Climate predictability on seasonal timescale over South America from NMME models through ANOVA

LUCIANO ANDRIAN; MARISOL OSMAN; CAROLINA VERA

Workshop; NOAA?s 45th Climate Diagnostics Prediction Workshop; 2020

In this study we assessed the predictability of South American climate on seasonal timescales depicted by the models participating in the North American Multi-Model Ensemble Project (NMME). The forecasts valid for March-April-May (MAM), June-July-August (JJA), September-October-November (SON) and December-January-February (DJF) issued on February, May, August, and November, respectively (lead 1) in the period 1982-2010 were considered.We performed an Analysis of Variance (ANOVA) to identify the sources of variability in the multi-model ensemble: forced signal, bias, structural model uncertainty and noise. ANOVA was applied first to the entire ensemble and then it was repeated by removing individual models from the ensemble. The predictability was estimated through the ratio between the forced signal and the sum of the forced signal and the noise. Overall, the predictability of temperature is maximum in MAM and DJF, and is minimum in SON. Predictability maximizes in the Tropics, over the Amazon region and presents a second maximum in the extratropics over SESA. Temperature predictability is greatly enhanced when the CANSipV2 model is removed from the ensemble. On the other hand, the precipitation is less predictable than temperature. The regions with the highest precipitation predictability varies according to the seasons, but a significant increase in predictability is observed when the CFSv2 model is removed from the ensemble, especially over the Amazon region in MAM and DJF.Results also show that overall the model biases are an important contributor to the total variance of the temperature while the noise is the most relevant for precipitation. The structural model uncertainty is of particular relevance because it describes the differences between the models in their responses to a common forcing. For temperature, this term is largest in southeastern South America (SESA), especially in SON and DJF. For precipitation, the interaction term becomes smaller when the CM2p1 model is removed from the ensemble. A first inspection on the reasons for these changes shows that CM2p1 may lack a proper representation of the variability associated with ENSO in northeastern Brazil and SESA.