Alterações de padrão de vazão decorrentes da operação de pequenos aproveitamentos hidrelétricos em cascata e previsão de série horária em afluentes do Pantanal

Abstract

Run-of-river small hydropower plants (SHP) have been built in many countries, based on the argument that they exert little influence on water resources systems. Nevertheless, recent studies have questioned their indiscriminate implementation, especially when these SHP are operated in ecosystems acknowledged for being fragile environments. In this regard, the present study aims to understand the short-term behavior of the streamflow signals in a system compounded by hydropower plants in a cascade arrangement. First, by taking into account streamgauges placed in the intermediate course of the river, one may study the effects of SHPs on the discharge cyclic patterns. For this purpose, the used approaches were the analysis of the hydrographs by the time domain and handling the signal processing technique with the multiresolution time-frequency domain, by the continuous wavelet transform. In a second moment, nonlinear regression models for forecasting hourly time series were developed, aiming at implementing a short-period warning system, focusing on minimum discharges. Therefore, the constructed models were the neural network, a relatively known technique in the scientific community, and the Gaussian Processes regression, which is still in broad development, mainly with regard to its usage in water resources problems. The viability of the time series decomposition was verified via discrete wavelet transform and via empirical mode decomposition. The univariate streamflow forecasting occurred with adjacent streamgauges, and also with the temporal and the spatial variability at the basin. Application of the methods were conducted at the Jauru river basin, Brazilian state of Mato Grosso, in the border region of the Pantanal wetland, where there are six SHPs and eight streamgauges under operation. The first group of results evidenced disturbance in the natural discharge patterns with increasing oscillations, intensifying the smallest cycles and dissolving the largest ones. With respect to streamflow forecasting, the hybrid model combining the Gaussian Processes and the wavelet decomposition demonstrated greater robustness. According to the identification of the disrupting of streamflow cycles in the first stage, the forecasting models were, to some extent, able of reproducing the series, especially at the five upstream gauges. Despite some limitations with the three downstream series forecasting, the results may serve as a subsidy for the creation of a warning system of critical discharges at the basin, enabling a better trade-off between the interests with the use of the water resources.