Abstract:
Simulation of complex hydrological responses in large watersheds over years has prompted the need for procedures for autocalibration. The commonly available models of watershed hydrology are of the event type applicable on a basin scale or continuous models applicable on a field scale. The Watershed Resources Management (WRM) model is a basin-scale model for continuous simulation. It is generally applicable in planning, forecasting and operational hydrology, to the study of environmental impacts of land-use change and to soil and water conservation planning. Empirical equations, derived from relating physical quantities experimentally and validated independently, are employed. In every hydrological simulation, there is always a need for optimization and the optimization is carried out by the best possible technique that will yield perfect or near-perfect values for selected calibration parameters. WRM model was originally applied to Canadian conditionsand was heuristically optimized for that application. In this work, a modified WRM version was embedded in normal and autocalibration modes. The normal mode does simulation without optimization of parameters, while the autocalibration mode runs with optimization of parameters. Theoptimization method adopted is Genetic Algorithm (GA), which is an Artificial Intelligence-based methodology for solving problemsemploying non-mathematical, non-deterministic, but stochastic process or algorithm. Four parameters with high sensitivity were usedinthe autocalibration process, namely, theManning roughness coefficient for land surface(MANN1),Manning roughness coefficient for stream surface (MANN2),Manning roughness coefficient for terrace surface (MANN3) and a surface retention parameter (KRET). These parameters were used for calibration using WRMGA and WRMGUI software developed in this study. Genomes were generated within specified ranges using random number generator. The generated values were stored in a file, Optimized. dat, which the WRMGA software calls up and uses to compute the best fit. Hydrograph plots of both the original heuristically calibrated simulations for Canadian watersheds and the autocalibrated simulations for the same watersheds were compared with measured hydrographs, and statistically analysed. WRM model originally calibrated to the watersheds gave a regression coefficient (R) of 34.8 % while the autocalibrated model gave 37 %. This result shows an improvement of 2.2 % by the autocalibration scheme. However, autocalibration involves a more objective procedure that can be employed by the non-expert in hydrologic modelling. To make for user-friendliness, the original WRM model coded in FORTRAN was translated to C-sharp (C#). The WRM model was successfully repackaged for autocalibration in this work.