ASSESSING GRID SIZE EFFECTS ON RUNOFF AND FLOW RESPONSE USING A GIS-BASED HYDROLOGIC MODEL Y. B. Liu 1,2, *, Y. Yi 1 , O. Batelaan 1 , F. De Smedt 1 1 Department of Hydrology and Hydraulic Engineering, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels, Belgium - yongbliu@vub.ac.be 2 Department of Geography, University of Guelph, Guelph, Ontario, Canada N1G 2W1 ABSTRACT The effects of grid size on model parameters and on the performance of a GIS-based hydrological model intended for flood simulation is assesses in this paper. The WetSpa model is used to estimate runoff and flow discharge in the 1176 km 2 Alzette river basin, Grand-duchy of Luxembourg. The GIS ArcView and its spatial analyst are used to extract hydrological model parameters from DEMs, soil and land use maps with cell sizes of 50, 100, 200, 400 and 800m. The value of the model parameter derived from the 50m data is used as the reference value, with estimates of model parameters of coarser resolutions compared against it. Results show that in this integrated model the grid size has a considerable effect on runoff and outflow hydrographs. However, coarser grid cell resolutions can be used for runoff simulations as long as parameters are appropriately calibrated. KEY WORDS: Grid size, Runoff, Flow response, GIS, WetSpa 1. INTRODUCTION Grid-based models are now widely used in simulating rainfall-runoff and other hydrologic processes, which operate by assigning a single value for each raster cell to represent terrain features, model parameters, and input/output variables. Typically, these models use a 1D representation of channel flow linked to some simple model of flow between grids of cells on the hillslope and floodplain. Thus the diffusive wave propagation approach can be applied in the raster-based model serving the equations governing fluid flow. Despite their crude representation of hydraulic processes, these models have been shown to give good results when compared to more complex approaches (Horritt and Bates, 2001). In the development of such models, there has been a trend among many modelers to increase the spatial resolution in the expectation of improved insight into temporal and spatial processes. As pointed out by Farajalla and Vieux (1995), there is a tendency to assume that an increase in the number of elements will improve the realism of the model’s predictive ability, owing to the heterogeneity of natural systems. However, the spatial resolution at which a model is applied affects the solution of the equations and thus the simulation results. Spatial heterogeneity affecting catchment response arises from three sources: variability, discontinuity, and process (Singh and Woolhiser, 2002). Spatial variability in climate inputs such as rainfall and hydrometorological variables, in soil characteristics such as hydraulic conductivity and porosity, in topography and in land use, encompass a space-time continuum. Discontinuities encompass the boundaries separating soil types, geologic formations, and land covers. Hydrologic processes such as surface retention, infiltration, overland flow, and evapotranspiration, are controlled by physical properties at different scales. As a result, the runoff and flow responses from a watershed are governed by local combinations of these factors. Raster based models work with grid cells which are assumed to be homogeneous with representative parameters for each cell. However, the size of a grid cell affects the homogeneity assumption, as larger sizes are likely to have variable conditions within the grid. Reducing the size and increasing the number of grid may improve the modeling accuracy, while increasing the input data preparation effort and the subsequent computational effort. On the other hand, using large grid size may result in inadequate representation of catchment characteristics and poor modeling results. Many studies have demonstrated the effects of grid size on derived terrain variables and the modeling results over the past years. Kienzle (2004) examined the effect of DEM raster resolution on first order, second order and compound terrain derivatives based on the statistical analysis of 3 small areas in Alberta, Canada, using a variety of grid cell sizes under terrain conditions ranging from moderately steep to very flat. The subsequent correlation analysis reveals that only elevation and local slope have a strong positive relationship while all other terrain derivatives are not represented realistically when derived from a coarser DEM. Valeo and Moin (2000) investigated the impacts of grid cell size on calibrated parameters and on the performance of a variable source area model for a small catchment with mild to moderate relief. The impacts were assessed by comparing the simulated model response from ten different grid cell size maps ranging from 10 to 100 m. Their study shows that the predicted processes based on calibrated parameters were dependent on grid resolution in this integrated model, and the most important parameter in determining the quantity of urban runoff was slightly affected by grid resolution. As cell size increased, contributions from urban areas increased and overland flow contributions from rural areas decreased. Horritt and Bates (2001) investigated the scaling properties of a raster-based flood flow model, LISFLOOD, for prediction of flood inundation on a 60 km length reach of the river Severn, UK. Models of resolution varying from 1000 to 10 m are tested with best performance at a resolution of 100 m, after which no improvement is seen with increasing resolution. Molnar and Julien (2000) evaluated the effects of grid cell size from 17 to 914 m on surface runoff modeling using a raster-based CASC2D hydrologic model for event-based simulation. Their findings indicate that coarser grid cell Liu, Y.B., Yi, Y., Batelaan, O. and De Smedt, F., Assessing grid size effects on runoff and flow response using a GIS-based hydrologic model, in S.N. Li and V. Tao (eds.), F050, pp1-8, Proceeding of the 13th International Conference on Geoinformatics, Toronto, Canada, Sept. 17-19, 2005.