Comparative probabilistic assessment of the hydrological performance of reconstructed and natural...

HYDROLOGICAL PROCESSESHydrol. Process. 24, 1333–1342 (2010)Published online 17 February 2010 in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/hyp.7596

Comparative probabilistic assessment of the hydrologicalperformance of reconstructed and natural watersheds

Nader Keshta,1 Amin Elshorbagy1* and Lee Barbour2

1 Center for Advanced Numerical Simulation (CANSIM), Department of Civil and Geological Engineering, University of Saskatchewan, Saskatoon,SK, Canada S7N 5A9

2 Department of Civil and Geological Engineering, University of Saskatchewan, Saskatoon, SK, Canada S7N 5A9

Abstract:

The oil sands industry has committed to returning the mine sites to a productive condition. The reconstructed soil covers musthave sufficient available water holding capacity (AWHC) to supply enough moisture over the growing season, to promotevegetation. In order to assess the sustainability of various soil cover alternatives, a generic, system dynamic watershed modelentitled GSDW was used along with the available historical meteorological records to estimate the maximum soil moisturedeficit and annual evapotranspiration fluxes. A probabilistic framework was adopted; consequently, frequency curves of themaximum annual moisture deficit values are constructed and used to assess the probability that various reconstructed andnatural watersheds can provide the associated moisture demands. In general, the study showed a tendency for the reconstructedwatershed to provide less moisture for evapotranspiration than natural systems. Watersheds of various soil types, layering,thicknesses and topography were studied. The gained knowledge was used to predict the possible performance of a hypotheticalreclamation cover. The results indicated that the hypothetical cover performed in a similar manner to the thickest existing soilcover which confirmed a high probability of that cover to survive under the same existing climatic conditions. Moreover, thisprobabilistic framework was found to be useful for integrating information gained from natural watersheds (e.g. the canopy ofmature natural systems and transfer the results to the reconstructed system). The results show that the canopy influenced themoisture deficit regime positively which signifies a greater possibility that reconstructed covers will adapt to vegetation type.In brief, the adopted approach enables better understanding of the response of reconstructed systems via multiple simulationsof ‘what-if’ scenarios using different soil/vegetation alternatives. Copyright 2010 John Wiley & Sons, Ltd.

KEY WORDS system dynamics; reconstructed watersheds; probabilistic approach; hydrologic models

Received 11 May 2009; Accepted 30 November 2009

INTRODUCTION

Oil sands in Alberta, Canada, were discovered beneatha complex boreal forest ecosystem comprised of aunique mosaic of forest, wetlands and lakes. Canada’sboreal forest is globally significant, representing one-quarter of the world’s remaining intact forests (Lee et al.,2003). In addition to the ecosystem services it provides,such as water cleansing, storing carbon and releasingoxygen, it is also home to a wide variety of wildlife.In order to gain access to the oil-bearing formations, theoverburden material is removed and stockpiled and thensalvaged topsoil is placed over it. The mining processresults in a large disturbance of ecological functions ofnature, such as the hosting of aquatic ecosystems andvegetation biomes (Haigh, 2000). The United NationsEnvironment Program has identified Alberta’s oil sandsmining projects as one of 100 key global ‘hotspots’ ofenvironmental degradation (UNEP, 2006). A measure ofthe level of potential environmental impact caused byoil sands mining is illustrated by the fact that one of the

* Correspondence to: Amin Elshorbagy, Center for Advanced NumericalSimulation (CANSIM), Department of Civil and Geological Engineering,University of Saskatchewan, Saskatoon, SK, Canada S7N 5A9.E-mail: [email protected]

largest daily producer of oil sand has disturbed 21 282 haof land since 1978 of which 4668 ha have been reclaimedover the same time period (Syncrude Canada Ltd., 2007).According to Alberta Environment, development of theoil sands presents overwhelming challenges for borealforest conservation and reclamation (Norah et al., 2001).The restoration of disturbed watersheds is mandatory togain government consent to commence mining activity.With the current rate of oil sands industry expansion,there is also a need to consider the combined andcumulative impacts arising from mining activities (e.g.future climate change and greenhouse gas emissions).

The emphasis in the very early years of land recla-mation was simply on ‘tidying up’. More recently, theemphasis has shifted to returning the land to produc-tive use. Sustainable land reclamation strategies employengineering measures to contain and control problemscaused by the disruption of natural systems. The processof re-establishing the disturbed landscape and developingsustainable soil–vegetation–water interactions to achieveland capability corresponding to the undisturbed con-dition is called land reclamation (Gilley et al., 1977).Selecting the texture, thickness and final landscape isvital for any reclamation scenario. The oil sands indus-try has adapted a method of evaluating reclamation

Copyright 2010 John Wiley & Sons, Ltd.

1334 N. KESHTA, A. ELSHORBAGY AND L. BARBOUR

soil prescriptions referred to as the Land CapabilityClassification System (LCCS) for forest ecosystems inthe oil sands (Leskiw, 2004). The LCCS classifies soilsinto several groups of ecosites based on available waterholding capacity (AWHC) to identify the soil moistureregime for a specific management scenario. The AWHCis the volume of water stored within the rooting zone asrepresented by the difference in water content betweenfield capacity and wilting point. The sites under consid-eration in this paper were targeted for upland forest (class‘d’ ecosites), and accordingly have a target AWHC valueof 160 mm.

More recently, oil sands creators have been consideringthe use of a risk-based method of managing environ-mental risks. The key to this approach is developing along-term plan to reduce the impact of projects by ensur-ing that the disturbed lands are returned to a stable andsafe condition that is capable of supporting biologicallyself-sustaining communities of plants and animals (Qual-izza et al., 2004). The hydrological performance of thesereconstructed watersheds needs to be assessed by cre-ating a landscape that sustains an integrated mosaic ofland uses. The performance is evaluated by the ability ofthose landscapes to store and release soil moisture dur-ing the growing season to support the vegetation duringthe growing season (Haigh, 2000). The large scope ofsustainable watershed management requires a commit-ment to continuous learning and improvement throughthe system known as adaptive management (Noss, 1993).Salwasser et al. (1993) quoted that the desired outcomeof any sustainable forest (watershed) management is abalance among different conditions that are economicallyviable, ecologically feasible and socially accepted.

