HYDROLOGICAL PROCESSES Hydrol. Process. 21, 242– 252 (2007) Published online 24 April 2006 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/hyp.6187 A distributed monthly hydrological model for integrating spatial variations of basin topography and rainfall Xi Chen,1 Yongqin David Chen2 * and Chong-yu Xu3 1 State Key Lab of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing, People’s Republic of China 2 Department of Geography and Resource Management, The Chinese University of Hong Kong, Shatin, Hong Kong 3 Department of Geosciences, University of Oslo, Oslo, Norway Abstract: Hydrological models at a monthly time-scale are important tools for hydrological analysis, such as in impact assessment of climate change and regional water resources planning. Traditionally, monthly models adopt a conceptual, lumped-parameter approach and cannot account for spatial variations of basin characteristics and climatic inputs. A large requirement for data often severely limits the utility of physically based, distributed-parameter models. Based on the variable-source-area concept, we considered basin topography and rainfall to be two major factors whose spatial variations play a dominant role in runoff generation and developed a monthly model that is able to account for their influences in the spatial and temporal dynamics of water balance. As a hybrid of the Xinanjiang model and TOPMODEL, the new model is constructed by innovatively making use of the highly acclaimed simulation techniques in the two existing models. A major contribution of this model development study is to adopt the technique of implicit representation of soil moisture characteristics in the Xinanjiang model and use the TOPMODEL concept to integrate terrain variations into runoff simulation. Specifically, the TOPMODEL topographic index lna/ tan ˇ is converted into an index of relative difficulty in runoff generation (IRDG) and then the cumulative frequency distribution of IRDG is used to substitute the parabolic curve, which represents the spatial variation of soil storage capacity in the Xinanjiang model. Digital elevation model data play a key role in the modelling procedures on a geographical information system platform, including basin segmentation, estimation of rainfall for each sub-basin and computation of terrain characteristics. Other monthly data for model calibration and validation are rainfall, pan evaporation and runoff. The new model has only three parameters to be estimated, i.e. watershed-average field capacity WM, pan coefficient and runoff generation coefficient ˛. Sensitivity analysis demonstrates that runoff is least sensitive to WM and, therefore, it can be determined by a prior estimation based on the climate and soil properties of the study basin. The other two parameters can be determined using optimization methods. Model testing was carried out in a number of nested sub-basins of two watersheds (Yuanjiang River and Dongjiang River) in the humid region in central and southern China. Simulation results show that the model is capable of describing spatial and temporal variations of water balance components, including soil moisture content, evapotranspiration and runoff, over the watershed. With a minimal requirement for input data and parameterization, this terrain-based distributed model is a valuable contribution to the ever-advancing technology of hydrological modelling. Copyright 2006 John Wiley & Sons, Ltd. KEY WORDS monthly hydrological model; digital elevation model; distributed model; Xinanjiang model; TOPMODEL Received 11 April 2005; Accepted 16 November 2005 INTRODUCTION Numerous hydrological models of watershed water balance with varying degrees of complexity have been developed and applied extensively around the world over the past half century. The earliest models are probably those using a simple accounting procedure of water budget to estimate soil moisture fluctuation and runoff production based on climatic inputs, i.e. rainfall and temperature (e.g. Thornthwaite, 1948; Thornthwaite and Mather, 1955, 1957). With the advent of digital computers in the early 1960s, hydrologists began to develop conceptual hydrological models into which more sophisticated water balance analysis could be incorporated. Later attempts were made to develop more physically based models that were supposedly able to keep track of water movement * Correspondence to: Yongqin David Chen, Department of Geography and Resource Management, The Chinese University of Hong Kong, Shatin, Hong Kong. E-mail: ydavidchen@cuhk.edu.hk Copyright 2006 John Wiley & Sons, Ltd. and budget in a discretized spatial domain using physical laws. The advancement of geoinformation technologies, such as geographical information systems and remote sensing, has offered a great impetus for the development of distributed hydrological models. However, the availability of basin characteristics and meteorological input data with sufficiently fine resolutions is usually still a critical constraint for applying distributed models. Over the history of model development, hydrological models have been adopted, modified, and applied to solve a wide spectrum of hydrological problems (e.g. Gabos and Gasparri, 1983; Alley, 1984; Vandewiele et al., 1992; Xu et al., 1996). In the past two decades, one major area of model applications is hydrological impact studies of climate change (e.g. Aston, 1984; Gleick, 1987; Arnell and Reynard, 1996; Guo et al., 2002). Hydrological models have also been widely utilized for long-range streamflow forecasting (e.g. Alley, 1985; Xu and Vandewiele, 1995). Hydrological model runs at monthly time-steps A DISTRIBUTED MODEL FOR INTEGRATING BASIN TOPOGRAPHY AND RAINFALL are often sufficient for assessing the long-term climatic change impact on water resources. The majority of the conceptual hydrological models based on the water balance concept were developed in the 1960s and 1970s. Examples