Impacts of deforestation and afforestation in the ...

Ž .Global and Planetary Change 30 2001 309–328www.elsevier.comrlocatergloplacha

Impacts of deforestation and afforestation in the Mediterraneanregion as simulated by the MPI atmospheric GCM

Lydia Dumenil Gates a,b,), Stefan Ließ a¨a Max-Planck-Institut fur Meteorologie, Bundesstrasse 55, D-20146 Hamburg, Germany¨

b Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, CalÕerton, MD 20705, USA

Received 21 December 1999; received in revised form 15 October 2000; accepted 26 October 2000

Abstract

For two reasons it is important to study the sensitivity of the global climate to changes in the vegetation cover over land.First, in the real world, changes in the vegetation cover may have regional and global implications. Second, in numericalsimulations, the sensitivity of the simulated climate may depend on the specific parameterization schemes employed in themodel and on the model’s large-scale systematic errors. The Max-Planck-Institute’s global general circulation modelECHAM4 has been used to study the sensitivity of the local and global climate during a full annual cycle to deforestationand afforestation in the Mediterranean region. The deforestation represents an extreme desertification scenario for thisregion. The changes in the afforestation experiment are based on the pattern of the vegetation cover 2000 years beforepresent when the climate in the Mediterranean was more humid. The comparison of the deforestation integration to thecontrol shows a slight cooling at the surface and reduced precipitation during the summer as a result of less evapotranspira-tion of plants and less evaporation from the assumption of eroded soils. There is no significant signal during the winterseason due to the stronger influence of the mid-latitude baroclinic disturbances. In general, the results of the afforestationexperiment are opposite to those of the deforestation case. A significant response was found in the vicinity of grid pointswhere the land surface characteristics were modified. The response in the Sahara in the afforestation experiment is inagreement with the results from other general circulation model studies. q 2001 Elsevier Science B.V. All rights reserved.

Keywords: climate; climate change; deforestation; afforestation

1. Introduction

Only recently have the impacts of land-use changeon the overlying atmosphere been measured in asystematic way in field experiments. The experiencefrom a measurement campaign that covered an Ama-

) Corresponding author. Lawrence Berkeley National Labora-tory, 1 Cyclotron Road, Mailstop 90-1116, Berkeley, CA 94720,USA. Tel.: q1-510-486-6642; fax: q1-510-486-5686.

Ž .E-mail address: [email protected] L. Dumenil Gates .¨

zonian forest area and a forest clearance is summa-Ž .rized by Gash et al. 1996 , while the climate in the

Mediterranean has been extensively studied in fieldŽ .experiments described by Bolle et al. 1993 . For

some land-use scenarios, field experiments would beimpossible to do—e.g. for future projections of land-use and for future potential changes due to projectedanthropogenic climate changes due to the increasedso-called greenhouse effect. The effects of such land-use or vegetation change scenarios, however, can betested in numerical simulations, which describe the

0921-8181r01r$ - see front matter q 2001 Elsevier Science B.V. All rights reserved.Ž .PII: S0921-8181 00 00091-6

( )L. Dumenil Gates, S. LießrGlobal and Planetary Change 30 2001 309–328¨310

physical relationships between the atmosphere, theland surface and ocean components of the climatesystem. Models might also be used to establish thepossible impact of historical land use changes. Usu-ally, the vegetation distribution over the continentsrepresents present-day conditions, but these bound-ary conditions may be changed to represent differentsurface or vegetation types and their associated phys-ical effects. Potential resulting impacts from vegeta-tion changes include changes in the radiation budgetvia the surface albedo and changes in the hydrologi-cal cycle in terms of evaporation, precipitation, runofffrom infiltration excess and drainage, as well aspossible interactions with the carbon cycle. Thestrength and regional variation of such impacts, how-ever, depends on the general architecture and sensi-tivity of the atmospheric model that is used. In thisstudy, we shall use the Max-Planck-Institute’s atmo-sphere general circulation model ECHAM4Ž .Roeckner et al., 1996 to simulate the response tospecified land-use changes in the Mediterranean re-gion.

The climate of the Mediterranean region in south-ern Europe and northern Africa is characterized bymild wet winters and warm dry summers. Theboundaries of the Mediterranean region are conve-niently defined as the area where the climate allows

Ž .olive trees to grow Rother, 1993 , the northernborder of which is approximately given by the 58CJanuary isotherm. In the south, the area is limited tothe coastal parts of North Africa. The dominantcirculation features of this region are the mid-latitudewesterlies in winter, and in summer, the subtropicalhigh-pressure belt associated with the seasonal shift

Ž .of the inter-tropical convergence zone ITCZ . Thechange between the two regimes occurs relativelyquickly, so that the spring and autumn seasons arenot pronounced. The Mediterranean region is gener-ally shielded from storms from the Atlantic as it issurrounded by mountain ranges. As a consequence,in the summer, this region is sensitive to localchanges via interactions with the surface.

