?Predictability of deep convection in Mendoza by combination of a statistical model with the ETA-PRM?

DIEGO ARANEO; JOAS GRASSI; FEDERICO A NORTE; JORGE RUBEN SANTOS

Montevideo

Conferencia; WCRP Conference for Latin America and the Caribbean: Developing, linking and applying climate knowledge; 2014

WRCP (World Research Climate Program )

A probabilistic Statistical Model (SM) was adjusted in order to predict the occurrence of deep convection events in the province of Mendoza. The resulting SM was then combined with 5-days forecasts from the regional ETA-PRM model in order to test the efficiency of the SM-ETA combination in forecasting the convective events over the region. Several rawinsondes at 12 UTC from Mendoza-Aero station (National Weather Service) were used to obtain the profiles of temperature (T), dew point (Td) and winds. Sets of 272 events from October-to-April of 1987-2005 and 371 events from October-to-March of 2006-2010 were used to adjust the SM and validate the predictions respectively. For the SM fit we defined a Convection Occurrence Index (C) according to the reported observations during 24 hours starting at 12 UTC. Firstly, the Principal Component Analysis (PCA) was applied to determine the main vertical profile patterns of T and Td. In order to build the SM, the resulting PC loadings, the helicity and the shear from different layers were used as predictors in a logistic multiple-regression fit taking the C index as the response variable. PCA resulted useful to obtain the vertical profile patterns of T and Td. The SM fit with profile patterns of T and Td as predictors reveals that the probability of convection increases (decreases) with a strong (weak) T lapse-rate between troposphere low and middle levels and a high (low) moisture content in the lower layers. The T/Td profiles result the best predictor variables for the SM. Taking 0.5 as the cut off probability (i.e. the probability that discriminates between cases of occurrence or nonoccurrence of deep convection), the SM effectiveness is about 73.6% using only T/Td profiles and slightly increase to 73.9% adding the helicity as predictor. The shear did not result a good predictor in any case. Combining the SM with ETA-PRM (i.e. using the T/Td profiles from 5-days runs of ETA as SM predictors), efficiencies go from about 72.7 to 64.4% for the 0 to 4 forecast-days respectively. The cut off probability can also be optimized to improve effectiveness. Taking 0.697 as the cut off probability, the SM effectiveness can be improved to about 75.5%. On the other hand, taking cut off probabilities from about 0.41 to 0.64, the effectiveness for the ETA-SM combination can be increased to about 73.4 to 66.9% for the 0 to 4 forecast-days respectively.