include the Stanford Watershed Model (Crawford and Linsley, 1964), the Sacramento Soil Moisture Accounting Model (Burnash et al., 1973), the HBV model (e.g. Bergström, 1976) and the Xinanjiang model (Zhao et al., 1980). The structures of these models have been improved by dividing runoff and state variables such as soil moisture storage into different components (e.g. Gleick, 1987; Mimikou et al., 1991; Mohseni and Stefan, 1998). Such models contain complex nonlinear functions defining moisture fluxes between a number of conceptual storage zones. They are usually operated on time-steps of a day or an even shorter period and require more data and parameters in comparison with monthly models. Models developed before the early 1980s have a lumped structure and the spatial variation of hydrological variables and model parameters is generally not taken into account. However, some models do consider heterogeneity of rainfall and basin characteristics such as infiltration capacity to some extent. Examples include the Hydrologic Simulation Program–FORTRAN, which accounts for the spatial variations by dividing the watershed into sub-basins (Bicknell et al., 1993), the Xinanjiang model (Zhao et al., 1980) and the VIC model (Liang et al., 1994), which can implicitly simulate hydrological heterogeneity by adopting a statistical distribution of soil water characteristics. Since the mid 1980s, given the need of explicit spatial representation of hydrological components and variables, distributed-parameter models based on physical laws describing water movement have been developed (e.g. the SHE model as described in Abbott et al. (1986a,b)). These models contain detailed simulation algorithms that are often partial differential equations and numerical solution of the equations requires a tremendous amount of input data. Model applicability is seriously limited by data availability, especially for large basins. Thus, a distributed hydrological model having the flexibility to deal with different conditions in a wide range of geographical regions is required. TOPMODEL (Beven and Kirkby, 1979) has provided hydrologists with a powerful tool to simulate analytically the hillslope response of site-specific topography without the need of making use of any finite-element model. A unique feature of TOPMODEL is its capability to operate at large watershed scales by using the statistics of the topography, rather than the details of the topography itself. A model comparison study by Guo et al. (2000) shows that the statistical curve of spatial field capacity (FC) distribution in the Xinanjiang model can be approximated by a curve derived from TOPMODEL’s topographic index. This approximation means that the spatial variability of hydrological components and variables may be simulated by the Xinanjiang model if terrain information can somehow be incorporated into the model. Moreover, at Copyright 2006 John Wiley & Sons, Ltd. 243 a monthly time-scale, the Xinanjiang model is computationally efficient in runoff simulation, and soil moisture content calculation based on a simple water balance is able to characterize spatial variations of FC. The main objective of this study was to develop a simple distributed water balance model that can make use of readily available meteorological and topographic data to simulate spatial distribution of hydrological variables for the purpose of planning and management of environment and water resources over large geographical regions. The simplicity here is not a virtue in itself, but is a pragmatic response to a desire to produce a modelling approach that is capable of being applied operationally, whilst reflecting the necessary accuracy and physical relevance. The new model characterizes and combines the TOPMODEL topographic index and the mechanism of runoff generation employed in the Xinanjiang model. The model was tested on two watersheds, namely Yuanjiang and Dongjiang, in the humid region in China. THE MODEL The following mass balance equation in a continuum form is essential for the development of hydrological models (Beven, 2002): ds D rq C p e dt 1 where s is a local mass storage, rq is the divergence in local mass flux, p is a local source term (such as precipitation) and e is a local loss term (such as evapotranspiration). More sophisticated mathematical equations for hydrological dynamics can be formulated if spatial variations of basin topography are taken into consideration. Based on the fundamental principle of mass balance, a set of equations describing water balance and movement in a three-dimensional space can be used to develop a watershed hydrological model in discrete space and time increments. The spatial discretization at the scale of finite elements or finite volumes represents spatially variable soil properties and hillslope profiles. This method requires a tremendous amount of data and involves a highly time-consuming process. An alternative approach is to find mathematical functions to describe the spatial variations. This method is relatively easy in application. However, selection of a proper mathematical function to represent the spatial distribution of hydrological components is difficult because of heterogeneity and variability of catchment characteristics and hydrological responses. Rodriguez-Iturbe (2000) pointed out that the final product from Equation (1) should be a probabilistic description of soil moisture at a point as a function of climate, soil, and vegetation. Soil moisture is a key variable in hydrological modelling. Viable attempts to link the spatial structure of the soil moisture field and inherent temporal fluctuations with organization and scaling have been made to simulate successfully the hydrological processes in an interlocked system of hillslopes and channels that make Hydrol. Process. 21, 242– 252 (2007) DOI: 10.1002/hyp 244 X. CHEN, Y. D. CHEN AND C.