In the Mediterranean, the precipitation is impor-tant for the maintenance of vegetation. A meanprecipitation of about 1 mmrday during the summerseason is required to maintain the Mediterranean

Ž .forests Reale and Shukla, 2000 , and a reduction inthe precipitation pattern due to climate change may

lead to the destruction of these forests. On the otherhand, reduction of the vegetation cover due to landuse changes may affect the local recycling of waterby evaporation and lead to less precipitation in thedry season, again resulting in a decrease of vegeta-tion. If there is more precipitation in the rainy sea-son, this would have the additional effect of enhanc-ing the fluvial erosion.

The natural distribution of vegetation types in aregion should be in equilibrium with the climate,water and nutrient limitations, but the distribution ofvegetation has often been modified by anthropogenicinfluences. In the Mediterranean region, deforesta-tion has been common practice for more than 2000years and its impacts are widespread, while defor-estation and afforestation scenarios are being dis-cussed for the future. We should note, however, thatthe influence of a given change in boundary condi-tions depends crucially on the local climate, and hasto be expected to be different in mid-latitudes and inthe tropics or subtropics. Thus, the physical descrip-tion of the impacts of vegetation change in differentparts of the world is not transferable, because differ-ent threshold values are involved. Vegetation–atmo-sphere interaction has also been studied in paleo-climate systems.

Several scientific questions arise in this context.The first question is how far regional and local

Ž .climate change e.g. due to land use change willaffect the global circulation patterns, or whether theywill have only a local impact. This question can onlybe addressed by global general circulation models.Detailed regional changes can then be estimatedfrom regional models. The second question arisesfrom the hypothesis of a more pluvial climate inRoman times as discussed by Reale and DirmeyerŽ .2000 . Has the historical anthropogenic deforesta-tion in the region lead to a desertification, or do wehave to consider other sources of atmospheric vari-ability? If this question is studied with a generalcirculation model, however, there are uncertaintiesbecause of the coarseness of its horizontal resolutionat which the relevant equations are solved and due tothe assumptions on which the parametrizations ofphysical processes that are represented by the modelare based. For each general circulation model, thesensitivity to certain changes needs to be established.A third question concerns the future conditions. Esti-

( )L. Dumenil Gates, S. LießrGlobal and Planetary Change 30 2001 309–328¨ 311

mates of the expected change in global mean temper-ature associated with an increase of the effects oflarge-scale global patterns of the anthropogenicemission of so-called greenhouse gases and aerosolsare reported to affect the hydrological cycle and heat

Ž .balance of the region IPCC, 1990, 1995 . This canbe viewed as a large-scale global climate changehaving consequences for the regional climate. Such aclimate change is expected to cause higher tempera-tures and changed availability of water, which maylead to an increase or decrease of the vegetationcover. Any such changes affecting the vegetationcover may feedback on the atmosphere and may leadto an enhancement or diminuition of the originalchanges. Modifications of vegetation patterns in thepast may already have interfered with temperaturechanges caused by other anthropogenic influences.

In this study, we shall focus on the first twoquestions. A control simulation with the Max-Planck-Institute’s ECHAM4 atmosphere general cir-

Ž . Ž .culation model GCM Roeckner et al., 1996 forpresent-day climate is the basis for two sensitivitystudies, one for a deforestation and one for an af-forestation scenario. While some of the realism ofsummer precipitation in terms of the intensity ofevents is missing in a such coarse resolution GCM,the use of the global model will enable us to see ifthere are any effects downstream from the areawhere the local changes are applied and may revealglobal teleconnections. Here we shall focus on thequality of the global climate simulation, on the gen-eral sensitivity of the ECHAM4 model, and on theconsistency of the changes in the simulated hydro-logical cycle and surface energy budget. A moredetailed analysis taking advantage of more realisticlevels of internal variability of the higher resolutionregional models PROMES and HIRHAM4 is made

Ž .elsewhere Polcher et al., 1999 .

2. Description of the model and the sensitivityexperiments

2.1. The model

The numerical experiments analysed in this studywere performed with the ECHAM4 atmospheric

Ž .GCM Roeckner et al., 1996 at T42 resolution,corresponding to a mesh size of about 2.88 in longi-tude and latitude. In the Mediterranean region, this isequal to about 250 km in the zonal and 300 km inthe meridional directions. The model has 19 verticallevels. The control simulation is a standard integra-tion of 35 years, the first 10 years of which werediscarded to eliminate the effects of model spin-up.The experimental simulations then start from 1 Jan-uary Northern Hemisphere winter conditions of year11 of this control simulation in order to providenear-saturation at least in northern mid-latitudes.Here, we found that it was sufficient if the first yearof the integration was discarded, because after thefirst year of the integration, the range of the annualcycle of soil moisture fell to within the range ofinter-annual variability of the remainder of the simu-lation. In the control simulation as well as the sensi-tivity experiments, climatological global sea surface

Ž .temperatures SST averaged over the period 1979–1988 were used to eliminate interannual variabilitydue to variations in sea surface temperature. Theatmospheric variability represented in such integra-tions is generally less than that in simulations with

Žinterannually varying boundary conditions Bengts-.son et al., 1996 .