The probability that a plant is in a healthy state isequivalent to the probability that its response, measuredby the soil moisture deficit, is less than a value repre-senting the maximum soil moisture deficit (Mukherjeeand Kottegoda, 1992). The values of soil moisture deficitand actual evapotranspiration (AET) vary with soil typeand soil texture (Shaw, 1994). A similar methodologyfor assessing the hydrological performance of the recon-structed soil covers was adopted by Elshorbagy and Bar-bour (2007) to assess the risk of failure of proposedreclaimed (reconstructed) sites under various climaticscenarios. Elshorbagy and Barbour (2007) used a cali-brated site-specific system dynamics watershed (SDW)model (Elshorbagy et al., 2005; Jutla, 2006), along withthe available historical meteorological records, to esti-mate the maximum soil moisture deficit the soil covercan sustain during the growing season. Frequency curvesfor maximum annual moisture deficit were used to assessthe probability that each cover can provide the desiredthreshold of moisture demand. This essentially createsa risk-based representation of the AWHC concept. Theadopted probabilistic approach can be used to evaluatethe life cycle costs of different proposed cover alterna-tives (thicknesses and texture). A similar analysis is quiteuseful in the design and formation of newly reconstructedwatersheds. A similar probabilistic approach was adopted

by Candela et al. (2005) to evaluate the effect of forestfires on flood frequency curves. Candela et al. (2005)adopted a methodology that was found to provide a use-ful framework for detecting changes in flood magnitudesin both pre- and post-fire conditions.

The aim of this paper is to build on Elshorbagy andBarbour’s (2007) probabilistic approach using a genericsystem dynamics watershed (GSDW) model developedby Keshta et al. (2009). The specific objectives of thispaper are: (i) to evaluate the current and future hydrologi-cal performance of existing reconstructed watersheds; (ii)to introduce an approach that allows for testing and opti-mizing the design of reconstructed soil cover alternativesbased on the knowledge gained from modelling the exist-ing reconstructed sites and (iii) to compare the hydro-logical performance of reconstructed watersheds withnatural watersheds, based on this comparison, knowl-edge can be transferred to assess the sustainability of therestored forests on the newly reclaimed sites. In orderto establish the link between reconstructed and naturalwatersheds, the probabilistic approach application willbe implemented on two natural and five reconstructedwatersheds.

CASE STUDIES

Reconstructed watersheds

An oil sands mining company operating north of FortMcMurray (57°390 N and 111°130W), northern Alberta,Canada, has been conducting experiments on its recon-structed watersheds (alternatively called experimentalcovers). The first location has three experimental cov-ers (D1, D2 and D3) of different thicknesses (0Ð5, 0Ð35and 1Ð0 m) that were constructed in 1999 over the saline-sodic overburden. Each cover comprised of a thin peatmineral mix layer (0Ð15–0Ð20 m thick) overlying a layerof glacial till. Each cover has an area of 1 ha (approx-imately 200 m long and 50 m wide), with a slope of5H:1V (Hill slope) and was initially sown with a barleynurse crop (Hordeum Jubatum) followed by plantings ofwhite spruce (Picea glauca) and aspen (Populus tremu-loides) tree seedlings (Boese, 2003).

The second location is a horizontal reconstructed coversystem, which is a located on the South Bison Hill (SBH).It was constructed in 2001 of 0Ð2 m of peat/mineral mixoverlying 0Ð8 m of glacial till. The area of the site isapproximately 2 km2, and is elevated 60 m above thesurrounding landscape (Hill top). The SBH vegetationis dominated by foxtail barley (Hordeum Jubatum) andminor species including fireweed (Epilobium angusti-folium) (Parasuraman et al., 2007).

The third location is the South West Sand Stor-age (SWSS), which was constructed of 0Ð2–0Ð4 m oftill/secondary cover material overlaying tailings sands.It is currently the largest operational tailings dam in theworld at approximately 40 m high. The vegetation variesfrom horsetail (Equisetum arvense), fireweed (Epilobiumangustifolia) and white and yellow clover (Melilotus

Copyright 2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 1333–1342 (2010)

HYDROLOGICAL PERFORMANCE OF RECONSTRUCTED AND NATURAL WATERSHEDS 1335

alba, Melilotus officinalis). The site was also plantedwith tree species including hybrid poplar (Populus sp.hybrid ), trembling aspen (Populus tremuloides), whitespruce (Picea glauca) and willow (Salix sp.) (Parasur-aman et al., 2007).

Intensive field instrumentation was established to mea-sure the meteorological and soil characteristics on thesesites from 1999 to date. The precipitation (mm), airtemperature (°C), net radiation (W/m2), soil temperature(°C) and volumetric soil moisture content measurementsare available four times daily. More details on the sitedescription and collected data are found in Boese (2003)and Barbour et al. (2004).