-Y. XU up a drainage basin. The description involves both the probability distribution of soil moisture content and its correlation structure in time. In this study, a spatially distributed hydrological model is developed by adopting and combining the techniques of two well-known hydrological models, namely the Xinanjiang model and TOPMODEL. Specifically, the model is composed of three major components. The first component uses the TOPMODEL concept to estimate the spatial distribution of soil moisture deficit from terrain characteristics. The second component simulates runoff based on the runoff generation theory adopted in the Xinanjiang model, i.e. runoff generation after filling up the FC of soils. In the third component, streamflow from a sub-basin outlet is routed by a simple storage routing technique. In total there are only three parameters that need to be determined. For the sake of completeness, some important equations of TOPMODEL, the Xinanjiang model and the newly developed model are briefly summarized below. TOPMODEL concept used in this study TOPMODEL, developed by Beven and Kirkby (1979), is a physically based watershed model based on the variable-source-area concept of streamflow generation. This model requires digital elevation model (DEM) data and a sequence of rainfall and potential evapotranspiration data for predicting, among others, stream discharge. Since the theoretical basis of TOPMODEL has been clearly reported in the literature (e.g. Beven and Kirkby, 1979; Beven and Wood, 1983; Beven, 1997b), we only discuss that part of TOPMODEL used in constructing the new model. TOPMODEL makes use of a topographic index of hydrological similarity based on an analysis of DEM data. Mathematically, the topographic index is equal to lna/ tan ˇ, in which a is the cumulative area drained through a unit length of contour line and ˇ is the slope of the unit area. In general, TOPMODEL predicts the catchment responses following a series of rainfall events and maintains a continuous accounting of the storage deficit, which identifies the saturated source areas within a catchment. This model involves a number of important hydrological equations and variables for its simulation purpose. The local soil moisture deficit is calculated by: Si D S C m[ lna/ tan ˇi ] The Xinanjiang model algorithms The Xinanjiang model was first developed in 1973 and published in English in 1980 (Zhao et al., 1980). It is a well-known lumped watershed model and has been widely used in China. In comparison with other lumped hydrological models, the Xinanjiang model describes watershed heterogeneity using a parabolic curve of FC distribution (Zhao et al., 1980): b f WM0 3 D1 1 F WMM where WM0 is the FC at a point, which varies from zero to the maximum of the whole watershed WMM, and b is a power index. The f/F versus WM0 curve is shown in Figure 1. WM, the watershed average FC, is the integral of (1 f/F) between WM0 D 0 and WM0 D WMM, obtaining WM D WMM/1 C b 4 Wt , the watershed-average soil moisture storage at time t, is the integral of 1 f/F between zero and WMŁt , a critical FC at time t (Figure 1): WMŁt f Wt D 1 dWM0 D WM F 0 WMŁt 1Cb ð 1 1 5 WMM Thus, the critical FC WMŁt corresponding to watershedaverage soil moisture storage Wt is 1/1Cb W t 6 WMŁt D WMM 1 1 WM Runoff occurs where soil moisture reaches FC. As shown in Figure 1, if the net rainfall amount (rainfall less WMM 2 where S is the average storage deficit, m is a scaling parameter, and is the areal average of lna/ tan ˇ. Equation (2) is used to predict the saturated contributing area at each time step. A negative value of Si indicates that the area is saturated and saturation overland flow is generated, whereas a positive value of Si indicates that the area is unsaturated. Unsaturated zone calculations are made for each lna/ tan ˇ increment. The subsurface flow rate per unit width of contour length qi , the vertical flow to the zone, and the outflow Copyright 2006 John Wiley & Sons, Ltd. from the saturated zone Qb are all dependent on the topographic index. Therefore, TOPMODEL is a distributed model and can calculate spatial variations of hydrological components based on the distribution of the topographic index. R P-E ∆Wt WM∗ Wt 0 f/F 1 Figure 1. FC curve of soil moisture and rainfall–runoff relationship. WMM is maximum FC in a watershed; f/F is a fraction of the watershed area in excess of FC; WMŁt is FC at a point in the watershed; Rt is runoff yield at time t; wt is soil moisture storage deficit at time t and is equal to WM Wt ; Wt is watershed-average soil moisture storage at time t Hydrol. Process. 21, 242–252 (2007) DOI: 10.1002/hyp 245 A DISTRIBUTED MODEL FOR INTEGRATING BASIN TOPOGRAPHY AND RAINFALL actual evapotranspiration) in a time interval [t 1, t] is Pt Et and initial watershed-average soil moisture (tension water) is Wt , then runoff yield in the time interval Rt can be calculated as follows. If Pt Et WMŁt t fi WMM Wt D 1 F N iD1 NWMŁ < WMM Rt D Pt Et Wt Pt Et CWMŁt f D Pt Et 1 dWM0 7 F WMŁt D Pt Et WM C Wt C WM[1 Pt Et WMŁt /WMM]1Cb If Pt Et WMŁt ½ WMM Rt D Pt Et WM C Wt 8 In Equation (3), the f/F versus WM0 curve describes the watershed heterogeneity of FC in a statistical way. Parameter b represents the spatial heterogeneity of FC (b D 0 for uniform distribution and large b for significant spatial variation) and it is usually determined by model calibration. Therefore, the Xinanjiang model is a lumped hydrological model used in a watershed where streamflow discharge and meteorological data are available. In a hilly mountain area, the topographic index can be used to represent the influences of terrain on the spatial variations of soil wetness. A larger topographic index in a local area means less soil moisture deficit or easier runoff generation in response to rainfall input. On the contrary, a larger WM0 means a larger soil moisture storage capacity in a local area and more difficult runoff generation. Comparing the