Local Mediterranean SSTs are also prescribed byclimatological values. This means that in both thecontrol simulation and the experiments the flow fromthe land to the sea is unable to modify the seasurface temperature. In the real world, the interactionwould cause a warming or cooling by direct heatingor upwelling due to the effects of friction, whichmay also change the water temperature at the sur-face. Fully coupled ocean models that also includethe Mediterranean are currently available at MPI, butlimited in horizontal resolution. We restricted thisinitial series of experiments to prescribed SSTs, be-cause the length of any coupled simulations wouldhave had to be extended considerably in order tostatistically interpret the results. The restriction toprescribed SSTs over the Mediterranean Sea, whichis always regrettable, however, does not pose aserious problem in this case. The 850 hPa wind fieldin Fig. 1 shows that the flow over the sea onlyaffects the southeastern part of the area where onlyfew grid points were taken into account in the aver-aging.


The parameterization of soil hydrology inECHAM4 comprises budget equations for the snowamount on the surface, for the water amount inter-cepted by the vegetation, and for the soil waterstorage. The time evolution of soil water contentconsiders evaporation, surface runoff and drainage.

Ž .The runoff scheme Dumenil and Todini, 1992 takes¨into account subgrid-scale variations of field capac-ity over inhomogeneous terrain in a grid-area, andtherefore allows runoff even if the bucket type soilreservoir is not fully saturated. Water in the soilwater reservoir is accessible for evaporation as afunction of the soil water amount and of the vegeta-tion fraction in a grid. Only a maximum value of 10cm is accessible for evaporation in the non-vegetated

Ž .part of a grid point bare soil . In the vegetated partthe difference between actual soil moisture and the

Ž .wilting point 35% of the maximum soil moisture isaccessible for evaporation.

2.2. The model control climate

In the Mediterranean, it is useful to combine therainy winter months of January, February and MarchŽ .JFM and the dry summer months July, August and

Ž .September JAS for averaging. Fig. 1 shows the sealevel pressure and 850 hPa wind field for winter andsummer in terms of mean values from the ECMWF

Žre-analysis European Centre for Medium Range.Weather Forecasts; Gibson et al., 1997 and the

25-year control simulation. The re-analysis data werecalculated at T106 resolution and were later trun-cated to T42 resolution for optimum comparisonwith our model results.

Although the quality of the model simulation isgenerally high during winter deviations from there-analysis occur in the western Mediterranean, wherethe Azores high is too strong and extends too far

Žeastward in the control run. For ECHAM4 there is

Ž . Ž .Fig. 1. Mean sea level pressure, 850 hPa wind field and streamlines for the ECMWF reanalysis 1978–1994 left and the ECHAM4Ž . Ž .25-year control simulation using climatological sea surface temperatures right for winter JanuaryrFebruaryrMarch and summer

Ž .JulyrAugustrSeptember averages.


Ž .Fig. 1 continued .

hardly any difference in this region between simula-tions using interannually varying or climatological

.SSTs. This results in a considerable underestimationof precipitation in the western Mediterranean, whichalso has repercussions for the spring and summerclimate, as the soil moisture reservoirs are not re-plenished. This systematic model error is insensitiveto changes in the formulation of the land surfaceprocesses and has to be attributed to the model’srepresentation of the general circulation. The sealevel pressure in summer is better represented, whilethe strength of the 850 hPa wind field is underesti-

Ž .mated. Precipitation maxima Fig. 2 are generallyunderrepresented due to the relatively coarse resolu-tion of the model. The largest precipitation deficien-cies occur along the Adriatic coast and near theCaspian Sea. Although our model simulation has aresolution of only T42, there is an indication ofrainfall in the Pyrenees in agreement with a higher

Ž .resolution climatology Hulme et al., 1995 . Precipi-tation in the Atlas mountain region, however, ismuch too high in the model simulation.

2.3. The experiments

The deforestation experiment simulates theMediterranean climate after all vegetation in theregion that is present today has been removed. In the

Ž .ECHAM4 model Roeckner et al., 1996 , the hori-zontal distribution of surface parameters was derivedfrom the high-resolution distribution of the main

Ž .ecosystems of Olson et al. 1983 and from the EarthŽ .Radiation Budget Experiment ERBE satellite data

Ž .Ramanathan et al., 1989 , while in the deforestationexperiment all points concerned were defined to beof the desert type, and were assigned the appropriatecharacteristics. The resulting changes in surface pa-rameters are described in Fig. 3. At the grid points


Ž . Ž . Ž .Fig. 2. Precipitation averaged over the summer season JulyrAugustrSeptember for: a the high-resolution climatology 1961–1990 ofŽ . Ž . Ž . Ž .Hulme et al. 1995 , b the ECMWF re-analysis at T106 resolution 1979–1988 , c the control simulation.


Ž .Fig. 3. Distribution of vegetation-related surface boundary conditions as used in the ECHAM4 atmosphere model; Roeckner et al., 1996for the control simulation and the deforestation and afforestation experiments.



where the original ECHAM4 surface dataset alreadycontained desert characteristics, the original values

were kept. A new surface roughness length wascomputed after vegetation was removed.



The maximum soil moisture is assumed to dependŽ .nearly linearly on rooting depth Patterson, 1990 .

