Grid-based Parameterization of Monthly Water Balance Model for Simulation of River Flow at Ungauged Sites in Abay River Basin

Abstract

This paper describes the development and validation of grid-based regional prediction equations of a conceptual monthly water balance model parameters. The purpose of the research is generally to provide gridded values of the model parameters using catchment characteristics for simulating monthly runoff at ungauged rivers, which at most suffer from lack of hydrological information. The catchment indices (characteristics) that are used in developing the prediction equations of the model parameters are mean annual rainfall, average slope, NDVI, TWI, aspect, flow length, flow direction, maximum soil moisture, elevation, etc. The monthly water balance model has three parameters; C, k and SC. Not all catchment characteristics are correlated to all parameters, but parameter C was better correlated with mean annual rainfall and NDVI; and parameter SC with mean annual rainfall, NDVI and average slope. Parameter k was not significantly correlated to any of the indices, and hence taken as the average of optimized values. The monthly water balance model was calibrated for a group of selected 28 gauged catchments using FORTRAN program that uses Shuffled Complex Evolution algorithm optimization technique. The model inputs used in optimizing the parameters in the calibration are areal precipitation, calculated using Thiessen polygon method, and monthly potential evapotranspiration calculated using Penman-Montieth and Thornthwait. The monthly potential evapotanspiration obtained using Thorthwait methed is underestimated and it was adjusted by multiplying each monthly value by mean monthly ratio of Penman to Thornthwaite. The R2, IVF and MSE were the criteria used to evaluate the optimization technique. The optimized parameter values were related to physical catchment indices using regression equations. The relationships were tested by comparing calibrated parameters values and the predicted values using the regression equation and evaluated using coefficient of determination, R 2, for the same catchments used in calibration. The result showed that predicted parameters are not much different from the optimized. After the parameters are optimized and evaluated for the selected catchments, the basin was descretized into 2131 grids of size 1 Okm by 1 Okm and the three model parameter values estimated for each grid using the regression equation. Then the vii model was tested at kessie gauging site and the result shows that the model is promising and its result will be improved by using grided inputs data. The correlation coefficient between observed and predicted discharges is 0.88. As the model is developed mainly for unguaged sites, it is also tested for ungauged site taking Muger sub-basin as unguaged catchment to show how the model works so that one can use the model to generate flow data any where in the regio