spatial distribution of WM0 and lna/ tan ˇ, Guo et al. (2000) demonstrated that f/F versus WM0 /WMM (normalized f/F versus WM0 curve in Figure 1) in Equation (3) can be substituted by a curve of f/F versus IRDG (normalized f/F versus lna/ tan ˇ curve) defined as an index of relative difficulty of runoff generation, calculated by max[lna/ tan ˇ] lna/ tan ˇ max[lna/ tan ˇ] min[lna/ tan ˇ] 9 where max[lna/ tan ˇ] and min[lna/ tan ˇ] represent the maximum and minimum topographic index respectively. Thus, curves of f/F versus WM0 /WMM in Equation (3) can be calculated from DEM data. If the FC of the basin varies from zero to WMM, which can be divided into N segments, then the FC WM0 corresponding to the ith segment is equal to iWMM/N and the fraction of area in excess of FC is fi /F. Then, WM, the watershed-average FC of the basin, in Equation (4) can be written as N WMM fi WM D 1 N iD1 F Copyright 2006 John Wiley & Sons, Ltd. 10 11 where NWMŁt is the NWMŁt th segment of WMM, and the WMŁt corresponding to that segment is given by WMŁt D NWMŁt WMM N 12 NWMŁt and WMŁt can be obtained from Equations (11) and (12) using an iteration method. When Pt Et > 0, runoff emerges and its amount is calculated using equations (13) and (14). Rt D DEM-based FC distribution and runoff generation IRDG D Then Equation (5) for determining the relationship between WMŁt and Wt is substituted by Pt Et CWMŁt WMŁt NP f fi WMM 0 dWM ³ F F N N Ł WM when Pt Et C WMŁt t < WMM 13 where NP is calculated by solving Equations (11) and (12) if WMŁt in these two equations is substituted for Pt Et C WMŁt . Rt D Pt Et WM Wt when Pt Et C WMŁt ½ WMM 14 where Wt is the watershed-average soil moisture storage at time t in the unsaturated zone and can be calculated by the following water balance equation (Zhao et al., 1980): Wt D Wt1 C Pt Et Rt 15 where Et is actual evapotranspiration and is estimated using Wt 1/Be 16 Et D Ep 1 1 WM where is a conversion coefficient from pan evaporation to potential evapotranspiration, Ep is pan evaporation and Be ³ 0Ð6 (Ripple et al., 1972). As illustrated above, using the terrain-based index IRDG as a surrogate to represent the spatial patterns of runoff generation is a major invention for the new model. Based on the distribution of rainfall stations and basin topography, the study basin will first be divided into sub-basins and then a rainfall amount and a curve of f/F versus WM0 /WMM can be obtained for each sub-basin. Depending on the distribution of rainfall stations, basin topography and river channel network as characterized by DEM data, a modeller may design a segmentation scheme to divide the basin into a certain number of sub-basins. Whereas the basin segmentation reflects the spatial distribution of rainfall, a curve of f/F versus WM0 /WMM for each sub-basin takes the effects of terrain on soil moisture storage and runoff yield into account. In other words, these procedures and algorithms have been developed to integrate the spatial variations of rainfall Hydrol. Process. 21, 242– 252 (2007) DOI: 10.1002/hyp 246 X. CHEN, Y. D. CHEN AND C.-Y. XU and basin topography into the distributed hydrological model. Storage routing of watershed Point or local runoff is regulated by watershed surface, subsurface and stream channel systems before it reaches watershed outlet. All surface runoff and a portion of subsurface runoff in a shallow layer will flow out of the watershed within a calculation time interval of 1 month; the rest will be stored in the subsurface soil and flow out in successive months. If the watershed regulation on a monthly scale is treated as a linear reservoir, then the following simple storage routing approach can be used to simulate streamflow: Qt D Qt1 e˛ C It 1 e˛ 17 It D Rt F/t 18 and where Qt and Qt1 are the discharges at time t and t 1 respectively, ˛ is a parameter used to describe the watershed regulation of monthly runoff, and F is watershed area. Model parameters There are only three parameters that need to be determined, i.e. WM, and ˛. The first two parameters influence runoff generation, and ˛ influences the slope of the streamflow hydrograph. Previous studies (Zhao, 1984; Zhao and Wang, 1988; Huang, 1993) indicate that the watershed-average FC WM is mainly dependent on climatic dryness or wetness. It is smaller in the humid region and larger in the dry region of China. Zhao (1984) demonstrated that runoff generation is insensitive to WM and a certain value can be set depending on the climatic zone. Approximate values of 120 mm for regions south of the Yangtze River and 160 mm for regions north of the Yanshan Mountains and northeastern China are recommended by Zhao (1984; Zhao and Wang, 1988). According to the Ministry of Water Resources (1992), the parameter varies between 0Ð72 and 1Ð00 for an evaporation pan of 80 cm in diameter and varies between 0Ð53 and 0Ð80 for an evaporation pan of 20 cm in diameter, with the larger values for the humid climate in the southeastern China and the smaller values for the dry climate in the northwestern China. Therefore, values of WM and are fairly consistent for basins located in the same climatic zone. As shown later, the results of model validation using the same parameters WM, , and ˛ for nested basins from upstream to downstream not only confirm the validity of the above findings and recommendations, but also demonstrate that the watershed regulation parameter ˛ varies very little spatially across a large basin on a monthly interval. STUDY WATERSHEDS AND DATA FOR MODEL TESTING Two watersheds were selected to test the model through calibration and validation. Yuanjiang watershed and Copyright 2006 John Wiley & Sons, Ltd. Dongjiang watershed are major tributaries of the Yangtze River (Changjiang) and the Pearl River (Zhujiang) respectively, two large rivers in China. As shown in Figure 2, both watersheds are located in southern China. Yuanjiang watershed Yuanjiang River, originating