hours before the satellite overpass could increase the soilmoisture content in the top 3 cm by ~5 vol.% (29 May, 7 and 26 Sep.) or ~17% (18 July).


To test this hypothesis, the soil moisture content that could explain the wet dayobservations was retrieved with the LUT. The result was that the top layer soil mois-ture content had to deviate from measured mv,3 cm with -30 vol.% (18 July), -10 vol.%,or +30 vol.% (7 and 26 September). These changes are not realistic, and therefore avertical skew distribution of the water in the top soil does not explain the deviationbetween Sradar and Sin-situ .

The second water reservoir in the forest was the vegetation itself. Assumed were aconstant water content, and no undergrowth. The water content per unit vegetation canbe assumed to be relatively stable, because the poplar roots reached till the shallowgroundwater level, and drought stress can be excluded (Dolman et al., 1998). Thecanopy was sparser and the undergrowth more lush in spring. In summer, the canopyclosed and the undergrowth started to perish. At 20 April, when the tree had no leaves,the radar observation was 1.5 dB lower than the model prediction. This difference wasin the qualitative interpretation of the measurements attributed to enhanced attenuationof soil backscatter by undergrowth. This undergrowth probably influenced theobservation of 9 May, when radar retrieved S was 0 mm, while in-situ measured S was0.19 mm. To compensate for the neglect of the undergrowth, the storage was retrievedfor a higher LAI. The 9 May radar retrieved storage increased till 0.14 mm (waterfilmthickness 0.02 mm) when using LAI = 2 instead of LAI = 0.6 as LUT-input. This valueis more in agreement with the in-situ measured S. It is unlikely that the presence of anundergrowth influenced the other observations, because Figure 5.6 shows that thesensitivity of σ0 to the waterfilm thickness ds is only strongly influenced by the leafdensity for LAI < 2, which was not the case. This discussion pleads for using satellitederived LAI as model input, because a satellite measures the LAI of both canopy andundergrowth.

The third water reservoir was the rainwater stored on the surface of leaves andbranches. In-situ measured was the amount of rainwater storage in the whole canopy.Radar estimated the waterfilm thickness on the surface of leaves. This waterfilmthickness was scaled up under the assumptions that: (i) storage was homogeneouslydistributed in the canopy; (ii) storage could be partitioned in storage on leaves, trunksand branches according to Equations (5.5)-(5.8). The scarce measurements and simu-lations of the vertical distribution of storage in the canopy, reviewed by Klaassen(2001), reveal that leaves in the upper canopy may dry faster than leaves in the lowercanopy, because of the better coupling with the atmosphere. For leaves in the lowercanopy, wooden parts can be read, because these are generally also sheltered. Dryupper leaves may attenuate the backscatter of sheltered wet leaves. An additionalmodel simulation was executed to check this hypothesis. Total storage was takenequal to in-situ measured storage of 18 July, when the canopy was drying. The appliedskewness was a 0.03-mm-thick waterfilm on leaves in the upper half of the canopy,and a 0.09-mm-thick waterfilm on leaves in the lower half of the canopy. The simu-lation equalled the 18 July observation. Similarly, the radar-retrieved storage might beoverestimated at the beginning of rain, as the upper parts of the canopy might bewetter than the lower parts of the canopy. This could explain the deviations betweenSradar and Sin-situ on 26 September and 4 October, when storage was still increasing


during rain, and on 7 September, when little rain was suspected to have been fallen,just before the satellite overpass. Summarising, a vertical skew distribution of theretained water in the canopy, and uncertainties in the partitioning of storage overleaves, branches and trunks are the most likely process to explain the deviationsbetween Sradar and Sin-situ.

Given the conclusion that satellite radar can be used to estimate the wetness of theupper canopy, what is the significance of this application for environmental research?The upper canopy is in close contact with the atmosphere; the turbulent mixing isstrong and it receives the most (photosynthetic active) radiation. The largest fluxes ofH2O, CO2 and depositing gasses between the forest and atmosphere can therefore beexpected in the upper canopy of the forest (e.g. Hinckley et al., 1978, De Jong andKlaassen, 1997). Infections of leaves in the upper canopy are also likely to have a sig-nificant effect on biological productivity (Butler, 1996). It is therefore justified to statethat the biosphere-atmosphere exchange processes in the upper canopy are importantfor ecosystem functioning. The nature of the atmosphere-forest boundary changeswhen leaves become wet. On the other hand, upper canopy leaves dry quicker thanleaves in the forest interior. How much faster is largely unknown. Radar remote sens-ing can estimate the spatial and temporal distribution of the upper-canopy wetness inforests, and can therefore be used for deriving a parameter that describes the temporaland spatial distribution of forest wetness.

5.6 Conclusion

The objective of this study was to evaluate the retrieval of the amount of rainwaterstorage from ERS satellite radar data. The experimental site was a 14-year-old poplarstand. The backscatter of the forest appeared to be influenced by the seasonalparameters: snowcover, forest phenology, soil moisture content and rainwater storagein the canopy. A method to retrieve the amount of rainwater storage in the canopy,after correcting for the contribution of soil backscatter, has been presented. The radar-retrieved amount of rainwater storage deviated maximal 50% with in-situ measuredstorage. It was argued that a reason for this deviation was the omission of the under-growth in the transformation of the radar signal to the amount of rainwater storage inspring, when the canopy is developing. Another reason was that radar was sensitive torainwater storage on the surface of leaves in the upper canopy, while in-situ measuredwater balance of the canopy resulted in the rainwater storage in the whole canopy. Theinteraction between the biosphere and atmosphere is intense in the upper canopies offorests. Radar remote sensing fills therefore an observational-scale gap between plotand catchment measurements in hydrological and geochemical studies of forests.

87

Chapter 6

Potential of Radar Sensing for Measurement of Forest Wetness

6.1 Introduction

Environmental issues, as sustainable fresh water resources and climate change, requiremodels that predict the large-scale water use of forests. Forest water cycling has twocomponents: interception of precipitation by the canopy, and loss of water by transpi-ration by the plants. Simulated large-scale rainfall interception is highly uncertain,even when the spatial distribution of rainfall is taken into account, because of thespatial variability in evaporation rate of wetted leaves and the water storage capacityof the vegetation.

The objective of this thesis was to determine the potential of satellite radar tomeasure the amount of rainwater storage in the canopies of forests. The amount ofrainwater storage in the canopy of a forest is a key factor in the partitioning of rainfallinto throughfall and interception. Radar backscatter is sensitive for the water contentof vegetation. Rainfall events over forests have been detected by satellite radar severaltimes. Radar backscatter of forest may originate from the trees and the underlying soil.It was expected that the contribution of soil backscatter might interfere with radarobservations of a forest canopy. Therefore, two hypotheses were formulated: (i) radarbackscatter was sensitive to the amount of rainwater storage in the canopy of forests;and (ii) the influence of the soil on recorded backscatter could removed. Bothhypotheses have been addressed, directly or indirectly, in several chapters of thisthesis. The conclusion regarding the hypotheses will be reviewed in this final chapter,after a short summary of the experimental radar observations of wet trees. This chapterends with a discussion on the potential of all types of radar, including satellite radar,for measurement of rainwater storage in a forest canopy.

6.2 Radar Observations of Wet Forests

The sensitivity of radar backscatter to the amount of rainwater storage was experi-mentally investigated using ground-based radar and satellite radar. The radar back-scatter of an ash tree without influence of backscatter from the soil was monitoredwith ground-based, upward-looking vv-polarised X-band radar (Chapter 3). The meas-ured difference in backscatter between a dry and a rain-saturated canopy was 2 dB.Observations of the same tree after leaffall, briefly described by de Jong et al. (2000b),but not included in this thesis, demonstrated that radar backscatter was insensitive torainwater storage on the surface of branches. Further observations with the SAR onboard of the ERS-satellites (vv-polarised, C-band) indicated that the backscatterchange between a wet and dry mixed forest (area ~120 km2) was between 0.7 and 2.5 dB,

Potential of Radar for Measurement of Forest Wetness 89

with an average of 1.3 dB (Chapter 2). The amount of rainwater storage in thecanopies of the trees was assessed by calculations taking explicitly into account thespatial distribution of rainfall. Soil moisture content did not change between observa-tions. Therefore, the observed backscatter change was attributed to rainwater storagein the canopy. For the same forest, the backscatter hardly changed between two meas-urements of a partially wet canopy. The ERS-SAR recorded backscatter of a 14-year-old poplar stand (~0.25 km2) changed with 1 dB due to storage in the canopy (Chapter5). This time, the wetness of the canopy was assessed from the in-situ monitored waterbalance of the canopy. The radar observations of the poplar forest contained somebackscatter from the soil, because the backscatter of the dry forest was correlated within-situ measured soil moisture content. The experimental observations were in linewith other observations of wet forests (Table 6.1).

6.3 Sensitivity of Radar Backscatter to Forest Wetness

The observed sensitivity of radar backscatter to rainwater storage in the canopies offorests was explained with the radiative transfer model of Karam et al. (1992, 1995).This model was adapted to account for storage by assigning a waterfilm on the surfaceof leaves, branches and stems. Three sets of simulations with increasing complexitywere executed. This section reviews the results of these simulations.

The first simulations (Chapter 2) were executed to provide an indication of the sen-sitivity of ERS-SAR backscatter to forest wetness. The simulations demonstrated theinvolvement of the backscatter mechanism: when the canopy became wet, the back-scatter of leaves and needles increased. Wet leaves and needles also attenuated morestrongly the backscatter from branches, stems and underlying soil. The simulatedmechanism was comparable with the simulations of Saich and Borgeaud (2000), whoused a slightly different model to simulate ERS-SAR backscatter from wet wheat. Thetrade-off between backscatter from, and attenuation by leaves and needles appeared tobe strongly depended upon forest type. The resulting sensitivity of radar backscatter torainwater storage was larger for deciduous forests than for coniferous forests. Thisdifference was confirmed by observations of rain induced C-band backscatter increaseof nearby deciduous and coniferous stands (Rignot et al., 1994; Proisy et al., 2000). Thesimulated backscatter increased maximal with 4 dB for a poplar forest. This value wasregarded as an upper limit, due to three model assumptions: (i) all storage was allo-cated on the surface of leaves or needles, and none on the surface of branches andtrunks; (ii) the canopy was closed, but in reality part of the backscatter arose directlyfrom the soil through gaps in the canopy, and this direct soil backscatter dampened thesensitivity to storage; (iii) storage was vertical homogeneous distributed over the can-opy. However, a lower storage on the upper leaves, e.g. due to drying or wind agita-tion, would reduce the backscatter sensitivity, because these dry leaves in the uppercanopy attenuated the backscatter from sheltered, wetter leaves in the lower canopy.

In the second set of simulations, the relation between the sensitivity of radarbackscatter to forest wetness and the radar type was investigated (Chapter 4).


