Prediction skill assessment of large-scale variability influence in summer southeastern South America rainfall in multi-model CMIP decadal predictions

DÍAZ, LEANDRO B.; VERA, CAROLINA S.; SAURRAL, RAMIRO I.; DOBLAS-REYES FRANCISCO

Boulder

Workshop; US Clivar Workshop on Societally-Relevant Multi-Year Climate Predictions Workshop.; 2022

US CLIVAR

The leading co-variability mode (SVD1) between summer southeastern South America (SESA) rainfall and tropical sea surface temperatures (SST) anomalies over the 1962-2013 period exhibits significant variability ranging from the interannual scale to long-term trends. It shows a clear global warming signal, mainly related to warming in the Pacific and Indian Oceans, in association with a rainfall increase in SESA. After detrending the series, the spatial distribution of both SST and SESA precipitation anomalies associated with the first mode resembles that typically related with El Niño-Southern Oscillation (ENSO). The objective of this work is to assess the prediction skill of SVD1 in decadal hindcast simulations from the Coupled Model Intercomparison Project Phase 5 (CMIP5) of the World Climate Research Programme, using different methodologies to deal with the multi-member/model ensemble.Four methodologies to perform SVD1 will be presented that made use of multi-member/model ensemble information for prediction skill: i) SVD1 computed for the multi-model mean anomalies; ii) SVD1 calculated after concatenating all ensemble members; iii) SVD1 computed projecting all ensemble members in the spatial modes obtained using the multi-model mean anomalies in methodology i); iv) SVD1 computed projecting all ensemble members in the spatial modes obtained using the observed anomalies. Methodologies ii), iii) and iv) allow to obtain probabilistic prediction information, so that internal variability uncertainties could be also assessed. The three methodologies were applied to both detrended and undetrended anomalies. It was found that initialized CMIP5 decadal hindcasts are able to represent SVD1 spatial structures with and without considering trends for the different methodologies, improving results from analogous uninitialized simulations. Although detrended SVD1 activity shows skill in the first two prediction years, differences between methodologies will be discussed. These facts represent a promising result for the predictions of rainfall in the SESA region on interannual and longer time scales.