from Yunwu Mountain in Guizhou province, has a length of 1033 km and a drainage area of 78 595 km2 above the Yuanling hydrological station (Changjiang Water Resources Commission, 2002). Mean annual rainfall, potential evapotranspiration (from 20 cm diameter evaporation pan) and runoff for the period 1971–85 are 1302 mm, 1152 mm and 695 mm respectively. Owing to the dominance of monsoon climate, more than 73% of the annual rainfall occurs in the wet season from April to September. Significant variations of topography from 90 m above the mean sea level in the downstream region to over 2000 m in the upstream mountainous areas (Figure 2) result in a remarkable spatial variation of annual rainfall from 927 mm to 1657 mm. Therefore, spatial variations of topography and rainfall play an essential role in runoff generation and distribution of hydrological processes throughout the watershed. The description of the spatial variations of rainfall is based on monthly precipitation data recorded at 121 precipitation stations from 1971 to 1985. Basin topography is characterized by DEM data with 50 m grid resolution. Monthly streamflow data recorded at four hydrological stations, i.e. Jingping, Anjiang, Pushi, and Yuanling from upstream to downstream, are used for model calibration and validation. Basic information for the four nested basins in the Yuanjiang watershed is presented in Table I. Dongjiang watershed Dongjiang River has a length of 439 km and drains an area of 25 325 km2 above Boluo, the lowest hydrological station of the study watershed. It is located in a subtropical region dominated by a monsoon climate with significant seasonal variations in rainfall and wind. Mean annual precipitation is 1768 mm. Spatial and temporal variations of rainfall are remarkable. About 76–86% of annual rainfall falls in the wet season (from April to September). Frontal rainfall mainly occurs from April to June, and rainfall brought by typhoons occurs from June to September mainly in the southern part of the region near the coast. Rainfall decreases from the southwest to the northeast within the watershed. For all of the rainfall stations, the coefficient of variation (CV) of annual rainfall falls between 0Ð20 and 0Ð25, and the ratio of maximum to minimum annul rainfall varies from 2Ð03 to 3Ð48. Monthly meteorological and hydrological data recorded from 1971 to 1985 were used for model testing, including rainfall at 52 stations, 80 cm diameter pan evaporation at three stations, and streamflow at four stations. DEM data of 50 m grid resolution are used to Hydrol. Process. 21, 242–252 (2007) DOI: 10.1002/hyp Copyright 2006 John Wiley & Sons, Ltd. 50 Jingping <200 m 200 ~ 400 m 400 ~ 600 m 600 ~ 800 m 800 ~ 1000 m >1000 m Elevation Yuanjiang Basin N 0 Anjiang Pushi Yuanling China Beijing Boluo N 10 0 10 20 30 40 50 km Dongjiang Basin Figure 2. Location, topography and nested basins of the Yuanjiang and Dongjiang watersheds 25 50 75 100 km N <100 m 100 ~ 200 m 200 ~ 300 m 300 ~ 400 m 400 ~ 600 m >600 m Elevation Heyuan Longchuan Fengshuba A DISTRIBUTED MODEL FOR INTEGRATING BASIN TOPOGRAPHY AND RAINFALL 247 Hydrol. Process. 21, 242– 252 (2007) DOI: 10.1002/hyp 248 X. CHEN, Y. D. CHEN AND C.-Y. XU Table I. Basin characteristics of the two study watersheds and their nested basins Yuanjiang watershed Dongjiang watershed Jingping Anjiang Pushi Yuanling Fengshuba Longchuan Heyuan Boluo 13 485 177 15 1 1 190 1 164 615 40 305 530 31 1 1 240 1 187 619 54 144 712 83 2 1 275 1 162 645 78 595 1 033 121 3 1 302 1 152 695 5 151 112 11 1 1 548 1 393 796 7 699 165 15 1 1 622 1 393 795 15 750 293 31 2 1 718 1 407 893 25 325 439 52 3 1 768 1 391 912 0.8 1.0 Area (km2 ) Length of mainstem (km) No. of rainfall stations No. of evaporation stations Mean annual rainfall (mm) Mean annual pan evaporation (mm) Mean annual runoff (mm) describe the spatial variations of topography (Figure 2). Basic information for the Dongjiang watershed and its nested basins is given in Table I. 1.0 IRDG 0.8 0.6 0.4 RESULTS AND DISCUSSION 0.2 Watershed division and calculation of IRDG 0.0 In the Yuanjiang and the Dongjiang watersheds, spatial variations of topography and rainfall, which dominate the rainfall–runoff relationship, can be described by dividing the two watersheds into a number of sub-basins. The Yuanjiang watershed was divided into 49 sub-basins and the Dongjiang watershed into 17 sub-basins. Monthly rainfall in each of the sub-basins was calculated from observation data of rainfall stations within the sub-basin, and IRDG was calculated using the DTM9704 program (Beven, 1997a,b) and a DEM at 50 m grid resolution. The calculated curves of f/F versus IRDG for all subbasins are shown in Figures 3 and 4 for the Yuanjiang and Dongjiang watersheds respectively. Model calibration and validation The model was calibrated and validated based on streamflow discharges observed in the whole watershed outlet in calibration and validation periods respectively. Further validation for different spatial scales was made based on streamflow discharges observed in nested watershed outlets. For the monthly model, the watershed-average FC of soil moisture WM is set as 130 mm in the Yuanjiang and 115 mm in the Dongjiang. The other two parameters need 1.0 IRDG 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Cumulative frequency (f/F) Figure 3. Cumulative frequency of IRDG for Yuanjiang watershed and its sub-basins. Each curve represents an f/F versus IRDG relationship for one of the 49 sub-basins. In the Xinanjiang model, f/F is a fraction of the watershed area in excess of FC Copyright 2006 John Wiley & Sons, Ltd. 0.0 0.2 0.4 0.6 Cumulative frequency (f/F) Figure 4. Cumulative frequency of IRDG for Dongjiang watershed and its sub-basins. Each curve represents an f/F versus IRDG relationship for one of the 17 sub-basins. In the Xinanjiang model, f/F is a fraction of the watershed area in excess of FC to be calibrated, i.e. pan