Estimation of water heat flux in small reservoirs: the role of neural networks and regression techniques

Water heat flux (WHF), which represents the heat stored or lost within a water body, plays a crucial role in analysing the surface energy balance at the water reservoirs. However, estimating WHF is often challenging due to the need for detailed vertical temperature profiles. This study evaluates the performance of artificial neural networks (ANNs) and regression modelling as alternative approaches for estimating WHF in the Ekbatan dam reservoir, a small-scale reservoir in Iran. Using water temperature data collected at various depths from Sep 26, 2018, to Sep 22, 2021, reference WHF values are calculated based on its fundamental equation. A multilayer perceptron (MLP) model is developed, featuring an input layer consisting of five variables (air temperature, water surface temperature, solar radiation, wind speed, and relative humidity) and two hidden layers. Additionally, a nonlinear regression model is formulated using dimensionless parameters. The MLP and nonlinear regression models’ results are compared with the reference WHF values. The MLP model shows strong performance, achieving a coefficient of determination (R2) of 0.968 and an RMSE of 18.88Wm-2, with water surface and air temperatures identified as the most influential predictors. The regression model also performed reliably, yielding an R2 value above 0.879 and an RMSE of less than 30.08Wm-2. While the regression model provides reliable results, artificial neural networks offer greater accuracy in WHF estimation, underscoring their potential for enhancing energy balance assessments in water reservoir

Subjects

Artificial neural networks

Regression modelling

Surface energy balance

Water heat flux

Water reservoirs

DDC Class

620: Engineering

Options

Estimation of water heat flux in small reservoirs: the role of neural networks and regression techniques