Note: This article narrowly missed the deadline for the June issue. The forecast was mentioned in the June "Summary of Forecasts" on the Internet, but not in the printed version. It is presented here even though the target period has passed, giving readers the unusual opportunity to verify a forecast right after reading about it.
 
 

The Ivory Coast (approximately 4-10N, 8-2W) lies at the western edge of a region (approximately south of 10N, 7.5W-7.5E) whose July-September rainfall total has a strong positive correlation (of the order r=0.7) with equatorial Atlantic sea-surface temperature (SST) (Ward et al. 1990; Janicot 1992; Rowell et al. 1995; Ward 1997). The association is robust, remaining steady through the complete historical record, beginning early this century. The July- September rains have not shown large multi-decadal fluctuations (Ward 1997).

The association is here exploited to make an experimental rainfall forecast for Ivory Coast. Investigations have shown that a robust predictor is the May SST anomaly averaged for the region 0-10S, 20W-10E (Aligbe et al. 1997).

First, a forecast is made for a rainfall index based on the 10 synoptic stations in the Ivory Coast with long records. Mean percentage of normal July-September rainfall is calculated for each year within 1955-90 and a regression is constructed with the May SST anomaly predictor. With one predictor, problems of overfitting are much less than is the case with multiple candidates. For independent verification we have fitted the model on the first half of the record and tested on the second half, and then reversed the process. Table 1 shows that the relationship holds well in both sub-periods.
 
 

Table 1. Forecasting the mean July-September percentage rainfall anomaly in Côte D'Ivoire. The first two columns give skill estimates for independent periods, the third column gives the model fit, the fourth column gives the 1997 forecast and the last column gives the standard error of the regression prediction.

Independent Period Skill, 1955-72	Independent Period Skill, 1973-90	Forecast Model Period (1955-90)	1997 Forecast,  Percent of Normal	1997 Forecast,  Standard Error (±)
r = 0.66	r = 0.74	r = 0.68	78.8	24.9

 
 
 
 
 

The May SST predictor in 1997 is -0.363C. This is considerably cooler than most recent years and the forecast is for a clearly below normal rainfall total (78.8% of the 1961-90 normal). The regression analysis has been repeated for individual stations (Table 2). The skill is generally better for the central and southern stations, and poorer for the northern and western stations, which are on the geographical edge of the teleconnection with SST.

The region normally experiences two rainfall peaks, one during March-June, the other during October -November, which are exploited for two crops. In years when the July-September rains are forecast to be good, there is the potential for innovative farmers to plant a second crop in early July, and have produce to take to market in September at a time when fresh produce is in short supply and prices are high.Table 2. Forecasting the July-September percentage rainfall anomaly at each of the 10 synoptic stations with a long record (listed by latitude). To assist interpretation of the forecast (last column), we also show the mean and standard deviation (SD) of the July-September rainfall total (in mm), independent skill estimates for 1955-72 and 1973-90 (correlation X100), and the model fit (correlation X100).

Location	1961-90  Mean (mm)	1961-90  SD (mm)	Indep. Skill  1955-72	Indep. Skill  1973-90	Forecast  Model Skill	1997 fcst  % normal
9.5N, 7.6W1	867.9	148.5	20	12	2	100.2
8.0N, 2.8W2	372.0	133.0	49	62	58	79.9
7.4N, 7.5W3	741.8	166.8	49	15	33	93.5
6.5N, 6.3W4	458.0	173.6	31	48	38	90.0
6.6N, 4.7W5	305.4	147.0	36	78	52	81.5
6.1N, 6.0W6	356.0	174.2	63	64	64	72.2
5.3N, 3.3W7	387.7	220.2	58	54	55	69.8
5.2N, 3.9W8	316.6	232.5	47	45	40	73.2
5.0N, 6.1W9	264.8	196.5	63	55	55	57.5
4.4N, 7.4W10	607.6	387.5	47	46	47	71.5

 
 
 

Station names: Odienne¹, Bondoukou², Man³, Daloa⁴, Dimbokro⁵, Gagnoa⁶, Adiake⁷,Abidjan⁸,Sassandra⁹, Tabou¹⁰.