Erosion and fossilization of soils after deforestationwill reduce the maximum soil moisture to even


lower values for non-vegetated areas than are gener-ally used in the model for bare soil, because the soilcharacteristics also change. The values in the defor-estation simulation are therefore rederived as a func-tion of a fictitious rooting depth of 0.1 m appropriate

Žfor extreme deserts Rowntree and Dumenil, 1995;¨.Abramopoulos et al., 1988 .

With a view to establishing if the model mayrespond to opposite changes in a consistent way, anafforestation experiment was also performed. Here,the vegetation was increased to the amount presentduring the Roman classical period in the first centuryB.C. These changes affect an area from the British

Ž .Isles to the Nile Fig. 3 and have a different regional

Ž .Fig. 4. Difference between a potential evaporation experiment irrigated continents and a control simulation using the MPI model at T21Ž .resolution, averaged over 5 years for July mean precipitation top and for the difference between a no-land-surface-evaporation experiment

Ž . Ž .desert continents and the control simulation bottom . Solid lines indicate positive differences, dashed lines indicate negative differences.Shading indicates that the differences are significant at the 97% level of a Student’s t-test.


extent than in the deforestation scenario because herea well-documented dataset is available. This dataset

Žwas derived from pollen analysis Huntley and Birks,.1983; Huntley and Webb, 1988 and was converted

to the appropriate biome types by Reale and DirmeyerŽ . Ž .2000 . A similar but summer-only experiment with

Ž .this dataset was made by Reale and Shukla 2000with the COLA GCM at a higher resolution thanused in this study, and in which the SSiB biosphere

Ž .model Xue et al., 1991 was implemented.

2.4. General land surface sensitiÕity of the MPImodel

The potential for atmospheric and hydrologicalchanges induced from modifications of the lowerboundary conditions is large. Experience over thelast decade has, however, illustrated that differentGCM experiments show a different sensitivity tochanges of the evapotranspiration, for example. InFig. 4, this sensitivity is shown for a member of theMax-Planck-Institute’s general circulation model hi-erarchy at the horizontal resolution of T21, which iscomparable to the one used in the study by Shukla

Ž .and Mintz 1982 for summer only simulations. Here,the MPI model was, however, integrated for five fullannual cycles. Results shown in Fig. 4 describe theMPI model’s response of July precipitation to theseauthors’ specification of Airrigated continentsB, i.e.the specification of potential evaporation at all landpoints, and a Adesert planetB where no evaporationwill occur over land during the annual cycle. Com-pared to their experiments which comprised 60-dayforecasts starting from initial conditions in the sum-

Ž .mer Shukla and Mintz, 1982 , our model’s simu-lated annual cycle shows less sensitivity. This sug-gests that allowing the variables to run through thesequence of evolution during a full annual cycle mayhave a decisive impact on the sensitivity to landsurface boundary conditions in a particular region,because reservoirs are replenished during the rainyseasons. If regions have characteristics in the pre-sent-day climate that are close to the extreme experi-mental boundary conditions, they stand outas regions where differences are small. For desertcontinents the precipitation is reduced over land inthe equatorial regions and in continental areas of theNorthern Hemisphere. For wet soils, a statistically

significant amount of local precipitation can form inthe subtropics and the adjacent Mediterranean cli-mates despite the large-scale sinking motion in theseregions. Analysis of other model variables and atother model levels showed that the bulk of thestatistically significant response is confined to thelower troposphere.

3. Impacts of deforestation and afforestation

3.1. Surface energy budget and near surface temper-ature

The change of vegetation cover in both experi-mental simulations considered here is much smallerthan the modifications applied in the experiments inFig. 4. The model response is therefore expected tobe smaller. The GCM gives a consistent descriptionof changes in the surface budgets for both the defor-estation and the afforestation simulations. Modifica-tions of vegetation that are relevant to the hydrologi-cal cycle are associated with changes of the latentheat flux, while modifications of the surface albedoresult in changes of solar net radiation. In the

Table 1Ž .Summer averages JulyrAugustrSeptember of the components

of the surface energy budget and the 2-m temperature averagedover grid points in the Atlas mountain range for the controlsimulation and the experiments

Atlas mountain range Deforestation Control Afforestationin JAS

Net solar radiation 185.2 194.0 206.0y2w xW m

Downward solar 262.0 257.0 251.7y2w xradiation W m

Upward solar y76.8 y63.0 y45.7y2w xradiation W m

Net thermal y97.5 y95.2 y96.1y2w xradiation W m

Downward thermal 378.3 381.5 386.0y2w xradiation W m

Upward thermal y475.8 y476.7 y482.0y2w xradiation W m

Latent heat flux y11.1 y23.4 y24.4y2w xW m

Sensible heat flux y45.8 y51.3 y58.2y2w xW m

2-m temperature 29.0 29.2 30.2w x8C


Mediterranean region, the components of the surfaceŽenergy budgets Table 1 for a representative example

.from the Atlas mountain range are in general asfollows: The turbulent fluxes of momentum, sensibleand latent heat are generally decreased for the de-

forestation experiment as the roughness length isreduced. The decrease in evaporation due to thereduced vegetation cover and the changed soil condi-tions results in a significantly decreased latent heatflux, which should be compensated by an increase in

Ž . Ž .Fig. 5. Differences between the experiments and the control of the summer JulyrAugustrSeptember average 2-m temperature for: aŽ .deforestation minus control, b afforestation minus control. Light shading indicates significance in a Student’s t-test at the 95% level, dark

shading at the 99% level.


sensible heat flux. As the surface albedo is increasedat the same time, the input from net solar radiation atthe surface is, however, reduced. This results in lesssensible heat flux and, as a consequence, a smalldecrease in near surface temperature, so that thethermal radiation at the surface does not changemuch. The cooler and drier surface reduces convec-tion, which suppresses locally produced precipita-tion. Afforestation generally acts in the oppositeway.