The improvement of this set of simulations compared to the previous simulations wasthat a realistic waterfilm thickness on the surface of leaves, branches and stems wasapplied. The modelled backscatter of a canopy was validated on ground-based radarmeasurements of the ash. These simulations indicated that radar backscatter washardly influenced by storage on surface of branches and stems, because wet branchesand stems retained only a small volume of rainwater relative to their own volume.Further simulations focused on common deciduous trees of temperate forests, as 2poplars species, beech, and ash. Variations in moisture content in the underlying soilwere also taken into account. The most suitable radar configuration for monitoringrainwater storage was co-polarised C- or X-band radar, because simulated backscatterfor these radar types combined a strong sensitivity to rainwater storage with a low sen-sitivity to tree species and soil moisture content. The simulated backscatter sensitivityof these radar types was 2-3 dB. The incidence angle of the radar appeared to be ofminor importance. Simulated cross-polarised backscatter appeared to be uncertain,probably because of the model limitations. Again, simulated backscatter sensitivitywas expected to be an overestimation of the backscatter sensitivity to rainwater stor-age, because direct backscatter from the soil through gaps in the canopy was not takeninto account.

The third set of simulations was executed to estimate the amount of rainwater stor-age from ERS-SAR observations, after correcting for the contribution of soil back-scatter (Chapter 5). The improvements of these simulations compared to the previoussimulations were: (i) simulations accounted for direct backscatter from the soilthrough gaps in the canopy; and (ii) a realistic value of the waterfilm thickness on thesurface of leaves was applied. The simulated forest was a poplar stand. The modelledsensitivity to rainwater storage in a fully developed canopy in summer (LAI = 4) var-ied between 1.4 dB above a water saturated soil till 1.8 dB above a dry soil. Whenfewer leaves were present, the backscatter sensitivity to soil moisture contentincreased and the sensitivity to canopy water storage decreased. The simulated sensi-tivity agreed with the observations.

The radar backscatter sensitivity to rainwater storage depended strongly on foresttype and used radar equipment. Co-polarised C- or X-band radar was the best choicefor monitoring of rainwater storage in deciduous forest, with a backscatter sensitivityto rainwater storage of 2-3 dB. Radar backscatter arising from the soil through gaps inthe canopy might reduce this radar backscatter sensitivity with 1 dB. For coniferousspecies, co-polarised C-band sensitivity to rainwater storage was low, and thereforenot a good choice. The question which radar type is most suitable for monitoringrainwater storage in coniferous forest remained unanswered.

6.4 Minimising the Influence of Soil Backscatter

Two approaches to reduce the influence of the soil backscatter to total backscatterhave been investigated. Firstly, the question whether a clever choice of the radar con-figuration could minimise the influence of soil backscatter was investigated (Chapter 4).


The method was a sensitivity analysis with the radar backscatter model. The influenceof soil backscatter was lowest for C- and X-band radar. The ERS-SAR, operating at C-band, was therefore expected to be suitable for retrieval of rainwater storage. Theobserved sensitivity of the ERS-SAR to changes in soil moisture content was less than1 dB for the 14-year-old poplar forest. This sensitivity was a fraction smaller than thesensitivity to rainwater storage. Therefore, the second approach focused on a methodto correct for soil backscatter (Chapter 5). The correction for soil backscatter was per-formed with the radar backscatter model, which needed soil moisture content asexplicit input variable. Soil moisture content was determined in two ways: (i) meas-ured in-situ; (ii) retrieved from another radar observation. Retrieval of soil moisturecontent from another radar observation was only possible when two conditions werefulfilled. The canopy had to be dry, which implicates that at least precipitation datashould be available, and the time span between the radar observations of the dry andwet canopy had to be short, e.g. 24 hours, to justify the assumption that soil moisturecontent was stable between the image acquisitions.

C- or X-band radar observations of deciduous forests were the least affected byradar backscatter arising from the soil. However, for C-band observations of a 14-year-old poplar forest, it was still needed to correct for the contribution of radar back-scatter from the soil, before estimating the amount of rainwater storage in the canopyof that forest. A correction for soil backscatter was possible when data on soil mois-ture content were available.

6.5 The Potential of Radar for Monitoring Rainwater Storage in Forest Canopy

The potentials of all types of radar for monitoring the amount of rainwater storage arediscussed in this final section. A feasible radar application is a compromise betweenthe capabilities and limitations of radar remote sensing. Therefore, the capabilities andlimitations are evaluated first.

The amount of rainwater storage in the canopy of deciduous forests could be esti-mated from observed radar backscatter, when sufficient additional data were available.This suggestion arose from the agreement between the radar-estimated waterfilmthickness on one side of a leaf in the upper canopy during rain, 0.09 mm for ash trees,and 0.11 mm for poplar trees. The radar retrieved waterfilm thickness on leaves waswithin the range of the laboratory observations of agitated leaves by Herwitz (1985):0.06 mm for Castanospermum australe, 0.07 mm for Aleuritis moluccana, 0.11 forToona australis, 0.11 for Dysoxylum pettigrewianum and 0.12 mm for Argyrodendronperalatum. For a species-specific comparison, the waterfilm thickness on ash leaveswas determined in the laboratory. This value, 0.06 mm, was indicative for the water-film thickness on the surface of leaves after rain has ceased and drainage stopped. Theobserved difference in storage during and after rain agreed with measurements of Grahand Wilson (Figure 1.1, 1944), Aston (1979) and Monson et al., (1992). The waterfilmthickness on poplar leaves was extrapolated to the amount of rainwater storage in thewhole canopy, and compared with measurements performed in-situ. The difference


between radar retrieved and in-situ measured storage was within the uncertainty limitsfor 4 of 6 analysed events. The 2 anomalous events could be explained. For the firstevent, it was attributed to the neglect of dense undergrowth in spring. For the secondevent, it was attributed to the occurrence of a small shower that only wetted the leavesin the upper canopy. Therefore, radar seems to be able to provide quantitativemeasurements of the amount of rainwater storage on leaves. However, four majorlimitations for retrieval of the amount of rainwater storage from satellite radar datawere identified. These limitations will be discussed in the following paragraphs.

The first limitation was the measurement accuracy of radar measurements. Thisaccuracy depended upon two processes. A stochastic variation in the receivedbackscatter was caused by interference of radiowaves with a different phase.Averaging over independent radar measurements reduced this variation. Spatialaveraging over 0.25 km2 appeared to be sufficient for ERS-SAR data (Chapter 5). Forthe ground-based radar, it was also recommended to average over spatiallyindependent samples, because temporal averaging over samples from one locationperformed poorly (Chapter 3). The second process that determined the accuracy ofradar measurements was the electronic stability of the radar. The stability of satelliteradar and that of the ground-based radar was in the order of 0.5 dB. The sensitivity ofradar backscatter to rainwater storage was only a few times larger. Therefore, themeasurement accuracy influenced the accuracy of the retrieved storage.

The second limitation for retrieval of the amount of rainwater storage from satelliteradar data was that the radar backscatter model was data demanding. This model wasessential, because it related radar backscatter to the waterfilm thickness on the surfaceof leaves, and corrected for the contribution of the backscatter from the soil. Itrequired data on tree structure, LAI, soil type, roughness and moisture content. Itmight be possible to collect some of these data from other sources, without extensivefieldwork, but such an approach would only be feasible for relatively homogeneoustypes of vegetation.

The third limitation was the poor temporal coverage of satellite radar. In the idealsituation, the amount of rainwater storage is retrieved from the deviation of the radarbackscatter from a ‘baseline’ that contains only radar observations from dry days. Thetime span between acquisition made by the ERS-1 satellite, approximately 17 days,was too long to provide a reliable ‘baseline’ of observations from dry days (Chapter5). Two acquisitions in 24 hour by the ERS-1 and ERS-2 satellites in tandem missionperformed better (Chapter 2, 5). However, when both observations were made from apartially wet canopy, both observations became unsuitable (Chapter 2).

The forth limitation was that satellite radar was mainly sensitive to rainwater stor-age on the surface of leaves in the upper canopy, because these leaves attenuated thesignal from the lower canopy. The upscaling of the waterfilm thickness on leaves inthe upper canopy to the amount of rainwater storage in the whole canopy appeared tobe difficult, because of uncertainties in the partitioning of storage over leaves andbranches and stems.

Due to these limitations, it was only possible to monitor the amount of rainwaterstorage in the canopy of well-known homogeneous forests. Therefore, it is not


recommended to monitor the amount of rainwater storage over areas withheterogeneous vegetation, and large-scale interception models should not be validatedwith satellite radar observations. However, radar has other qualities that might beuseful for future applications.

A quality of radar sensing by satellite was the orientation on the upper canopy of aforest. Atmosphere-surface exchange processes depend strongly upon the wetness ofthe upper canopy, because of the close contact between the upper canopy and theatmosphere. A potential application of satellite radar is demonstrated on the basis ofFigure 1.4. The area behind the rainstorms with increased backscatter wasapproximately 6 km wide. This increased backscatter could be attributed to rainwaterstorage in the upper canopy. When the rainstorms moved with a speed of 5 m.s-1, theupper canopy of the forest would remain wet for 6 km / 5 m.s-1 = 20 minutes after therainstorm. During this period, transpiration rates were reduced, and the depositionrates of hydrophilic gasses were increased. In this example, the radar observationswere used qualitatively, without retrieving the waterfilm thickness on leaves from theobserved backscatter. Using the proposed approach, a stochastic distribution ofwetness might be assessed on a regional scale, on condition of sufficient temporalcoverage of satellite observations, and the availability of precipitation and wind data.This stochastic distribution of wetness can be used to assess spatial variability inatmosphere-surface exchange processes.

It was demonstrated that the amount of water storage on deciduous leaves could beestimated for single trees or homogeneous forest plots. This capacity of radar might beuseful in answering questions regarding the interception process. In this respect, espe-cially ground-based radar is suitable, because it can continuously monitor wetness. Forexample, the partitioning of water storage on the surface of leaves, branches and stemscan be studied. Total storage in a small tree can be measured by the weight increase.The difference between the measurements should be an estimate of the rainwater stor-age on the surface of branches and stems. In this way, the wetting and drying of vari-ous parts of the canopy might be studied. This approach could be extended to anumber of trees, to compare the differences in wetting and drying between species.

A most promising application combines the excellent qualities of satellite radar andground-based radar. A ground-based radar that is placed at a high position should beable to monitor the temporal and spatial distribution of wetness in the upper canopy ofthe surrounding forests. This application could have a high potential in hilly ormountainous landscape, because the spatial distribution of rainfall and evaporationdepend strongly upon relief. For example, variation in winter rainfall in Colorado wasexplained for 30% by altitude, and for 88% by a combination of altitude, slope,exposure and orientation of the location (Spreen, 1947). Evaporation rate dependsupon available energy and aerodynamic roughness. These parameters show strongspatial variability in a hilly landscape (e.g. Sellers et al., 1997b, De Jong et al., 1999).Vegetation structure and leaf wettability may adapt to the process of wetting anddrying. Plants, growing in a moist location, had a lower wettability than plantsgrowing in a dryer microhabitat (Brewer and Smith, 1997). Therefore, the temporaland spatial distribution of wetness in rough terrain might vary strongly over distances


that can be monitored with one radar. The wetting and drying of forest patches on ahill slope or in a valley could be monitored with radar placed at a hilltop. Such anexperiment might additionally provide information on the spatial and temporaldistribution of dew and intercepted cloud water.

