| description abstract | When reservoirs are constructed on rivers, river velocity decreases and the river system gradually evolves into a reservoir system. This phenomenon is particularly pronounced in the case of megareservoirs. Previous studies on this problem mostly relied on physical models, analyzing changes in water temperature. However, traditional physical process models are limited by data availability on hydrology, meteorology, and topography. Conversely, data-driven models offer the advantages of a simple structure, ability to use remote sensing data as primary data, easy acquisition, and efficiency in parameter tuning. This study constructed a hybrid artificial neural network data-driven water temperature model using a mutual information screening model to drive factors and dividing the dataset using the hold-out method. Taking the Xiaowan Reservoir as an example, the vertical distribution of water temperature across 20 layers was simulated and predicted. The results are as follows: (1) The Hybrid Artificial Neural Network (H-ANN) model enhanced the accuracy of simulating vertical water temperature in the reservoir by taking into account the correlation between water temperatures at different depths, effectively overcoming the challenges of traditional physical models (acquisition of experimental data and difficulties in model parameter tuning). (2) The water temperature simulated using the H-ANN model showed good agreement with observed water temperature in the Xiaowan Reservoir. The average Nash-Sutcliffe efficiency (NSE), correlation coefficient (r), root mean square error (RMSE), mean squared error (MSE), and mean absolute error (MAE) for water temperature in the 1–200 m layer were 0.94, 0.98, 0.23°C, 0.1°C, and 0.13°C, respectively, during the training period, and 0.9, 0.96, 0.32°C, 0.16°C, and 0.19°C, respectively, during the testing period. Overall, the model showed a high degree of conformity between simulated and observed series, indicating the suitability of the mutual information-based and concatenated multilayer ANN data-driven model for simulating vertical water temperature. (3) The Xiaowan Reservoir is a typical stratified reservoir with evident seasonal thermal stratification, where the epilimnion ranges from 1 to 15 m, the metalimnion ranges from 15 to 80 m, and the hypolimnion lies below 80 m. | |
