Biologia 64/3: 589—593, 2009 Section Botany DOI: 10.2478/s11756-009-0086-7
The limitations of assessing impacts of land use changes on runoff with a distributed hydrological model: case study of the Hron River Kamila Hlavčová, Ján Szolgay, Silvia Kohnová & Oliver Horvát Department of Land and Water Resources Management, Slovak University of Technology, Radlinského 11, SK-81368 Bratislava, Slovakia; e-mail:
[email protected]
Abstract: A distributed hydrological model was applied for estimating changes in a runoff regime due to land use changes. The upper Hron river basin, which has an area of 1766 km2 and is located in central Slovakia, was selected as the pilot basin. A physically-based rainfall-runoff model with distributed parameters was used for modelling runoff from rainfall and melting snow. The parameters of the model were estimated using climate data from 1981–2000 and from three digital map layers: a land-use map, soil map and digital elevation model. Several scenarios of changes in land use were prepared, and the runoff under the new land use conditions was simulated. Long-term mean annual runoff components and the design maximal mean daily discharges with a return period from 5 to 100 years under the previous and changed land uses were estimated and compared. The simulated runoff changes were confronted with expert judgments and estimates from the literature. Limitations of the use of distributed models for estimating land use changes are discussed. Key words: distributed parameters; land use change; rainfall-runoff model
Introduction Changes in runoff generation due to changes in land use, particularly those connected to agricultural and forest management, have often been documented in the literature, e.g., the removal of forest cover is known to change stream flow as a result of changes in interception, infiltration, surface roughness and water tables (e.g., Mattheusen et al. 2000; Brown et al. 2005; Kostka & Holko 2006). Distributed rainfall-runoff model simulations are often used to evaluate the impact of such changes on runoff generation. These models have the advantage of reflecting the effects of land use in spatially distributed model parameters. Moreover, the present-day availability of spatially distributed data such as digital elevation models, land use, and soil information makes the use of distributed models much easier. In connection with this the representation of runoff processes and land use changes in distributed models is frequently discussed (e.g., Niehoff et al. 2002; Ott & Uhlenbrook 2004; Bloeschl et al. 2007). Due to the complexity of the processes involved, the magnitude of their impact on runoff generation and subsequent discharges into a river system is still highly uncertain (Niehoff et al. 2002). In distributed models, land cover properties have to be characterized by plant-specific parameters, but reliable results of modelling can only be obtained if the parameter values for the land covers involved are known with some degree of accuracy. A review of the literature (Eckhardt et al. 2003) shows a high degree of uncertainty in the parameterisation of land covers. This
c 2009 Institute of Botany, Slovak Academy of Sciences
uncertainty is caused by the problematic observation of some parameters and difficulties with the regionalization of point measurements because of the natural variability of plant characteristics due to the particular climate, soil, stand age, phenophase, etc. In this paper, simulated changes in the runoff regime in the Horn River basins due to land use changes were estimated using the modified WetSpa model (e.g., Bahremand & De Smedt 2008). The runoff changes were confronted with expert judgments and estimates from the literature. The limitations of the distributed models for estimating land use changes were discussed. Material and methods The rainfall-runoff model used in this study is based on the structure of the physically-based WetSpa distributed model. Several of its components were modified in order to make it more appropriate for modelling runoff from rainfall and snowmelt in the mountainous upper Hron River basin (Hlavčová et al. 2007). The model accounts for the hydrological balance and the runoff generation on a grid scale. The balance components considered are: liquid and solid precipitation, interception, soil moisture, infiltration, actual evapotranspiration, surface runoff, interflow in the root zone, percolation into the groundwater and groundwater recharge in the saturated zone. The surface and subsurface runoff is routed by an approximation of a diffusive wave model using the geometric and hydraulic characteristics of the hillslopes and stream network. Further, a digital elevation model (DEM), a map of the land use types and a map of the soil types are needed to parameterise the model. From these maps other physiographical characteristics are derived
