Probabilistic flood forecast for a small Pantanal watershed

Abstract

Monitoring for flood forecasting in small hydrographic basins is of great importance in view of the relationship of water resources with society, as it can guarantee the sustainable use of urban communities in cities bordering the basin. The Aquidauana River, classified as a small basin, belongs to the Paraguay River basin and is an affluent of the Miranda River, forming part of the Pantanal plain, being inserted in the mapping of rivers vulnerable to flooding in the Central-West region of Brazil. This study deals with the monitoring of the river Aquidauana and uses artificial neural networks (RNAs) of the MultiLayer Perceptron type (training back-propagation) with parameters optimized by Genetic Algorithms. The RNA was trained and tested based on hydrological data between 1995 and 2014, accumulated rainfall (mm) and river level (cm) upstream. The forecast was 1 to 5 days, with the best performance of the model for 1 day of forecast, with a coefficient of determination and mean square error of 0.93 and 30 (cm), respectively.