The general conclusion of the research described in this thesis is that radar canprovide qualitative information on the wetness of leaves in the canopy of deciduoustrees. The observational scales may range from a single tree to a region, dependingupon the observational platform. Quantitative measurements of the amount ofrainwater storage on leaves are only feasible when detailed ground information isavailable. Therefore, satellite radar is not suitable for estimating the amount ofrainwater storage in the canopy of large heterogeneous forests. Satellite radar might beuseful for estimating atmosphere-surface exchange processes that depend upon thewetness of the land surface, because the downward-looking radar is sensitive towetness of leaves in the upper canopy. A most promising application of radar is tomonitor the temporal and spatial distribution of forest wetness in rough terrain byscanning the surrounding landscape from a high location, e.g. the top of a hill.

95

Literature

Ahern, F. J., D. J. Leckie, and J. A. Drieman, 1993: “Seasonal changes in relative C-band backscatter ofnorthern forest cover types”, IEEE Trans. Geosci. Remote Sens. 31: 668-680.

Alexandratos, N., 1999: “World food and agriculture: outlook for the medium and longer term”, Proc.Natl. Acad. Sci. USA 96: 5908-1914.

Anderson, F. O., K. H. Feger, R. F. Hüttl, N. Krãuchi, L. Mattsson, O. Sallnãs, and K. Sjöberg, 2000: “Forest ecosystem research-Priorities for Europe”, For. Ecol. Management 132: 111-119.

Anonymous, 1997: LKN, Landscape ecological atlas of the Netherlands, University of Wageningen, theNetherlands, CD-ROM.

Aobal, J. R., M. S. Jiménez, D. Morales, and J. M. Hernández, 1999: “Rainfall interception in laurelforest in the Canary Island”, Agric. For. Meteorol. 97: 73-86.

Arnell, N. W., 1999: “Climate change and global water resources”, Glob. Environ. Change 9: 31-49.Aston, A. R., 1979: “Rainfall interception by eight small trees”, J. Hydrol. 42: 383-396.Ataroff, M., and F. Rada, 2000: “Deforestation impact on water dynamics in a Venezualan Andean cloud

forest”, Ambio 29: 440-444.Bengtsson, L., 1997: “A numerical simulation of antropogenic climate change”, Ambio 26: 58-65.Bergen, K. M., Dobson, M. C., Pierce, L. E., and Ulaby, F. T., 1997: “Effects of within season dielectric

variations on terrain classification using Sir-C/X-Sar,” in: Proc. IGARSS 1997, Singapore, pp 1072-1074.

Bernard, R., and D. Vidal-Madjar, 1989: “C-band radar cross section of the Guyana Rain Forest, possibleuse as a reference target for spaceborne radars”, Remote Sens. Environ. 27: 25-36.

Blöschl, G., and M. Sivapalan, 1995: “Scale issues in hydrological modelling: a review,” in (J. D. Kalmaet al., Eds.): Scale issues in hydrological modelling, John Wiley & Sons, Chichester, UK, Chap. 1.

Borgeaud, M. and N. Floury, 2000: “On the retrieval of soil moisture of bare soils with ERS SAR data,”in: Proc. IGARSS 2000, Honolulu, Hawaii, pp. 1687-1689.

Bosch, J. M., and J. D. Hewlett, 1982: “A review of catchment experiments to determine the effect ofvegetation changes on water yield and evapotranspiration”, J. Hydrol. 55: 3-23.

Bouten, W., P. J. F. Swart, and E. Water, 1991: “Microwave transmission, a new tool in foresthydrological research”, J. Hydrol. 124: 119-130.

Bouten, W., G. Heuvelink, S. Huisman, and R. Wilms, 1995: “Interceptieverdamping in Nederlandsebossen” Landscape and Environmental Research Group, University of Amsterdam, The Netherlands.

Bouten, W., M. G. Schaap, J. Aerts, and A. W. M. Vermetten, 1996: “Monitoring and modelling canopywater storage amounts in support of atmospheric deposition studies”, J. Hydrol. 181: 305-321.

Brewer, C. A., and W. K. Smith, 1997: “Pattern of leaf wetness for montane and subalpine plants”, PlantCell Environ. 20: 1-11.

Brooks, C. E. P., 1928: “The influence of forests on rainfall and run-off”, Quart. J. R.. Meteorol. Soc. 54:1-17.

Bruijnzeel, L. A., and K. F Wiersum, 1987: “Rainfall interception by a young Acacia Auriculiformis (A.cunn) plantation fores in West Java: application of Gash’s analysitcal model”, Hydrol. Process. 1:309-319.

Butler, D. R., 1986: “Evaporation from raindrops on leaves in a cereal canopy: a simulation model”,Boundary-Layer Meteorol. 36: 39-51.

Butler, D. R., 1996: “The presence of water on leaf surfaces and its importance for microbes and insects,”in (G. Kertiens, Ed.): Plant Cuticles, BIOS scientific publishers, Oxford, p 267-282.

Calder, I. R., and Wright, 1986: “Gamma ray attenuation studies of interception from sitka spruce: someevidence for an additional transport mechanism”, Water Resources Res. 22: 409-417.

Calder, I. R., 1986: “A stochastic model of rainfall interception”, J. Hydrol. 89: 65-71.Calder, I. R., 1996: “Dependence of rainfall interception on drop size: 1 development of the two layer

stochastic model”, J. Hydrol. 185: 363-378.Calder, I. R., R. L. Hall, P. T. W. Rosier, H. G. Bastable, and K. T. Prasanna, 1996: “Dependence of

rainfall interception on drop size: 2 Experimental determination of the wetting functions and two-layer stochastic model parameters for five tropical tree species”, J. Hydrol. 185: 379-388.


Calder, I. R., and R. L. Hall, 1997: “Comment on: ‘interception of tropical rain forest: performance of acanopy water balance model’ by V. G. Jetten, Hydrological Processes, Vol. 10, 671-685 (1996)”,Hydrol. Process. 11: 225-226.

Calvet, J. C., J. P. Wigneron, E. Mougin, Y. H. Kerr, and J. L. S. Brito, 1994: “Plant water content andtemprature of the Amazon forest from satellite microwave radiometry”, IEEE Trans. Geosci. RemoteSens. 32: 397-408.

Cape, J. N., 1983: “Contact angle of water droplets on needles of scots pine (Pinus sylvestris) growing inpolluted atmospheres”, New Phytol. 93: 293-299.

Carlyle-Moses, D. E., and A. G. Price, 1999: “An evaluation of the Gash interception model in a northernhardwood stand”, J. Hydrol. 214: 103-110.

Cassie, A. B. D., and S. Baxter, 1945: “Large contact angles of plant and animal surfaces”, Nature, 155:21-22.

Choudhury, B. J., N. E. DiGirolamo, J. Susskind, W. L. Darnell, S. K. Gupta, and G. Asrar, 1998: “Abiophysical process-based estimate of global land surface evaporation using satellite and ancillarydata II. Regional and global patterns of seasonal and annual variations”, J. Hydrol. 205: 186-204.

Cochard, H., M. Peiffer, K. Le Gall, and A. Garnier, 1997: “Developmental control of xylem hydraulicresistances and vulnerability to embolism in Fraxinus excelsior L.: impacts on water relations”, J.Exp. Botany 48: 655-663.

Crockford, R. H., and D. P. Richardson, 2000: “Partitioning of rainfall into throughfall, stemflow andinterception: effect of forest type, ground cover and climate”, Hydrol. Process. 14: 2903-2920.

Davidson, M., T. Le Toan, F. Mattia, T. Manninen, P. Borderies, I. Chenerie, and M. Borgeaud, 1998: “Avalidation of multi-scale roughness description for the modelling of radar backscatter from bare soilsurfaces,” in (M. Borgeaud and T. D. Guyenne, Eds.): Second international workshop on theretrieval of bio-and geo-physical parameters from Sar data for land applications, ESA/ESTEC,Noordwijk, pp. 395-400.

De Jong, J. J. M., and W. Klaassen, 1997: ”Simulated dry deposition of nitric acid near forest edges”,Atmospheric Environ. 31: 3681-3691.

De Jong, J. J. M., A. C. de Vries, and W. Klaassen, 1999: “Influence of obstacles on the aerodynamicroughness of the Netherlands”, Boundary-Layer Meteorol. 91: 51-64.

De Jong, J. J. M., W. Klaassen, and A. Ballast, 2000a: “Rain storage detected with ERS tandem missionSAR”, Remote Sens. Environ. 72: 170-180.

De Jong, J. J. M., H. W. de Groot, W. Klaassen, and P. J. C. Kuiper, 2000b: “Radar backscatter changefrom an ash in relation to its hydrological properties,” in: Proc. IGARSS 2000, Honolulu, Hawaii, pp.2927-2929.

Deardorff, J. W., 1978: “Efficient prediction of ground surface temperature and moisture with inclusionof a layer of vegetation”, J. Geophysic. Res. 83: 1889-1903.

Dekker, S., 2000: Modelling and monitoring forest evapotranspiration, PhD thesis, University ofAmsterdam, 120 pp.

Dickinson, R. E., 1984: “Modelling evapotranspiration for three dimensional global climate models,” inClimate processes and climate sensitivity, Geoph. Monogr. No 29, American Geoph. Union, pp 58-72.

Dobson, M. C., F. W. Ulaby, L. E. Pierce, T. L. Sharik, K. M. Bergen, J. Kellndorfer, J. R. Kendra, E. Li,Y. C. Lin, A. Nashashibi, K. Sarabandi, and P. Siqueira, 1995: “Estimation of forest biophysicalcharacteristics in Northern Michigan with SIR-C/X-SAR”, IEEE Trans. Geosci. Remote Sens. 33:877-895.

Dobson, M. C., L. Pierce, K. McDonald, and T. Sharik, 1991: “Seasonal change in radar backscatter frommixed conifer and hardwood forest in Northern Michigan,” in: Proc. IGARSS 91, Espoo, Finland, pp.1121-1124.

Dolman, A. J., 1987: “ Summer and winter rainfall interception in an oak forest. Predictions with ananalytical and a numerical simulation model”, J. Hydrol. 90: 1-9.

Dolman, A.J., and Gregory, D., 1992: “The parameterization of rainfall interception in GCMs”, Quart. J.R. Meteorol. Soc. 188: 445-467.

Dolman, A. J., E. J. Moors, J. A. Elbers, and W. Snijders, 1998: “Evaporation and surface conductance ofthree temperate forests in the Netherlands”, Ann. Sci. For. 55: 255-270.

Literature 97

Dunin, F. X., E. M. O’Loughlin, and W. Reyenga, 1988: “Interception loss from eucalypt forest:lysimeter determination of hourly rates for long term evaluation”, Hydrol. Process. 2: 315-329.

