Water Resour ManageDOI 10.1007/s11269-006-9089-0

WetSpa Model Application for Assessing ReforestationImpacts on Floods in Margecany–HornadWatershed, Slovakia

A. Bahremand · F. De Smedt · J. Corluy · Y. B. Liu ·J. Poorova · L. Velcicka · E. Kunikova

Received: 29 August 2005 / Accepted: 31 August 2006© Springer Science + Business Media B.V. 2006

Abstract The spatially distributed hydrologic model WetSpa working on a dailyor hourly time scale combines elevation, soil and landuse data within GIS, topredict flood hydrographs and spatial distribution of hydrologic characteristics in awatershed. The model is applied to the Margecany–Hornad river basin (1,131 km2)located in Slovakia. Daily hydrometeorological data from 1991 to 2000, includingprecipitation data from nine stations, temperature data from four stations andevaporation data measured at one station are used as input to the model. Three basemaps, i.e., DEM, landuse and soil types are prepared in GIS form, using 100×100 mcell size. Results of the simulations show good agreement between calculated andmeasured hydrographs at the outlet of the basin. The model predicts the daily/hourlyhydrographs with good accuracy, between 75–80% according to the Nash–Sutcliffcriteria. For assessing the impact of forests on floods, the calibrated model is appliedfor a reforestation scenario using the hourly data of the summer of 2001. Thescenario considers a 50% increase of forest areas. The model results show that thereforestation scenario decreases the peak discharge by 12%. Investigation of peakdischarges from the whole simulation period, shows that the scenario results arereduced by 18% on average. Also, the time to peak of the simulated hydrographof the reforestation scenario is 14 h longer than for the present landuse. The resultsshow that the effect of land cover on flood is strongly related to storm characteristicsand antecedent soil moisture.

A. Bahremand (B) · F. De Smedt · J. Corluy · Y. B. LiuDepartment of Hydrology and Hydraulic Engineering, Vrije Universiteit Brussel, Pleinlaan 2,B-1050 Brussels, Belgiume-mail: Abdolreza.Bahremand@vub.ac.be

J. Poorova · L. VelcickaSlovak Hydrometeorological Institute, Bratislava, Slovakia

E. KunikovaSlovak Water Research Institute, Bratislava, Slovakia

Water Resour Manage

Key words flood prediction · landuse impacts · reforestation · WetSpa model ·hydrological modeling

1 Introduction

Major changes in landuse that affect hydrology are deforestation, intensification ofagriculture, drainage of wetlands, and urbanization. The most obvious influence oflanduse on the water balance of a catchment is on the evapotranspiration process(Calder, 1993). Different landuse types have different evapotranspiration rates,because different crops have different vegetation cover, leaf area indices, root depthsand albedo. Also, interception rates are different although the influence of inter-ception is noticeable only during small storms (Calder, 1993; Ward and Robinson,1990). But landuse influences especially infiltration and soil water redistribution. Anextreme example is the influence of build up areas on overland flow. Finally, landuseand land management influences surface roughness, which controls overland andfloodplain flow rates (De Roo et al., 2003).

Computer simulation modelling has been used for at least 35 years to studythe effects of landuse changes within catchments see, e.g., Onstad and Jamieson(1970); Bultot et al. (1990); Ewen and Parkin (1996); Hillman and Verschuren (1998);Niehoff et al. (2002). Rainfall-runoff generation by distributed hydrological modelsand GIS techniques has become increasingly possible, practical and popular. Themodels are becoming more capable for predicting flood, landuse impacts on floods,and decision making in watershed management. Starting from a digital elevationmodel, hydrologic features of the terrain can be determined using standard GISfunctions that operate on raster terrain data. Also, estimation of surface and soilrelated parameters becomes feasible by combining soil type and land use data inraster format. Flow routing can be achieved by tracking the water throughout thecell network along topographic flow paths. The WetSpa model used in this study isa grid-based distributed runoff and water balance simulation model that runs on anhourly or daily time step. It predicts hourly or daily overland flow occurring at anypoint in a watershed, and provides spatially distributed hydrologic characteristicsin the basin. Inputs to the model include digital elevation data, soil type, land use,precipitation and potential evaporation time series. Stream discharge data is optionalfor model calibration. In this paper, an application of the WetSpa model is presentedfor a rather large catchment located in Slovakia for assessing the impacts of landusechanges on floods.

2 WetSpa Model

The WetSpa model was originally developed by Wang et al. (1997) and adaptedfor flood prediction by De Smedt et al. (2000) and Liu et al. (2003). For each gridcell, four layers are considered in the vertical direction: a canopy layer, the rootzone, a transmission zone and the groundwater reservoir. The hydrological processesconsidered in the model are precipitation, interception, depression storage, surfacerunoff, infiltration, evapotranspiration, percolation, interflow, ground water flow,and water balance in each layer. The total water balance for each raster cell is

