PROBABILISTIC QUANTITATIVE PRECIPITATION FORECAST CALIBRATION OVER SOUTH AMERICA: EXPERIMENTS WITH A SHORT RANGE ENSEMBLE. Juan J. Ruiz 1,2 , Celeste Saulo 1,2 , Eugenia Kalnay 3 1 University of Buenos Aires, Buenos Aires, Argentina 2 Centro de Investigaciones del Mar y de la Atmósfera (CONICET-UBA), Buenos Aires, Argentina 3 University of Maryland, College Park, USA. Abstract Different techniques for obtaining probabilistic quantitative precipitation forecasts (PQPFs) over South America are tested during the 2002-2003 warm season. Some of the techniques are based on a parametric representation of the conditional probability of precipitation over a particular threshold given a certain value of forecasted precipitation and the other uses a non-parametric estimation of the probabilities. The results are also compared with the calibration algorithm based on the rank histogram. A number of experiments were performed to compute a reasonable size for the training period. The PQPFs of a short range ensemble forecast system (SREF) based on the WRF model and the breeding technique were calibrated using the different approaches and the resulting PQPFs scores were compared. The calibration is performed using two different data sets, one derived from a high resolution rain gauge network and the other one based on passive microwave satellite estimates. This is done in order to evaluate if the use of satellite derived rainfall produces a significant degradation of PQPFs reliability and resolution. 1. INTRODUCTION Quantitative precipitation forecast (QPF) is one of the most difficult and least accurate products available from numerical weather prediction (NWP) (Ebert 2001). Continuous efforts are devoted to improve forecast quality, ensemble forecasting being an example of one possible strategy to deal with errors arising from uncertainties in the initial and boundary conditions. An interesting characteristic of ensemble systems is that probability forecasts can easily be created, leading to the generation of PQPFs. Different methodologies for obtaining PQPFs, and corresponding measures to quantify their usefulness, have been developed. Of particular interest is how to obtain a reliable PQPF i.e., a system where the forecasted frequency of a particular weather phenomenon is close to the observed probability. The importance of PQPF reliability is directly related to its effect upon the economic value of the forecast, as discussed by Zhu et al. (2002). Several techniques have been developed to generate reliable PQPFs. For example, Hamill and Colucci (1998), Gallus et al. Corresponding author address: Juan J. Ruiz, CIMA/University of Buenos Aires, Ciudad Universitaria, Buenos Aires, Argentina. E-mail: jruiz@cima.fcen.uba.ar 2007, Stensrud and Yussouf 2007, Sloughter et al (2007) (hereafter S2007) among many others introduced different techniques for forecast calibration with the aim of improving forecast quality. The main objective of this work is to evaluate the skill of different calibration algorithms over South America. In a former study (Ruiz et al. 2008), some of these strategies have been tested for a regional ensemble based on the SLAF technique and on a multi model ensemble. Yet, there were other alternatives that we wanted to test –both for ensemble generation and calibration- and also use a denser precipitation network for validation and calibration that became available after that previous work. Accordingly, this work progresses on the previous one, including more methods for PQPF calibration (i.e. the one based on the paper of S2007) applied to an alternate regional ensemble system as will be explained subsequently. To compare the skill of the different calibration strategies, 2 months of accumulated precipitation obtained from 48-hr ensemble forecasts have been analyzed. The ensemble system is a regional ensemble with perturbations in the initial conditions obtained trought the Breeding of the Growing Modes method (Toth and Kalnay, 1993). PQPFs obtained via the combination of the above mentioned calibration techniques are analyzed through the computation of the Brier Skill Score (hereafter BSS) and its components (Wilks 1995) which also allows for comparison with results obtained in previous works. 4.7