Neural Network-Based Predictions of Hydrological Patterns in the Amazon Basin

Código

I-EBHE0141

Autores

João Henrique Macedo Sá, Ana Isabel Baião Ramos de Oliveira, Tiago Cunha Brito Ramos, João Albino Soares de Nascimento, Ramiro Joaquim de Jesus Neves, DAVIDE FRANCO

Tema

WG 2.2: Participatory Water Systems

Resumo

The Amazon Basin is one of the largest river basins in the world, encompassing a vast territorial expanse and playing a fundamental role in regional and global climate. Enhanced understanding of hydrological patterns and their predictions are crucial for informed decision-making in various activities and contexts, such as urban planning, agriculture, navigation, and biodiversity conservation. In this study, we investigate and predict the hydrological behavior of this important region in terms of flow. Based on previous studies, this work aims to test the capability of a neural network model for flow calculation in different basins of the Amazon region. This task will be performed using a Convolutional Neural Network model previously developed and successfully tested for a river basin with completely different characteristics and rainfall patterns. Thus, using observed data from streamflow and rainfall stations provided by the National Water Agency (Agência Nacional de Águas e Saneamento Básico, ANA), the model was trained, validated, and tested. For each elected streamflow station, the rainfall for the corresponding drained area was considered as input while the streamflow values represent the model?s outputs. Initially, we collected and processed data from more than 1500 streamflow stations located in the Amazon river basin. Considering flow data recorded over a period of at least 10 years, only those with a percentage of daily value gaps below 25% were maintained. Then, a second filtering process was performed based on criteria such as geographical representativeness and rainfall data availability. This study aims to demonstrate the effectiveness of neural networks in modeling and predicting river flows at different points in the Amazon basin, allowing for the capture of seasonal and interannual patterns, as well as extreme events. Furthermore, the analysis of neural networks allowed for the identification of hydrological connectivity patterns among the different stations, revealing the complex dynamics of the Amazonian river system. These results have significant implications for the management and conservation of water resources in the Amazon basin. Additionally, the use of advanced machine learning techniques, such as neural networks, opens new perspectives for the analysis and monitoring of complex systems worldwide.