Water Resour Manage

composed of a separate water balance for the vegetated, bare-soil, open water andimpervious part of each cell. This allows to account for the non-uniformity of theland use per cell, which is dependent on the resolution of the grid. A mixture ofphysical and empirical relationships is used to describe the hydrological processesin the model. The model predicts peak discharges and hydrographs in any locationof the channel network and the spatial distribution of hydrological characteristicsin each cell. Hydrological processes in each grid cell are set in a cascading way,starting from a precipitation event. Incident rainfall first encounters the plant canopy,which intercepts all or part of the rainfall until the interception storage capacityis reached. Excess water reaches the soil surface and can infiltrate the soil zone,enters depression storage, or is diverted as surface runoff. The depression storageis subject to evaporation and further infiltration. The sum of the interception anddepression storage forms the initial losses at the beginning of a storm, and does notcontribute to the storm flow. A fraction of the infiltrated water percolates to thegroundwater storage and some is diverted as interflow. Soil water is also subjectedto evapotranspiration depending on the potential evapotranspiration rate and theavailable soil moisture. Groundwater discharges to the nearest channel proportionalto the groundwater storage and the recession coefficient. Evapotranspiration fromgroundwater storage is also accounted for. Total runoff from a grid cell is thesummation of surface runoff, interflow and groundwater discharge. For each gridcell, the root zone water balance is modelled continuously by equating inputs andoutputs:

D�θ

�t= P − I − V − E − R − F (1)

where D [L] is the root depth, �θ [L3L−3] the change in soil moisture, �t [T] thetime interval, I [LT−1] the initial abstraction including interception and depressionlosses within time step �t, V [LT−1] the surface runoff or rainfall excess, E [LT−1]the actual evapotranspiration from the soil, R [LT−1] the percolation out of the rootzone, and F [LT−1] the amount of interflow in depth over time. The rainfall excess iscalculated using a moisture-related modified rational method with a potential runoffcoefficient depending on the land cover, soil type, slope, the magnitude of rainfall,and the antecedent soil moisture:

V = C(P − I)(

θ

θs

)α

(2)

where θ s [L3L−3] is the soil porosity, C [-] is the potential runoff coefficient. Thevalues of C are taken from literature and a lookup table, linking values to slope,soil type and landuse classes (Liu and De Smedt, 2004). The exponent α [−] inthe formula is a variable reflecting the effect of rainfall intensity on the rainfallexcess coefficient. The value is higher for low rainfall intensities resulting less surfacerunoff, and approaches to one for high rainfall intensities. The threshold value can bedefined during model calibration. If α = 1, a linear relationship is assumed betweenrainfall excess and soil moisture. The effect of rainfall duration is also accounted bythe soil moisture content, in which more excess produces due to the increased soilmoisture content. The difference between net precipitation and excess rainfall is the

Water Resour Manage

amount of infiltration into the soil. Evapotranspiration from the soil and vegetation iscalculated based on the relationship developed by Thornthwaite and Mather (1955),as a function of potential evapotranspiration, vegetation type, stage of growth andsoil moisture content. For the surface layer, actual evapotranspiration is computedas the area-weighted mean of the land use percentage, for which transpiration occursfrom the vegetated parts and evaporation from the soil, while there is no evaporationfrom impervious areas. A portion of the remaining potential evapotranspirationis transpirated from the groundwater as a proportion of the groundwater storage.Finally, the total evapotranspiration is calculated as the sum of evaporation frominterception storage, depression storage, and the evapotranspiration from soil andgroundwater storage. Percolation and interflow are assumed to be gravity driven. Thepercolation out of the root zone is equated as the hydraulic conductivity dependingon the moisture content as a function of the soil pore size distribution index(Eagleson, 1978). Interflow is assumed to occur in the root zone after percolationand becomes significant only when the soil moisture is higher than field capacity.Darcy’s law and a kinematic wave approximation are used to determine the amountof interflow generated from each cell, in function of hydraulic conductivity, themoisture content, slope angle, and the root depth. The routing of overland flow andchannel flow is implemented by the method of the diffusive wave approximation ofthe St. Venant equation:

∂ Q∂t

+ c∂ Q∂x

− d∂2 Q∂t2 = 0 (3)

where Q [L3T−1] is the discharge, t [T] is the time, x [L] is the distance along the flowdirection, c [LT−1] is the location dependent kinematic wave celerity, is interpretedas the velocity by which a disturbance travels along the flow path, and d [L2T−1] isthe location dependent dispersion coefficient, which measures the tendency of thedisturbance to disperse longitudinally as it travels downstream. Assuming that thewater level gradient equals the bottom slope and the hydraulic radius approachesthe average flow depth for overland flow, c and d can be estimated by c = (5/3)v, andd = (vH)/(2S0) (Henderson, 1966), where v [LT−1] is the flow velocity calculated bythe Manning equation, and H [L] is the hydraulic radius or average flow depth. Anapproximate solution to the diffusive wave equation in the form of a first passagetime distribution is applied (Liu et al., 2003), relating the discharge at the end of aflow path to the available runoff at the beginning.

U(t) = 1

σ

√2π t3/t3

0

exp[− (t − t0)2

2σ 2t/t0

](4)

Where U(t) [T−1] is the flow path unit response function, serving as an instantaneousunit hydrograph (IUH) of the flow path, which makes it possible to route watersurplus from any grid cell to the basin outlet or any downstream convergent point,t0 [T] is the flow time, and σ [T] is the standard deviation of the average flowtime. The parameters t0 and σ are spatially distributed, and can be obtained by

Water Resour Manage

integration along the topographic determined flow paths as a function of flow celerityand dispersion coefficient.

t0 =∫

c−1dx (5)

and

σ =√∫

(2d/c3)dx (6)

Because, groundwater movement is much slower than the movement of water inthe surface and near surface water system, and little is known about the bedrock,groundwater flow is simplified as a lumped linear reservoir on small GIS derivedsubcatchment scale. Considering the river damping effect for all flow components,overland flow and interflow are routed firstly from each grid cell to the mainchannel, and joined with groundwater flow at the subcatchment outlet. Then thetotal hydrograph is routed to the basin outlet by the channel response functionderived from Eq. 4. The total discharge is the sum of overland flow, interflow andgroundwater flow, and is obtained by convolution of the flow responses from all gridcells. One advantage of this approach is that it allows the spatially distributed runoffand hydrological parameters of the basin to be used as inputs to the model. Inputsto the model include digital elevation data, soil type, land use data, and measuredclimatological data. Stream discharge data is optional for model calibration. Allhydrological processes are simulated within a GIS framework.