The regional pattern of differences in the 2-mtemperature averaged over the summer months of thecontrol simulation and the experiments is given inFig. 5. There are no significant global scale impactsof the changes in Mediterranean vegetation on thedynamics and the hydrological cycle outside the

Ž .Sahara not shown .In both the deforestation and the afforestation

cases, the regional pattern of the response is notuniform over the Mediterranean and the surroundingregions. The changes induced by the deforestation orafforestation is due to local changes of the surfaceparameters as given in Fig. 3, whereas there are nosignificant changes in the cloud fields. In the northof the area under consideration, we find two limita-tions: first, the meteorological conditions are highlyvariable during the wintertime. This may cause ahighly irregular pattern of surface conditions trailingfrom the early spring into the summer. Results inthese regions are therefore expected to be less con-sistent. Second, the horizontal model resolution istoo coarse to allow firm conclusions about small-scaleremote differences in a highly heterogeneous region.

In the deforestation case colder temperatures arefound in the region of the Iberian peninsula, northernAfrica and the Middle East, while the differences inthe Adriatic are positive. A significant response tothe modification of land-surface characteristics ismainly found on the regional scale in the vicinity ofthe grid points where changes were made. The differ-ences reach a maximum of 18C and are therefore ofthe order of 20% of the projected impact of theadditional so-called greenhouse effect in that regionfor doubling of CO . In the afforestation case, the2

differences are generally of the opposite sign and arestatistically significant in the Mediterranean. Signifi-cant changes occur in the Sahara, where the anoma-lies are very similar to the patterns found by Reale

Ž . Ž .and Shukla 2000 in their higher resolution af-forestation experiment for an ensemble of summerintegrations.

The annual cycle of near-surface temperature isshown in Fig. 6 in terms of the differences betweenthe 10-year afforestation or deforestation experi-ments and the 25-year control simulation averagedover all land points that were modified in the experi-ments. For deforestation, there are fewer grid pointsthan for afforestation, as in the latter case, they also

Fig. 6. The seasonal cycle of monthly averaged 2-m-temperatureŽ .for: a the control simulation and the deforestation experiment,

Ž .b the control and the afforestation experiment. Bars indicate theinterannual variability as represented in the control simulation and

Ž .the experiments all using climatological sea surface temperatures .


Ž . Ž . Ž .Fig. 7. Evaporation averaged over the summer season JulyrAugustrSeptember for: a the control simulation, b the deforestationŽ .experiment, c the afforestation experiment.


comprise points at higher latitudes to the north of408N where almost no changes occur, and along the

Ž .Nile valley Fig. 4 . Fig. 6 shows an increase intemperatures for all months of the year. This isconsistent with the results shown in Table 1.

3.2. The hydrological cycle

Due to their critical importance for water re-sources in the region and the unique capability ofECHAM4 to simulate surface runoff from infiltrationexcess and drainage, we will pay particular attentionto the modification of the components of the hydro-logical cycle due to deforestation and afforestation inthe following. This will be limited to a regionalanalysis. Further details for sub-regions of the

Ž .Mediterranean are given by Ließ and Dumenil 1999 .¨

3.2.1. EÕaporationThe most immediate response to vegetation

changes is found in the evaporation. Deforestationgenerally reduces the evaporation in the summerŽ .Fig. 7 , where evaporation is already low. Suchregions are the Pyrenees and the whole Iberianpeninsula, the Balkan coast, Greece, Turkey and theMiddle East. Especially over the Iberian peninsula, adistinct decrease of evaporation is caused by defor-estation and an increase is found following afforesta-tion. Changes in northern Africa are restricted to thewestern parts of the Mediterranean coast, in particu-lar the Atlas mountains. In the control simulation,evaporation is at a maximum in July in this regiondue to the erroneously high precipitation simulatedby the model as discussed above. When the vegeta-tion is removed, evaporation is reduced. In general,the region north of 458N shows a less consistentresponse due to the dominance of the mid-latitudeflow.

In the afforestation case, evaporation increases inthe Pyrenees, while evaporation in the Atlas moun-tain range is hardly affected because the experimen-tal changes applied here are relatively smaller. Largedifferences also occur over modified land points inthe corridor between the Black Sea and Caspian Sea.

Ž .In the annual cycle Fig. 8 , there is a clear differ-ence in the evaporation between the deforestationand afforestation cases, with the former showing amajor decrease in the summer months and the lattera smaller but persistent increase throughout the year.