K. Hlavčová et al.
590
Table 1. Long-term mean annual discharges [m3 s−1 ] and their components for the actual land use and the considered scenarios of land use changes. Land use Present-day land use 1 Natural land use 2 Change in forest composition 3 Grass over forest 4 Grass over farmland 5 Afforestation of critical hydrotopes
Surface discharge
Interflow discharge
Baseflow discharge
Total discharge
3.863 0.320 4.014 3.894 1.811 3.703
7.702 3.656 7.917 7.468 7.765 7.692
13.819 13.782 13.919 14.671 13.999 13.716
25.383 17.758 25.850 26.033 23.575 25.111
change in Q [%]
20 0 surface
-20
interflow
-40
baseflow
-60
total
-80 -100 natural land use
change in forest composition
grass over forest
grass over farmland
afforestation of critical hydrotopes
Fig. 2. A comparison of changes in the long-term mean annual discharges and its components [%] for the considered scenarios of land use changes. Fig. 1. Location of the upper Hron River basin in Slovakia.
as digital maps: e.g., maps of the soil and land cover parameters, flow accumulation, flow direction, the stream network, slopes and the hydraulic radius. The model requires the calibration of 11 additional parameters. For this study the upper Hron River basin in central Slovakia (Fig. 1), which has an area of 1766 km2 , was selected. The minimum elevation of the basin is 340 m a.s.l.; the maximum elevation is 2004 m a.s.l.; and the mean elevation is 850 m a.s.l. Seventy percent of the basin’s area is covered by forest, 10 % by grasslands, 17 % by agricultural land and 3% by urban areas. The DEM used had a resolution of 100 × 100 m, and reference elevation points were added to improve the quality of the topography of the river valleys. The land use map was derived from LANDSAT satellite images made in the year 2000, and their resolution was 30 × 30 m. Daily precipitation totals were collected from 20 rain gauge stations, while the mean daily temperature was used from 6 climatic stations. The rainfall-runoff model was calibrated on discharges at the outlet gauging station at Banská Bystrica on data from the period 1981–2000 in daily time steps. The model’s performance was assessed by the Nash-Sutcliffe coefficient; the NS value achieved for the whole 20-year period was 0.732.
Results Over the course of exploring the changes in land use in the Hron River basin, different land use scenarios were created: (1) Natural land use – a scenario representing the land use closest to that of a potential natural, pristine landscape, with almost the whole basin area covered by forest; (2) change in forest composition – a scenario of changing land use towards the natural land use, which would be possible respecting the existing land use, i.e., urban land, farm land, etc.; (3) grass over forest – a scenario suggesting the forest becoming grassland; (4) grass over farmland – a scenario suggesting that arable land be left as grass; (5) afforestation
of critical hydrotopes – a scenario where the areas that generate the most runoff (according to a combination of the various geophysical characteristics defined for the hydrotopes) would be afforested, if possible. Using these scenarios, runoff from rainfall and snowmelt was simulated in daily steps for the 1981–2000 period. The resulting changes in runoff were evaluated by comparing the simulated mean daily discharges and their statistical characteristics for the existing land use and land use scenarios, as well as plotting the runoff changes spatially on a map. The comparison between mean daily discharges for each scenario and the existing land use was expressed by the values of the longterm mean daily discharges and their components and the long-term mean annual discharges and their components (Table 1). The percentage changes in the longterm mean annual discharges for each scenario as opposed to the existing land use state are illustrated in Fig. 2. The comparison between the mean daily discharges for the “natural land use” scenario and the existing land use suggests that the almost complete afforestation of the basin can indeed lead to a very significant decrease in mean daily values. The average annual discharge has decreased by 7.625 m3 s−1 , which represents a difference of –30% from the existing state. Of the runoff components, the decrease was largest for the surface runoff, i.e. 3.543 m3 s−1 (–92%), compared to the existing state and the interflow, which had a decrease of 4.045 m3 s−1 (–53%) from the present state. For this scenario, there was no apparent change in the baseflow. From the results of the mean daily discharges simulated for the “change in forest composition“ scenario compared to the actual state, it can be seen that the change in forest composition in the Hron basin is having little to no effect on the runoff. The total change in
Impacts of land use changes on runoff
591
450 400 350
Actual land use Natural land use Grass over forest Change in forest composition Grass over farmland Afforestation of critical hydrotopes
250
3
-1
Qmax [m . s ]
300
200 150 100 50 0 0
10
20
30
40
50
60
70
80
90
100
N [years]
Fig. 3. N-year values of maximum mean daily discharges.
