INSTITUTO NACIONAL DE PESQUISAS ESPACIAIS CENTRO DE PREVISÃO DE TEMPO E ESTUDOS CLIMÁTICOS Summer rainfall forecast skill over South America derived from WCRP SMIP-2 models results. Lincoln M. Alves, Paulo Nobre, Luciano Pezzi, José F. Pesquero Climate Prediction Group lincoln@cptec.inpe.br WCRP Workshop on Seasonal Prediction Barcelona Spain June 4-7, 2007

OUTLINE Part I – A brief discussion about the main climatological features of South America rainfall Part II – Methodology and data Part III – Skill of the models simulation Part IV – Summary and Conclusions

OVERVIEW Distribution of Climatological Precipitation for DJF and JJA (mm/day) Diminuir purpose para uma frase… PURPOSE This work analyses the forecast skill of summer (DJF) rainfall for several SMIP-2 AGCMs over South America, relative to CPTEC’s both atmospheric and coupled ocean-atmosphere GCMs.

WCRP SMIP-2 prescribed_SST Datasets DATA AND METHODOLOGY WCRP SMIP-2 prescribed_SST Datasets CPTEC AGCM: Centro de Previsão de Tempo e Estudos Climáticos Resolution: 1.875x1.864677; Members: 5; HMC: Hydrometcentre of Russia Resolution: 2.5x2.5; Members: 6; IAP: Institute of Atmospheric Physics, China Resolution: 2.8125x3.050848; Members: 6; KMA: Korea Meteorological Agency MGO: Main Geophysical Observatory, Russia MRI: Meteorological Research Institute (Japan) SCRIPPS: Scripps Institute of Oceanography Resolution: 1.875x1.904129; Members: 10; CPTEC CGCM: Centro de Previsão de Tempo e Estudos Climáticos Resolution: 1.875x1.875; Members: 10;

DATA AND METHODOLOGY Time: [DJF 1979-1999]; season Data used for model evaluation: The Global Precipitation Climatology Project (GPCP) Skill Measures: Ensemble means are used to compare the models results with observational datasets Bias Correlation of seasonal rainfall anomalies Interannual variability of observed and modeled normalized rainfall departures in several regions of South America The multi-model simulations are combined by two methods: a simple average and the superensemble (Krishnamurti, 2000, 2001a, 2001b) Conventional Superensemble methodology is given by:

MODEL PRECIP. CLIMATOLOGY: DJF (mm/day) CPTEC AGCM HMC IAP KMA MGO MRI SCRIPPS SIMPLE AVERAGE CPTEC CGCM OBS

DJF MODEL’S BIAS CPTEC AGCM HMC IAP KMA MGO MRI SCRIPPS SIMPLE AVERAGE CPTEC CGCM Bias na AM, forte bias nao é um bom modelo… erro sistematico sobre a AM…. Wet bias over south… sul da Argentina. Dry bias no sul do Brazil… diferencao de parametrizao no modelo acoplado RAS as you can see here…

Seasonal Anomaly Correlation CPTEC AGCM HMC IAP KMA MGO MRI SCRIPPS SIMPLE AVERAGE SUPERENSEMBLE CPTEC CGCM Potential predictabily Couple and AGCM model actual prediction predictability… skill variavel nos varios modelos no sudeste… keeps both features Deficience em nordeste because parametrization

Seasonal Anomaly Correlation Models minus Superensemble CPTEC AGCM HMC IAP KMA MGO MRI SCRIPPS SIMPLE AVERAGE CPTEC CGCM worse

CPTEC’s AGCM & CGCM DJF PRECIP FORECAST ACC DIFFERENCE CGCM Actual predict one month lead to predict djf… bom skill para prever enso variability and AGCM

INTERANNUAL VARIABILITY

INTERANNUAL VARIABILITY

SUMMARY The models simulate the patterns of DJF precipitation climatology reasonably well, despite a general dry bias over the Amazon. A common feature of all AGCMs was the higher predictability over Northern Amazon, Northeast, Southern Brazil, and northern Argentina, and low predictability over Central and SE Brazil. It is not possible to point out a model with characteristics outstandingly better than the others, on the contrary, each model has strong and weak points in different regions. A multi-model fcst may contribute to more skillful predictions than any single AGCM; Although a more sophisticated combination method does not necessarily result better fcst everywhere; CGCM presented marginally better skill over the South-SW Tropical Atlantic and the SACZ region: seasonal rainfall predictability dependent on coupled interactions?