Fig. 8. As in Fig. 6 but for evaporation.

3.2.2. PrecipitationWhile the Mediterranean region is generally clas-

Žsified as a winter-rainfall climatic regime Koppen,¨.1931 , it shows a large heterogeneity of climatic

flows and conditions mainly due to the orographicstructure of the terrain. In each of the sub-regions,the effects of deforestation and afforestation willtherefore be somewhat different. There is also aninteraction with model systematic errors that have aregional structure.

The regional patterns of the precipitation changeduring the summer months are shown in Fig. 9. Ingeneral, the afforestation and deforestation experi-ments change the precipitation over the Mediter-


Fig. 9. As in Fig. 5 but for precipitation.

ranean region in opposite ways. The strongest de-crease in precipitation due to deforestation is foundin the Atlas mountain range. In the European part ofthe region, the greatest decrease occurs over theIberian peninsula and along the Balkan coast. All ofthese differences are significant. In the afforestation

case, the simulated response in the Mediterraneanregion is smaller and of little statistical significance.

The annual cycle of monthly mean changes ofsimulated precipitation is shown in Fig. 10. Eachtime series is an average over all land points whereland surface characteristics were modified. Although


Fig. 10. As in Fig. 6 but for precipitation.

the ECHAM4 model’s simulated precipitation hasŽsystematic errors with respect to the observed see

.Fig. 2 , there is a predominant decrease of summerprecipitation in the deforested case and a smaller butpersistent increase of precipitation throughout theyear in the afforested case.

The erroneously large rainfall in the summer isdue to a few land points in the Atlas mountain rangein northern Africa. Here, a large amount of water isstored in the interception reservoirs of the vegetationand is re-evaporated later during the annual cycle.This creates local rainfall that is unrealistic. Thelarge amount of vegetation at several grid points in

the ECHAM4 model in North Africa is not sup-ported by the latest state-of-the-art vegetation datasets

Žobtained from satellite measurements Hagemann et.al., 1999 . An indication of what effects a smaller

vegetation fraction would have is found in the defor-estation experiment, where the erroneously highroughness of the terrain and the recycling of water inthis region is reduced and leads to better agreementwith observed precipitation.

In the afforestation experiment, the region wherechanges were made to the land surface charac-teristics is different, and also comprises points in

Ž .mid-latitudes Fig. 3 . This is reflected in a lesspronounced range of the observed annual cycle of

Ž .precipitation Fig. 10 of the average of all modifiedland points. Precipitation increases in both winterand in summer, but larger differences occur in thesummer. In the Mediterranean the response of thehydrological cycle is in general opposite to that ofthe deforestation experiment. The regional impact is

Ž .less pronounced Fig. 9 , because the applied changesŽ .are more moderate Fig. 3 .

In the afforestation experiment, precipitationanomalies occur in the eastern central part of the

Ž .Sahara Fig. 9 . In the afforestation experiment apositive significant anomaly occurs near and down-stream from the modifications in the Nile valley.Due to the more restricted mask of changed landpoints, this anomaly has no counterpart in the defor-estation experiment. The anomaly patterns in theSahara agree with those found by Reale and ShuklaŽ .2000 and those in Fig. 5. They are consistent with

Ž .the arguments of Charney 1975 and Charney et al.Ž .1977 . Due to the moistening of the soil and theresulting change of the Bowen ration the surface

Ž .temperature in this region decreases Fig. 5 .

3.2.3. Surface runoffThe model-simulated surface runoff and drainage

for the deforestation experiment are shown in Fig.11, but there are no observations for either of thesequantities. The model-simulated runoff and drainagegenerally follow the precipitation, and can occurbefore the soil water reservoir reaches saturationŽ .Dumenil and Todini, 1992 . Runoff from thunder-¨storm events in the summer is not represented in themodel due to its coarse resolution. Therefore, the


Ž .Fig. 11. The seasonal cycle of monthly mean values of a surfaceŽ .runoff and b drainage for the deforestation and control simula-

tions averaged over deforested land points.

strongest impact on runoff occurs in winter, whenprecipitation exceeds evaporation, because soil waterreservoirs have been set to minimum values in the

Ž .experiment cf. Fig. 3 and runoff can occur morefrequently. This result is, of course, model-depen-dent and has to be seen against the background ofthe systematic model errors discussed above. The

Ž .drainage Fig. 11b shows similar behaviour, but thechanges are smaller.

The changes of surface runoff and drainage causedŽ .by afforestation not shown are generally smaller

than those caused by deforestation because of the

relatively small increase in maximum soil moistureas compared to the control simulation.

4. Conclusions

The general circulation model ECHAM4 was usedto identify the sensitivity of the local and globalclimate to desertification and afforestation in theMediterranean. The deforestation scenario representsan extreme desert, which would be the worst casescenario for this region. The changes in the afforesta-tion experiment are spread over a relatively largerarea according to the pattern of the vegetation cover2000 years before present as given by Reale and

Ž .Shukla 2000 , and include some points at mid-lati-tudes.