coefficient of evapotranspiration and watershed regulation coefficient of the basin ˛. The model parameters are calibrated automatically using the simplex acceleration technique with respect to the objective function of maximizing the Nash–Sutcliffe efficiency coefficient (NSC) between observed and simulated monthly discharges. Another measure of model performance used in the study is the root-mean-squared error (RMSE) of monthly discharge. For the Yuanjiang watershed, the calibration period is 1971–79 and the validation period is 1980–85. The simulated streamflow from 1971 to 1985 in each of the three observation stations (Jingping, Anjiang and Pushi) was further compared with the observed data. Model calibration and validation results are shown in Table II. The calibrated model parameter values are 0Ð55 and 0Ð991 for and ˛ respectively. The NSC values are 0Ð91 and 0Ð87 respectively for the calibration period and the validation period for Yuanling. NSC values in the three nested watersheds of Anjiang, Pushi and Jingping are 0Ð75, 0Ð89 and 0Ð89 respectively. For illustrative purposes, monthly simulated and observed runoff in Yuanling for the calibration period 1971–79 and for the validation period 1980–85 are shown in Figures 5 and 6 respectively. These results demonstrate that the model is capable of reproducing both the magnitude and the dynamics of the monthly discharge at different spatial scales. For the Dongjiang watershed, the model calibration was done for the period 1960–74 and the validation for 1975–88. Model validation at spatial scales was executed for the three observation stations (Fengshuba, Longchuan and Heyuan) for the period 1960–88. Hydrol. Process. 21, 242–252 (2007) DOI: 10.1002/hyp 249 A DISTRIBUTED MODEL FOR INTEGRATING BASIN TOPOGRAPHY AND RAINFALL 250 450 Simulation Observation Runoff (mm) Runoff (mm) 200 150 100 400 Simulation 350 Observation 300 250 200 150 100 50 0 1971 50 1972 1973 1974 1975 1976 1977 1978 0 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 Time (year) 1979 Time (year) Figure 5. Monthly calibrated and observed runoff of Yuanjiang watershed for the period 1971– 79 Figure 7. Monthly calibrated and observed runoff of Dongjiang watershed for the period 1960– 74 400 Simulation Observation 350 Simulation Observation Runoff (mm) Runoff (mm) 450 200 180 160 140 120 100 80 60 40 20 0 1980 300 250 200 150 100 50 0 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 Time (year) 1981 1982 1983 1984 Figure 8. Monthly validated and observed runoff of Dongjiang watershed for the period 1975– 88 1985 Time (year) Figure 6. Monthly validated and observed runoff of Yuanjiang watershed for the period 1980– 85 The calibrated model parameter values at Boluo are 0Ð67 and 0Ð917 for and ˛ respectively. NSC is 0Ð89 in the calibration period and 0Ð90 in the validation period (Table II). The NSC values in the three nested watersheds Fengshuba, Longchuan and Heyuan are 0Ð81, 0Ð85 and 0Ð88 respectively. Monthly observed and simulated runoffs at the Boluo station for the calibration period 1960–74 and for the validation period 1975–88 are shown in Figures 7 and 8 respectively. Again, the results demonstrate that the model is reliable for runoff simulation at different spatial scales. Successful application of the model in the nested watersheds using the same values of WM, and ˛ indicates that these parameters are less spatially variable for the monthly hydrological model applied at the watershed scale. each parameter (the ratio of parameter changes to the calibrated model parameter dP/P) are used as an indicator of the sensitivity of runoff to parameter changes. For illustrative purposes, the relationship between parameter changes and the corresponding streamflow changes in the Dongjiang watershed is shown in Figure 9. For the three parameters, ˛ has the most significant effect on streamflow. A 10% change in parameter ˛ results in approximately the same percentage of streamflow changes, compared with 5Ð8% for and only 0Ð5% for WM. The results show that parameter WM is least sensitive to streamflow, and using fixed values of 130 mm and 115 mm respectively for the Yuanjiang and the Dongjiang watersheds is warranted. Similar results were obtained in the Yuanjiang watershed. The effect of change of rainfall amount on model output was also tested; as expected, rainfall had the most significant effect on annual streamflow. A 5% change in rainfall changed streamflow by approximately 10% in the two watersheds. Sensitivity analysis Sensitivity analysis was carried out on both watersheds to evaluate and quantify the effect of the parameter variations on model output. Relative changes of annual streamflow (the ratio of streamflow changes to annual mean streamflow dR/R) resulting from the relative changes of Spatial variation of hydrological components Basin topography, soil and vegetation cover affect hydrological processes; consequently, they influence the energy balance, which is a key component in global climate models. Numerous efforts have been made to Table II. Results of model calibration and validation of the two study watersheds and their nested basins Yuanjiang watershed Yuanling Annual runoff (mm) Simulated 716a Observed 710a NSC 0Ð91a RMSE (mm) 16Ð2a a Calibration Dongjiang watershed Anjiang Pushi Jingping 642 647 0Ð75 22Ð8 620 620 0Ð89 15Ð7 642 646 0Ð89 15Ð8 664 678 0Ð87 17Ð0 Boluo 890a 880a 0Ð89a 26Ð7a 986 993 0Ð90 22Ð6 Fengshuba Longchuan Heyuan 806 797 0Ð81 26Ð7 823 830 0Ð85 24Ð0 912 936 0Ð88 25Ð7 results. Copyright 2006 John Wiley & Sons, Ltd. Hydrol. Process. 21, 242– 252 (2007) DOI: 10.1002/hyp 250 X. CHEN, Y. D. CHEN AND C.