Because a large part of the annual precipitation is in the form of snow, a modelbased on daily temperature data is used to simulate snow melt. The conceptualtemperature index or degree-day method is used in this study, because it is simplebut nevertheless has a strong physical foundation. The method replaces the fullenergy balance with a term linked to air temperature. It is physically sound in theabsence of shortwave radiation when much of the energy supplied to the snowpackis atmospheric long wave radiation (Leavesley, 1989; Parajka, 1999). The equationcan be expressed as:

M = max[0, Csnow(T − T0) + Crain P(T − T0)] (7)

Where M [LT−1] is the daily snowmelt, T [◦C] is the daily mean temperature, T0 [◦C]is a threshold melt temperature, Csnow is a melt-rate factor [L◦C−1T−1], and Crain

is a degree-day coefficient taking into account the heat contribution from rainfall[◦C−1T−1]. The critical melt temperature, T0, is often intuitively set to 0◦C. The melt-rate factor, Csnow, is an effective parameter and may vary with location and snowcharacteristics; however, in this study it is assumed constant for model simplicity.

3 Application

3.1 Study Area

The Hornad river is located in Slovakia. The drainage area of the river up toMargecany station is 1,131 km2 (in this paper simply named Margecany catchment).Margecany station is located just above a multi purpose reservoir called Ruzin.

Water Resour Manage

Figure 1 Hydrologic network of Hornad catchment, topography of Margecany subcatchment, andlocation of gauging stations and Ruzin reservoir.

Figure 1 shows the Hornad catchment, the topography of Margecany subcatchment,and location of gauging stations and the Ruzin reservoir. The Margecany basin isa mountainous catchment, with elevations ranging from 339 to 1,556 m. The meanelevation of the catchment is 670 m and the mean slope is about 18%. Some of thephysiographical and hydrological characteristics of the basin are given in Table I.

The DEM for the river basin was obtained from the Slovak HydrometeorologicalInstitute (SHMU), and converted to a 100 m grid size DEM, from which the drainagesystem and area were determined (Figure 1). Land cover data were obtained from30 m Landsat-7 Enhanced Thematic Mapper (ETM) satellite data, acquired onAugust 20th, 2000. The final landuse map (Figure 2) for this study has a 100×100 mcell size and is composed of six different types of land cover: 49.8% of the basinis covered by forest (26.8% coniferous forest and 23.0% mixed-deciduous forest),25.5% grassland and pasture, 22.8% agriculture areas, 1.8% urban area and about0.1% water surfaces which are mainly reservoirs. There are 10 different soil texturesin the catchment. The dominant soil texture is loam, which covers about 42% of thebasin, and sandy loam and silt loam about 24 and 23%, respectively.

Table I Physiographical andhydrological characteristics ofthe Margecany watershed

Area (km2) 1,131Perimeter (km) 204Gravelius coef. = 0.28×P/A1/2 1.69Min. elevation (m) 339Max. elevation (m) 1,556Mean elevation (m) 670Mean basin slope 18%Mean slope of streams order≥2 1.25%Stream order at outlet 6Max. flow length (km) 90

Water Resour Manage

Figure 2 Landuse map of the Margecany watershed.

The basin has a northern temperate climate with four distinct seasons. January isthe coldest month and July is the warmest month of the year. The highest amount ofprecipitation occurs in the period from May to August while in January and Februarythere is usually only snow. The mean annual precipitation of the watershed based on10 years data of nine stations within the basin is 678 mm, ranging from about 640 mmin the valley to more than 1,000 mm in the mountains. The mean temperature ofthe catchment based on a 40-year period isothermal map is about 6◦C. The annualpotential evapotranspiration based on 10 years data of one station (Spisske Vlachy)is about 518 mm.

For this study, precipitation, temperature and discharge data were obtained fromSHMU, whereas the potential evapotranspiration (PET) data were obtained fromthe Water Research Institute of Slovakia. The sets include daily precipitation for ninestations, temperature for four stations, PET at one station, and daily discharge dataat eight gauging stations. All these data are available for a 10-year period from 1991to 2000. Flow stations containing daily discharge data at eight locations are availableinside the catchment (Bahremand et al., 2005; Corluy et al., 2005). The watershedoutlet station, Margecany, and one of the upstream stations, i.e., Hrabusice are usedfor model calibration and validation.