Natural watersheds

The study areas with natural conditions are wellinstrumented and part of the area of the former BorealAtmosphere Exchange Study (BOREAS), which coversa large portion of Saskatchewan and Manitoba with anarea of 1000 km ð 1000 km. This region includes youngand old aspen, old jack pine and old black spruce forests.Two natural sites are considered in this study. The firstis the Old Aspen site (OA) located near the south endof Prince Albert National Park, Saskatchewan (53Ð63 °N,106Ð19 °W; 600Ð6 m). The field instrumentation of theOA site has been providing continuous measurementssince 1997 as part of the Boreal Ecosystem Researchand Monitoring Sites (BERMS) program. The soil iswell drained loam to clay loam. The top 0Ð1 m is anorganic layer (leaf litter, plus fermentation layer) andthe underlying mineral soil consists of an upper layer of0Ð07–0Ð3 m of till mixed with sand and clay, overlying alayer of 0Ð45 m derived from gravely and clay enrichedtill. The forest canopy is dominated by trembling aspenwith an average height of 21 m and the understory canopyconsists of 2 m height beaked hazelnut (Corylus cornuta)(Balland et al., 2006). Thermocouple sensors are used tomeasure soil temperature every 30 min at depths of 0Ð02,0Ð05, 0Ð10, 0Ð20, 0Ð50 and 1Ð00 m below the moss layer.Campbell Scientific CS615 soil moisture probes are usedto measure the volumetric moisture content of the soil at0Ð08, 0Ð23, 0Ð45 and 1Ð05 m below the ground surface.Net radiation (NR) is measured using Middleton CNR-1net radiometer above the canopy. Measurements of thelatent heat fluxes are made with the eddy covariance(EC) technique and reported at 30 min intervals. Leafarea index (LAI) is measured near the flux tower usinga plant canopy analyser (PCA) (model LAI-2000) andfound to be 5Ð5 m2/m2. Additional information regardingthe saturated hydraulic conductivity and the soil waterretention function can be obtained from Cuenca et al.(1997).

The second site is the Old Jack Pine (OJP) water-shed, located north-east of Prince Albert, Saskatchewan(53Ð91 °N, 104Ð69 °W; Elev. 579Ð27 m) (BOREAS coor-dinates). The site is dominated by mature jack pine (Pinusbanksiana Lamb) established in 1914 with heights from12 to 15 m. The understory consists of sparse green alder

(Alnus crispa) and predominantly lichens (Cladina spp.)ground cover. The canopy LAI is 1Ð7 m2/m2. The soil iswell drained sand of 1Ð05 m in depth. The field instru-mentation of the OJP site has been providing continuousmeasurements from 1998 to 2002. The soil moisture mea-surements are taken every 4 h, at depths of 0Ð15, 0Ð30,0Ð60, 0Ð90, 1Ð20 and 1Ð50 m below ground surface. Ther-mocouple sensors are set to measure the soil temperatureevery 30 min, at depths of 0Ð02, 0Ð05, 0Ð10, 0Ð20, 0Ð50and 1Ð00 m below ground surface. Meteorological mea-surements at both natural sites are sampled every 5 s,and outputs are averaged every 30 min, year round andare available from BOREAS/BERMS database (Ballandet al., 2006). The data was aggregated in a daily scalebases for executing the GSDW model.

HYDROLOGICAL MODELLING

Model description

The GSDW model is a mechanistic, semi-empirical,lumped watershed model capable of simulating variouscomponents of watershed hydrology, such as canopyinterception, evapotranspiration, surface runoff, lateralinterflow, infiltration and soil moisture redistribution inunsaturated/saturated layers, based on the surface energyand water balances. The model requires daily meteoro-logical data (e.g. precipitation, air temperature, relativehumidity, wind speed and net radiation), plant commu-nity data (LAI and canopy storage), soil properties andrudimentary site descriptors (e.g. field capacity, porosity,saturated water content, residual moisture content, vanGenuchten empirical parameters for suction calculations,gradient, texture and layer thicknesses). This model isan upgrade/generalization of a site-specific SDW modelwhich was initially developed by Elshorbagy et al. (2005)and Jutla (2006). The main drawbacks of the previousSDW model version were that it had no flexibility inchoosing the number of soil layers, thickness and topo-graphic inclination. Also, soil water characteristic curves(SWCC) were pre-requisite for the SDW model to iden-tify movement of soil moisture in unsaturated condi-tions. The modified version of the model was adapted toselect the number of soil layers, stratification and topo-graphic inclination of the site. A provision was madein the model for interflow and runoff components toaccount for the effect of topography (surface slope). Thevan Genuchten (1980) equation was implemented in themodel to describe the SWCC. The SDW model did notconsider canopy interception in spite of the importanceof canopy interception losses, which can vary from 10 to40% of the total precipitation, depending on plant com-munities (Dingman, 2002). A canopy interception modulewas developed and added to the GSDW model to over-come this problem.

The system dynamics (SD) simulation environment(STELLA) (HPS, 2001) was used to model the watershedas a dynamic system, based on the concept of stock-flow, in a user-friendly environment. The SD approach



Figure 1. A flow chart of the procedure of which the GSDW model follows to accommodate various hydrological processes in top layer

tolerates building a tentative knowledge of the relationbetween two parameters by incorporating a qualitativerelationship between those parameters, as well as havingthe potential of implementing a combination of empiricalformulations and physically based concepts (Elshorbagyet al., 2007). The model uses empirical (fitted-parameter)formulations to calculate AET, van Dijk and Bruijnzeel(2001) analytical model for canopy interception and thevan Genuchten (1980) equation to account for matricsuction–soil moisture relationship. Also, the conceptualmodel of Valente et al. (1997) was incorporated intothe model as another alternative to simulate interceptionlosses. In addition to the previously mentioned empiri-cal equations, a set of physically based formulations wasincorporated to the model (e.g. Green-Ampt for infiltra-tion and soil moisture redistribution; Penman equationfor potential evapotranspiration). Figure 1 shows a flowchart of the procedures that the GSDW model follows

to simulate the various hydrological processes in the toplayer. The subsequent layers follow the same proceduresdescribed for the top layer subject to the constraint that nodownward moisture movement is allowed if the suctionof the upper layer is greater than that of the lower layer.More details of the model formulations can be found inKeshta et al. (2009).

Methodology

The hydrologic performance of reconstructed water-sheds is assessed based on their ability to store and releaseenough moisture to maintain land–atmospheric fluxesand vegetation growth. The calibrated and validatedGSDW model was used to conduct long-term hydrolog-ical simulations using a period of 50 years (1953–2002)of historical meteorological input data. The long-termmeteorological dataset included the consequences of highprobabilities of dry or wet climatic conditions.