-Y. XU describe the spatial variations of hydrological components, and parameterizations of hydrological models are made for a successful assessment of climate-change impacts on hydrology and water resources and estimation of rainfall-runoff in ungauged areas. For hydrological modelling, the distribution of the three water budget components, i.e. soil moisture content, actual evapotranspiration and runoff, is vital for water resources management, environmental protection and assessment of land use and climate-change impacts on hydrology. A major strength of the monthly model is its capability of simulating the spatial variations of these components. The newly developed model was executed for each of the segmented sub-basins with rainfall input and IRDG curves. The hydrological components, i.e. soil moisture 0 15 Figure 11. Spatial variation of the simulated actual evapotranspiration in Yuanjiang watershed in November 1973 Soil moisture content 75 - 80mm 80 - 85mm 85 - 90mm 90 - 95mm 95 - 100mm 100 - 105mm 105 - 110mm 110 - 115mm 115 - 120mm 120 - 125mm 125 - 130mm 130 - 135mm 10 5 dP/P (%) 0 -20 -15 -10 -5 0 5 10 15 N 100 Kilometers dR/R (%) 20 WM α η Actual evaporation 18 - 22mm 22 - 25mm 25 - 28mm 28 - 31mm 31 - 34mm 34 - 37mm 37 - 40mm 40 - 43mm 20 -5 -10 0 100 Kilometers N -15 Figure 12. Spatial variation of the simulated soil moisture storage in the Yuanjiang watershed in November 1973 -20 17 0 0 16 00 Figure 9. Sensitivity of the model simulated discharge to the change of model parameter values for Dongjiang watershed. The Y-axis is the relative change in the simulated discharge and the X-axis is the relative change in parameter values 1400 00 0 15 130 16 00 150 0 13 00 1300 00 14 0 140 1300 00 11 1200 0 00 150 1400 0 0 130 140 0 150 1700 0 1400 14 1000 1100 1200 N 100 Kilometers Figure 10. Spatial variation of the mean annual rainfall in the Yuanjiang watershed for the period 1971– 85 Copyright 2006 John Wiley & Sons, Ltd. content, actual evapotranspiration and runoff, are simulated for each sub-basin with their spatial variations taken into account. For illustrative purposes, spatial variations of mean annual precipitation (1971–85), actual evapotranspiration (November 1973), soil moisture content (November 1973) and mean annual runoff (1971–85) simulated by the model for the Yuanjiang watershed are shown in Figures 10–13 respectively. Figure 10 shows that precipitation is greater in the high mountain area of the upper stream (the highest elevation is 2570 m) and smaller in the western centre of the watershed. Spatial variations of simulated actual evapotranspiration are similar to those of the soil moisture content (Figure 11). Simulation results indicate that soil moisture content in November 1973 is greater in the south and smaller in the western centre and the northern areas of the watershed (Figure 12). The runoff distribution in Figure 13 demonstrates that greater runoff occurs in the high mountain areas with greater precipitation and steeper slopes. CONCLUSIONS The Xinanjiang model is well known for its implicit description of watershed heterogeneity using a parabolic curve of FC. Traditionally, the parabolic curve is calibrated on the basis of observed stream discharges, and Hydrol. Process. 21, 242–252 (2007) DOI: 10.1002/hyp 10 00 11 00 A DISTRIBUTED MODEL FOR INTEGRATING BASIN TOPOGRAPHY AND RAINFALL TOPMODEL and the Xinanjiang model have been successfully used in humid regions worldwide. However, the use of the proposed modelling approach in arid regions is yet to be tested. 0 90 700 800 0 90 700 100 0 800 251 ACKNOWLEDGEMENTS 800 900 This research was substantially supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (project no. CUHK4247/ 03H) and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry, China. 500 600 700 800 900 0 900 800 900 700 400 600 500 800 N 100 Kilometers Figure 13. Spatial variation of the simulated mean annual runoff in the Yuanjiang watershed for the period 1971– 85 thus the model could not be applied in an ungauged watershed. We consider rainfall and terrain to be the two dominant spatially variable factors that influence the runoff generation processes. Segmentation of a study basin into sub-basins will reflect the spatial distribution of rainfall. A surrogate for the FC curve using a DEM-based curve of index of relative difficulty of runoff generation has been successfully adopted to develop a distributed monthly water balance model. The new model incorporates the TOPMODEL topographic index into the mechanism of runoff generation of the Xinanjiang model, making it a distributed model with small data requirements and high applicability. The model has three parameters, only two of which need to be calibrated. These two parameters are less spatially variable for watersheds of different sizes located in the same climate zone, although they are highly sensitive to the basin water yield. Therefore, the model can be applied successfully in those areas where rainfall and topography distribution dominate the spatial variation of runoff generation. The model’s accuracy and reliability have been tested in four nested basins from upstream to downstream in the Dongjiang watershed and the Yuanjiang watershed respectively using monthly meteorological and hydrological series. Simulation results demonstrate that not only is the model capable of reproducing both the magnitude and the dynamics of the monthly discharge for basins of different sizes, but it is also able to produce reasonable spatial variations of major water balance components, such as soil moisture storage and actual evapotranspiration. The study shows that, with prudent simplification, a distributed hydrological model based on terrain analysis is appropriate for finding a reasonable solution of regional hydrological problems associated with planning, optimal allocation and management of water resources. Based on the results of model testing, we believe that the conclusions reached in this paper can be extended to other basins in humid regions. This is because both Copyright 2006 John Wiley & Sons, Ltd. REFERENCES Abbott MB, Bathurst JC, Cunge JA, O’Connell PE, Rasmussen J. 1986a. An introduction to the European Hydrological