Aplicações de redes neurais artificiais e satélite TRMM na modelagem chuva-vazão da bacia hidrográfica do rio Paraguaçu/BA.

Abstract

The growing imbalance between water demand and availability is subject to water resources planning, conservation policies, water resources programs and hydrologic simulations. In this context, the aim of this study is to assess the application of artificial neural networks and Tropical Rainfall Measuring Mission (TRMM) satellite in the rainfall-runoff modeling in the Paraguaçu River Basin/BA. The results revealed that the basin is not very prone to flooding. The main river has a low slope, with transitional forms regular and irregular in its course. More than 50% of the basin area is between 200 and 600 meters, the lowest altitudes are located to the east and elevations above 1000 meters are in the western portion. The basin is relatively flat with smooth slope type. The predominant vegetation consists of secondary vegetation and agricultural activities. Orographic and/or convective rainfall causes the sub and overestimation of the TRMM values. The neural network chosen for modeling rainfall-runoff used the algorithm of Levenberg-Marquardt Backpropagation with 90 neurons in the hidden layer transfer function and hyperbolic tangent sigmoid (tansig) in the intermediate and output layers in its architecture. At the stage of the network prediction overestimated the higher peaks flow. The results show that artificial neural networks are capable of predicting flow. Estimates of rainfall by TRMM can be used with caution in hydrologic analyzes. But it is an interesting alternative in places where there is no availability of hydrological information.