the long-term mean annual discharge is 0.466 m3 s−1 , which is insignificant. Similar small changes can be observed for the partial discharge components. The results of the mean daily discharges for the “grass over forest“ scenario suggest that, compared to the actual state, the overall runoff can be slightly higher. This increase can be seen in the mean daily discharges as well as the increase in the long-term mean annual discharge. The average annual discharge has increased by 0.650 m3 s−1 (+2.5%). The increase in total runoff is mainly caused by a substantial increase in the baseflowup 0.853 m3 s−1 (+6%). Surface runoff remains mostly unchanged, and interflow has decreased to 0.234 m3 s−1 (–3%). The difference in mean daily discharges for the “grass over farmland“ scenario compared to the actual state means that for this scenario, a decrease in runoff is to be expected. The total long-term mean annual discharge has decreased 1.808 m3 s−1 , i.e., a –7% decrease from the actual state. The total decrease has primarily been caused by the decrease in surface discharge; its average annual value has declined 2.051 m3 s−1 (–53 % from the actual state). There was only an insignificant change in the other partial discharge components. From a comparison between the mean daily discharges for the “afforestation of critical hydrotopes“ scenario and the actual state, it can be observed that almost no change in runoff occurred for this scenario. The total long-term mean annual discharge was 0.272 m3 s−1 bellow the actual state, accounting for a –1% change from the actual state. These results are probably caused by the critical areas in the Hron basin (steep slopes, less-permeable soils) being largely afforested in reality, which therefore makes the theoretical scenario for this basin already close to the actual state. In order to account for the flood regime changes due to the land use changes in the simulated scenarios, a design flood analysis was undertaken. For each land use scenario and the actual state, design maximal mean
daily discharges for 2 to 100–year values were calculated using the DVWK/101 (1999) method. The generalised extreme value (GEV) distribution with parameter estimation using the method of probability weighted moments was the best and most applied distribution. The comparison of the estimated N-year flood values is in Fig. 3. The largest decrease in the design maximal mean daily values as opposed to the actual land use can be seen for the “natural land use” scenario for all the return periods, for the return period of 100 years, it represents −75% decrease from the actual state. Also, for the “grass over farmland” scenario, a decrease in the design maximum mean daily discharges of –12 % can be indicated for the 100-year return period. Changes in design floods for all the other scenarios are not significant; all the values of the maximal mean daily discharges are very close to the values for the actual land use, with slight increases or decreases. Discussion The reliability of the results depends not only upon the availability and quality of the input data, but also on the conceptualisation and parameterisation of the processes represented by the model. In the absence of direct experiments with land use changes in the modelled catchments the modelling results can only be confronted with results from experimental catchments other modelling studies and expert judgments. In the Hron basin the results of simulating changes in runoff due to land use changes show that the scenario representing the land use closest to that of a natural, pristine landscape, with almost the whole basin area covered by forest, has the most significant effect on changes in runoff and design floods in the upper Hron river basin. For this scenario, a significant decrease in total, surface and sub-surface runoff and interflow as well as design maximal mean daily discharges
592 was indicated. A smaller, but still significant, decrease in runoff was indicated under the scenario suggesting that arable land be left as grass. The afforestation of the critical hydrotopes indicated almost no change in the runoff regime. This result is probably caused by the critical areas in the Hron basin (steep slopes, lesspermeable soils) being largely afforested in reality. Also, other scenarios (grass over forest and change in forest composition) resulted in insignificant changes in runoff. But it is necessary to realize that these consequences can be attributed to the parameterization of the forest and grassland types of land use in the model (root depth, interception capacity, roughness), which can affect the