Modelagem hidrológica mensal chuva–vazão utilizando dados de precipitação TRMM e mapas auto-organizáveis para a bacia do alto Rio São Francisco

• Autor(a) principal: Nascimento, Thiago Victor Medeiros do
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFPB
• Texto Completo: https://repositorio.ufpb.br/jspui/handle/123456789/25262
• topic: Modelos chuva-vazão; RNA; SOM; TRMM; Rainfall-streamflow models; CNPQ::ENGENHARIAS::ENGENHARIA CIVIL

The good management of water resources is an essential activity for the achievement of an efficient management of it. Therefore, studies and research in the area are essential, and for which the existence of long and long series of hydrological data is required, e.g., precipitation, and streamflow. However, it is common to have historical series with missing data, such as rainfall and flow in hydrographic basins. Within this aspect of monitoring hydrological variables, rainfall-streamflow modeling is an alternative for filling, and even forecasting, affluent flow data in rivers and reservoirs where there is limited data available. And in this respect, as an alternative for rainfall-streamflow modeling, there is also the use of Artificial Neural Networks (ANN). The present work uses an RNA called self-organizing maps (SOM), in rainfall-streamflow modeling, using precipitation products from the Tropical Rainfall Measurement Mission (TRMM) satellite, affluent natural flows from a reservoir the Alto São Francisco basin. They are worked with three sets of data, being carried through three different models, with periods for training and testing of the different results, within the total period of available data 1998-2019. Finally, statistical metrics are calculated for the results, in order to analyze the performance of the models, as well as to analyze the influence of data homogeneity in the calibration and testing of a SOM network. The results showed that the homogeneity of the data influenced the rainfall-flow modeling. For the calibration step, the models showed relatively high efficiency indexes, and for the test, the results varied depending on the dataset and the type of modeling. The worst values of the metrics found were for the modeling that made use of 70% of the initial data of the series in the calibration, and 30% of the final in the test for dataset # 1, which is the non-homogeneous. The remaining eight models presented relatively good to excellent metrics, i.e., Nash-Sutcliffe (NSE) and Determination coefficient (DC)> 0.70, and (Relative bias) RB and Normalize root-mean-square error (NRMSE) close to zero. Therefore, the possibility of using TRMM precipitation data and SOM networks in rain-flow modeling at a monthly level is clear.