The GSDW model was manually calibrated by adjust-ing the model parameters and executing a series of sim-ulations until an acceptable level of match is obtainedbetween the observed and simulated soil moisture val-ues and evapotranspiration fluxes. Model error measures(e.g. the mean absolute relative error (MARE), correla-tion coefficient (R), and root mean squared error (RMSE))were used to check the simulation accuracy in addition tovisual inspection of soil moisture patterns and evapotran-spiration fluxes (Keshta et al., 2009). It was found that themodel was sensitive to lambda coefficients, a main fac-tor in the AET formulation, and infiltration coefficients,which directly affect the moisture distribution in each soillayer. In the GSDW model, soil moisture movement isrestricted to or from that layer once the soil temperaturein this particular layer is 0 °C or below. The historical setof data offered by the Environment Canada website doesnot include soil temperature values for the modelled sites,whereas soil temperature measurements are needed in theGSDW model to track soil moisture movement betweenlayers. The literature suggests that there is a high correla-tion between the air temperature and ground temperaturefor the skin layer (Zheng et al., 1993; Paul et al., 2004).In this study, a Genetic Programming (GP) model wasdeveloped as a tool for predicting the soil temperaturebased on a set of measured data during the extensivemonitoring program for each site. Monitored values ofprecipitation and air temperature were used along withthe GPLAB, a genetic programming toolbox for MAT-LAB, developed by Silva (2003). The total number ofgenerations was 200 with a population size of 25 mem-bers to simulate the top layer soil temperature. For thesubsequent layer, the same set of input data was usedalong with the preceding soil temperature and the GPLABto predict soil temperature for this layer. The deducedequations from GPLAB for each layer were used to sim-ulate the soil temperature for the long-term simulationperiod for that layer.

The probabilistic framework proposed by Elshorbagyand Barbour (2007) was adopted to assess the proba-bility of various reconstructed and natural watershedsto provide the desired moisture threshold over 50 yearsof simulations. For each soil cover, the daily volumet-ric moisture content of the cover layers was predictedusing the validated GSDW model along with historicalmeteorological records. Soil moisture content values ineach cover were converted to soil moisture content (mm)stored in each layer by multiplying those values by therespective thicknesses of the soil layer in each cover. Thesoil moisture values of all layers within the root zone ofa soil cover were added to get the daily stored soil watervolume within each cover (St). The daily moisture deficit(Dt), which could be attributed to evapotranspiration, wasthen calculated as follows:

S D St � StC1 �1�

Dt D S � �I C P� �2�

where S is the difference in soil moisture contentbetween the current day (St) and the immediate subse-quent day (StC1), I is the interflow, and P is percolationbelow the cover depth. A positive value of Dt indicatesthe depletion of soil moisture due to evapotranspiration.In other words, Dt represents the amount of water that thesoil cover stored and released for vegetation in time t. Thedaily Dt values over the growing season were accumu-lated, and the maximum cumulative Dt obtained in a yearis called maximum annual soil moisture deficit (Dm). TheDm values reflect the performance of the sub-watershedconsidering the wetness and dryness of the year as well asthe precipitation variability. A positive value of Dm rep-resents the maximum amount of moisture that was storedand released in any year. A negative value indicates thatthere was a water surplus, which is of less importanceregarding the design of the soil cover. The annual Dm

values, as well as the annual AET values (50 values rep-resenting 50 years) were treated as a random variableand probability distributions were derived and fitted toboth values for each soil cover with the help of @RiskBestfit 4Ð5Ð5 software (Palisade Corporation, 2005). Thefrequency curves were used to define a particular proba-bility of non-exceedance for moisture stored and releasedfor vegetation. This probabilistic assessment approachprovides an insightful visualization of the hydrologicalsustainability of the reconstructed watershed. In order toevaluate the performance of different vegetation alterna-tives and soil textures on the soil moisture deficit regime,and to establish a link between reconstructed sites andnatural watersheds, long-term simulations were carriedout on the natural watershed as well. The same, previ-ously described, probabilistic framework was adopted toobtain Dm and AET flux frequency curves for the naturalwatersheds.

In general, the long-term simulations were carriedout to help in understanding the relationship betweenreconstructed and natural systems. The generated sce-narios would help to address the following queries:‘Could we use the knowledge gained from existing recon-structed covers to suggest an approach to optimize futurereclaimed sites?’. Another query is, ‘What would be thehydrological effect of introducing mature canopies suchas those in mature natural watersheds to the reconstructedwatersheds?’. Long-term simulations on both the recon-structed and natural systems were carried out using thesame climatic data as for the reconstructed sites. Therewere two reasons for this approach: (i) both reconstructedand natural watersheds are located in a semi-arid regionhaving almost the same climate constrains and (ii) inensures that the comparative hydrological performanceanalysis of reconstructed watersheds and natural water-sheds are based on a common climatic basis. Owing tothe uncertainty in the estimation of maximum moisturedeficit from input data, model structure, model parametersand formulations, a margin of safety should be taken intoconsideration while relying on this probabilistic approachfor designing the soil covers.



RESULTS AND DISCUSSIONS

Hydrological performance assessment for reconstructedinclined watersheds

The maximum annual soil moisture deficit, Dm, wasobtained and treated as a random variable to fit thecumulative probability distribution curve. More than 15distributions were tested and the best fit distributionswere selected for each cover using Chi-squared goodnessof fit test. The best distributions were found to beLogistic (˛, ˇ) for the D1 cover, Normal (�, �) forD2 cover and InvGauss (�, �) for the D3 cover. Inthe Logistic distribution, ˛ and ˇ are the continuouslocation and shape parameters. In the Normal distribution,� and � are mean and standard deviation, respectively.The parameters in the InvGauss, inverse Gaussian (orWald) distribution are mean (�) and shape parameter (�)(Palisade Corporation, 2005). The fitted distributions arepresented in Figure 2a–c for the covers D1, D2 and D3,respectively.