process of forming partial runoff components in the model and which can have different manifestations during various seasons of the year and in areas in the catchment. It seems obvious that actual land use and scenario 5 gave comparable results regarding changes in all the runoff components due to the very similar land use structures. However, one could question why the change in forest composition in scenario 2 did not lead to a reduction of total runoff as in the case of scenario 1. These results would suggest that the change in the forest composition alone would not lead to changes in runoff in the basin [however, results from experimental catchments in temperate zones reported different runoff changes after reductions in forests for deciduous and coniferous forests, see, e.g., Brown et al. (2005)] and that a great amount of surface runoff and interflow are generated in anthropogenically exploited areas (arable land and urban areas) near the river network. This behaviour of the model is supported by the results of scenario 3, where the slight increase in runoff is attributed to baseflow. While the increase in baseflow after deforestation seems to be correct (a decrease in evapotranspiration), results from experimental catchments reported different runoff changes after reductions in forests and grasslands. Some of these paradoxes could be partially explained by the parameterisation of the runoff generation in the model, e.g., those in coniferous forests and on grasslands, which have similar parameters in this model. Consequently the runoff change in scenario 1 was more pronounced, because both the arable land and urban areas were afforested with deciduous forest, which has different parameters in the model (the root zone depth for coniferous and deciduous forests was considered as 1.5 and 2.0 m, respectively). Moreover, the afforested areas are located in river valleys and can be regarded as variable source areas for runoff generation. In this respect, the parameter sensitivity analysis of the original WetSpa model performed by Bahremand & De Smedt (2008) using the well known model-independent parameter estimator PEST (Doherty 2001) showed that the correction factor for calculating the actual evapotranspiration from the potential evaporation had the highest relative sensitivity. Such results, however, do necessarily imply why the model structure and its parameterization may not be adequate
K. Hlavčová et al. in a given case; they only help to gain insight as to how the particular model structure functions in a particular model experiment. Since the actual evapotranspiration is controlled by the soil moisture, the spatial properties of soil characteristics may also play a role in the case described (Gusev & Novák 2007). In this respect a description of the sensitivity of soil-water content profiles in the root zone to extraction functions based on different root morphological parameters may be particularly important (Himmelbauer et al. 2007). A possible overestimation of the overland flow from arable land is in scenario 4, which decreased dramatically. The changes in the design floods exhibit a similar paradox regarding surface runoff generation. It seems that changes in floods are mainly generated by surface runoff, since only in one case (scenario 1) was a change accompanied by a significant change in interflow. Although, the situation may be different during particular floods, the average model behaviour signals that the amount of quick runoff components may not be adequately taken into account, since the interflow should play a more pronounced role in flood generation. The change from forest to grassland does not seem to alter the surface runoff and generation; therefore, the design floods did not change. In the WetSpa model, storm runoff is generated in ways that assume a saturationexcess response to rainfall. This scheme may fail to predict variable source areas correctly where runoff could be typically generated in certain catchments and under a given land use. Therefore, based on experimental knowledge and model experiments, conceptualization of the model could modified to distribute overland flow more properly in each catchment separately (Easton et al. 2008). Conclusion As the results from this case study indicate, the calls for novel types of hydrological process models and an advancement of the underlying fundamental concepts are justified, especially with regard to the conceptualization and parameterization of the processes and representation of spatial heterogeneity. A model parameter could be investigated as interrelated entities instead of as individually