Evaluation of a SubX Multimodel Ensemble to predict an intraseasonal index for South America based on Outgoing Long Wave Radiation

MARIANO S. ALVAREZ; CAROLINA S. VERA

Virtual

Workshop; NOAA's 45th Climate Diagnostics & Prediction Workshop; 2020

Climate Prediction Center

The Seasonal-IntraSeasonal (SIS) pattern is defined as the leading EOF of the 10-90-day filtered OLRanomalies in a region over South America (40 ◦ S-5 ◦ N,70 ◦ W-32.5 ◦ W). Previous studies have shown that ISvariability of rainfall and circulation in the region can be studied and monitored through the SIS patternactivity, indicated by the time series of the first principal component (named SIS index). During the extendedaustral summer (October-April) the SIS pattern is a dipole with centers of action over the South AtlanticConvergence Zone (SACZ) region and Southeastern South America (SESA). The broad impacts of the SISpattern activity in South America and its relation to the organization of hemispheric and regional circulationanomalies makes it desirable to be able to monitor and forecast this mode of variability.In this study, the OLR hindcast of a group of the SubX Project models were used: the GEM (Environmentand Climate Change Canada), the GEFS (Environmental Modeling Center-NCEP-NOAA), the FIM (EarthSystem Research Laboratory-NOAA), the GEOS (Goddard Modeling and Assimilation Office-NASA), theNESM (Naval Research Laboratory) and the CCSM (Rosentiel School of Marine and Atmospheric Science).The ensemble mean of each model was used separately, and also to create a Multimodel Ensemble (MME)mean by averaging all available initializations between Saturdays and Fridays; that is, defining the weeks bytarget. The hindcast of each model and their lead-dependent climatology were downloaded from IRI?s DataLibrary.First, the predictability of the weekly OLR anomalies was assessed for each model and the MME mean.Averaged over all seasons, for all target weeks (1 to 4), the Anomaly Correlation Coefficient (ACC) for theMME mean shows higher values over the SACZ region and lower values over SESA. Also, the Root MeanSquared Error for the MME mean is highest within the SIS pattern centers of action, as expected, as thevariability there is also higher. The Mean Error for the MME mean maximizes between both centers ofaction, probably associated with displacements in the location of the predicted anomalies. An analysis ofthe ACC for each season averaged over the SACZ and SESA regions showed that the ACC is always higherover SACZ than over SESA for weeks 1 and 2 and also in most seasons for week 3.As the OLR anomalies can not be band-passed filtered on real time conditions, the methodology tocompute the SIS index was adapted: low frequency variability is removed by subtracting the last 40 daysmean of the OLR anomalies and afterwards two consecutive centered 5-points running mean is applied tosmooth the high frequency. This adaptation was proven to be successful when comparing both SIS indexesderived from observations. Simulating real-time conditions, the OLR anomalies in the hindcast for each startdate were joined with the low-frequency-filtered observed OLR anomalies and the high frequency smoothingwas applied on observations and forecast. Then, those real-time filtered anomalies were projected overthe SIS pattern to obtained the SIS index forecast for each model ensemble mean and the MME mean.Preliminary results show that in some cases the MME mean SIS index forecast and the ensemble spreadare able to capture the SIS index variability, though extreme events (which are the ones associated to theorganization of the regional circulation) seem to be harder to detect.