3.2 Model Simulation

Once the required data are collected and processed for use in the WetSpa model,identification of spatial model parameters is undertaken. Terrain features at eachgrid cell including elevation, flow direction, flow accumulation, stream network,stream link, stream order, slope, and hydraulic radius are first extracted from theDEM. The threshold for delineating the stream network is set to 10, i.e., a cell isconsidered being drained by a stream when the upstream drained area is greater than0.1 km2. The threshold value for determining subcatchments is set to 1,000, by which69 subcatchments are identified with an average subcatchment area of 16.6 km2.When creating the grid of surface slope, a threshold value of minimum slope of0.01% is considered; if the calculated slope is less than this threshold value the slopeis set to 0.01% in order to avoid stagnant water or extreme low velocities. The grid

Water Resour Manage

Figure 3 Potential runoff coefficient map of the Margecany watershed.

of hydraulic radius is generated with an exceeding frequency of 0.5 (2-year returnperiod) resulting in an average hydraulic radius of 0.005 m for the upland cells and1.5 m at the outlet of the main river channel (The related formula and callculationmethod are discussed in the documentation and user manual of the model by (Liuand De Smedt, 2004)). Next, the grids of soil hydraulic conductivity, porosity, fieldcapacity, residual moisture, pore size distribution index, and plant wilting point arereclassified based on the soil texture grid by means of an attribute lookup table.Similarly, the grids of root depth, interception storage capacity, and Manning’sroughness coefficient are reclassified from the land use grid, in which the Manning’scoefficient for channels is linearly interpolated based on the stream order grid with0.055 m−1/3 s for the lowest order and 0.025 m−1/3 s for the highest order. The gridsof potential runoff coefficient and depression storage capacity are obtained by meansof attribute tables combining the grids of elevation, soil and land use, for which thepercentage of impervious area within an urban cell is set to 30%. The results areshown in Figure 3. From this figure it follows that non-forested and steeper areashave a very high potential runoff coefficient, whereas the forested and gentle slopesgenerate less surface runoff. The calculated average potential runoff coefficient is0.43 for the entire catchment. The grids for precipitation, temperature and PET arecreated based on the geographical coordinates of each measuring station and thecatchment boundary using the Thiessen polygon extension of the ArcView SpatialAnalyst.

Finally, the grids of flow velocity, travel time to the basin outlet, as well as thestandard deviation are generated, which enable to calculate the IUH from each gridcell to the basin outlet. Figure 4 shows the estimated average flow time from eachgrid cell to the basin outlet; the flow time for the most remote area is around 62 h,while the mean travel time for the entire catchment is 25 h.

3.3 Calibration and Validation

The 10 years (1991–2000) measured daily precipitation, temperature, PET, anddischarge data are used for model calibration and validation. The first 5 years of

Water Resour Manage

Figure 4 Flow travel time to the Margecany watershed outlet.

the 10-year period is chosen for model calibration and the second 5 years for modelvalidation. The calibration process is performed manually for the global modelparameters only, whereas the spatial model parameters are kept as they are. Initialglobal model parameters are specifically chosen according to the basin characteristicsas discussed in the documentation and user manual of the model (Liu and De Smedt,2004). The simulation results are then compared to the observed hydrograph atMargecany both graphically and statistically. The parameters of base temperatureand degree-day coefficients are adjusted independently in order to get a proper fit ofsnowmelt, occurring normally in late February or early March. The groundwater flowrecession coefficient is adjusted by fitting the baseflow, which is separated from theobserved hydrograph using linear or non-linear reservoir model algorithms used inthe model are described in Chapman (1999), Tallaksen (1995), and Wittenberg andSivapalan (1999a,b). The interflow scaling factor, which is sensitive for high flows, isadjusted for the recession part after flood peak. The two parameters controlling theamount of surface runoff, i.e., the surface runoff exponent for a near zero rainfallintensity and the rainfall intensity corresponding to a surface runoff exponent of 1,are adjusted mainly for small storms, for which the actual runoff coefficients are smalldue to the low rainfall intensity. Initial soil moisture and active groundwater storageare adjusted by comparison of the hydrograph and water balance for the initial phase.The maximum active groundwater storage controls the amount of transpiration fromthe groundwater, and therefore can be adjusted by comparison of the flow volumeduring dry periods.

Figure 5 gives a graphical comparison between observed and calculated daily flowat Margecany for the year 1991 from the calibration period. For model validation,the calibrated global parameters are used to simulate the daily stream flow forthe second part of the data, i.e., from 1996 to 2000. Figure 6 gives a graphicalcomparison between observed and calculated daily flow at Margecany for the year1997 from the validation period. Figure 5 and Figure 6 show that both the springand summer flood hydrographs are well reproduced by the model. The simulation

Water Resour Manage

0

20

40

60

1 31 61 91 121 151 181 211 241 271 301 331 361

0

10

20

30

40

PrecipitationQ simulatedQ measured

Figure 5 Graphical comparison between observed and calculated daily flow at Margecany for theyear 1991 of the calibration period.

of snowmelt flood is important in this study, as it contributes to the results of modelevaluation. The calibrated base temperature and degree-day coefficient are 0◦C and2.0 mm/d/◦C, whereas the heat contribution from rainfall to the snowmelt is notimportant for this case study. The calibrated groundwater flow recession coefficient

0

20

40

60

80

1 31 61 91 121 151 181 211 241 271 301 331 361

0

10

20

30

40

PrecipitationQ simulatedQ measured

Figure 6 Graphical comparison between observed and calculated daily flow at Margecany for theyear 1997 of the validation period.

