PREDICTION OF SURFACE FLOWS IN SEMI-ARID RIVERS
Linear regression model. Response surface. Flood wave propagation. Nash-Sutcliffe coefficient.
This work investigated the use of multivariate statistical techniques combined with a computational model based on the Saint-Venant wave propagation equation. The objective was to find a model that uses a minimum number of parameters, so that its implementation cost is lower, favoring its adoption by public water resources management agencies. Events from two hydrographic basins were analyzed, the first is the MHSJ with an intermittent river, with low flows and fast transitions, the second basin is the SBHPCI located on crystalline rock, high and perennial flows. The statistical methods used were multiple linear regression and the surface-response model combined with a computational model based on the Saint-Venant wave propagation equation. The parameters used were the inlet hydrographs (upstream) and hydraulic load, where it was sought to obtain the outlet hydrograph (downstream) in the basin's exhutories. The efficiency of the investigated models was measured using the Nash-Sutcliffe coefficient (NSE) and the standard deviation. The models performed well, but an average flow threshold was found for each study area, from which the results showed better NSEs. The pure statistical models better simulated the events in both basins, highlighting the multiple linear regression model.