Avaliação de séries sintéticas de vazões para o dimensionamento da capacidade de reservatórios de regularização de vazões

Abstract

As a complement to forest conservation and restoration, flow regulation reservoirs are important alternatives to improve the flow of water in rivers, being also an important instrument for storing water destined for the most diverse uses, especially in places where demand exceeds water availability in periods of drought. However, the unavailability of historical series of flow in hydrographic basins makes the adequate dimensioning of reservoirs difficult. Therefore, the present study aimed to apply and evaluate two methods for the generation of synthetic series of flows for the dimensioning of the capacity of flow regulation reservoirs, in conditions of limited data. A monitored hydrological region (Itapemirim River basin, Espírito Santo) and synthetic flow series generated according to the proposals by Rodrigues (2017) and Cesconetto (2021) were used. The methodologies were tested by cross-validation, and the results were analyzed using graphs and statistical indicators, in comparison with the results obtained with observed data, measured in fluviometric reference stations. After analyzing statistical indicators, such as the coefficient of determination (R²), the Nash-Sutcliffe efficiency index (NSE) and the reservoir design quality index (IQD), it was concluded that the method proposed by Cesconetto (2021) showed a tendency to undersize the capacities of the reservoirs, mainly for lower flows to be regularized (β<0.75). The method proposed by Rodrigues (2017) performed better than the method proposed by Cesconetto (2021), for the dimensioning of the capacity of flow regulation reservoirs, under conditions of limited data, in the Itapemirim river basin. Although the method proposed by Rodrigues (2017) has shown better results, it depends, for its application, on a pre-existing flow regionalization study in the hydrological region of interest, which makes it possible to estimate the long-term average flow. The method proposed by Cesconetto (2021) depends only on the existence of rainfall and flow data (historical series), so that an artificial neural network can be adjusted (trained) and generated synthetic flow series.