Dybing, C. D., and H. B. Currier, 1961: “Foliar penetration by chemical”, Plant Physiol. 36: 169-174.Dykes, A. P., 1997: “Rainfall interception from a lowland tropical rainforest in Brunei”, J. Hydrol. 200:

260-279.Easterling, D. R., G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and L. O. Mearns, 2000:

“Climate extremes: observations, modelling, and impact”, Science 289: 2068-2074.Ebermayer, E., 1873: Die physikalischen einwirking des walden auf luft und boden und seine

klimatologische und hygienische bedeutung, Wiegandt, Hempel & Parey, Berlin.Elbers, J. A., A. J. Dolman, E. J. Moors, and W. Snijder, 1996: Hydrologie en waterhuishouding van

bosgebieden in Nederland, fase 2: meetopzet en eerste resultaten, Staring centrum, Wageningen, Rep.333.2.

El-Rayes, M. A., and F. T. Ulaby, 1987: “Microwave dielectric spectrum of vegetation-part 1:Experimental observations”, IEEE Trans. Geosci. Remote Sens. 25: 541-549.

Eltahir, E. A. B., and R. L. Bras, 1993: “A description of rainfall interception over large areas”, J. Clim.,6: 1002-1008.

FAO, 2000: State of the world’s forests 1999, FAO, Rome, 200 pp.Field, T. S., M. A. Zwieniecki, M. J. Donoghue, and N. M. Holbrook, 1998: “Stomatal plugs of Drimys

winteri (Winteraceae) protect leaves from mist but not drought”, Proc. Natl. Acad. Sci. USA 95:14256-14259.

Fogg, G. E., 1944: “Diurnal variation in a physical property of leaf cuticle”, Nature 154: 515.Fogg, G. E., 1947: “Quantitative studies of the wetting of leaves by water”, Proc. R. Soc. London (B) 134:

503-522.Fung, A. K., Z. Li, and K. S. Chen, 1992: “Backscattering from a randomly rough dielectric surface”,

IEEE Trans. Geosci. Remote Sens. 30: 799-784.Gash, J. H. C., 1979: “An analytical model of rainfall interception by forests”, Quart. J. R. Meteorol. Soc.

105: 43-55.Gash, J. H. C., C. R. Lloyd, and G. Lachaud, 1995: “Estimating sparse forest rainfall interception with an

analytical model”, J. Hydrol. 170: 79-86.Gash, J. H. C., F. Valente, and J. S. David, 1999: “Estimates and measurements of evaporation from wet

sparse pine forest in Portugal”, Agric. For. Meteorol. 94: 149-158.Gates, W. L., A. Henderson-Sellers, G. J. Boer, C. K. Folland, A. Kitoh, B. J. McAvaney, F. Semazzi, N.

Smith, A. J. Weaver, and Q. C. Zeng, 1996: “Climate Models-Evaluation,” in (IntergovernmentalPanel on Climate Change, Eds.): Climate Change 1995, The science of climate change, CambridgeUniversity Press, Cambridge, pp. 279-284.

Ghan, S. J., J. C. Liljegren, W. J. Shaw, J. H. Hubbe, and J. C. Doran, 1997: “Influence of subgridvariability on surface hydrology”, J. Clim. 10: 3157-3166.

Gordon, A. M., C. Chourmouzis, and A. G. Gordon, 2000: “Nutrient inputs in litterfall and rainwaterfluxes in 27-year old red, black and white spruce plantations in Central Ontario, Canada”, For. Ecol.Management 138: 65-78.

Grah, R. A., and C. C. Wilson, 1944: “Some components of rainfall interception”, J. For. 42: 890-898.Guentzel, J. L., W. M. Landing, G. A. Gill, and C. D. Pollman, 1998: “Mercury and major ions in rainfall,

throughfall and foliage from the Florida Everglades”, Sci. Total Environ. 213: 43-51.Hallikainen, H. T., F. T. Ulaby, M. C. Dobson, M. A. El-Rayes, and L. K. Wu, 1985: “Microwave

dielectric behaviour of wet soils”, IEEE Trans. Geosci. Remote Sens. 23: 25-34.Hancock, N. H., and J. M. Crowther, 1979: “A technique for the direct measurement of water storage on a

forest canopy”, J. Hydrol. 41: 105-122.Helvey, J. D., and J. H. Patric, 1965: “Canopy and litter interception of rainfall by hardwoods of eastern

United States”, Water Resour. Res. 1: 193-206.Herwitz, S. R. 1985: “Interception storage capacities of tropical rain forest trees”, J. Hydrol. 77: 237-252.Herwitz, S. R., and R. E. Slye, 1995: “Three dimensional modeling of canopy tree interception of wind

driven rainfall”, J. Hydrol. 168: 205-226.Hess, L. L., J. M. Melack, and D. S. Simonette, 1990: “Radar detection of flooding beneath the forest

canopy: a review”, Int. J. Remote Sens. 5: 1313-1325.


Hinckley, T. M., J. P. Lassoie, and S. W. Running, 1978: Temporal and spatial variations in the waterstatus of forest trees, Forest Science Monograph 20, Society of American Foresters, WashingtonD.C., 70 pp.

Hobbs, S. E., W. Ang, C. Seynat, 1998: “Wind and rain effects on SAR backscatter from crops,” in (M.Borgeaud and T. D. Guyenne, Eds.): Second international workshop on the retrieval of bio-and geo-physical parameters from Sar data for land applications, ESA/ESTEC, Noordwijk, pp. 185-189.

Hoekman, D. H., M. van der Linden, and J. J. van der Sanden, 1995: “Application of ERS-1 in forestmanagement,” in (G. J. A. Nieuwenhuis, and W. W. L. van Rooij, Eds.): Application of ERS-1 SARdata in agriculture and forestry, rep. 94-34, Delft, The Netherlands: BCRS, pp. 105-121.

Hope, J. R. 1966: “Path of heavy rainfall photographed from space”, Bull. Am. Meteorol. Soc. 47: 371-373.

Hörmann, G., A. Branding, T. Clemen, M. Herbst, A. Hinrichs, and F. Thamm, 1996: “Calculation andsimulation of wind controlled canopy interception of a beach forest in Northern Germany”, Agric.For. Meteorol. 79: 131-148.

Horton, R. E., 1919: “Rainfall interception”, Montly Weather Rev. 47: 603-623.Huneycutt, B. and J. Zuzek, 2000: “Frequency use and needs of spaceborne active sensors,” in: Proc.

IGARSS 2000, Honolulu, Hawaii, pp. 2457-2463.Hutchings, N. J. M., R. Miline, and J. M. Crowther, 1988: “Canopy storage capacity and its vertical

distribution in a Sitka spruce canopy”, J. Hydrol. 104: 161-171.Ishibashi, M. and I. Terashima, 1995: “Effects of continuous leaf wetness on photosynthesis: adverse

aspects of rainfall”, Plant, Cell, Environ. 18: 431-438.Jackson, N. A., 2000: “Measured and modelled rainfall intercepttion loss from an agroforestry system in

Kenya”, Agric. For. Meteorol. 100: 323-336.Jones, J. A. A., 1999: “Climate change and sustainable water resources: placing the threat of global

warming in perspective”, Hydrol. Sci. J. 44: 541-557.Kändler, G., 1986: “Die ermittlung von bestandesparametern als eingangsgrössen für

interzeptionsmodelle mit hilfe aerophotogrammetrischer verfahren”, Mitteilungen des Frostl.Versuchs- und Forsungsanstalt Baden-Würtenberg, Freiburg, Germany, no. 127.

Karam, M. A., A. K. Fung, R. H. Lang, and N. S. Chauhan, 1992: “A Microwave scattering model forlayered vegetation”, IEEE Trans. Geosci. Remote Sens. 30: 767-783.

Karam, M. A., F. Amar, A. K. Fung, E. Mougin, A. Lopes, D. M. LeVine, and A. Beaudoin, 1995: “Amicrowave polarimetric scattering model for forest canopies based on vector radiative transfertheory”, Remote Sens. Environ. 53: 16-30.

Kasischke, E. S., J. Melack, and M. C. Dobson, 1997: “The use of imaging radars for ecologicalapplications -a review”, Remote Sens. Environ. 59: 141-156.

Kerstiens, G., 1996: “Cuticular water permeability and its physiological significance”, J. Exp. Bot. 47:1813-1832.

Klaassen, W., M. van der Linden, and A. Ballast, 1997: “SAR sensing of vegetationwetness: the firstresults,” in: Third ERS symposium on space at the service of our environment, Florence, Italy, pp 103-106.

Klaassen, W., F. Bosveld, and E. de Water, 1998: “On water storage and evaporation as constituents ofrainfall interception”, J. Hydrol. 212: 36-50.

Klaassen, W., 2001: “Evaporation from rain-wetted forest in relation to canopy wetness, canopy cover,and net radiation”, Wat. Res. Research, in press.

Klugman, D., K. Heinsohn, and H.-J. Kirtzel, 1996: “A low cost 24 GHz FM-CW Doppler radar rainprofiler”, Contrib. Atmos. Physics 69: 247-253.

Lankreijer, H. J. M., M. J. Hendriks, and W. Klaassen, 1993: “A comparison of models simulatingrainfall interception of forests”, Agric. For. Meteorol. 64: 187-199.

Lankreijer, H., A. Lundberg, A. Grelle, A. Lindroth, and J. Seibert, 1999: “Evaporation and storage ofintercepted rain analysed by comparing two models applied to a boreal forest”, Agric. For. Meteorol.98-99: 595-604.

Larsson, S., 1981: “Influence of intercepted water on transpiration and evaporation of Salix”, Agric.Meteorol. 23: 331-338.

Literature 99

Laur, H., P. Bally, P. Meadows, J. Sanchez, B. Schaettler, and E. Lopinto, 1997: Derivation of thebackscattering coefficient σ0 in ESA ERS SAR PRI products, ESA, Noordwijk, The Netherlands,Document ES-TN-RS-PM-HL09, Issue 2, Rev. 4.

Leonard, R. E., 1967: “Mathematical theory of interception,” in (E. W. Sopper and H. W. Lull, Eds.):International symposium on forest hydology, Pergamon press, Pennsylvania.

Leyton L., E. R. C. Reynold, and F. B. Thompson, 1967: “Rainfall interception in forests and moorland,”in (E. W. Sopper and H. W. Lull, Eds.): International symposium on forest hydology, Pergamon press,Pensylvania.

Lichtenegger, J., 1996: “ERS-1 SAR images-mirror of thunderstorms,” Earth Observ. Quart. 53: 7-9.Linacre, E. and B. Geerts, 1997: Climates & weather explained, Routledge, London, New York, p 98.Liu, S., 1997: “A new model for the prediction of rainfall interception in forest canopies”, Ecol.

Modelling 99: 151-159.Liu, S., 1998: “Estimation of rainfall storage capacity in the canopies of cypress wetlands and slash pine

uplands in North-Central Florida”, J. Hydrol. 207: 32-41.Llorens, P., R. Poch, J. Latron, and F. Gallart, 1997: “Rainfall interception by a Pinus sylvestris forest

patch overgrown in a Mediterranean mountainous abandoned area. I Monitoring design and resultsdown to the event scale”, J. Hydrol. 199: 331-345.

