Machine learning models applied to estimate the water temperature of rivers and reservoirs




measurement water temperature, machine learning, neural networks, statistical models


Water temperature in rivers and reservoirs plays a crucial role in aquatic ecology, as inadequate conditions can promote the overgrowth of harmful algae and bacteria, resulting in the production of harmful toxins for human and animal health, and affecting water quality. To effectively manage water resources, continuous monitoring of these bodies is crucial. However, existing technological devices rarely offer continuous and real-time data collection, necessitating an alternative approach. The aim of this study was to compare the performance of four machine learning models (Linear Regression, Stochastic Model, Extra Tree, and Multilayer Perceptron Neural Network) in estimating water temperature in Pernambuco, Brazil's rivers and reservoirs. Statistical metrics showed that all models achieved a satisfactory capacity, with the Multilayer Perceptron Neural Network demonstrating slightly superior performance in reservoirs and rivers where it obtained the best result with a Mean Squared Error: 0.343, Root Mean Squared Error: 0.585, Mean Absolute Error: 0.445 and Coefficient of Determination: 0.595. Consequently, the MLPNN model was chosen for the development of virtual sensors. In addition to an interface that allows users to access a map and obtain estimated water temperature information for various locations, facilitating informed decision-making and resource management.






Research Papers