Previsões Sub-Sazonais de chuva vazão para bacias hidrográficas do nordeste brasileiro

Abstract

Brazil, being a country of vast extension, faces challenges related to the scarcity of rainfall in several regions. However, it is in the Northeast region of Brazil (NEB) that this condition manifests most frequently. The Hydrographic Basin of the São Francisco River, located mostly in the NEB, and the basins of the Northern Northeast of Brazil, have historically dealt with water scarcity issues. The impacts of climate variability are felt in sectors such as agriculture, agribusiness, livestock farming, energy generation, irrigation, human health, and job creation. In this context, monitoring and forecasting hydrological variables are of utmost importance to guide the implementation of mitigation and adaptation strategies. To that end, studies were conducted to analyze sub-seasonal precipitation variability in the region; assess the performance of sub-seasonal precipitation forecast models; calibrate the SMAP hydrological model; and develop hybrid neural network models based on wavelets for streamflow prediction. The analysis of precipitation variability revealed the existence of short-duration, high-frequency fluctuations within a rainy season, highlighting a consistent sub-seasonal variability across all studied regions. The Madden-Julian Oscillation associated with the El Niño Southern Oscillation demonstrated a significant influence on the variability of these precipitations. Among the analyzed precipitation forecast models, the results indicate that the CFSv2 and GEFSv2 models exhibited superior performance. It was also noted that the CPTEC and ESRL models showed performance limitations, especially in simulations of the Ceará basins. Finally, studies related to the implementation of hybrid neural networks have demonstrated promising results. In the Três Marias and Sobradinho basins, the hybrid neural network achieved a Nash efficiency coefficient close to 0.9 for short-term forecasts. In the Xingó, Luiz Gonzaga, and Paulo Afonso basins, the model managed to capture the trend of flow series and showed significant improvements compared to traditional neural networks. However, in the Ceará basins, the lack of quality data considerably hindered the model’s performance. In comparative studies, the forecasts made by the hybrid neural network outperformed the performance of SMAP and, in general, exhibited better statistical indices than the models used by ONS.