Llorens, P., and F. Gallart, 2000: “A simplified method for forest water storage capacity measurements”,J. Hydrol. 240: 131-144.

Lundberg, A., M. Erikson, S. Halldin, E. Kellner, and J. Seibert, 1997: “New approach to themeasurement of interception evaporation”, J. Atmos. Ocean. Tech. 14: 1023-1035.

Madeira, A. C., T. J. Gillespie, and C. L. Duke, 2001: “Effect of wetness on turfgrass canopyreflectance”, Agric. For. Meteorol. 107: 117-130.

Marloth, R., 1903: “Results of experiments on the table montain for ascertaining the amount of moisturedeposited from the South-East clouds”, Trans. S. Afr. Philos. Soc. 14: 403-409.

Massman, J. W., 1983: “The derivation and validation of a new model for the interception of rainfall byforests”, Agric. Meteorol. 28: 261-286.

McDonald, K. C., M. C. Dobson, and F. T. Ulaby, 1991 “Modeling multifrequency diurnal backscatterfrom a walnut orchard”, IEEE Trans. Geosci. Remote Sens. 29: 852-863.

Mizutani, K., K. Yamanoi, T. Ikeda, and T. Watanabe, 1997: “Applicability of the eddy correlationmethod to measure sensibile heat transfer to forest under rainfall conditions”, Agric. For. Meteorol.86: 193-203.

Moghaddam, M. and S. S. Saatchi, 1999: “Monitoring tree moisture using an estimation algorithmapplied to SAR data from Boreas”, IEEE Trans. Geosci. Remote Sens. 37: 901-916.

Moghaddam, M. and S. S. Saatchi, 2000: “Estimation of vegetation variables using AIRSAR datacontaining multiple scattering mechanisms,” in: Proc. IGARSS 2000, pp. 1408-1410.

Monson, R. K., M. C. Grant, C. H. Jaeger, and A. N. Schoettle, 1992: “Morphological issues for theretention of precipitation in the crown of alpine plants”, Environ. Exp. Bot. 32: 319-327.

Monteith, J. L., 1965: “Evaporation and environment”, Symp. Soc. Expl. Biology 19: 205-234.Mulder, J. P. M., 1985: “Simulating interception loss using standard meteorological data,” in (B. A.

Hutchison and B. B. Hicks, Eds.): The forest atmosphere interaction, Reidel, pp 177-196.Munich Re, 1999: Press release millenium review, 20 December 1999, Münchener Rückversicherungs

Gesellschaft, Munich.Murakami, S., Y. Tsuboyama, T. Shimizu, M. Fujieda, and S. Noguchi, 2000: “Variations of

evaportranspiration with stand age and climate in a small Japanese forested catchment”, J. Hydrol.227: 114-127.

Noilhan, J., and S. Planton, 1989: “A simple parameterisation of land surface processes formeteorological models”, Monthly Weather Rev. 117: 537-549.

Noilhan, J., P. Lacarrère, A. J. Dolman, and E. M. Blyth, 1997: “Defining area-average parameters inmeteorological models for land surfaces with mesoscale heterogeneity”, J. Hydrol. 190: 302-316.

Olszyczka, B., and J. M. Crowther, 1981: “The application of gamma ray attenuation to the determinationof canopy mass and canopy surface water storage”, J. Hydrol. 49: 355-368.

Penman, H. L., 1948: “Natural evaporation from open water, bare soil and grass”, Proc. R. Soc. London(A) 193: 120-145.


Peters, G., 1995: “The potential of a micro radar as rain gauge,” in: Meteohytec 21, Geneva 22-26 May,WMO/TD 672, pp. 65-68, WMO, Geneva.

Phillips, J., 1926: “Rainfall interception by plants”, Nature 118: 837-847.Pitman, A. P., A. Henderson-Sellers, and Z. L. Yang, 1990: “Sensitivity of regional climates to localized

precipitation in global models”, Nature 346: 734-737Proisy, C., E. Mougin, E. Dufrene, and V. le Dantec, 2000: “Monitoring seasonal changes of a mixed

temperate forest using ERS-SAR observations”, IEEE Trans. Geosci. Remote Sens. 38: 540-552.Pulliainen, J. T., K. Heiska, J. Hyyppä, and M. Hallikainen, 1994: “Backscattering properties of boreal

forest at the C- and X-bands”, IEEE Trans. Geosci. Remote Sens. 32: 1041-1050.Pulliainen, J. T., P. T. Mikkela, M. T. Hallikainen,, and J. P Ikonen, 1996: “Seasonal dynamics of C-

band backscatter of boreal forest with applications to biomass and soil moisture estimation”, IEEETrans. Geosci. Remote Sens. 34: 758-769.

Putuhena, W. M., and I. Cordery, 2000: “Some hydrological effects of changing forest cover fromeucalypts to Pinus radiata”, Agric. For. Meteorol. 100: 59-72.

Quesney, A., S. Le Hégarat-Mascle, O. Taconet, D. Vidal-Madjar, J. P. Wigneron, C. Loumagne, and M.Normand, 2000: “Estimation of watershed soil moisture index from ERS/SAR data”, Remote Sens.Environ. 72: 290-303.

Rignot, E., J. B. Way, K. McDonald, L. Viereck, C. Williams, P. Adams, C. Payne, W. Wood, and J. Shi,1994: “Monitoring of environmental conditions in Taiga Forest using ERS-1 SAR”, Remote Sens.Environ. 49: 145-154.

Roberts, J., 2000: “The influence of physical and physiological charesteristics of vegetation on theirhydrological response”, Hydrol. Process. 14: 2885-2901.

Robinson, D., 2001: ”δ15N as an integrator of the nitrogen cycle”, Trends Ecol. Evol. 16: 153-162.Röhrig, E., and B. Ulrich, 1991: Temperate Deciduous Forests Elsevier, Amsterdam.Rutter, A. J., 1963: “Studies in the water relations of pinus sylvestris in plantation condition, I:

measurements of rainfall and interception”, J. Ecol. 51: 191-203.Rutter, A. J., K. A. Kershaw, P. C. Robins, and A. J. Morton, 1971: “A predictive model of rainfall

interception in forests: I. Derivation of the model from observations in a plantation of Corsican pine”,Agric. Meteorol. 9: 367-384.

Saich, P., and M. Borgeaud, 2000: “Interpreting ERS SAR signatures of agricultural crops in Flevoland,1993-1996”, IEEE Trans. Geosci. Remote Sens. 38: 651-657.

Schellekens, J., F. N. Scatena, L. A. Bruijnzell, and A. J. Wickel, 1999: “Modelling rainfall interceptionby a lowland tropical rain forest in Northeastern Puerto Rico”, J. Hydrol. 225: 168-184.

Schönherr, J., and M. J. Bukovac, 1972: “Penetration of stomata by liquids”, Plant Physiol. 49: 813-819.Schowengerdt, R. A., 1983: Techniques for Image Processing and Classification in Remote Sensing,

Academic Press, New York.Schulze, R., 2000: “Transcending scales of space and time in impact studies of climate and climate

change on agrohydrological responses”, Agric. Ecosys. Environ. 82: 185-212.Sellers, P. J., Y.Mintz, Y. C. Sud, and A. Dalcher, 1986: “A simple biosphere model (SiB) for use within

general circulation models”, J. Atmos. Sci. 43: 505-531.Sellers, P. J., D. A. Randall, G. J. Collatz, J. A. Berry, C. B. Field, D. A. Dazlich, C. Zhang, G. D.

Colleleo, and L. Bounoua, 1996: “A revised land surface paramterization (SiB2) for atmosphericGCMs. Part I: Model formulation”, J. Clim. 9: 676-705.

Sellers, P. J., R. E. Dickinson, D. A. Randall, A. K. Betts, J. A. Berry, G. J. Collatz, A. S. Denning, H. A.Mooney, C. A. Nobre, N. Sato, C. B. Field, and A. Henderson-Sellers, 1997a: “Modeling theexchanges of energy, water, and carbon between continents and the atmosphere”, Science 275: 502-509.

Sellers, P. J., M. D. Heiser, F. G. Hall, S. B. Verma, R. L. Desjardins, P. M Schuepp, and J. I.MacPherson, 1997b: “The impact of using area-averaged land surface properties –topography,vegetation condition, soil wetness- in calculations of intermediate scale (approximately 10 km2)surface-atmosphere heat and moisture fluxes”, J. Hydrol. 190: 269-301.

Shiklomanov, I. A. (Ed.), 1997: Comprehensive assessment of the freshwater resources of the World :assessment of water resources and water availability in the World, UNSECO, Paris.

Literature 101

Shuttleworth, J. W., 1975: “Experimental evidence for the failure of the Penman-Monteith equation inpartially wet conditions”, Boundary-Layer Meteorol. 10: 91-94.

Shuttleworth, J. W., 1988: “Evaporation from Amazonian rain forest”, Proc. R. Soc. London (B) 233:321-346.

Spreen, W. C., 1947: “A determination of the effect of topography upon precipitation”, TransactionsAGU 2: 285-290.

Stewart, J. B., 1977: “Evaporation from the wet canopy of a pine forest”, Water Resour. Res. 13: 915-921.Stewart, J. B., E. T. Engman, R. A. Feddes, and Y. H. Kerr, 1998: “Scaling up in hydrology using remote

sensing: summary of a workshop”, Int. J. Remote Sens. 19: 181-194.Stogryn, A., 1971: “Equations for calculating the dielectric constant of saline water”, IEEE Trans

Microwave Theory Tech. 19: 733-736.Swank, W. T., J. M. Vose, and K. J. Elliott, 2001: “Long-term hydrologic and water quality responses

following commercial clearcutting of mixed hardwoods on a southern Appalachian catchment”, For.Ecol. Manag. 143: 163-178.

Tan, H. S., 1981: “Microwave measurements and modelling of the permittivity of tropical vegetationsamples”, Appl. Phys. 25: 351-355.

Ten Houte de Lange, S.M., 1977: Rapport van het Veluwe onderzoek, Centre of Agricultural Publicationsand Agricultural Documentation, Wageningen.

Theklehaimanot, Z., and P. G. Jarvis, 1991: “Direct measurement of evaporation of intercepted waterfrom forest canopies”, J. Appl. Ecol. 28: 603-618.

Tian, Y., Y. Zhang, Y. Knyazikhin, R. B. Myneni, J. M. Glassy, G. Dedieu, and S. W. Running, 2000:“Prototyping of MODIS LAI and FPAR algortihm with LASUR and LANDSAT data”, IEEE Trans.Geosci. Remote Sens. 38: 2387-2401.

Tilman, D., J. Fargione, B. Wolff, C. D’Antonio, A. Dobson, R. Howarth, D. Schindler, W. H.Schlesinger, D. Simberloff, and D. Swackhamer, 2001: “Forecasting agricultural driven globalenvironmental change”, Science 292: 281-284.

Ulaby, F. T., R. K. Moore, and A. K. Fung, 1981: Microwave remote sensing, active and passive, volumeI, II, and III, Artech House, Norwood.

