jats:pThe energy of water masses is a first-order factor that controls the essential physicochemical dynamics of a water body. Its study allows one to understand the roots of the processes that occur at the water-mass, water-atmosphere and water-sediment interfaces. The analysis of the Thomas Pond in the Brenne region gives a valuable overview of energy stock evolution on a yearly scale. It highlights the direct impact of this evolution on thermal stratification and the potential for evaporation and exchange with the atmosphere. The study of evaporation remains challenging due to the complexity of the energy processes and factors involved. Its estimation using formulas, which are mostly empirical, is one of the most used means for studying the process. The studied pond shows a natural stratification during the summer season, however often fragile and disturbed by other climatic factors such as wind and precipitation. This disruption leads to increased exchanges between the pond and the atmosphere. The methods used to estimate pond-atmosphere exchanges, namely evaporation, vary in values ranging between 1 mm/d to > 15 mm/d. Among these methods, three stand out and seem to give reasonable values. This observation is based on the noticeable drop of the pond’s water level during the period of non-communication with the outside, which corresponds to 65 mm. The energy required for this evaporation varies between 600 W/m2 and 1500 W/m2, except for the Smith model, that slightly overestimates this parameter. The regulation of ponds’ water volumes by managers, the increased duration of bungs closure and the intermittence of precipitations in recent years exacerbate the reduction of direct inputs to ponds and the aggravates the impacts of a changing climate. Under the effect of increasing air temperatures, losses by evaporation will also increase significantly. If we generalise the results obtained to all of the Brenne Park water bodies (4500 ponds of the park), losses by evaporation will lead to a significant water deficit of the Loire basin. From this study, the use of deep learning ensemble models was found to provide better short-term predictions (RMSE between 0.003 and 0.006 for all methods), thus confirming the effectiveness of these methods for similar applications.</jats:p>