Water Resour Manage

Table II Evaluation criteriafor the assessment of modelperformance

Criteria Calibration Validation

Model bias for water balance 0.018 0.004Nash–Sutcliffe efficiency 0.758 0.802Model efficiency for low flows 0.726 0.806Model efficiency for high flows 0.817 0.816

at Margecany is 0.0085 d−1, and gives a good estimation for all base flows. Peakdischarges, concentration time, and flow volumes are especially well predicted forthe three summer floods of 1997.

The model performance is satisfactory for both calibration and validation periods.Evaluation criteria for the calibration and validation period are given in Table II.Four evaluation criteria used by Hoffmann et al. (2004) are selected. The model biasfor water balance criterion evaluates the ability of the model to reproduce the waterbalance (which should be close to zero). The accuracy of the model to simulate the

Line 1:

1

1

10

100

1 10 100

Measured discharge (m3/s)

Sim

ulat

ed d

isch

arge

(m

3 /s)

Figure 7 Logarithmic scatter plot of observed versus simulated flows for the validation period and95% confidence limits.

Water Resour Manage

discharge is evaluated through the Nash–Sutcliffe criterion, and two adapted Nash–Sutcliffe efficiencies to assess the model’s performance for low flows and high flows(which should be close to 1). It turns out that the criteria are better for the validationperiod than for the calibration period. For the validation period, the flow volume is0.4% over-estimated, the Nash–Sutcliffe efficiency is 0.802, and the modified Nash–Sutcliffe efficiency is 0.806 and 0.816, respectively, for high and low flows (Table II).A logarithmic scatter plot of measured versus simulated flows for the validationperiod is given in Figure 7. The model validation outputs also show that 7.8% of theprecipitation is intercepted by the plant canopy, 84.5% infiltrates to the soil, 67.9%evapotranspirates to the atmosphere, 23.5% recharges to the groundwater reservoir,and 31.3% becomes runoff, of which direct flow, interflow, and groundwater flowcontribute 5.7, 3.8 and 21.8% , respectively. These values are reasonable in view ofthe catchment hydrological characteristics.

A spatially distributed model should be able to simulate reasonably well thedischarge at the outlet of a series of embedded catchments with the same modelstructure and parameter sets. Therefore, the calibrated model is applied for anupstream subcatchment with a size of 221.9 km2. The outlet of this subcatchmentis Hrabusice station, shown in Figure 1, for which 10 years daily data from 1991 to2000 are available. The same parameter set as previous simulation are used for thesimulation. Figure 8 gives a graphical comparison between observed and calculateddaily flows at Hrabusice for the year 1997. For the 10-year simulation period, theflow volume is 0.6% over-estimated, the Nash–Sutcliffe efficiency is 67%, and themodified Nash–Sutcliffe efficiency is 78 and 62%, respectively, for high and low flows.Similar results with the model were obtained for the Hornad watershed at Zdanastation (Figure 1) and for the neighboring Torysa catchment at Kosicke Olsanystation (Figure 1) with the same model structure and setup, similar parameter sets,

0

5

10

15

1 31 61 91 121 151 181 211 241 271 301 331 361

0

10

20

30

40

PrecipitationQ simulatedQ measured

Figure 8 Graphical comparison between observed and calculated daily flow at Hrabusice for theyear 1997.

Water Resour Manage

and for the same simulation period. These results were published by Bahremandet al. (2005),Bahremand and De Smedt (2005), and Bahremand (2006). In thesepublications, the model efficiency for the Hornad simulation 80% and for the Torysasimulation 73% has been reported.

The model has been applied in several other regions, e.g., Barebeek catchment inBelgium (De Smedt et al., 2000), Alzette river basin in Luxembourg (Liu et al., 2003),and Tisza watershed in Hungary (Corluy et al., 2005). The results of the mentionedstudies as well as the results of this study indicate that the model is able to simulatehydrologic processes in a spatially realistic manner based on topography, land useand soil type, resulting in a fairly high accuracy for both high and low flows.

Distributed hydrological models should pass through a careful calibration pro-cedure backed by sensitivity, uncertainty and predictive analysis before they areutilized as a decision making aid in watershed management and scenario studies. Thishas been done by Bahremand (2006), in which, sensitivity, uncertainty and predictiveanalysis of the model parameters have been analyzed using a model-independentparameter estimator, PEST. This work proves that the parameter uncertainty ofthe model does not result in a significant level of predictive uncertainty, and thatthe calibrated model can be applied for further predictions and scenario studies inthe Hornad watershed.

3.4 Hourly Flow Simulation for Assessing Impact of Reforestation

Landuse has a great influence on the rainfall runoff process. Since the spatialdistribution of hydrologic characteristics can be obtained from the model outputs ateach time step, the model has a great potential to analyze the effects of topography,soil type, land cover and the effects of landuse changes on the hydrologic behaviorof a river basin. However, for the present study area impacts of landuse changeson flood can not be assessed and captured using daily data because the time stepis too coarse. Hence for this reason hourly data are needed. For this part of themodeling study, less than 5 months of hourly precipitation data in the summer of 2001were obtained from SHMU from five stations within the catchment, together withhourly discharge data at Margecany station. Hourly evaporation data for this periodwere calculated based on monthly data measured at Spisske Vlachy (Figure 1) in thesummer of 2001 and an empirical curve which was developed based on 10 years dailyobserved PET data, presented in the documentation and user manual of the model(Liu and De Smedt, 2004). Figure 9 gives a graphical comparison between observedand calculated hourly flows at Margecany during the summer of 2001. No calibrationwas performed to obtain these results, i.e., the same global parameter values as forthe daily model were used. The simulated hydrograph shows a good agreement withthe measured hydrograph. The model efficiency (Nash–Sutcliffe criterion) is 86%.

