Two hybrid data-driven models for modeling water-air temperature relationship in rivers

Area of Science:

• Environmental Science
• Hydrology
• Computational Intelligence

Background:

• Accurate river water temperature (RWT) forecasting is vital for effective stream ecology management.
• Existing forecasting models often struggle with the complexities of river thermal dynamics, especially during extreme events like heatwaves.

Purpose of the Study:

• To propose and evaluate novel hybrid models for RWT prediction by coupling wavelet transformation (WT) with artificial intelligence (AI) techniques.
• To compare the performance of these hybrid models against conventional methods, including multilayer perceptron neural network (MLPNN), adaptive neural-fuzzy inference system (ANFIS), and multiple linear regression (MLR).

Main Methods:

• Development of hybrid WT-MLPNN and WT-ANFIS models for RWT forecasting.
• Application and comparison of models at two river stations in the Drava River, Croatia.
• Performance evaluation using statistical metrics: coefficient of correlation (R), Willmott index of agreement (d), root mean squared error (RMSE), and mean absolute error (MAE).

Main Results:

• Hybrid WT-AI models (WTMLPNN, WTANFIS) demonstrated superior performance in RWT simulation compared to conventional models.
• MLPNN and ANFIS models generally outperformed MLR models for RWT forecasting.
• The study confirmed the significant influence of the day of the year (DOY) on river thermal dynamics.

Conclusions:

• Coupling wavelet transformation with AI techniques (MLPNN, ANFIS) offers a promising approach for accurate RWT forecasting.
• Hybrid models provide enhanced simulation capabilities for both regular periods and heatwave events.
• The findings highlight the potential for improved water resource management and ecological monitoring through advanced forecasting methods.