Detection of land use/cover change effect on watershed’s response in generating runoff using computational intelligence approaches

Quantifying the potential impacts of changing land use/cover in watersheds especially via the rainfall-runoff modeling remains among the more challenging problems in hydrology. This study presents a new method in detection, quantification and prediction of long term land use/cover change of the South Fork Eel River watershed, based on linear conceptual rainfall-runoff modeling and computational intelligence tools. The methodology was presented via a two-phase modeling procedure: event-based and continuous-based modeling phases. In the first phase, a linear conceptual rainfall-runoff model was proposed based on unequal cascade of reservoirs in which the drainage basin was divided into several sub-basins in a sequence based on the stream gauge locations and drainage network. This model includes two parameters of linear reservoir and channel lag times which were computed using relations based on the watershed’s geomorphology. The parameters were calibrated using observed rainfall-runoff events via genetic algorithm. In the second phase, calibrated lag times of events during each month were averaged to produce monthly time series of lag time to be used for training of classic artificial neural network and hybrid Wavelet-ANN for prediction of land use/cover change effect on the watershed’s response in generating runoff. Results indicate the good capability of such two-phase modeling procedure in quantification and prediction of land use change effects on hydrological response of the watershed. Results show, autoregressive and seasonal patterns of lag time, evapotranspiration and temperature are the most important factors which could affect rainfall-runoff modeling of the watershed. The land use change effect results show a slight improvement in the vegetation cover during long period for the South Fork Eel River watershed.