For assessing the impact of landuse, in particular forestation on floods, thecalibrated WetSpa model was applied for a reforestation scenario using the hourlydata of the summer 2001. For the reforestation scenario it is assumed that forest willincrease by 50%. This is achieved by converting grassland to forest, because grasslandis almost everywhere situated between forest and agriculture areas. Currently abouthalf (49.8%) of the catchment is covered by forest, hence a 50% increase in forestswill yield a forested area of 75.3% in total as shown in Table III.

Water Resour Manage

0

40

80

120

1 480 959 1438 1917 2396 2875

0

10

20

30

PrecipitationQ measuredQ simulated

1/5/2001 15/9/2001

Figure 9 Observed and calculated hourly flow at Margecany for the period 1/5/ to 15/9/2001.

Based on this landuse change scenario, the model was run to estimate themodified flows. It is worth to mention how landuse changes affect the simulationparameters in the model. After changing the landuse map the following maps haveto be recalculated: root depth, interception, Manning coefficient, depression storage,runoff coefficient, velocity, flow travel time, and the standard deviation of travel time.The changes in the flow travel time and the standard deviation consequently alsochange the IUH of each cell and the watershed IUH. Figure 10 gives the simulatedflood hydrograph (for the period from 17/7/2001 to 5/8/2001) for the present and thereforestation scenario for the flood event that occurred in July 2001. The simulatedpeak discharge for the present land use is 84.8 m3/s, and for the reforestation scenariothis becomes 74.9 m3/s, which means the scenario decreases the peak discharge by12%. Investigation of peak discharges for the whole simulation period, shows thatthe reforestation scenario results in a reduction of 18% on average. In addition tothe difference in magnitude of the simulated peak discharges, differences in time topeak are also observed, as the time to peak of the reforestation scenario is 14 h longerthan for the present landuse.

Table III Percentages of eachlanduse type in the presentcondition and for thereforestation scenario in theMargecany watershed

Landuse type Present Reforestation

Agriculture 22.8 22.8Grassland/Pasture 25.5 0Forest 49.8 75.3Urban area 1.8 1.8Open water 0.1 0.1

Water Resour Manage

Figure 10 Simulated hydrographs for reforestation scenario and present landuse for the flood thatoccurred in July 2001.

Additionally, the effects of land use change on flood volume, runoff composition,evapotranspiration and soil moisture can also be evaluated quantitatively from themodel results. Reforestation results in decreasing surface runoff and flood volume,but increasing amounts of interflow and baseflow, as well as soil moisture andevapotranspiration. The magnitude of changes varies from one storm to the otherdepending upon the storm intensity and duration, and the antecedent soil moistureconditions.

The following test shows how storm characteristics and antecedent soil moistureconditions affect the impact of forests on flood reduction. The results of the testshows that the influence of landuse on storm-runoff generation depends on therainfall event characteristics, i.e., the influence of landuse on storm-runoff generationis stronger for convective storm events with high precipitation intensities than forlong advective storm events with low precipitation intensities (Niehoff et al., 2002).The test considers three different rain duration scenarios. The total rain amountremains the same, i.e., 80 mm, but the duration is varied from 5 to 10 and 20 h. Forpresent conditions these duration scenarios produce flood peaks varying from 188 to128 m3/s as shown in Figure 11. Then, the reforestation scenarios is considered forthese scenarios. The results show that the 50% increase of forest in the watershedreduces the peaks from 133 to 106 m3/s, respectively, (Figure 11). The reforestationscenario reduces the peak by 29% for the shortest rain duration and only 17% for thelongest one. These results show that for a 50% increase of forest in the basin there isa significant reduction of peak flows of about 24% on average, but the effect becomesless if the rain continues for a longer time.

Comparing and analyzing the test’s result with the reforestation scenario simula-tion in July 2001, confirms that the effect of land cover on flood is strongly related

Water Resour Manage

100

120

140

160

180

200

5 10 20

Present

Reforestation

Figure 11 Rain duration scenarios and resulting peak discharges: peak discharge produced by eachrain scenario for the present landuse and the reforestation scenario.

to antecedent soil moisture and storm amount and duration. The effect of vegetationis very significant for the beginning of the storm especially when the soil is dry, butlater on, the effect will become less.

4 Discussion and Conclusions

Distributed models have proven to be useful in scenario analyses because of theirability to predict the effect of spatially changing variables, like landuse change.This paper at first defines a method for estimating flood runoff in the Margecanybasin by using detailed basin characteristics together with meteorological data as aninput to the WetSpa spatially distributed model. To avoid the complexity inherentin estimating surface runoff, a simple but effective approach is presented where thewhole basin is divided into grid cells, giving the possibility to simulate the hydrologicprocesses at reasonably small scale.

The generation of runoff depends upon rainfall intensity and soil moisture andis calculated as the net precipitation times a runoff coefficient, which depends uponslope, land use and soil type and is further continued by antecedent soil moisture.Overland flow is routed through the basin along flow paths determined by thetopography using a diffusive wave transfer model, while interflow and groundwaterrecharge are simulated using Darcy’s law and the kinematic approximation. Modelparameters based on surface slope, land use, soil type and their combinationsare collected from literature, which can be prepared easily using standard GIStechniques.