The details of the Dm distributions for the D1, D2 andD3 covers are as follows:

Soil cover D1 (50 cm); Logistic(52Ð1, 10Ð9)Soil cover D2 (35 cm); Normal(42Ð3, 15Ð8)Soil cover D3 (100 cm); InvGauss(351Ð1, 23998Ð6) Shift

D �271Ð6

The negative values of Dm indicate a water surpluswhere positive values indicate the ability of the soilcover to store-and-release moisture for vegetation. In theD3 cover, there are a few instances where the coverhas negative values of Dm, suggesting that water mayconsequently exit in the form of either interflow, ordeep percolation. The positive values of Dm contribute tothe water requirements such as vegetation requirementsand evaporation needs. The cover D1 is capable ofstoring and releasing up to 100 mm of moisture if certainclimatic conditions mandate so (99% non-exceedanceprobability). At the same time, the covers D2 and D3have Dm values of 79 and 189 mm, respectively, at 99%non-exceedance probabilities. Hence, when the covers areunder the stress of high water requirements, the D3 covermight be able to release the moisture needed while coversD1 and D2 may fail to release adequate moisture. The D3cover able to store and release up to 135 mm of moistureat a non-exceedance probability of 90%, indicating thatthis thickest cover may only be required to store-and-release this amount of moisture ten times in 100 years.

To evaluate the performance of the GSDW model andits utility for the design of future soil covers, a hypothet-ical cover (Dhyp) was used which consisted of 20 cmof peat mineral mix layer overlying 50 cm of glacial till.Estimating the Dhyp model parameters was a challengingtask as there were no observed data to calibrate the model.To overcome this problem, a descriptive analysis for theexisting D-cover model parameters was conducted. Theresults showed a concurrence in selecting model param-eters and cover thickness, which affects the store and

Maximum moisture deficit (mm)


-20 0 20 40 60 80 100 120 140 160 180 200

Simulated distributionFitted distribution

0

0.2

0.4

0.6

0.8

1(a)

(b)

(c)

0 10 20 30 40 50 60 70 80 90 100


0 10 20 30 40 50 60 70 80

Non

-exc

eeda

nce

prob

abili

ty

0

0.2

0.4

0.6

0.8

1

Non

-exc

eeda

nce

prob

abili

ty

0

0.2

0.4

0.6

0.8

1

Non

-exc

eeda

nce

prob

abili

ty

Figure 2. Frequency curve of maximum moisture deficit of soil covers:(a) D1 cover, (b) D2 cover and (c) D3 cover

release abilities of each cover, particular as it affectsAET and infiltration-related parameters. Based on thisanalysis, the soil properties of Dhyp were consideredsimilar to the existing D-covers. Consequently, the cali-bration parameters were selected as averages of the D1and D3 calibration parameters. Figure 3 shows the prob-abilistic hydrological performance of D1, D2, D3 andDhyp watersheds relative to each others. The details ofthe Dm distribution for the hypothetical cover were foundto be BetaGeneral (˛1, ˛2, min, max). The Dm frequencycurve indicates that the maximum ability of this cover tostore and release moisture for vegetation is around 150and 110 mm at 99 and 90% non-exceedance probabili-ties, respectively. The D3 curve is situated to the right ofthe Dhyp indicating its higher ability to store and releasewater as compared to the Dhyp cover.

As mentioned in the introduction, the ultimate goal forany sustainable watershed management alternative is a



0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100 120 140 160 180 200


Non

-exc

eeda

nce

prob

abili

ty

D1D2D3Dhyp

220

Figure 3. Stochastic comparison between the hydrological performancesof the D1, D2, D3 and Dhyp sub-watersheds

balance between being ecologically feasible and econom-ically viable. The Dhyp cover soil moisture deficit regimeshowed that it has a comparable hydrologic performanceas the D3 cover, which suggests that this hypotheticalcover could survive under the same climatic conditionsas the existing D3 cover. Conversely, the Dhyp cover is30 cm thinner than the D3 cover and consequently wouldsubstantially reduce reclamation costs.

Hydrological performance assessment of reconstructedand natural horizontal watersheds

The best fit probability distribution of the Dm valuesfor the SBH (Hill top) was found to be a Weibull (˛,ˇ), where ˛ is the shape parameter and ˇ is scaleparameter. Figure 4 shows that the SBH soil cover iscapable of storing and releasing around 117 mm ofmoisture under the 99% non-exceedance probability, and88 mm at a non-exceedance probability of 90%. Forthe Natural OA site, the best distribution to fit the Dm

values was found to be Weibull (˛, ˇ), as well. The OAsite is capable of releasing around 182 mm of moistureat 99% non-exceedance probability, and 130 mm at anon-exceedance probability of 90%. Figure 4 shows theDm frequency curves of both SBH and OA (horizontal)sites. The reconstructed SBH cover appears to have alower store and release ability than the OA site. The

0

0.2

0.4

0.6

0.8

1

0 25 50 75 100 125 150 175 200 225 250

Maximum soil Moisture Deficit (mm)

Non

-exc

eeda

nce

prob

abili

ty

Old JacK Pine

SWSSSBH

Old aspen

Figure 4. Stochastic comparison between the hydrologic performances ofthe OA, SBH, OJP and SWSS sites

OA site was able to store and release 130 mm about10% of the time for the same climatic conditions. Thisobservation should be treated with caution for severalreasons: first, because of the uncertainties associated withthe model; second, because the canopy distribution canalter the land–atmosphere interaction by keeping the sub-soil moisture trapped under the canopy for longer period(Entekhabi et al., 1996) and third, the reconstructed SBHcould still be evolving toward more mature conditions.

The reconstructed SWSS and the natural OJP water-sheds were treated as comparable since both of the siteshave comparable soil textures being mostly comprisedof sand. The best fit probability distribution of the Dm

values for both sites was found to be a Normal (�, �),which is consistent with our initial assumption that theyare comparable. Figure 4 shows that the OJP seems ableto store and release around 190 mm of water at a 99%non-exceedance probability, whereas the SWSS site, wasonly able to store and release 160 mm of water underthe same meteorological conditions. This higher abilityto store and release moisture in both the OJP and SWSSsites is attributed mainly to the soil texture associatedwith the well drained sandy soil.

