A comparison of statistical downscaling techniques for daily precipitation: Results from the CORDEX Flagship Pilot Study in South America

BETTOLLI MARIA LAURA; GUTIERREZ LLORENTE, JM; ITURBIDE, M; BAÑO MEDINA, J; HUTH, R; SOLMAN, S.; FERNANDEZ, JESUS; PORFIRIO DA ROCHA, R; LLOPART, M; LAVIN GULLON, A; COPPOLA, E; CHOU, S; DOYLE, M; OLMO, MATIAS; FEIJOO, M

Beijing

Conferencia; ICRC-CORDEX 2019; 2019

CORDEX

Southeast South America (SESA) is one of the regions of the planet where extreme precipitation eventsoccur and have high impact on human activities. These extreme events result from the complexinteractions of a broad range of scales, therefore their study, modelling and projections in a changingclimate continue to be a challenging task. The CORDEX Flagship Pilot Study in South America (FPSSESA)addresses this topic in order to advance in the understanding and modelling of extremeprecipitation events based on coordinated experiments using different downscaling approaches. Inthis work we present the results from the collaborative action to intercompare different statisticaldownscaling techniques in simulating daily precipitation in SESA with special focus on extremes. Tothis end, seven statistical downscaling models based on the regression and analog families wereevaluated over SESA. The sensitivity to the different predictor and predictand datasets were testedusing two reanalyses (ECMWF ERA-Interim and Japanese 55-year Reanalysis JRA-55) and two dailyprecipitation (station data and MSWEP) datasets. The models were calibrated and cross-validatedduring the 1979-2009 period and also evaluated in the independent warm season of 2009-2010. Thisseason, with record of extreme precipitation events, is the target season chosen in the FPS-SESA toperform the dynamical downscaling simulations as well, and therefore it allows for comparisonsbetween both approaches. The results show that the methods are more skillful when combinedpredictors including circulation variables at middle levels and local humidity at low levels of theatmosphere are considered. The performance of the models is also sensitive to reanalysis choice. Themethods show overall good performance in simulating daily precipitation characteristics over theregion, but no single model performs best over all validation metrics and aspects evaluated.