We propose a stochastic optimization framework to perform water management in coolingconstrained power plants. The approach determines optimal set-points to maximize power output in the presence of uncertain weather conditions and water intake constraints. Weather uncertainty is quantified in the form of ensembles using the state-of-the-art numerical weather prediction model WRF. The framework enables us to handle first-principles black-box simulation models and to construct empirical distributions from limited samples obtained from WRF. Using these capabilities, we investigate the effects of cooling capacity constraints and weather conditions on generation capacity. In a pulverized coal power plant study we have found that weather fluctuations make the maximum plant output vary in the range of 5-10% of the nominal capacity in intraday operations. In addition, we have found that stochastic optimization can lead to daily capacity gains of as much as 245 MWh over current practice and enables more robust bidding procedures. We demonstrate that the framework is computationally feasible.