It is noteworthy that though both SBH (Hill top) andthe D3 cover (Hill slope), are similar in texture, thicknessand soil stratification; the SBH site has a tendency tostore and release less moisture than the D3 cover. Thisfinding could be attributed to the topographic inclinationof the D3 cover. This inclination affects the soil moisturemovement rates within the soil layer. It was also observedby field personnel that the SBH is poorly drained withfree water sometimes trapped in the soil profile.

The details of the Dm distributions for the OA, SBH,OJP, and SWSS sites are as follows:

SWSS site; Normal(61Ð2, 43Ð2)SBH site; Weibull(2Ð7, 70Ð1) Shift D �7Ð7OA site; Weibull(2Ð2080, 96Ð325) Shift D �10Ð373OJP site; Normal(71Ð7, 51Ð1)

Hydrological performance assessment of vegetationalternatives

The GSDW model was also used to establish a relationbetween the performance of the existing reclaimed sitesand natural sites with regard to their ability to hold andrelease soil moisture for different vegetation alternatives.To assess the comparative hydrological performance ofthose covers under the same climate and vegetation, twosimulation scenarios were generated for the reconstructedSBH watershed. In the first scenario, the mature veg-etation of the natural OA site was brought on to thereconstructed system (SBH site) by replacing the vegeta-tion model parameters of the reconstructed system withthe parameters of the natural system (e.g. LAI, maxi-mum leaf storage). The same procedure was repeated bybringing the OJP watershed vegetation parameters to thereconstructed system.



The long-term scenarios and consequent Dm frequencycurves give an initial idea of the ability of the recon-structed watershed to supporting different mature vege-tation, in terms of its long-term hydrological response.Figure 5 shows the Dm frequency curves of both SBH,OJP and OA sites along with the two generated scenar-ios of the reconstructed system with the natural systemvegetation alternatives.

The SBH soil cover was capable of releasing around117 mm of moisture at 99% non-exceedance probabil-ity, and 88 mm occurs at a non-exceedance probabil-ity of 90% with its existing immature vegetation. Byintroducing the OA canopy to the reconstructed site,the reconstructed system showed a Dm value of about150 mm at 99% non-exceedance probability and 110 mmat 90% non-exceedance probability with a value of LAI5Ð5 m2/m2. Whereas, the SBH site was capable of releas-ing 134 mm at 99% non-exceedance probability and96 mm at 90% non-exceedance probability with the OJPcanopy parameters with a LAI-value of 1Ð7 m2/m2. Theprevious Dm values indicated that the model performancewas correlated to the vegetation parameters of the naturalsystems. The evolution of the ability of the reconstructedSBH watershed to store and release water was enhancedusing mature vegetation. This supports the idea that thissite might modify its storing/releasing abilities with thedevelopment of vegetation over time. The Dm values forthe SBH site with the existing canopy and the SBH alongwith the natural system’s canopy were statistically tested.As the three sets of Dm values have different distribu-tions, a non parametric test was conducted on a multiplerelated sample using Kendall’s test for the analysis. Theanalysis showed a Kendall’s coefficient of concordanceof 0Ð212, a Chi-squared value of 21Ð16 with two degreesof freedom, and a p-value of zero at 95% confidencelevel. This shows a significant difference in the meansand the variance of the three Dm sets, which reflects sta-tistically significant differences in their moisture storingand releasing abilities.

The Dm distributions were fitted to demonstrate theeffect of using different vegetation alternatives for the

0

0.2

0.4

0.6

0.8

1

0 25 50 75 100 125 150 175 200 225 250

Maximum soil Moisture Deficit (mm)

Non

-exc

eeda

nce

prob

abili

ty

SBH with OJP canopySBH with OA canopySBHOld AspenOld Jack Pine

Figure 5. Stochastic comparison between the hydrological performancesof the SBH site using different vegetation alternatives

reconstructed SBH site. The details of the distributionsare presented as follows:

SBH site with OJP site canopy; Logistic(58Ð0, 17Ð2)SBH site with OA site canopy; Weibull(2Ð4, 90Ð1) Shift

D �17Ð8Hydrological performance assessment of AET fluxesfor different sites

Infiltration, soil moisture redistribution and AET arethe main hydrological processes affecting the behaviorof natural and reconstructed watersheds in arid and semi-arid regions where runoff is limited. Hydrological pro-cesses, such as soil moisture redistribution and AET, areintrinsically linked; therefore, the understanding of theirmutual interaction could lead to more accurate simula-tions of the water balance. In fact, vegetation dynamicsare controlled by climate and soil moisture, whereas veg-etation in turn modulates the total water balance (Arora,2002). Vegetation acts as the transitional link betweensoil and the atmosphere via the evapotranspiration pro-cess. As well, it affects the soil hydraulic and mechanicalproperties and surface energy budget (soil temperature)(Falkenmark, 1997).

The effect of vegetation was shown in the previoussection on the soil moisture regime of the reconstructedSBH site. In this section, the AET fluxes distributionwas constructed to identify the ability of each watershedto release water for evapotranspiration. The simulateddaily AET values were summed for each growing seasonand the total annual values were treated as a randomvariable out of the 50 years of simulation, and used to fitprobability distribution curves of AET fluxes.

Figure 6 shows the fitted distributions of the annualgrowing season AET fluxes. At 90% non-exceedanceprobability, transpiration at the natural watersheds (boththe OJP and OA sites) was 360 and 410 mm, respectively.For the reconstructed watersheds (SBH and SWSS)about 323 and 330 mm of moisture, respectively, wasreleased under the same meteorological conditions. Moreimportantly, under extreme conditions (e.g. 99% non-exceedance probability), the natural system may allow

0

0.2

0.4

0.6

0.8

1

150 175 200 225 250 275 300 350 325 375 400 425450 475 500

Evapotranspiration (mm)

Non

-exc

eeda

nce

prob

abili

ty

OJP

SWSSOA

SBH

Figure 6. Frequency curve of growing season evapotranspiration fluxes



for as much as 410 and 500 mm of evapotranspirationfor the OJP and the OA, respectively. The reconstructedsites provided around 360 mm of evaporative flux forboth sites. The reason for this difference is attributed tothe fact that the reconstructed system’s vegetation is stillimmature (mostly grass), whereas the vegetation for thenatural watersheds is mature. Also, the soil properties ofthe natural watershed are stable, while the reconstructedwatersheds are likely still evolving in terms of theirhydrologic behavior and properties.

