The prediction of river water temperature is of key importance in the field of environmental science. Water temperature datasets for low-order rivers are often in short supply, leaving environmental modelers with the challenge of extracting as much information as possible from existing datasets. Therefore, identifying a suitable modeling solution for the prediction of river water temperature with a large scarcity of forcing datasets is of great importance. In this study, five models, forced with the meteorological datasets obtained from the fifth-generation atmospheric reanalysis, ERA5-Land, are used to predict the water temperature of 83 rivers (with 98 % missing data): three machine learning algorithms (random forest, artificial neural network and support vector regression), the hybrid Air2stream model with all available parameterizations and a multiple regression. The machine learning hyperparameters were optimized with a tree-structured Parzen estimator, and an oversampling–undersampling technique was used to generate synthetic training datasets. In general terms, the results of the study demonstrate the vital importance of hyperparameter optimization and suggest that, from a practical modeling perspective, when the number of predictor variables and observed river water temperature values are limited, the application of all the models considered in this study is crucial. Basically, all the models tested proved to be the best for at least one station. The root mean square error (RMSE) and the Nash–Sutcliffe efficiency (NSE) values obtained for the ensemble of all model results were 2.75±1.00 and 0.56±0.48 ^∘C, respectively. The model that performed the best overall was random forest (annual mean – RMSE: 3.18±1.06 ^∘C; NSE: 0.52±0.23). With the application of the oversampling–undersampling technique, the RMSE values obtained with the random forest model were reduced from 0.00 % to 21.89 % (μ=8.57 %; σ=8.21 %) and the NSE values increased from 1.1 % to 217.0 % (μ=40 %; σ=63 %). These results suggest that the solution proposed has the potential to significantly improve the modeling of water temperature in rivers with machine learning methods, as well as providing increased scope for its application to larger training datasets and the prediction of other types of dependent variables. The results also revealed the existence of a logarithmic correlation among the RMSE between the observed and predicted river water temperature and the watershed time of concentration. The RMSE increases by an average of 0.1 ^∘C with a 1 h increase in the watershed time of concentration (watershed area: μ=106 km²; σ=153).

The complexity of the state-of-the-art climate models requires high computational resources and imposes rather simplified parameterization of inland waters. The effect of lakes and reservoirs on the local and regional climate is commonly parameterized in regional or global climate modeling as a function of surface water temperature estimated by atmosphere-coupled one-dimensional lake models. The latter typically neglect one of the major transport mechanisms specific to artificial reservoirs: heat and mass advection due to inflows and outflows. Incorporation of these essentially two-dimensional processes into lake parameterizations requires a trade-off between computational efficiency and physical soundness, which is addressed in this study. We evaluated the performance of the two most used lake parameterization schemes and a machine-learning approach on high-resolution historical water temperature records from 24 reservoirs. Simulations were also performed at both variable and constant water level to explore the thermal structure differences between lakes and reservoirs. Our results highlight the need to include anthropogenic inflow and outflow controls in regional and global climate models. Our findings also highlight the efficiency of the machine-learning approach, which may overperform process-based physical models in both accuracy and computational requirements if applied to reservoirs with long-term observations available. Overall, results suggest that the combined use of process-based physical models and machine-learning models will considerably improve the modeling of air–lake heat and moisture fluxes. A relationship between mean water retention times and the importance of inflows and outflows is established: reservoirs with a retention time shorter than ∼ 100 d, if simulated without inflow and outflow effects, tend to exhibit a statistically significant deviation in the computed surface temperatures regardless of their morphological characteristics.

Changes in water temperature and stratification dynamics can have a significant effect on hydrodynamics and water quality in reservoirs. Therefore, to assess future climate impacts, projections of three regional climate models for Europe, under the IPCC A1B emission scenario (2081–2100), were used with the CE-QUAL-W2 water quality model to evaluate changes in the thermal regime of 24 Portuguese reservoirs, representing different geographic regions, morphologies, volumes and hydrological regimes. Simulation results were compared with reference simulations for the period 1989–2008 and changes in water temperature and thermal stratification characteristics were evaluated. Future inflow scenarios were estimated from precipitation-runoff non-linear correlations and outflows were estimated considering present water uses, including hydropower, water supply and irrigation. Results suggest a significant increment in the mean water temperature of the reservoirs for the entire water volume and at water surface of 2.3 and 2.5 °C, respectively, associated with a runoff reduction of approximately 23%. Overall, variations in annual stratification patterns are characterized by changes in the mean annual length of stratification anomaly that ranged from −21 to +39 days. Results also show the influence of depth and volume over the reservoir's temperature anomaly, highlighting the importance of future water uses and operation rule curves optimization for reservoirs.