After deforestation, evaporation shows the largestsignificant response in terms of a decrease, while thelocal climate is subjected to a slight cooling at thesurface. Locally produced summer precipitation isreduced consistent with less evapotranspiration andless bare soil evaporation, but the results vary re-gionally. In general, the results of the afforestationexperiment are opposite in sign to those found in thedeforestation case, but the response is smaller be-cause the modifications are smaller than in the defor-estation case. The response in the Sahara in the caseof afforestation is very similar to the response found

Ž .by Reale and Shukla 2000 and is statistically sig-nificant.

In general, our model results agree with the re-Ž .sults of Reale and Shukla 2000 indicating that the

larger vegetation cover during the Roman ClassicalPeriod may have allowed more precipitation to formlocally in the summer, and that the vegetation mighthave been sustained by this precipitation. The modelsensitivity also indicates that land use change overthe centuries may have had an effect on near surfacetemperature in some regions of the Mediterranean,which should be taken into account if long-termtimes series of observations are examined.

The specific parameterization schemes used in ourmodel make the results model-dependent. While theECHAM4 model at T42 resolution is, in principle,capable of representing an atmospheric circulationresembling the observed, model systematic errors inthe winter circulation in the domain of this study


cause an unrealistically dry climate in the Mediter-ranean area and too much precipitation in the Atlasmountain region. Nevertheless, the GCM is capableof giving a consistent description of the change ofphysical processes at the land surface. Both themodel-simulated hydrological cycle and surface en-ergy budget are influenced by deforestation or af-forestation in a consistent way. The model system-atic error of reduced winter precipitation in theMediterranean makes our results highly model-de-pendent: If there is less than observed precipitationin the winter, the soil will be too dry in the earlyspring. This may suppress the local evaporation and

Ževapotranspiration re-evaporation of rainfall inter-cepted by the vegetation is not affected, because of

.small precipitation levels . This may affect the rela-tive importance of the hydrological changes relativeto radiation responses driven by albedo change. Sincethe evaporation and albedo changes tend to exertopposite effects, suppressing one, but not the other,could affect the magnitude or even the sign of thenet response.

We should also note that the results of sensitivitytests depend crucially on the climate conditions ofthe region where they are performed. In experimentaldeforestation simulations for humid regions such asthe tropical rainforests, the response is quite differ-

Žent. There, the surface temperature increases e.g..Lean et al., 1996 because the reduction of evapora-

tion more than compensates the changes of net solarradiation and a net heating is available for heatingthe soil. Such results have also been shown to bemodel-dependent, but it is reassuring to see that theresults for Amazonian deforestation found by otherswere reproduced in a sensitivity experiment with theECHAM4 model.

In parallel to the present GCM study, a regionalstudy of deforestation in the Mediterranean has beenundertaken with the PROMES and HIRHAM re-

Ž .gional models Polcher et al., 1999 in order to studyif more realistic levels of internal variability have aninfluence on the model response. In addition work isin progress to investigate how a more realistic de-scription of the winter circulation would change theresponse found in the present experiments.

Considering the high sensitivity of the model andits tendency to create self-sustaining conditions as inthe Atlas mountain range, it is of utmost importance

that the correct vegetation boundary conditions bespecified when present-day climate conditions are tobe simulated. To minimize unrealistic feedback inthe model, new surface parameters derived fromsatellite data have been specified for future ECHAMand HIRHAM model development in order to repre-sent the local present-day climate more faithfullyŽ .Hagemann et al., 1999 .

Acknowledgements

The study presented here is part of the EU-ProjectALand-Surface Processes and Climate ResponseB No.ENV4-CT95-0112-PL950189. We would like tothank Oreste Reale for providing the data set used inthe afforestation experiment, Yongkang Xue for pro-viding the soil parameters from SSiB. We thankLennart Bengtsson and Hartmut Graßl for their sup-port and the staff at MPI for technical assistance andscientific discussions. We are grateful to RichardBetts and an anonymous reviewer for their valuablecomments. The re-analyses data were supplied byECMWF and DWD in cooperation with DKRZ.

References

Abramopoulos, F., Rosenzweig, C., Choudhury, B., 1988. Im-proved ground hydrology calculations for global climate mod-

Ž .els GCMs : soil water movement and evaporation. J. Clim. 1,921–941.

Bengtsson, L., Arpe, K., Roeckner, E., Schulzweida, U., 1996.Climate predictability experiments with a general circulationmodel. Clim. Dyn. 12, 261–278.

Bolle, H.J. et al., 1993. EFEDA: European field experiment in adesertification-threatened area. Ann. Geophys. 11, 173–189.

Charney, J., 1975. Dynamics of deserts and drought in the Sahel.Q. J. R. Meteorol. Soc. 101, 193–202.

Charney, J., Quirk, W.J., Chow, S.H., Kornfield, J., 1977. Acomparative study of the effects of albedo change on droughtin semi-arid regions. J. Atmos. Sci. 34, 1366–1385.

Dumenil, L., Todini, E., 1992. A rainfall-runoff scheme for use in¨Ž .the Hamburg climate model. In: O’Kane, J.P. Ed. , Advances

in Theoretical Hydrology, vol. 1. Elsevier, Amsterdam, pp.129–157.