Ulaby, F. T., and M. A. EL-Rayes, 1987: “Microwave dielectric spectrum of vegetation -Part II: dual-dispersion model”, IEEE Trans. Geosci. Remote Sens. 25: 550-557.

UNESCO, 2000: Report of the status of preparation of the IHP-VI draft plan, Unesco, Paris, 53 pp.Vertessy, R. A., F. G. R. Watson, and S. K. O’Sullivan, 2001: “Factors determining relations between

stand age and catchment water balance in mountain ash forests”, For. Ecol. Management 143: 13-26.Vissers, M. A. M., and J. J. van der Sanden, 1993: Groundtruth collection for the JPL-SAR and ERS-1

campaign in Flevoland and the Veluwe, BCRS, Delft.Voight, G. K., and M. J. Zwolinski, 1964: “Absorption of stemflow by bark of young red and white

pines”, For. Science 10: 277-282.Vörösmarty, C. J., P. Green, J. Salisbury, and R. B. Lammers, 2000: “Global water resources:

vulnerability from climate change and population growth”, Science 289: 284-288.Wang, Y., L. L. Hess, S. Filoso, and J. M. Melack, 1995: “Understanding the radar backscatter from

flooded and nonflooded Amazonian forests: results from canopy backscatter modeling”, RemoteSens. Environ. 54: 324-332.

Wang, G., and E. A. B. Eltahir, 2000: “Biosphere-atmosphere interactions over West Africa. I:Development and validation of coupled dynamic model”, Quart. J. R. Meteorol. Soc. 126: 1239-1260.

Watanabe, T., and K. Mizutani, 1996: “Model study on micrometeorological aspects of rainfallinterception over an evergreen broad-leaved forest”, Agric. For. Meteorol. 80: 195-214.

Way, J., J. Paris, E. Kasischke, C. Slaughter, L. Viereck, N. Christensen, M. C. Dobson, F. Ulaby, J.Richards, A. Milne, A. Sieber, F. J. Ahern, D. Simonett, R. Hoffer, M. Imhoff, and J. Weber, 1990:“The effect of changing environmental conditions on microwave signatures of forest ecosystems:preliminary results of the March 1988 Alaskan aircraft SAR experiment”, Int. J. Remote Sens. 11:1119-1144.

Weimann, A., M. von Schönermark, A. Schumann, P. Jörn, and R. Günther, 1998: “Soil moistureestimation with ERS-1 SAR data in the East-German less soil area”, Int. J. Remote Sens. 19: 237-243.


Wesely, M. L., and B. B. Hicks, 2000: “A review of the current status of knowledge on dry deposition”,Atmospheric Environ. 34: 2261-2282.

Wessels, J. H. G., 1997: “Hydrophobins; proteins that change the nature of the fungal surface”, Adv.Microb. Physiol. 38: 1-45.

Whitehead, P. G., and M. Robinson, 1993: “Experimental basin studies -an international historicalperspective of forest impacts”, J. Hydrol. 145: 217-230.

Wigneron, J. P., J. C. Calvet, and Y. Kerr, 1996: “Monitoring water interception by crop fields frompassive microwave observations”, Agric. For. Meteorol. 80: 177-194.

Wilm, H. G., and C. H. Niederhof, 1941: “Interception of rainfall by mature lodgepole pine”, Trans.Amer. Geophys Un. 22: 660-665.

Wilson, K. B, P. J. Hanson, P. J. Mulholland, D. B. Baldocchi, and S. D. Wullschleger, 2001: “Acomparison of methods for determining forest evapotranspiration and its components: sap-flow, soilwater budget, eddy covariance and catchment water balance”, Agric. For. Meteorol. 106: 153-168.

Xiao, Q., E. G. McPherson, S. L. Ustin, and M. E. Grismer, 2000: “A new appraoch to modeling treerainfall interception”, J. Geoph. Res.(D) 105: 29173-29188.

Zeng, N., J. W. Shuttleworth, and J. H. C. Gash, 2000: “Influence of temporal variability of rainfall oninterception loss. Part I. Point analysis”, J. Hydrol. 228: 228-241.

Zhang, Y, Y. Tian, Y. Knyazikhn, J. V. Martonchik, D. J. Diner, M. Leroy, and R. B. Myneni, 2000:“Prototyping of MISR LAI and FPAR Algorthm with POLDER data of Africa”, IEEE Trans. Geosci.Remote Sens. 38: 2402-2418.

Zink, M. and R. Bamler, 1995: “X-SAR radiometric calibration and data quality”, IEEE Trans. Geosci.Remote Sens. 33: 840-847.

Zzribi, M., V. Ciarletti, O. Taconet, J. Paillé, and P. Boissard, 2000: “Characterisation of the soil structureand microwave backscatter based on numerical three dimensional surface representation: analysiswith a fractional brownian model”, Remote Sens. Environ. 72: 159-169.

103

List of Symbols and Abbreviations

ε dielectric constant F.m-1

ρ density of air kg.m-3

σ ionic conductivity Sβ angle of leaf, branch or stem with the normal °λ latent heat of vaporisation J.kg-1

γ psychometric constant Pa.K-1

θ radar incidence angle °∆ slope of saturated specific humidity-temperature curve Pa.K-1

σ0 radar cross section m2.m-2

εb dielectric constant of bound water F.m-1

εf dielectric constant of free water F.m-1

δq specific humidity deficit Paεr residual dielectric constant vegetation F.m-1

σv Doppler velocity standard deviation m.s-1

A available energy W.m-2

Ai one-sided surface area (i = leaf, branch or trunk) m2

C radar constant W.m2

D drainage rate from leaves mm.h-1

d thickness leaf-waterfilm entity mdlf thickness leaf mds thickness waterfilm mE evaporation rate mm.h-1

Ep potential evaporation rate mm.h-1

F stemflow rate mm.h-1

f frequency GHzga aerodynamic conductance m.s-1

h atmospheric humidity %j denotes imaginairy part of number -k empirical factor -l length needle mLAI leaf area index m2.m-2

Mg vegetation gravimetric water content g.g-1

mv soil volumetric moisture content vol.%N received power numbersNb background power numbersNr reference power numbersp canopy gap fraction m2.m-2

P precipitation rate mm.h-1

P4 hour 4 hour cumulative precipitation mm


P24 hour 24 hour cumulative precipitation mmPc cumulative precipitation mmR canopy runoff rate mm.h-1

r range mrd radius dry needle mrw radius wet needle mS storage mmS0,i specific storage capacity (i = leaf, branch or trunk) mmSmax maximum storage capacity mmSsat saturation storage mmT throughfall rate mm.h-1

t temperature °Ku windspeed m.s-1

vb volume fraction bound water m3.m-3

vf volume fraction free water m3.m-3

vs volume stored water m3

ADC analog digital converterAOI area of interestERS European remote sensingFM-CW frequency modulated, continuos waveGIS geographic information systemhh horizontal polarisationhv cross-polarisationLUT look-up tablepdf probability density functionSAR synthetic aperture radarvv vertical polarised

105

Samenvatting

Hoofdstuk 1: Algemene inleidingDit proefschrift gaat over het nat worden van een bos in de regen. Als meetmethodewordt een radar gebruikt. Zo toont dit proefschrift aan dat een radar aan boord van eensatelliet op 800 km boven de aarde een regenwaterfilmpje met een dikte van slechts0.1 mm op het oppervlak van boombladeren kan detecteren. Wat is het belang van dittype onderzoek voor de kennis over het functioneren van bos in de kringloop vanwater op aarde? En wat is het praktische nut? Om met dit laatste te beginnen, drink-water voor veel grote steden, maar ook het bevloeiingswater voor landbouw, is oor-spronkelijk als regen op bos gevallen, voordat het via rivieren naar de plaats vangebruik is gestroomd. Verwacht wordt dat rond 2050 de helft van de wereld bevolkingwoont in landen met een tekort aan zoet water. Tegelijkertijd wordt verwacht dat hetklimaat zodanig verandert dat extreme regenval, en daaraan gerelateerde overstromin-gen, vaker zullen voorkomen.

Een kwart van het landoppervlak is bedekt met bos. De wateropbrengst van eenbosgebied is grofweg het verschil tussen de hoeveelheid neerslag en verdamping.Bossen verbruiken relatief veel water omdat de boomkruinen in nauw contact staanmet de omringende lucht. Dit waterverbruik bestaat uit twee componenten. De bomenworden nat door regen. Een deel van dit regenwater verdampt direct vanaf de nattebomen, nog voordat de regen de bosgrond heeft bereikt en in de bodem infiltreert.Deze hoeveelheid neerslag wordt interceptie genoemd. Daarnaast verbruikt een bosbodemwater doordat water met de wortels wordt opgenomen waarna het opgenomenwater via de bladeren verdampt. Om de wateropbrengst van grote gebieden te kunnenvoorspellen, is het wenselijk de hoeveelheid interceptie van bossen expliciet teschatten.

De interceptie van regen in een bos is niet eenduidig bepaald voor een enkele bui.Als het regent, dan blijft deze regen als druppels of als een laagje water op de blade-ren, takken en stammen liggen. De hoeveelheid regenwater die op deze manier in dekruin aanwezig is, wordt berging genoemd. Op een gegeven moment is de boomkruinverzadigd met regenwater. Alle extra neerslag zal van de bladeren en takken druipenof langs de stam naar beneden stromen. De maximale hoeveelheid berging van regen-water in de boomkruin neemt toe met de hoeveelheid water die een blad of een tak kanbergen en het aantal bladeren en takken in het bos. De maximale hoeveelheid bergingvan regenwater in de boomkruin is groot als het langdurig en zacht regent, en vermin-dert sterk bij harde wind of bij stortregen. Daarnaast speelt de snelheid waarmee hetgeborgen regenwater tijdens en na de bui verdampt een essentiële rol. De genoemdebiologische en atmosferische factoren variëren in ruimte en tijd. Hierdoor is het voor-spellen van interceptie over grote gebieden lastig. Metingen van de hoeveelheid water-berging in boomkruinen kunnen een hulpmiddel zijn om het waterverbruik van bossente schatten, en zo mogelijk voorspellingen te verbeteren. De praktijk wijst uit dat hetmoeilijk is de hoeveelheid berging van regenwater in boomkruinen te meten.


De hoeveelheid berging van regenwater in boomkruinen kan wellicht gemetenworden met satellietradar, dat de radarreflectie van een groot gebied meet. DeEuropese Remote Sensing (ERS) satellieten zijn sinds 1991 operationeel. Deze satel-lieten hebben een bepaald type radar (SAR) aan boord waarmee objecten tot enkeletientallen meters in doorsnede gedetecteerd kunnen worden, onafhankelijk van deatmosferische omstandigheden (nacht, wolken, en regen). Zowel de ERS-SAR alsvergelijkbare radars aan boord van vliegtuigen hebben een versterkte radarreflectievan natte bossen gemeten. Het is echter onduidelijk wat de precieze oorzaak van dezeversterkte radarreflectie was. De gemeten radarreflectie hangt namelijk van een grootaantal factoren af. Een belangrijke factor is de hoeveelheid water in en op het object.Andere factoren zijn het type vegetatie en de samenstelling en ruwheid van de bodem.De radargolf gaat namelijk gedeeltelijk door bomen heen, en wordt dan gereflecteerddoor de bodem. Naast de genoemde factoren die voor elk type bos uniek zijn, heeft hettype radar veel invloed, omdat de uitgezonden radargolf kan verschillen in golflengte,polarisatierichting en ‘kijkhoek’.

