Pre- and post-dam river water temperature alteration prediction using advanced machine learning models

Dams alter river hydrology by changing flow patterns, affecting both low and high flows. This impacts downstream water quality, including temperature. Accurate temperature prediction is essential for agricultural planning and aquatic habitat preservation. While dams are known to modify flow regimes, the precise effects on water temperature remain unclear. Prior research has shown that dam construction influences downstream ecosystems, but the extent of temperature changes is less understood. This gap motivated a study to assess how machine learning can predict temperature changes. No prior work had resolved how different models compare in this context. This paper contributes by evaluating machine learning models for water temperature prediction. The study focuses on the Yangtze River at Cuntan.

The goal was to predict daily water temperature changes in the Yangtze River at Cuntan before and after dam construction. This helps in managing water resources and protecting aquatic habitats. The study aimed to compare machine learning models for their accuracy in temperature prediction. A specific problem was the lack of reliable temperature forecasting methods in dam-impacted rivers. The motivation was to find cost-effective and accurate models for real-time use. This work builds on prior knowledge of dam effects on flow but extends it to temperature. The study addresses a gap in predictive modeling for water temperature. The findings could support better dam management and ecological planning.

The study used four machine learning models: M5P, RF, RSS, and REPTree. Input variables were selected based on correlation coefficients. The focus was on predicting water temperature using lagged water level data. The models were trained and tested using historical temperature data. Goodness-of-fit criteria were used to evaluate model performance. Graphical analysis compared predicted and recorded temperatures. The comparison included R², PCC, MAE, RMSE, RAE, and RRSE metrics. The best-performing model was identified for pre- and post-dam conditions.

The M5P model showed the highest accuracy in predicting water temperature. It achieved an R² of 0.9920 and 0.9708 for pre- and post-dam conditions. The PCC values were 0.9960 and 0.9853, respectively. MAE and RMSE were 0.2387 and 0.3449 for pre-dam and 0.4285 for post-dam. RAE and RRSE values indicated strong model fit. These results suggest M5P outperformed other models in this context. The study found that machine learning can reliably estimate temperature changes. The models used a day lag time input for water level data.

The study found that machine learning models can accurately predict river water temperature changes. M5P was the most effective model for both pre- and post-dam conditions. These findings support the use of machine learning in water resource management. The models provide cost-effective forecasting methods for dam-impacted rivers. The study highlights the potential of M5P for real-time temperature prediction. The results align with the authors’ claim that dams alter hydrologic regimes. The authors propose that these models can aid in ecological planning. The findings suggest that machine learning is a reliable tool for this application.