Gash, J.H.C., Nobre, C.A., Roberts, J.M., Victoria, R.L., 1996.An overview of ABRACOS. In: Gash, J.H.C., Nobre, C.A.,

Ž .Roberts, J.M., Victoria, R.L. Eds. , Amazonian Deforestationand Climate. Wiley, Chichester, UK, pp. 1–15.

Gibson, J.K., Kallberg, P., Uppala, S., Hernandez, A., Nomura,Ž .A., Serrano, E., 1997. The ECMWF Re-Analysis ERA : 1.

ERA description. ECMWF Re-Analysis Project Report Series


No. 1. European Centre for Medium-Range Weather Fore-casts, Reading, UK, 71 pp.

Hagemann, S., Botzet, M., Dumenil, L., Machenhauer, B., 1999.¨Derivation of global GCM boundary conditions from 1 kmland use satellite data. Report No. 289, Max-Planck-Institut fur¨Meteorologie, Hamburg, 34 pp.

Hulme, M., Conway, D., Jones, P.D., Barrow, E.M., Jiang, T.,Turney, C., 1995. Construction of a 1961–90 European clima-tology for climate change, modelling and impacts applications.Int. J. Climatol. 15, 1333–1363.

Huntley, B., Birks, H., 1983. An Atlas of Past and Present PollenMaps for Europe: 0–13,000 Years Ago. Cambridge Univ.Press, Cambridge, 667 pp.

Ž .Huntley, B., Webb III, T. Eds. , 1988. Vegetation History.Handbook of Vegetation Science, vol. 7. Kluwer AcademicPublishing, Dordrecht, 803 pp.

IPCC, 1990. Scientific Assessment of Climate Change. CambridgeUniv. Press, Cambridge, UK, 366 pp.

IPCC, 1995. Climate Change 1995—The Science of ClimateChange. Cambridge Univ. Press, Cambridge, UK, 572 pp.

Koppen, W., 1931. Grundriss der Klimakunde. Die Klimate der¨Erde. W. de Gruyter and Co., Berlin, 388 pp.

Lean, J., Bunton, C., Nobre, C., Rowntree, P., 1996. The simu-lated impact of Amazonian deforestation on climate usingmeasured ABRACOS vegetation characteristics. In: Gash, J.,

Ž .Nobre, C., Roberts, J., Victoria, R. Eds. , Amazonian Defor-estation and Climate. Wiley, Chichester, UK, pp. 549–576.

Ließ, S., Dumenil, L., 1999. The response of ECHAM to the¨Ž .land-surface change. In: Polcher, J. Ed. , Final Report for EU

contract ALand-surface Processes and Climate ResponseB.LMD, Paris, 235 pp.

Olson, J., Watts, J., Allison, L., 1983. Carbon in live vegetation ofmajor world ecosystems. Report ORNL-5862, Oak Ridge Na-tional Laboratory, Oak Ridge, TN, 397 pp.

Patterson, K., 1990. Global distributions of total and total-availa-ble soil water-holding capacities. Dissertation, Dept. of Geog-raphy, University of Delaware, 119 pp.

Polcher, J. et al., 1999. Final Report for EU Contract ALand-surfaceProcesses and Climate ResponseB. LMD, Paris, 235 pp.

Ramanathan, V., Cess, R., Harrison, E., Minis, P., Ahmad, B.B.E.,Hartmann, D., 1989. Cloud radiative forcing and climate:results from the Earth Radiation Budget Experiment. Science243, 57–63.

Reale, O., Dirmeyer, P., 2000. Modeling the effects of vegetationon Mediterranean climate during the Roman Classical Period.Part I: Climate history and model sensitivity. Global and

Ž .Planetary Change 25 3–4 , 163–184.Reale, O., Shukla, J., 2000. Modeling the effects of vegetation on

Mediterranean climate during the Roman classical period. PartŽ .II: Model simulation. Global Planet. Change 25 3–4 , 185–

214.Roeckner, E., Arpe, K., Bengtsson, L., Christoph, M., Claussen,

M., Dumenil, L., Esch, M., Giorgetta, M., Schlese, U.,Schulzweida, U., 1996. The atmospheric general circulationmodel ECHAM4: Model description and simulation of pre-sent-day climate. Report No. 218, Max-Planck-Institut fur¨Meteorologie, Hamburg, 90 pp.

Rother, K., 1993. Der Mittelmeerraum. Teubner Studienbucher¨der Geographie, Stuttgart, 212 pp.

Rowntree, P., Dumenil, L., 1995. Hydrology in climate models¨Ž .and effects on climate. In: Oliver, H., Oliver, S. Eds. , The

Role of Water and the Hydrological Cycle in Global Change.Springer Verlag, Berlin, pp. 59–104.

Shukla, J., Mintz, Y., 1982. Influence of land-surface evapotran-spiration on the Earth’s climate. Science 215, 1498–1501.

Xue, Y., Sellers, P., Kinter, J., Shukla, J., 1991. A simplifiedbiosphere model for global climate studies. J. Clim. 4, 345–360.

Impacts of deforestation and afforestation in the ...

Documents

Transcript of Impacts of deforestation and afforestation in the ...