Experimental Forecast of East African Rainfall for October-December 2000

contributed by Andrew Colman

Seasonal Forecasting Group, Ocean Applications, The Met. Office, Bracknell, UK

Introduction

The Met. Office is conducting research into the effects of sea surface temperatures and other climatic variables on tropical rainfall. As part of this research, experimental seasonal rainfall forecasts have been made for the Sahel and adjacent regions in tropical NW Africa since 1986, and for the Nordeste region of Brazil since 1987. Using similar statistical methods, forecasts for tropical East Africa October-December rainfall (the 'short rains') have been issued since 1994 and appear in previous September issues of this Bulletin. A long-lead forecast for East African rainfall using observed data up to the end of July has already been produced and was contributed to the Greater Horn of Africa Climate Outlook Forum (GHACOF6 ). This forecast uses observed data up to the end of August.

The region covered by the East Africa prediction is between 5N and 15S and between 30E and the Indian Ocean coast. These forecasts for E Africa were produced using statistical methods and by using The Met. Office's Atmospheric General Circulation Model (AGCM). This year we have introduced forecasts for 2.5^o latitude x 3.75^o longitude rectangular grid box regions. Skill is not so high at this higher resolution but these forecasts give an indication of rainfall distribution.

The statistical forecast is made by using linear regression and discriminant analysis techniques, with three indices of global sea surface temperature (SST) anomaly patterns (Appendix, Figure 4a, 4b and 4c). The forecast model is derived from historical rainfall and SST information.

The AGCM forecast was extracted from a nine member ensemble of AGCM predictions using sea temperatures and atmospheric conditions observed just prior to when the forecast was run (September 1^st). The historical rainfall record is divided into 5 equi-probable categories. Based on 1961-1990 rainfall, the category boundaries (as percentages of mean rainfall) are:

Forecast Skill - Performance of Trial Forecasts for the Past 50 Years

The statistical and dynamical forecasts were tested using trial forecasts over the period 1948 to 1997. The assessment measure used is correlation. Statistical linear regression forecasts were assessed using a method where a trial forecast is made for each year using a regression equation calculated using data for the remaining years. This assessment provides a good measure of forecast skill from minimal data. To provide an indication of AGCM skill, the performance of a long term AGCM run forced with observed SST in simulating rainfall is measured.

Statistical forecast skill correlation = 0.50

AGCM simulation skill correlation = 0.65

These correlations are statistically significant at the 5% level.

Performance of Real Time Empirical Forecasts

Forecasts have been made for this region since 1994. The forecasts for 1994 and 1995 were strongly influenced by above and below average SST in the NW Pacific respectively and the forecasts for 1997 and 1998 where influenced by the 1997 El Niño and the 1998 La Niña events.

Year	1994	1995	1996	1997	1998	1999
Forecast Category	Very Wet	Dry	Average	Wet	Average Dry	Dry
Observed Category	Wet	Dry	Very Dry	Very Wet	Very Dry	Average

Note: The categories used for the 1994-1998 forecasts are based on a 1951-1980 climatology. For the 1999 and 2000 forecast, categories based on the 1961-1990 climatology are used as 1961-1990 is the accepted WMO standard climatology period and is used by most forecasters. The 1961-1990 rainfall average is 104% of the 1951-1980 average.

Forecasts for the 2000 Season

Statistical Forecast:

The La Niña event that was associated with recent dryness in parts of East Africa has ended. Warm SST anomalies in the NW Pacific are favoring above average rainfall in E Africa this year. The regression forecast is 106% of the 1961-1990 average and is in the WET category. The discriminant analysis technique gives the following probabilities for the 5 (1961-1990 based) categories:

Very Dry	Dry	Average	Wet	Very Wet
0.10	0.22	0.21	0.29	0.18

AGCM Dynamical Forecast:

The ensemble mean prediction provides a best estimate rainfall forecast which is: 83% of the 1961-1990 model climatology. Based on the performance of AGCM ensemble simulations of rainfall from 1961 to 1990, the AGCM ensemble forecast is presented as probabilities of 5 (1961-1990 based) observed rainfall categories which are:

Grid Box Forecasts

The grid box forecasts are expressed as probabilities of terciles which are climatologically equiprobable over 1961-1990. This is in order to make the forecasts compatible with GHACOF forecasts which are expressed in the same way. Figure 1 shows the skill of the empirical forecasts. The empirical and dynamical forecasts are shown in figure 2 and figure 3 respectively.

Two forecast maps are shown for each category, one includes all grid boxes for which there is data, the second (skill mask) version includes only grid boxes where independent test correlation skill is significant. To be included on the skill mask map, hindcasts for the box are must pass at least 1 of these 3 tests:

1) Correlation between independent hindcasts and observations over 1981-1998 are significant at the 5% level (shown in figure 1)

2) Correlation between independent hindcasts and observations in mature La Niña events (la Niña event at least 1 year old and continuing) during 1981-1998 are significant at the 5% level

3) Correlation between independent hindcasts of this years forecast tercile and observations during 1981-1998 are significant at the 5% level

Overall Best Estimate:

This year, there is a clear conflict between the signal from the AGCM forecasts which are indicating below average rainfall to be likely and the empirical forecasts which are indicating above average rainfall. The empirical forecast anomalies are weaker. Whilst skill levels of the dynamical and empirical forecasts are broadly similar, the statistical forecast is dominated by a link between SST anomalies in the NW Pacific and African rainfall which has not been fully explained physically. The dynamical model is considered more physically reliable; hence our best estimate is based on the dynamical input and is for the DRY category.

Reference:

Mutai, C.C., Ward, M.N and Colman, A.W. Prediction of East Africa seasonal "short rainfall" rooted in evidence for widespread SST-forced variability during October-December. I.J.Climatol.18 975-997 (1998).

Acknowledgements:

Thanks to David Rowell for providing output from the HADAM3 model. Thanks to Robin Clark and Richard Graham for supplying dynamical forecast output.

Appendix:

Predictor patterns used for empirical forecast. The pattern shown in the third panel of fig. 4 is the most important predictor contributing to over 50% of the forecast variance.

Figure 1: Correlation Skill of Empirical Regression Forecasts

Figure 2: Probability Forecasts by Empirical Method

Figure 3: Probability Forecasts From AGCM Dynamical Forecast

Figure 4a, b, c: Empirical predictor patterns of SSTA.