The details of the AET probability distributions arepresented below:

SBH site; Normal(271Ð7, 39Ð9)OA site; Weibull(1Ð5416, 122Ð56) Shift D C200Ð10SWSS site; Weibull(2Ð4, 97Ð7) Shift D C192Ð8OJP site; Logistic(314Ð7, 21Ð1)

The OJP watershed was capable of storing and releas-ing more soil moisture than the OA watershed under highstress demands. This is only valid for the watershedsunder consideration, keeping in mind that soil textureof both sites are different and this may alter the moistureregimes. Moreover, Balland et al. (2006) mentioned thatthe OJP watershed canopy coverage provides a thermalinsulation to the soil during summer, which allows forbetter moisture storing abilities. To evaluate the effectof vegetation individually, the SBH long-term simulationwas revisited using the vegetation of the natural water-shed. The moisture regimes showed a response to thevegetation canopy of both natural systems with differentfrequencies under the same meteorological conditions.This difference in response was attributed to vegetationrelated parameters (e.g. LAI and leaf storage capacity). Ingeneral, the study showed that all the reconstructed water-sheds had less store-and-release abilities, and allow forless evapotranspiration fluxes than the natural watersheds.In addition, the analysis shows that the reconstructedwatersheds were able to respond positively by provid-ing more moisture for the mature vegetation. This couldreflect a reasonable success of the design of the currentreconstructed watersheds.

CONCLUSIONS

The concept of the hydrological failure of reconstructedwatersheds is redefined here because the performance ofthe watershed is subject to a number of hydrological andphysical factors that cannot justifiably be modelled as adeterministic function. These factors arise mainly fromthe following: precipitation, evapotranspiration and soiltexture. Due to the numerous factors that are involvedin reconstructed sites during the growing season, itis almost impossible to account for the influence ofeach factor individually. Therefore, the technique ofusing the maximum soil moisture deficit is used togeneralize the combined effect of all factors. In thisstudy, an attempt was made to assess the hydrological

performance of three reconstructed soil covers (35, 50and 100 cm thick) based on their soil moisture holdingcapacity using a system dynamics approach. Using long-term climatic data, the hydrological performance of thereconstructed covers was assessed with the help ofa probabilistic approach. The results indicate that thethickest soil cover is able to release more moistureunder high water demands, while the other two coversmay fail to release the same amount of moisture underhigh stress levels. The proposed hypothetical alternative(Dhyp) with intermediate thickness seems to releasesimilar amounts of moisture as the D3 cover. It wascapable of releasing 150 mm at 99% non-exceedanceprobability compared to 190 mm for the D3 cover. TheDhyp cover soil moisture regime is close to that ofthe thickest (D3) cover, which gives an indication thatthis hypothetical cover could survive under the sameclimatic conditions as the existing D3 cover. Adoptingthis hypothetical cover for future design purpose, oreven adopting the same probabilistic technique to suggestother cover alternatives, could enhance our knowledge inchoosing the optimum sustainable reclaimed cover whichis ecologically feasible and economically viable.

In general, the model also showed that the naturalsystems are storing/releasing moisture, especially underextreme conditions, better than the reconstructed systems.It is noteworthy that the SBH cover responded positivelyby the introduction of the canopy of a mature naturalsystem, suggesting that there is a possibility for thereconstructed covers to adapt to vegetation type. Oncethe level of protection to be provided by the margin ofsafety is specified by decision makers, this probabilisticapproach could be applied to verify the success of aparticular management action, e.g. the use of certainvegetation type, texture and landscape topography (flat orinclined). Obviously, the more information that is gainedfrom natural and reconstructed watersheds, with differentvegetation, textures, stratifications and thicknesses, theless uncertain the results will be. It is also recommendedthat the simulated sites be revisited in future to revalidatethe findings of this study.

ACKNOWLEDGEMENTS

The authors acknowledge the financial support of theNatural Science and Engineering Research Council(NSERC) of Canada, Cumulative Environmental Man-agement Association (CEMA), the Egyptian governmentscholarship program and the Department of Civil andGeological Engineering, University of Saskatchewan,Canada.

REFERENCES

Arora V. 2002. Modeling vegetation as a dynamic component insoil-vegetation-atmosphere transfer scheme and hydrological models.Reviews of Geophysics 40: 1006. DOI:10.1029/2001RG000103.

Balland V, Bhatti J, Errington R, Castonguay M, ARP PA. 2006.Modeling snowpack and soil temperature and moisture conditions in a



jack pine, black spruce and aspen forest stand in central Saskatchewan(BOREAS SSA). Canadian Journal of Soil Science 86: 203–217.

Barbour SL, Chapman D, Qualizza C, Kessler S, Boese C, Shurniak R,Meiers G, O’Kane M, Hendry J, Wall S. 2004. Tracking theevolution of reclaimed landscapes through the use of instrumentedwatersheds—A brief history of the Syncrude Southwest 30Overburden Reclamation Research Program. Proceedings of theInternational Symposium on Instrumented Watershed. Edmonton,Canada. www.rr.ualberta.ca/oilsands/IIWS.htm.

Boese K. 2003. The Design and Installation of a Field InstrumentationProgram for the Evaluation of Soil-atmosphere Water Fluxes in aVegetated Cover Over Saline/sodic Shale Overburden. MSc thesis,University of Saskatchewan, Saskatoon, Canada.

Candela A, Aronica G, Santoro M. 2005. Effects of forest fires on floodfrequency curves in a Mediterranean catchment. Hydrological Sciences50(2): 193–206.