independent values. This could help to partly overcome the inadequacies in the representation and parameterization of the processes and could result in sets of transferable process conceptualizations and related parameters (Zehe & Sivapalan 2007). These and other aspects of model building could be better understood through experimental research, which should accompany the development of tools used for the assessment of impacts (Pekárová et al. 2005; Strelcová et al. 2006; Tesař et al. 2006). The impact of remote sensing data on the reliability of the estimating water balance in the model could also be investigated with the aim of improving the model’s performance through an enhanced land cover representation and corresponding model modifications. The same applies to the data on the topography of the modelled basins, which serves as the basis for the estimation of a number of model
Impacts of land use changes on runoff parameters. Further, in order to improve the models, a sensitivity analysis to verify the model structures could be conducted based on tracer studies (Sieber & Uhlenbrook 2005). Acknowledgements This work was supported by the Slovak Research and Development Agency under Contract No. APVT-0378-07, LPP0254–07 and VEGA Agency Project No. 2/0096/08. References Bahremand A. & De Smedt F. 2008. Distributed Hydrological Modeling and Sensitivity Analysis in Torysa Watershed, Slovakia. Water Resour. Manag. 22: 393–408. Bl¨ oschl G., Ardoin-Bardin S., Bonell M., Dorninger M, Goodrich D., Gutknecht D., Matamoros D., Merz B., Shand P. & Szolgay J. 2007. At what scales do climate variability & land cover change impact on flooding and low flows? Hydrological Proc. 21: 1241–1247. Brown A.E., Zhang L., McMahon T.A., Western A.W. & Vertessy R.A. 2005. A review of paired catchment studies for determining changes in water yield resulting from alterations in vegetation. J. Hydrol. 310: 28–61. Doherty J. 2001. PEST-ASP Users’ manual. Watermark Numerical Computing, Brisbane, Australia. DVWK (Deutscher Verband f¨ ur Wasserwirtschaft und Kulturbau) 1999. Wahl des Bemessungshochwassers. DVWK Schriften, Heft 101. Verlag Paul Parey, Hamburg. Easton Z.M., Fuka D.R., Walter M.T., Cowan D.M., Schneiderman E.M. & Steenhuis, T.S. 2008. Re-conceptualizing the soil and water assessment tool (SWAT) model to predict runoff from variable source areas. J. Hydrol. 348: 279–291. Eckhardt K., Breuer L. & Frede H.G. 2003. Parameter uncertainty and the significance of simulated land use change effects. J. Hydrol. 273: 164–176.
593 Gusev Y. & Novák V. 2007. Soil water – main water resources for terrestrial ecosystems of the biosphere, J. Hydrol. Hydromech. 55: 3–15. Himmelbauer M.L., Novák V. & Majerčák J. 2008. Sensitivity of soil water content profiles in the root zone to extraction functions based on different root morphological parameters, J. Hydrol. Hydromech. 56: 34–44. Hlavčová K., Horvát O., Szolgay J., Danko M. & Kohnová S. 2007. Scenarios of land use changes and simulations of hydrological responses in the Poprad river basin. Meteorological J. 10: 199–203. Kostka Z. & Holko L. 2006. Role of forest in hydrological cycleforest and runoff. Meteorological J. 9: 143–148. Mattheusen B., Kirschbaum R.L., Goodman I.A., O’Donnell G.M. & Lettenmaier D.P. 2000. Effects of land cover change on streamflow in the interior Columbia River Basin (USA and Canada). Hydrol. Process. 14: 867–885. Niehoff D., Fritsch U. & Bronstert A. 2002. Land-use impacts on storm-runoff generation: scenarios of land-use change and simulation of hydrological response in a meso-scale catchment in SW-Germany. J. Hydrol. 267: 80–93. Ott B. & Uhlenbrook S. 2004. Quantifying the impact of land use changes at the event and seasonal time scale using a processoriented catchment model. Hydrol. Earth Syst. Sci. 8: 62–78. Pekárová P., Koníček A. & Miklánek P. 2005. Impact of land use on runoff regime in experimental microbasins of IH SAS. Bratislava, Veda, 216 pp. (In Slovak) Sieber A. & Uhlenbrook S. 2005. Sensitivity analyses of a distributed catchment model to verify the model structure. J. Hydrol. 310: 216–235. Střelcová K., Mindáš J. & Škvarenina J. 2006. Influence of tree transpiration on mass water balance of mixed mountain forests of the West Carpathians. Biologia 61(Suppl. 19): S305–S310. Tesař M., Šír M. & Lichner Ľ. 2006. Influence of vegetation cover on thermal regime of mountainous catchments. Biologia 61(Suppl. 19): S311–S314. Zehe E. & Sivapalan M. 2007. Towards a new generation of hydrological process models for the meso-scale: an introduction. Hydrol. Earth Syst. Sci. 10: 1–7. Received December 1, 2008 Accepted January 22, 2009