This paper presents an application of optimally adaptive linear combinations (OALC) to forecast daily discharges and water levels for different forecast horizons (one day to two weeks ahead). To calibrate the OALC, we have used the forecast mean squared error. The optimal length of the calibration period and the optimal number of predictors were established using the performance criteria of the Hydrometeorological Centre of Russia as the goal function (Dominguez et al., 2011). The percentage of successful forecast for error levels lower or equal than 5, 10, and 20% were also determined to find the most efficient OALC. The forecasting technique presented here was implemented in Colombia to predict daily streamflows to the Betania hydropower reservoir located in the upper Magdalena River, as well as to forecast the water levels at the hydrometric station “El Banco” located within the navigable sector of the same river. Finally, we discuss the possibility of using OALC as a deterministic kernel to simulate the evolution of conditioned probability density curves (CPDC) through the multidimensional Fokker-Planck-Kolmogorov (FPK) equation to provide users with probabilistic forecasts of river water levels. Developed forecast method, and its potential probabilistic enhancement, can be useful for forecasting river water levels under climate change conditions, where uncertainty maximizes and extremes flows are expected to occur more frequently. The OALC, being a simple linear method, can be solve by means of Kolmogorov’s optimal interpolation or by least square methods as it is done in this paper. OALC proved to be efficient for lead times “T” from 1 to 10 but it is expected to increase computational time requirements when issuing forecasts for lead times from 11 to 24 or more.