Het doel van het in dit proefschrift beschreven onderzoek is het vaststellen van degeschiktheid van satellietradar in het meten van de hoeveelheid regenwater die tijdensen na een bui in boomkruinen aanwezig is. Het onderzoek is gefaseerd uitgevoerd, enis beschreven in de volgende vier hoofdstukken.

Hoofdstuk 2: De berging van regenwater in bossen gedetecteerd met de ERS-tandemmissie SARHet onderzoek begon met het vaststellen van de haalbaarheid om met radar de hoe-veelheid berging van regenwater in boomkruinen te meten. De haalbaarheid is inge-schat met een bestaand model, dat de radarreflectie van droge bossen op basis vannatuurkundige principes simuleert. Dit model is aangepast voor dit onderzoek door hetaanbrengen van een waterlaagje op het oppervlak van alle bladeren, takken en stam-men. De berekeningen met dit model tonen aan dat bij een zelfde hoeveelheid regen-water in de boomkruinen de radarreflectie van loofbossen sterker is verhoogd dan deradarreflectie van naaldbossen. De berekende versterking van de radarreflectie is ver-geleken met ERS-SAR radaropnamen van de Veluwe, waarbij de boomkruinen de enedag nat, en de andere dag droog waren. Deze natheid van de boomkruinen is geschatop basis van de neerslag op, en de daarop volgende verdamping van de kruin. Deneerslagmetingen zijn gemaakt met de regenradar van het KNMI. De verdamping isafgeleid van meteorologische metingen, die tegelijkertijd zijn uitgevoerd door hetonderzoeksinstituut Alterra boven een bos bij Kootwijk. Door de overeenkomst tussende radarmeting en de theoretische verwachting wordt het haalbaar geacht om metradar de hoeveelheid berging van regenwater te meten. Een beperking is dat het typebos, loofbos of naaldbos, op voorhand bekend moet zijn.

Samenvatting 107

Hoofdstuk 3: Radar metingen van de berging van regenwater in een loofboomDe veranderende radarreflectie van een loofboom tijdens het nat worden is onderzochtmet behulp van een kleine radar. Om radarreflectie van de bodem uit te sluiten, is deradar op de grond geplaatst en schuin omhoog gericht naar de kruin van de boom. Deboom in kwestie was een es. De meetgegevens van een regenachtige week zijngeanalyseerd. Direct na het begin van een bui wordt al een versterkte radarreflectiegemeten. Na 2 mm neerslag stabiliseert de radarreflectie zich. Deze stabilisatie wordtwaarschijnlijk veroorzaakt door verzadiging met regenwater van de kruin. Dit patroonherhaalde zich tijdens alle 14 buien in de desbetreffende week. De radarreflectie vaneen boom was dus inderdaad gevoelig voor de hoeveelheid waterberging in de kruin.

Hoofdstuk 4: Het monitoren van de berging van regenwater in bossen met satelliet-radarDe stap van kleinschalige waarnemingen met grondradar naar grootschalige waar-nemingen met satellietradar wordt in dit hoofdstuk met behulp van het radarreflectiemodel gemaakt. De betrouwbaarheid van het model is getest aan de hand van demetingen uit het vorige hoofdstuk. Om de metingen van de grondradar te verklaren,moest tijdens regen een 0.09 mm dik waterlaagje op het oppervlak van de bladerenaanwezig zijn. De dikte van het waterlaagje is ook in het laboratorium bepaald. Degebruikte methode, het meten van de gewichtstoename van bladeren als deze natworden, is indicatief voor de hoeveelheid water op het blad na regen, als al het over-tollige water is afgedrupt. Het verschil tussen radar- en laboratoriumbepaling was30%. Dit verschil komt overeen met het te verwachten verschil tussen de hoeveelheidberging tijdens en na regen. Het model is dus betrouwbaar. Het model is vervolgensgebruikt om te onderzoeken welk type radar het meest geschikt is om de hoeveelheidberging te bepalen. De prestatie van radar is berekend voor verschilde golflengtes,polarisatierichtingen en ‘kijkhoeken’. Omdat geen enkel type bos hetzelfde is, zijndeze simulaties uitgevoerd voor twee populier variëteiten, een beukenbos en eenessenbos. Ook het vochtgehalte in de bodem is gevarieerd. Het meest geschikt voorhet meten van de hoeveelheid berging van regenwater in de kruin van een loofbosblijkt een radar met een horizontaal of verticaal gepolariseerde radargolf met eengolflengte tussen de 5 en 10 cm en een ‘kijkhoek’ van 20 tot 40 graden. Een dergelijkeradar wordt geschikt geacht omdat de gesimuleerde radarreflectie sterk wordt ver-hoogd door de berging van regenwater, terwijl de radarreflectie relatief weinig wordtbeïnvloed door de boomsoort of variaties in bodemvocht. De ERS-SAR behoort juisttot dit type radar.


Hoofdstuk 5: Schattingen van de hoeveelheid berging van regenwater in de kruin vanloofbomen met satellietradarIn dit hoofdstuk werd de ERS-SAR aan een praktijktest onderworpen. Gebruiktworden metingen van een 14 jaar oude populierenopstand in Flevoland. De radar-reflectie wordt met behulp van het model vertaald naar de hoeveelheid berging vanregenwater. Het model corrigeert tegelijkertijd de invloed van de radarreflectie doorde bodem. Voor deze correctie is het nodig te weten hoe nat de bodem was. Hetvochtgehalte van de bodem is op twee manieren bepaald. Door meting ter plekke, endoor bepaling uit radarmetingen van de ERS- SAR, een dag voor of na de radarmetingvan de natte kruin. Essentieel is de korte tijd tussen de twee radarmetingen, omdat hetvochtgehalte van de bodem niet drastisch mag veranderen. De geschatte berging vanregenwater in de boomkruinen is vergeleken met de hoeveelheid berging die terplekke is bepaald uit het verschil in neerslag boven de kruin en onder de kruin. Dezemetingen zijn uitgevoerd door het onderzoeksinstituut Alterra. De radarmeting engrondmeting van de berging komen overeen voor 4 van de 6 geanalyseerde radar-observaties van natte kruinen. De verschillen voor de twee overige observaties wordentoegeschreven aan het verwaarlozen van ondergroei in het vroege voorjaar, wanneer ernog weinig blad aan de boom zit, en doordat een satellietradar het sterkst reageert opnatte bladeren in boomtoppen.

Hoofdstuk 6: Het potentieel van radar om de natheid van bossen te metenTenslotte werd het onderzoek geëvalueerd. Het is mogelijk om de dikte van het water-laagje op bladeren te meten met zowel grondradar als satellietradar, op voorwaarde datonder meer de structuur van de bomen en de natheid van de bodem bekend zijn. Eensatellietradar meet voornamelijk de natheid van bladeren in het bovenste deel van hetbos. Boomtoppen zijn belangrijk in de uitwisselingprocessen tussen landoppervlak enatmosfeer. Vervuilende gassen als zwaveldioxide en ozon kunnen bijvoorbeeld snellerneerslaan op natte bladeren dan op droge bladeren. De opname van het broeikasgaskoolstofdioxide door natte bladeren is lager dan door droge bladeren, omdat vloeibaarwater een vrijwel ondoordringbare barrière over de huidmondjes legt. Hierdoor isradar vooral geschikt om de uitwisselingprocessen tussen het landoppervlak en deatmosfeer te bestuderen. Een nadeel van satellietradar is dat de satelliet maar enkelekeren per maand overkomt. Een grondradar kan continu metingen verrichten. Desterke punten van een grondradar en een satellietradar kunnen gecombineerd wordendoor bijvoorbeeld met een grondradar vanaf een heuveltop de natheid van het lagerliggende landschap continue te meten.

Remote Sensing of Wet Forests 109

Nawoord

Dit proefschrift komt niet – zoals regen – plotseling uit de lucht vallen. De Groningsevakgroep Fysische Geografie onderzocht onder leiding van wijlen Arthur Veen dewaterhuishouding van bossen sinds eind jaren 70. Helaas werd deze vakgroep metingang van het jaar 2000 opgeheven. Het boshydrologisch onderzoek kon daarnaworden voortgezet onder de vlag van de vakgroep Plantenfysiologie. Veel mensenhebben direct of indirect een bijdrage geleverd aan dit proefschrift. Mede door dezesteun kijk ik op een prettige manier op de afgelopen jaren terug. Ik wil in de eerste plaats mijn promotor, Piet Kuiper, en mijn referent, WimKlaassen, bedanken voor hun inzet en betrokkenheid. Zelfs na het nemen van zijnemeritaat, bleef Piet enthousiast en vol ideeën. Zijn adviezen op het gebied vanschrijven en zijn brede kijk op ecologische en evolutionaire aspecten van de water-huishouding van bossen waren leerzaam en inspirerend. Wim heeft dit projectgeïnitieerd en de financiering verworven. Hij heeft mij ingewijd in de bos-hydrologische en meteorologische modellen. Zijn redactionele vaardigheden hebbende leesbaarheid van dit proefschrift sterk verbeterd. Henk de Groot had ook eenbelangrijke rol. Tijdens het experimentele werk met de grondradar in de tuin van hetBiologisch Centrum in Haren stond hij mij met raad en daad ter zijde. Vervolgensanalyseerden we gedurende ettelijke weken het vocht- en suikergehalte van boom-bladeren. Helaas heeft dit laatste niet tot het gewenste resultaat geleidt. Zonder debijdrage van deze drie keien, had dit proefschrift niet voor u gelegen.

Gedurende de analyses en het experimentele werk is voortborduurd op de ervaringvan anderen. Marco van der Linden heeft een eerste analyse van de radarbeeldenuitgevoerd. Sander Lensink en Mark Durenkamp hebben de kinderziektes uit deexperimentele radar gehaald. Albert Ballast heeft op een zeer theoretische (en voor mijvrijwel onbegrijpelijke) wijze de validiteit van het model uit hoofdstuk 2 van ditproefschrift bewezen. Daarnaast wordt in dit proefschrift gebruik gemaakt van data diedoor de volgende personen ter beschikking zijn gesteld: de hydrometeorologischegegevens van de Veluwe (hoofdstuk 2) en het Fleditebos (hoofdstuk 5) zijn afkomstigvan Eddy Moors en Han Dolman (ALTERRA). De regenradar gegevens (hoofdstuk 2)werden geleverd door Rudmer Jilderda (KNMI). Job Verkaik (KNMI) leverde wind-gegevens van vliegveld Eelde (hoofdstuk 3). Paul Saich en Maurice Borgeaud (ESA-ESTEC) leverden voorbewerkte radardata (Hoofdstuk 5). Ik ben hen allen zeererkentelijk.

