International audienceThis paper investigates three categories of models that are derived from the equilibrium temperature concept to estimate water temperatures in the Loire River in France and the sensitivity to changes in hydrology and climate. We test the models' individual performances for simulating water temperatures and assess the variability of the thermal responses under the extreme changing climate scenarios that are projected for 2081-2100. We attempt to identify the most reliable models for studying the impact of climate change on river temperature (Tw). Six models are based on a linear relationship between air temperatures (Ta) and equilibrium temperatures (Te), six depend on a logistic relationship, and six rely on the closure of heat budgets. For each category, three approaches that account for the river's thermal exchange coefficient are tested. In addition to air temperatures, an index of day length is incorporated to compute equilibrium temperatures. Each model is analysed in terms of its ability to simulate the seasonal patterns of river temperatures and heat peaks. We found that including the day length as a covariate in regression-based approaches improves the performance in comparison with classical approaches that use only Ta. Moreover, the regression-based models that rely on the logistic relationship between Te and Ta exhibit root mean square errors comparable (0.90 °C) with those obtained with a classical five-term heat budget model (0.82 °C), despite a small number of required forcing variables. In contrast, the regressive models that are based on a linear relationship Te = f(Ta) fail to simulate the heat peaks and are not advisable for climate change studies. The regression-based approaches that are based on a logistic relationship and the heat balance approaches generate notably similar responses to the projected climate changes scenarios. This similarity suggests that sophisticated thermal models are not preferable to cruder ones, which are less time-consuming and require fewer input data

No abstract