Brazilian tropical woodlands, such as the wooded Cerrado, perform hydrological functions that need to be well understood by field data acquisition and mathematical modeling. Here, we aimed to assess the water partitioning behavior and variability of a wooded Cerrado fragment located in Southeastern Brazil by (i) measuring fluxes using eddy covariance; (ii) applying machine learning techniques to obtain a model to estimate the evapotranspiration (ET) using meteorological data as input: solar radiation (Rg), wind speed (WS), temperature (T), relative humidity (RH), and rainfall (P); and (iii) simulating a long-term water balance using stochastic climate generator inputs and the previously calibrated ET model. The average observed ET was 3.12±0.93 mm d) and P (1227±208 mm yr^-1) have similar standard deviations; differently from the ET (1054±46 mm yr^-1), which presented higher annual rates with a small variability throughout the simulation.