Cuenca RH, Stangel DE, Kelly SF. 1997. Soil water balance in a borealforest. Journal of Geophysical Research 102(D24): 29355–29365.

van Dijk AIJM, Bruijnzeel LA. 2001. Modelling rainfall interceptionby vegetation of variable density using an adapted analyticalmodel. Part 1: model description. Journal of Hydrology 247:230–238.

Dingman SL. 2002. Physical Hydrology . 2nd edn, Prentice-Hall: UpperSaddle River, New Jersey; p 646.

Elshorbagy A Barbour SL. 2007. A probabilistic approach for designand hydrologic performance assessment of reconstructed watersheds.Journal of Geotechnical and Geoenvironmental Engineering:, ASCE133(9): 1110–1118.

Elshorbagy A, Jutla A, Barbour L, Kells J. 2005. System dynamicsapproach to assess the sustainability of reclamation of disturbedwatersheds. Canadian Journal of Civil Engineering 32: 144–158.

Elshorbagy A, Jutla A, Kells J. 2007. Simulation of the hydrologicalprocesses on reconstructed watersheds using system dynamics.Hydrological Sciences Journal 52(3): 538–561.

Entekhabi D, Rodriguez-Iturbe I, Castelli F. 1996. Mutual interaction ofsoil moisture state and atmospheric processes. Journal of Hydrology184: 3–17.

Falkenmark M. 1997. Society’s interaction with the water cycle: aconceptual framework for a more holistic approach. HydrologicalSciences 42: 451–466.

van Genuchten MTh. 1980. A closed-form equation for predicting thehydraulic conductivity of unsaturated soils. Soil Science Society ofAmerica Journal 44: 892–898.

Gilley JE, Gee GW, Bauer A, Willis WO, Young RA. 1977. Runoff anderosion characteristics of surface mined sites in western North Dakota.Transactions of American Society of Agricultural Engineers 20(4):697–704.

Haigh MJ. 2000. The aims of land reclamation. Land Reconstructionand Management , vol. 1. A.A. Balkema Publishers: Rotterdam, TheNetherlands; 1–20.

High Performance Systems Inc (HPS). 2001. Tutorial and TechnicalDocumentation: STELLA II . High Performance Systems Inc.:Hanover, NH (now ISES Systems, Lebanon, NH).

Jutla A. 2006. Hydrological Modeling of Reconstructed Watershedsusing the System Dynamics Approach. MS thesis, University ofSaskatchewan, Saskatoon, Saskatchewan, Canada.

Keshta N, Elshorbagy A Carey S. 2009. A generic system dynamicsmodel for simulating and evaluating the hydrological performance ofreconstructed watersheds. Hydrology and Earth System Sciences 13(6):865–881.

Leskiw LA. 2004. Land Capability Classification for Forest Ecosystemsin the Oil Sands . Paragon Soil and Environmental Consulting Inc.:Alberta, Canada.

Lee P, Aksenov D, Laestadius L, Nogueron R, Smith W. 2003.Canada’s Large Intact Forest Landscapes . Global Forest WatchCanada: Edmonton, Alberta; 84 pp.

Mukherjee D, Kottegoda NT. 1992. Stochastic model for soil moisturedeficit in irrigated lands. Journal of Irrigation and DrainageEngineering 118(4): 527–542.

Noss RF. 1993. Sustainable forestry or sustainable forest? In DefiningSustainable Forestry . Aplet G, Johnson N, Olson J, Sample A. (eds).Island Press: Covelo, CA.

Norah M, Fluet C, Davidson D, Krogman N, Ross M. 2001. IntegratedResource Management in Alberta’s Boreal Forest: Opportunities andConstraints . http://www.ualberta.ca/sfm. ISBN 1-55261-132-9.

Palisade Corporation Inc. 2005. Guide to Using @RISK. Advanced RiskAnalysis for Spreadsheets . Palisade Corporation: New York.

Parasuraman K, Elshorbagy A, Carey SK. 2007. Modelling the dynamicsof the evapotranspiration process using genetic programming.Hydrological Sciences Journal 52(3): 563–578.

Paul KI, Polglase PJ, Smethurst PJ, O’Connell AM, Carlyle CJ, KhannaPK. 2004. Soil temperature under forests: a simple model for predictingsoil temperature under a range of forest types. Agricultural and ForestMeteorology 121(3): 167–182.

Qualizza C, Chapman D, Barbour SL, Purdy B. 2004. ReclamationResearch at Syncrude Canada’s Mining Operation in Alberta’sAthabasca Oil Sands Region. International Conference on EcologicalRestoration SER2004. Victoria, August 24–26.

Salwasser H, MacCleery DW, Snellgrove TA. 1993. An ecosystemperspective on sustainable forestry and new directions for the USnational forest system. In Defining Sustainable Forestry . Aplet G,Johnson N, Olson J, Sample A. (eds). Island Press: Covelo, CA.

Shaw EM. 1994. Hydrology in Practice. 3rd ed. Chapman and Hall:London, UK; p 569.

Silva S. 2003. GPLAB, A genetic programming toolbox for Matlab.ITQB/UNL (http://gplab.sourceforge.net).

Syncrude Canada Ltd. Online: 2007. Sustainability report. (http://www.syncrude.ca/users/folder.asp?FolderIDD5909).

United Nations Environment Programme (UNEP). 2006. Flying aroundthe globe on a time machine. Online: (http://www.unep.org/Documents.Multilingual/Default.asp?DocumentIDD487&ArticleIDD5350&lDen).

Valente F, David JS, Gash JHC. 1997. Modeling interception loss for twosparse eucalypt and pine forests in central Portugal using reformulatedRuttter and Gash analytical models. Journal of Hydrology 190:141–162.

Zheng D, Hunt Jr ER, Running SW. 1993. A daily soil temperaturemodel based on air temperature and precipitation for continentalapplications. Climate Research 2: 183–191.


Comparative probabilistic assessment of the hydrological performance of reconstructed and natural...

Documents

Transcript of Comparative probabilistic assessment of the hydrological performance of reconstructed and natural...