Averaging the results of different models and/or model realizations has been suggested as a way to improve the estimations of environmental models, especially in meteorology and climate sciences. We applied ensemble modelling to study the hydrodynamic and thermal behaviour of the Mediterranean reservoir of Bimont (France). We used a single-model ensemble composed of three realizations of the same hydrodynamic model (Dynamic Reservoir Simulation Model, DYRESM) with different calibration parameter values as follows: 1) default parameter values, 2) parameter values obtained by manual calibration, and 3) parameter values corresponding to the best performing member of a set of behavioural parameter sets obtained using the GLUE (Generalized Likelihood Uncertainty Estimation) method. The results confirmed the good performance of the ensemble average for lake water temperature simulations. A loss of signal was present because of the smoothing induced by model averaging, especially for the simulation of surface temperature. Single ensembles are a good alternative to obtain consistently better simulation results in hydrodynamic simulations, as well as providing a qualitative estimation of uncertainty. However, awareness of its limitations is necessary when interpreting the results.