Integração de dados de múltiplas estações para a previsão de eventos hidrológicos extremos na Bacia do Rio Negro com aprendizado profundo

Abstract

I applied convolutional neural networks with subnet-based architectures to predict extreme floods and droughts in the Negro River basin, located in the northern region of South America. The basin is characterized by diverse geological formations, including a variety of rock types that influence water infiltration and flow patterns, alongside climatic factors that shape precipitation dynamics. Additionally, the basin’s hydrodynamics, driven by interactions between its geological formations and water flow, as well as its topography, which govern the direction and speed of runoff, play crucial roles in its hydrological behavior. Rainfall regime and seasonality further dictate the temporal distribution of water levels. Land cover types also play a significant role in altering infiltration rates and surface runoff. By analyzing hydrological data from five stations, namely Cucuí, Serrinha, Caracaraí, Santa Maria do Boiaçú, and Moura, this research aims to model and forecast extreme hydrological events. The application of deep learning approach underscores the importance of integrating multi-station data to capture localized variations and regional differences. These findings highlight the potential of artificial intelligence as a complementary tool to existing models used by competent authorities responsible for monitoring, aiding in mitigating the impacts of extreme floods and droughts. While not a standalone solution, the model provides valuable insights and emphasizes the need for further improvements, particularly through hyperparameter tuning, to enhance its reliability and accuracy.