ESTIMATION OF RAINFALL PROBABILITY, THROUGH THE USE OF NON PARAMETRIC STATISTICAL TECHNIQUES, APPLIED TO NUMERICAL SIMULATIONS OF WRF. A CASE OF STUDY

In this paper was used the kernel density estimation (KDE), a nonparametric method to estimate the probability density function of a random variable, to obtain a probabilistic precipitation forecast, from an ensemble prediction with the WRF model. The nine members of the prediction were obtained by varying the convective parameterization of the model, for a heavy precipitation event in southern Brazil. Evaluating the results, the estimated probabilities obtained for periods of 3 and 24 hours, and various thresholds of precipitation, were compared with the estimated precipitation of the TRMM, without showing a clear morphological correspondence between them. For accumulated in 24 hours, it was possible to compare the specific values of the observations of INMET, finding better coherence between the observations and the predicted probabilities. Skill scores were calculated from contingency tables, for different ranks of probabilities, and the forecast of heavy rain had higher proportion correct in all ranks of probabilities, and forecasted precipitation with probability of 75%, for any threshold, did not produce false alarms. Furthermore, the precipitation of lower intensity with marginal probability was over-forecasted, showing also higher index of false alarms.