With this application in mind, we have made an additional set of forecasts that can be viewed as forecasts for the "rainfall season's characteristics that are relevant to agriculture", including such features as dry spell lengths. For this, we calculated the Frere- Popov (1979) crop model to calculate a crop index in each year in 1955-90, using 10-day rainfall totals for each station. The index effectively measures the amount of water stress suffered by the crop (either/both excess and/or deficit), with a value of 100 for no stress. Thus, if there is a 30-day dry spell, the index will drop dramatically, and it cannot recover if there are subsequently very heavy rains--in fact, the index may fall further due to excess water. Results are shown here assuming 100-day maize is planted on July 1 in each year, and that each year has a potential evapotran-spiration of 50mm per 10 day period. Other crops and other choices of evapotranspiration have also been studied (Aligbe et al. 1997) and the skillful results (Table 3) found to be robust. To make a forecast, we make a regression between the crop index and the May SST anomaly index. An example of the relationship is shown in Fig. 1, demonstrating that a linear assumption is not bad, though more sophisticated regressions are possible. Results in Table 3 are presented in the same form as in Table 2. On average, there is more skill in predicting the crop index than in predicting the seasonal rainfall total. With the cool Equatorial Atlantic SST index this year, the forecasts do not encourage experimenting with cropping through JAS 1997.

Acknowledgments: The authors are grateful to Dr. Roger Stern (formerly ICRISAT, Niamey) for supervision on using the crop model and to another person, whose name escapes us, for assistance with accessing the daily precipitation data held at ANAM, Abidjan. The work was supported by the WMO project CLIPS and the African Centre of Meteorological Applications for Development, Niamey, Niger.

Table 3. Forecast of 100 day maize crop index (100 = crop experiences no stress from deficit or excess of water) at each of the 10 synoptic stations with long record; assumes maize is planted on July 1st. The forecast is shown in the last column. Other columns follow the format of Table 2.

Location	1961-90  Mean (mm)	1961-90  SD (mm)	Indep. Skill  1955-72	Indep. Skill  1973-90	Forecast  Model Skill	1997 fcst  % normal
9.5N, 7.6W1	94.9	4.1	16	1	5	95.2
8.0N, 2.8W2	76.3	18.0	51	22	44	67.7
7.4N, 7.5W3	94.5	4.7	76	43	60	89.1
6.5N, 6.3W4	83.4	14.8	52	59	55	75.0
6.6N, 4.7W5	70.7	23.2	62	71	66	56.9
6.1N, 6.0W6	73.0	20.2	59	49	53	62.8
5.3N, 3.3W7	68.2	18.2	86	71	78	52.0
5.2N, 3.9W8	48.7	19.0	63	64	64	35.8
5.0N, 6.1W9	45.0	16.3	80	66	70	33.6
4.4N, 7.4W10	81.4	15.8	29	22	26	77.4

 
 
 

Station names: Odienne¹, Bondoukou², Man³, Daloa⁴, Dimbokro⁵, Gagnoa⁶, Adiake⁷,Abidjan⁸,Sassandra⁹, Tabou¹⁰.
 
 

Aligbe, O., Y. Berte and M.N. Ward, 1997: The procedure for forecasting seasonal rainfall amount and crop performance in the Guinea Coast (West Africa), using Equatorial Atlantic Sea Surface Temperature Anomalies. African Centre of Meteorological Applications for Development. Niamey, Niger. Research Report, 50pp. (French version available from Y. Berte).

Frere, M. And G. Popov, 1979: Agrometeorological crop monitoring and forecasting FAO Plant Production and Protection Paper no. 17, 64pp.

Janicot, S., 1992: Spatiotemporal variability of West African rainfall. Part II: Associated surface and airmass characteristics. J. Climate, 5, 499-511.

Rowell, D.P., C.K. Folland, K. Maskell and M.N. Ward, 1995: Variability of summer rainfall over tropical North Africa (1906-92): Observations and modelling. Quart. J. Roy. Meteor. Soc., 121, 669-704.

Ward, M.N., J.A. Owen, C.K. Folland G. Farmer, 1990: The relationship between sea surface temperature anomalies and summer rainfall in Africa 4-20N. Long Range Forecasting and Climate Memorandum No. 32.Available from the National Meteorological Library, Meteorological Office, Bracknell, Berkshire, UK.

Ward, M.N., 1997: Diagnosis and short-lead time prediction of summer rainfall in tropical North Africa at interannual and multi-decadal timescales. J. Climate, submitted.

Fig. 1. Frere-Popov maize crop index for Jul-Aug-Sep at Adiake (5N, 3W) versus May SSTA in the equatorial south Atlantic. Daily rainfall data 1955-90 from ANAM. Ivory Coast Met. Service; Research at ACMAD Jan-Mar 1997 (Berte, Aligbe and Ward).