Eerdere versies van de hoofdstukken zijn door diverse mensen gelezen enbecommentarieerd, waardoor de kwaliteit van de tekst is toegenomen. Hoofdstuk 2 isverbeterd door de suggesties door Jan Delvigne, Bernard Hoenders en twee anoniemereferenten. Hoofdstuk 3 is gelezen door Jan Delvigne, Bert van den Broek (TNO-FEL), Gerhard Peters, en Hans-Jurgen Kirtzel (beide METEK GmbH). Hoofdstuk 4 isbecommentarieerd door Christophe Proisy (CESBIO) en twee anonieme referenten.Hoofdstuk 5 is verbeterd op aanwijzingen van de co-auteurs, Eddy Moors(ALTERRA) en Maurice Borgeaud (ESA-ESTEC), terwijl Paul Saich (University


College, London) een waardevolle bijdrage leverde aan de eerste analyse. PeterHoogeboom (TNO-FEL, TU-Delft) heeft tenslotte commentaar geleverd op hetgeheel.

Tijdens de uitvoering van het onderzoek moesten veel praktische obstakels over-wonnen worden. Onmisbare hulp bij het oplossen van computer-, GIS- en weerstation-problemen was afkomstig van Jan van den Burg. Marjan Boertje-van den Bergh, en ineen later stadium, Ger Telkamp en Jannie Nuijten-Tjalkes hebben de projectadministratie beheerd. Cees Rappoldt loste mijn FORTRAN problemen op. TheoElzenga heeft na het ter ziele gaan van de vakgroep Fysische Geografie eenbemiddelende rol gespeeld en gezorgd dat ik kon rekenen op de organisatorische steunvan de vakgroep Plantenfysiologie. Allen bedankt!

Stipt om tien uur werd er koffie gedronken door de collega’s van de bovenste gangvan vleugel C. Deze traditie bleef gehandhaafd na het opheffen van de vakgroepFysische Geografie, en vormde een prettige afleiding. In de loop der jaren zatendiverse mensen aan de koffietafel. Een groot aantal werd hierboven al genoemd.Daarnaast wil ik de volgende mensen nog bedanken voor het aangenaam verpozen:Peter van Breugel, Carole Elling, Margit Gosen, Mtinkheni van Himbergen, JanSpieksma, Jo Vergoossen, Arjen de Vries, en Henk Zemmelink.

Tenslotte zou dit proefschrift nooit zijn afgerond zonder de interesse en steun vanmijn ouders en van mijn vrouw, Brenda. Door de flexibele opstelling van Brendawaren we in staat beiden te blijven werken en tegelijkertijd onze dochter Mieke telaten opgroeien tot een vrolijke peuter. Daarnaast heeft ze meegeleefd met de toppenen dalen die bij een onderzoek horen, en heeft ze de kaft van dit proefschriftontworpen.

Möge nun das Werk vielleicht auch in andere Ländern, insbesondere innördlich und südlich gelegen, Veranlassung geben, durch ähnlicheForschungen die Arbeit zu vervollständigen und dadurch zur allgemeinerenErkenntniss der volkswirthschaftlichen Bedeutung der Wälder beitragen!(Ebermayer, 1873)

Remote Sensing of Wet Forests 111

Curriculum Vitae

Nadat de auteur zijn middelbare schoolopleiding op het St. Ludger college te Doetinchemin 1987 had afgerond, ging hij werktuigbouwkunde studeren aan de Universiteit Twente.Hij vertrok in 1991 voor langere tijd naar India. Geïnteresseerd geraakt in de verhoudingtussen de mens en zijn omgeving werden cursussen milieukunde, geologie en biologie vande Open Universiteit gevolgd. Hij vervulde zijn vervangende dienstplicht bij het Neder-lands Instituut voor Onderzoek der Zee (NIOZ) te Texel in 1993. Zijn onderzoek betrof derelatie tussen de nutriëntenuitstroom van de Rijn en de op satellietbeelden zichtbare algen-bloeien op de Noordzee. Vervolgens ging hij milieukunde studeren aan het Van HallInstituut, de Groningse Hogere Agrarische School. Gedurende deze opleiding werdpraktijkervaring opgedaan bij de Rijksuniversiteit Groningen (RUG), de NederlandseAardolie Maatschappij (NAM) en het RijksInstituut voor Kust en Zee (RIKZ).Onderwerpen waren het modelleren van de depositie van salpeterzuurgas bij bosranden,het inschatten van de verstoring van vogels door aardgasboringen in de Waddenzee, en hetmonitoren van kweldervegetatie met remote sensing technieken. Hij verkreeg deingenieurstitel in 1997. Op dat moment werkte hij aan diverse projecten bij het RIKZ. In1998 begon hij bij de RUG met het onderzoek dat is beschreven in dit proefschrift.

Publicaties

NAM, 1996, “Verstoring van wadlopers door continu- en piekgeluid”, in: aanvullingMER exploratieboringen Waddenzee, NAM, Assen, pp. 53-62.

J. J. M. de Jong, R. Sam, C. L. M. Van de Ven, en M. G. Vroom, 1997:Gegevenscatalogus Waddenzee, rapport RIKZ/AB-97.606x, RIKZ, Haren, 12 pp.

J. J. M. de Jong en W. Klaassen, 1997: “Simulated dry deposition of nitric acid nearforest edges”, Atmospheric Environment, 31, pp. 3681-3691.

J. J. M. de Jong, W. Klaassen, en M. van der Linden, 1998: “SAR sensing ofvegetation wetness”, in: Proceedings of the second international workshop on theretrieval of bio- and geophysical parameters from SAR data for land applications,21-23 October 1998, Noordwijk, pp. 341-345.

J. J. M. de Jong, A. C. de Vries, en W. Klaassen, 1999: “Influence of obstacles on theaerodynamic roughness of the Netherlands”, Boundary-Layer Meteorology, 91, pp.51-64.

W. Klaassen en J. J. M. de Jong, 1999: “Fast recycling of rain water by interception”,in: Proceedings of third international conference on the global energy and watercycle, 16-19 June 1999, Beijing, China, pp. 318-319.

J. J. M. de Jong, W. Klaassen, en A. Ballast, 2000: “Rain storage in forests detectedwith ERS tandem mission SAR”, Remote Sensing of Environment, 72, pp. 170-180.

J. J. M. de Jong, H. W. de Groot, W. Klaassen, en P. J. C. Kuiper, 2000: “Radarbackscatter change from an ash in relation to its hydrological properties”, in:Proceedings IGARSS 2000, 23-28 July 2000, Honolulu, pp. 2927-2929.

Stellingen

Behorende bij het proefschrift:Remote Sensing of Wet Forests

door Joost de Jong

1. De invloed van de regenintensiteit en de wind op de regeninterceptie van bossen is,ondanks 150 jaar onderzoek, nog steeds onzeker (Ebermayer, 1873, Horton, 1919,Philips, 1926, Calder, 1996, Hörmann et al., 1996, Herwitz en Slye, 1996). Hetontbreken van een eenvoudig toepasbare meetmethode om de berging van regen-water in boomkronen te bepalen is een belangrijke reden voor dit probleem(Lundberg et al., 1997, Calder en Hall, 1997, Crockford en Richarson, 2000,Dekker, 2000).

2. De sterkte van de radarreflectie door bossen wordt sterk beïnvloed door de

diëlectrische constante van bladeren en naalden. De diëlectrische constante van eennat blad of naald kan bepaald worden door het meten van de diëlectrische constanteen het volume van een droog blad of naald, en de diëlectrische constante en hetvolume van het aanhangende water. De diëlectrische constante van een nat blad ofnaald kan vervolgens worden berekend door het volumetrisch gewogen gemiddeldevan de diëlectrische constanten te nemen (dit proefschrift).

3. De horizontaal of verticaal gepolariseerde C- of X-band radarreflectie van een nat

loofbos is circa 2 dB hoger dan de radarreflectie van een droog loofbos. Door dezegevoeligheid kan een radar gemonteerd op een satelliet onderscheid maken tusseneen nat en een droog bos. Deze stelling geldt niet voor naaldbossen (ditproefschrift).

4. Het monitoren van de natheid van bossen is gebaat bij dagelijkse satelliet-

waarnemingen van elke locatie (dit proefschrift). 5. Een op de grond staande radar kan het nat worden en opdrogen van boombladeren

continu monitoren (dit proefschrift). 6. Regenpatronen en soortensamenstelling bepalen voornamelijk de primaire productie

van een biome (A. K. Knapp en M. D. Smith, Science 291: 481-484, 2001).

7. Een boom absorbeert relatief veel energie om water – zowel opgenomen via dewortels als opgevangen regenwater – te laten verdampen. Een boom kan daarombeschouwd worden als een ideale airco, slechts geschikt voor buiten gebruik,aangedreven door zonne-energie, en volledig gemaakt van recyclebaar materiaal.Het maakt geen lawaai, vergt een minimum aan onderhoudt, en heeft een levensduurvan tientallen jaren (naar Pokorný, Renewable Energy 24: 641-645, 2001).

8. Bij interdisciplinair onderzoek wordt kennis uit verschillende vakgebieden op

logische wijze aan elkaar gekoppeld. De gebeurt meestal op een simpele wijzeomdat er in tenminste één van de vakgebieden onzekerheden bestaan over deparameters die aan elkaar gekoppeld worden. Specialisten uit het andere vakgebiedkijken hierdoor met gekromde tenen toe.

9. Natuurrestauratie op niet rendabele landbouwgronden in Nederland is mogelijk

doordat tegelijkertijd natuur wordt omgevormd tot landbouwgronden in de zichontwikkelende wereld.

10. Staatsbosbeheer geeft blijk van een vooruitziende visie op natuurbeheer door in

enkele proefgebieden het van de bospaden afdwalen te stimuleren: respect voor denatuur wordt niet geleerd door de toegang te ontzeggen.

11. Een multifunctioneel rokkostuum heeft afritsbare panden. 12. Wetenschap is de titanische poging van het menselijke intellect zich uit zijn kos-

mische isolement te verlossen door te begrijpen (W. F. Hermans in Nooit meerslapen).

13. De eerste regel van het algoritme voor ontdekking, “Slow down to explore”

(Paydarfar en Schwartz, Science 292: 13, 2001), komt in de verdrukking doordatAIO projecten te ambitieus worden opgezet en de deadline bij voorbaat vast staat.

14. Wetenschap zal aan populariteit winnen als er een Quizcitatie-index wordt

ingevoerd. 15. Met dit proefschrift wordt ruim 20 jaar Gronings boshydrologisch onderzoek

afgesloten. Gezien de maatschappelijke en politieke belangstelling voor bossen enwater is dit opmerkelijk.

Remote Sensing of Wet Forests Proefschrift · Remote Sensing of Wet Forests Proefschrift ter...

Documents

Transcript of Remote Sensing of Wet Forests Proefschrift · Remote Sensing of Wet Forests Proefschrift ter...