Welcome to Eprints@IARI

Abstract

Artificial Neural Network (ANN) and regression models were developed using watershed-scale geomorphologic parameters to predict surface runoff and sediment losses of the St. Esprit watershed, Quebec, Canada. Geomorphological parameters describing the land surface drainage characteristics and surface water flow behaviour were empirically associated with measured rainfall and runoff data and used as input to a three-layered back-propagation feed-forward neural network model. Morphological parameters such as bifurcation ratio, area ratio, channel length ratio, drainage factor and relief ratio were selected using the Multivariate Adaptive Regression Splines tool, based on their relative importance in prediction of runoff and sediment yield. Regression models were developed using the curve-fitting toolbox of MATLAB software and compared with the results obtained from ANN models. The coefficient of determination (R2) and model efficiency factor (E) were estimated to ascertain the model performance. Geomorphology-based ANN model validation statistics resulted in R2 values ranging from 0.85 to 0.95 and E values from 0.74 to 0.82 for peak runoff rate and R2 values from 0.78 to 0.93 and E values from 0.71 to 0.76 for sediment loss. Using geomorphology-based regression models, R2 values for the same dataset varied from 0.78 to 0.88 (0.74 > E > 0.69) for peak runoff rate prediction and 0.39 to 0.54 (0.53 > E > 0.46) for sediment prediction. When morphological parameters were not associated with rainfall depth and peak runoff rate, prediction error statistical parameter values (R2 and E) were less for both neural network and regression models. Thus, associating selected geomorphological parameters with rainfall depth and peak runoff rate enhances the accuracy of runoff rate and sediment loss predictions from the watershed. Furthermore, ANN models performed better than the regression equations.