Seasonal-to-Interannual Variability of Ethiopia/Horn of Africa Monsoon. Part II Statistical Multimodel Ensemble Rainfall Predictions

• Published in: Journal of climate Vol. 28; no. 9; pp. 3511 - 3536
• Main Authors: Segele, Zewdu T., Richman, Michael B., Leslie, Lance M., Lamb, Peter J.
• Format: Journal Article
• Language: English
• Published: Boston American Meteorological Society 01.05.2015
• Subjects: Anomalies; Atmospheric models; Atmospherics; Brackish; Climate change; Climatology; Correlation; Correlations; Drought; Ensemble forecasting; Error reduction; Global temperatures; Interannual variability; Investigations; Marine; Mathematical models; Meteorology; Modeling; Monsoons; Monthly rainfall; Oceans; Precipitation; Predictions; Probability theory; Rain; Rainfall; Rainfall anomalies; Rainfall forecasting; Rainfall variability; Regional development; Regression analysis; Regression models; Sea surface; Sea surface temperature; Seasonal variability; Seasons; Standard deviation; Statistical analysis; Statistical median; Statistical models; Statistical prediction; Studies; Surface temperature; Variability; Wind; Africa; Ethiopia; Addis Ababa Ethiopia
• ISSN: 0894-8755; 1520-0442
• DOI: 10.1175/JCLI-D-14-00476.1

An ensemble-based multiple linear regression technique is developed to assess the predictability of regional and national June–September (JJAS) anomalies and local monthly rainfall totals for Ethiopia. The ensemble prediction approach captures potential predictive signals in regional circulations and global sea surface temperatures (SSTs) two to three months in advance of the monsoon season. Sets of 20 potential predictors are selected from visual assessments of correlation maps that relate rainfall with regional and global predictors. Individual predictors in each set are utilized to initialize specific forward stepwise regression models to develop ensembles of equal number of statistical model estimates, which allow quantifying prediction uncertainties related to individual predictors and models. Prediction skill improvement is achieved through error minimization afforded by the ensemble. For retroactive validation (RV), the ensemble predictions reproduce well the observed all-Ethiopian JJAS rainfall variability two months in advance. The ensemble mean prediction outperforms climatology, with mean square error reduction (SSClim) of 62%. The skill of the prediction remains high for leave-one-out cross validation (LOOCV), with the observed–predicted correlationr(SSClim) being +0.81 (65%) for 1970–2002. For tercile predictions (below, near, and above normal), the ranked probability skill score is 0.45, indicating improvement compared to climatological forecasts. Similarly high prediction skill is found for local prediction of monthly rainfall total at Addis Ababa (r= +0.72) and Combolcha (r= +0.68), and for regional prediction of JJAS standardized rainfall anomalies for northeastern Ethiopia (r= +0.80). Compared to the previous generation of rainfall forecasts, the ensemble predictions developed in this paper show substantial value to benefit society.