Water Resour Manage

The model is tested on the Margecany mountainous catchment in Slovakia with10 years of observed daily rainfall and evaporation data. Good agreement withthe measured hydrograph is achieved. Since the spatial distribution of hydrologiccharacteristics can be obtained from the model outputs at each time step, the modelis especially useful to analyze the effects of topography, soil type, and landuse on thehydrologic behavior of a river basin. For assessing the impact of landuse changes,the calibrated WetSpa model is applied for a reforestation scenario using the hourlydata of the summer of 2001. A 50% increase of forests in the catchment shows only amoderate reduction in discharge. The results show the reforestation scenario reducesthe discharge by 18% on average. The time to peak of the simulated hydrographin the reforestation scenario becomes 14 h longer than for the present landuse.Gebremeskel (2003) and Liu and De Smedt (2004) applied the WetSpa model forassessing the hydrological effects of land use changes on floods in the Alzette basinlocated in Luxembourg. For their study they assumed an afforestation scenario with53% increase of forested area. It is necessary to note that in the present landusescenario, the urban area proportion is 21% of the basin area, and for forest this is29%. The simulation of Liu and De Smedt (2004) shows the afforestation decreasesthe peak by 8%. Comparing their results with the results of this paper, it can beconcluded that the extent of the simulated impacts of landuse changes on flood variesfrom one particular area to another based on many factors such as the variety oflanduse composition and their percentages, the physiographical characteristics of thebasin, the magnitude of the simulated flood, etc. Moreover it should be mentionedthat the results of a scenario study is very much based on the extent of the assumedlanduse change in the scenario and the model which is applied for the simulation.Similar studies, using different models, by Lahmer et al. (2001) and Niehoff etal. (2002) also mentioned the importance of land use change and forestation onflood generation. Lahmer et al. (2001) presented a forestation scenario assuming theconversion of all the arable land (about 66% of the total basin area) to forest thatreduced peak flow by 42.3%.

This study shows that the model can be useful for investigating the effect ofvegetation in different rain condition and antecedent soil moisture. It presents theapplication of the WetSpa model for rain storm scenarios and possibility of assessingthe effects of land covers under different rain condition. A test using three rain sce-narios proves how the effect of the reforestation scenario differs from one storm toanother depending upon the storm amount and duration. In similar study, Niehoff etal. (2002) examined different landuse scenarios on flooding, using a modified versionof the physically based hydrological model WaSiM-ETH in a meso-scale catchmentin SW Germany, to show that the influence of land-use conditions on storm-runoffgeneration depends greatly on the rainfall event characteristics and on the relatedspatial scale. They concluded that the influence is only relevant for convectivestorm events with high precipitation intensities in contrast to long-lasting advectivestorm events with low precipitation intensities. However, convective events-and thuslanduse conditions-are of very minor relevance for the formation of floods in largeriver basins because this type of rainfall event is usually restricted to small-scaleoccurrence (Niehoff et al., 2002).

As pointed out by Ward and Trimble (2004), on small watersheds, forests canreduce flooding. Thunderstorms with high intensities, covering small areas andlasting a short time, can create severe flooding in streams with other land uses. Storms

Water Resour Manage

with high intensities can create Horton overland flow on land used for agricultureand grazing, but interception and infiltration capacities of forests are great enoughthat overland flow will not occur even in a thunderstorm. Pathways for stormflowin forests are slower than agricultural, range, and urban lands, as streams respondonly to long-duration events. This scenario study shows reforestation can be a way offlood control in case of flood events, when short but intensive rainfalls cause disasters(e.g., in watersheds endangered by flash-flood events such as flooding in small riversin the mountains of Italy, Switzerland, Slovakia, Czech Republic etc.).

5 Recommendation

The spatial and temporal variability of rainfall, temperature, PET, and snow ac-cumulation in the western mountains of the Hornad River basin is an importantbut poorly documented phenomenon. It governs the hydrological processes such asrunoff generation, soil moisture condition, etc., and is one of the major uncertaintiesto the modeling results. Moreover, local variability in input time series results fromvariations in latitude, aspect, orographic effects, wind speed and direction, humidity,as well as sun radiation, is not taken into account in this case study. In particular,the snow pack is assumed to be accumulated from the solid precipitation withoutconsidering the wind-induced snow drift. These local factors have considerableeffects on the spatial distribution of input time series and the modeling results aswell when conducting water studies for a large mountainous catchment. Furtherresearches may be undertaken on this subject once the relevant data are available. Afuture work, also, can improve the model landuse categories. The land use categoriesare grouped, for which some of the categories might be somewhat ambiguous. Forinstance, the category agriculture may include farmsteads, lawns, disturbed areas,and other land uses that are not identifiable as one of the other specified land usecategories. Also, variable travel time into flow routing schemes, for which the flowvelocity is estimated as time variant variable depending upon the channel geometryand runoff volumes should be incorporated into the model. This may overcome theshortage that flow velocity is assumed time invariant for a flood event in the currentmodelling approach. Improvements of the interflow redistribution and combinationwith a physically based distributed groundwater model in order to evaluate the effectof groundwater table to the generation of surface runoff are the next steps in thisstudy.