System–Système Hydrologique European, ‘SHE’, 1: history and philosophy of a physically-based distributed modelling system. Journal of Hydrology 87: 45– 59. Abbott MB, Bathurst JC, Cunge JA, O’Connell PE, Rasmussen J. 1986b. An introduction to the European Hydrological System–Système Hydrologique European, ‘SHE’, 2: structure of a physically-based distributed modelling system. Journal of Hydrology 87: 61–77. Alley WM. 1984. On the treatment of evapotranspiration, soil moisture accounting and aquifer recharge in monthly water balance models. Water Resources Research 20(8): 1137– 1149. Alley WM. 1985. Water balance models in one-month-ahead stream flow forecasting. Water Resources Research 21(4): 597– 606. Arnell NW, Reynard NS. 1996. The effects of climate change due to global warming on river flows in Great Britain. Journal of Hydrology 183: 397–424. Aston AR. 1984. The effect of doubling atmospheric CO2 on streamflow: a simulation. Journal of Hydrology 67: 273–280. Bergström S. 1976. Development and application of a conceptual runoff model for Scandinavian catchments. Department of Water Resources Engineering, Lund Institute of Technology, Bulletin Series A-52, Swedish Meteorological and Hydrological Institute, Norrköping, Sweden. Beven KJ. 1997a. TOPMODEL user notes (Windows version). Centre for Research on Environmental Systems and Statistics. Lancaster University, UK. Beven KJ. 1997b. TOPMODEL: a critique. Hydrological Processes 11: 1069– 1085. Beven KJ. 2002. Towards an alternative blueprint for a physically based digitally simulated hydrologic response modelling system. Hydrological Processes 16: 189–206. Beven KJ, Kirkby MJ. 1979. A physically based variable contributing area model of basin hydrology. Hydrological Sciences Journal 24: 43– 69. Beven KJ, Wood EF. 1983. Catchment geomorphology and the dynamics of runoff contributing areas. Journal of Hydrology 65: 139–158. Bicknell BR, Imhoff JC, Kittle Jr JL, Donigian AS, Johanson RC. 1993. Hydrologic Simulation Program–FORTRAN (HSPF): users manual for release 10 . EPA/600/R-93/174, US Environmental Protection Agency, Athens, GA. Burnash RJC, Ferral RL, Maguire RA. 1973. A generalized streamflow simulation system: conceptual models for digital computer, Joint FedState River Forecast Center, Sacramento, CA. Changjiang Water Resources Commission. 2002. Flood and Drought Disasters in Changjiang Basin. China Water and Power Press: Beijing (in Chinese). Crawford NH, Linsley RK. 1964. A conceptual model of hydrologic cycle. In Commission on Subterranean Waters. IAHS Publication No. 63. IAHS Press: Wallingford; 573– 587. Gabos A, Gasparri L. 1983. Monthly runoff model for regional planning. Water International 8: 42–45. Gleick PH. 1987. The development and testing of a water balance model for climate impact assessment: modeling the Sacramento basin. Water Resources Research 23(6): 1049– 1061. Hydrol. Process. 21, 242– 252 (2007) DOI: 10.1002/hyp 252 X. CHEN, Y. D. CHEN AND C.-Y. XU Guo F, Liu XR, Ren LL. 2000. Topography based watershed hydrological model—TOPOMODEL and its application. Advances in Water Science 11(3): 296–301 (in Chinese). Guo SL, Wang JX, Xiong LH, Ying AW, Li DF. 2002. A macro-scale and semi-distributed monthly water balance model to predict climate change impacts in China. Journal of Hydrology 268: 1–15. Huang YK. 1993. Calculation methods on runoff generation and flow concentration in ungauged regions. PhD dissertation, Hohai University, Nanjing (in Chinese). Liang X, Lettenmaier DP, Wood EF, Burges SJ. 1994. A simple hydrological based model of land surface water and energy fluxes for general circulation models. Journal of Geophysical Research 99(D7): 14 415– 14 428. Ministry of Water Resources. 1992. Water Resources Assessment for China. China Water and Power Press: Beijing (in Chinese). Mimikou M, Kouvopoulos Y, Cavadias G, Vayianos N. 1991. Regional hydrological effects of climate change. Journal of Hydrology 123: 119– 146. Mohseni O, Stefan HG. 1998. A monthly streamflow model. Water Resources Research 34(5): 1287– 1298. Ripple CD, Rubin JT, van Hylckama TEA. 1972. Estimating SteadyState Evaporation Rates from Bare Soils under Conditions of High Water Tables. US Geological Survey, Water-Supply Paper 2019-A. US Geological Survey: Washington, DC. Rodriguez-Iturbe I. 2000. Ecohydrology: a hydrologic perspective of climate–soil–vegetation dynamics. Water Resources Research 36(1): 3–9. Copyright 2006 John Wiley & Sons, Ltd. Thornthwaite CW. 1948. An approach toward a rational classification of climate. Geographical Review 38: 55–94. Thornthwaite CW, Mather JR. 1955. The water balance. Publications in Climatology, Laboratory of Climatology, Drexel Institute of Technology 8(1): 1–104. Thornthwaite CW, Mather JR. 1957. Instructions and tables for computing potential evapotranspiration and the water balance. Publications in Climatology, Laboratory of Climatology, Drexel Institute of Technology 10(3): 185–311. Vandewiele GL, Xu CY, Ni-Lar-Win. 1992. Methodology and comparative study of monthly water balance models in Belgium, China and Burma. Journal of Hydrology 134: 315–347. Xu CY, Vandewiele GL. 1995. Parsimonious monthly rainfall-runoff models for humid basins with different input requirements. Advance in Water Resources 18: 39–48. Xu CY, Seibert J, Halldin S. 1996. Regional water balance modelling in the NOPEX area—development and application of monthly water balance models. Journal of Hydrology 180: 211–236. Zhao RJ. 1984. Water Hydrological Modeling—Xinanjiang Model and Shanbei Model . China Water Resources and Hydropower Publishing House: Beijing (in Chinese). Zhao RJ, Wang PL. 1988. Parameter determination of Xinanjiang model. Hydrology 6: 2–9 (in Chinese). Zhao RJ, Zhuang YL, Fang LR, Liu XR, Zhang QS. 1980. The Xinanjiang model. In Hydrological Forecasting, IAHS Publication No. 129. IAHS Press: Wallingford; 351– 356. Hydrol. Process. 21, 242–252 (2007) DOI: 10.1002/hyp