Acknowledgments The data has been provided by the Slovak Hydrometeorological Institute,SHMU and the Water Research Institute of Slovakia, VUVH, for the Tisza River EC-5th frameworkinternational project. More detailed information about the project can be found at the followingwebsite: http://www.tiszariver.com

References

Bahremand A (2006) Simulating the effects of reforestation on floods using spatially distributedhydrologic modeling and GIS. PhD thesis, Vrije Universiteit Brussel, Belgium

Bahremand A, Corluy J, Liu YB, De Smedt F, Poorova J, Velcicka L (2005) Stream flow simulationby WetSpa model in Hornad river basin, Slovakia. In: van Alphen J, van Beek E, Taal M (eds)Floods, from defence to management. Taylor–Francis Group, ISBN 0 415 38050 2, London,pp 67–74

Water Resour Manage

Bahremand A, De Smedt F (2005) Distributed hydrological modeling and sensitivity analysis inTorysa watershed, Slovakia. Water Resour Res (Under review).

Bultot F, Dupriez GL, Gellens D (1990) Simulation of land use changes and impacts on the waterbalance: a case study for Belgium. J Hydrol 114:327–348

Calder IR (1993) Hydrologic effects of land-use change. In: Maidment DR (ed) Handbook ofhydrology. McGraw–Hill, New York, p 50

Chapman TG (1999) A comparison of algorithms for stream flow recession and baseflow separation.Hydrol Process 13:701–714

Corluy J, Bahremand A, Liu YB, De Smedt F (2005) Development and application of the GIS-based distributed precipitation–runoff model WetSpa in the Tisza River Basin. In: Annex 2Aof the Tisza river project report, Department of Hydrology and Hydraulic Engineering, VrijeUniversiteit Brussel, Belgium

De Roo A, Schmuck G, Perdigao V, et al. (2003) The influence of historic land use changesand future planned land use scenarios on floods in the Oder catchment. Phys Chem Earth28:1291–1300

De Smedt F, Liu YB, Gebremeskel S (2000) Hydrological modeling on a catchment scale using GISand remote sensed land use information. In: Brebbia CA (ed) WTI, Boston, pp 295–304

Eagleson PS (1978) Climate, soil, and vegetation, a simplified model of soil moisture movement inliquid phase. Water Resour Res 14(5):722–730

Ewen J, Parkin G (1996) Validation of catchment models for predicting land-use and climate changeimpacts. 1. Method. J Hydrol 175:583–594

Gebremeskel S (2003) Modelling the effect of climate and land-use changes on hydrologicalprocesses: An integrated GIS and distributed modelling approach. PhD thesis, Vrije UniversiteitBrussel, Belgium

Henderson FM (1966) Open channel flow. McMillan, New York, p 522Hillman GR, Verschuren JP (1998) Simulation of the effects of forest cover, and its removal, on

subsurface water. Water Resour Res 24:305–314Hoffmann L, El Idrissi A, Pfister L, Hingray B, Guex F, Musy A, Humbert J, Drogue G, Leviandier

T (2004) Development of regionalized hydrological models in an area with short hydrologicalobservation series. River Res Applic 20(3):243–254

Lahmer W, Pftzner B, Strbl B (2001) Possible impacts of climate and land use changes on thewater balance of semi-arid German river basins. In: 19th European regional ICID conferencesustainable use of land and water (conference proceeding), Prague

Leavesley GH (1989) Problems of snowmelt runoff modelling for a variety of physiographic andclimatic conditions. Hydrol Sci J 34(6):617–634

Liu YB, De Smedt F (2004) WetSpa extension, documentation and user manual, Department ofHydrology and Hydraulic Engineering, Vrije Universiteit Brussel, Belgium

Liu YB, Gebremeskel S, De Smedt F et al. (2003) A diffusive transport approach for flow routing inGIS-based flood modeling. J Hydrol 283:91–106

Niehoff D, Fritsch U, Bronstert A (2002) Land-use impacts on storm-runoff generation: scenar-ios of land-use change and simulation of hydrological response in a meso-scale catchment inSW-Germany. J Hydrol 267:80–93

Onstad CA, Jamieson DG (1970) Modeling the effect of land use modifications on runoff. WaterResour Res 6:1287–1295

Parajka J (1999) Mapping long-term mean annual precipitation in Slovakia using geostatisticalprocedures. In: Problems in fluid mechanics and hydrology, Institute of Hydrodynamics, ASCR,Prague, Czech Republic

Tallaksen LM (1995) A review of baseflow recession analysis. J Hydrol 165:349–370Thornthwaite CW, Mather JR (1955) The water balance: publications in climatology, No.8, Labora-

tory of Climatology, Centerton, New JerseyWang Z, Batelaan O, De Smedt F (1997) A distributed model for water and energy transfer between

soil, plants and atmosphere (WetSpa). Phys Chem Earth 21:189–193Ward AD, Trimble SW (2004) Environmental hydrology, 2nd edn. Lewis publishers, Washington,

District of ColumbiaWard RC, Robinson M (1990) Principles of hydrology, 3rd edn. McGraw Hill, London, p 365Wittenberg H, Sivapalan M (1999a) Watershed groundwater balance estimation using streamflow

recession analysis and baseflow separation. J Hydrol 219:20–33Wittenberg H, Sivapalan M (1999b) Baseflow recession and recharge as nonlinear storage processes.

Hydrol Process 13:715–726