[Previous Article] [Next Article]
Statistical Models for Seasonal Forecasts
of Southern Africa Summer Rainfall
contributed by Mark Jury
Center for Environmental Studies, Geography Department, University of Zululand, South Africa
A statistical system to objectively forecast southern African summer rainfall one season in advance
was described in the December 1996 issue of this Bulletin. This is briefly reviewed here, and the
current forecasts shown. Long-range predictive skill for southern Africa exists because of links
with the global El Niño-Southern Oscillation (ENSO) and regional climate teleconnections.
ENSO signals near Africa (Jury et al. 1994) and hemisphere-scale precursor patterns (Jury 1996)
have been described using correlational statistics. Global data sets include sea surface temperature
(SST) from COADS and UKMO data sets, upper level winds from Climate Prediction Center
(CPC) and ECMWF gridded products, surface air pressure and winds from COADS, ENSO and
quasi-biennial oscillation (QBO) indices from CPC, and satellite cloud estimates via outgoing
longwave radiation (OLR) and highly reflective cloud albedo from CPC from 1971 to the present.
Key areas identified by Rocha (1992) and Pathack (1993) offering high correlation at one season
lead time were considered. Rotated principal component patterns for global SST were analyzed
from normalized anomalies. Because of the repeated selection of satellite OLR indices in
preliminary analyses, model training periods were limited to 1971-1993.
Because forecasts are required in the austral spring for management purposes, predictors for the
July to November period were considered. To reduce noise, 3-month means were used.
Exploratory analyses identified >100 candidate predictors. Target rainfall data were obtained from
national weather services in the region. Emphasis was placed on areas with significant water
resources and rain-fed agricultural production. Rainfall station data were spatially grouped
according to annual total, seasonal cycle, elevation, and intercorrelation structures (e.g. principal
components). Area-averaged targets were about 300 km diameter as in Landman (1995), and
were shown in Figure 1 in the December 1996 issue of this Bulletin. It was considered that a
larger area-average improves the signal-to-noise ratio and thus the predictability potential, up to a
spatial limit defined by local users. Monthly station rainfall normalized departures were averaged
for early and late summer seasons: November to mid-January and mid-January to March. This
compromises between "noisy" single month targets and users' need for guidance on the temporal
distribution of rainfall within the summer.
Of the >100 candidate predictors developed, 49 predictors were selected by 15 area-specific
models. Of these 49, about one-third are significantly correlated with the Southern Oscillation
Index and eastern Pacific (Niño 3) SST, another one-third (some overlapping) describe the
transition of the Indian NE monsoon, and about one-quarter reflect conditions over the South
Atlantic and persistence of standing westerly waves in the subtropics. Lead times of one season
(i.e., no "skipped" time before the target season) were considered for the 1971-1993 training
period.
Multiple linear regression models were developed using a forward stepwise procedure on the
spring pre-dictors. To control over-fitting, a maximum of 5 predic-tors was imposed. Models
having the highest hindcast adjusted (for shrinkage) r-squared values were deter-mined. Models
with co-linear pairs of predictors having oppositely-signed coefficients were screened out. A
number of potential candidate models were developed for each target area and then subjected to
skill valida-tion tests using jack-knifing, the results of which lead to final model selection. The
predictors participating most heavily in the final models, described more fully in the December
1996 issue, include SSTs, OLR, ENSO or QBO indices, and wind/pressure variables. Correlation
skills between model forecasts and rainfall observations exceed 0.7 for most models. It should be
noted that in the jack-knife technique mentioned above, each year was held out in turn and the
predictor term coefficients (but not the choice of predictors) were determined from the remaining
years and used to forecast the withheld year. However, since the predictor combinations had
already been selected using the entire available sample, skills in real-time forecasts are expected to
be lower by an unknown (possibly small) amount than that shown by the jack-knife exercises.
Southern Africa tends to receive plentiful summer rains, especially in the latter half of summer,
when the following climate conditions are present: (1) a cold phase of ENSO, (2) an easterly
tropical Atlantic 200 hPa wind anomaly, (3) west QBO phase, (4) negative SST anomaly in the
central (near 60E at equator) Indian Ocean, (5) easterly surface winds (in JAS) in the eastern
(near 80E) tropical Indian Ocean, (6) positive SST anomaly (in JAS) in east (near 80E) tropical
Indian Ocean, (7) negative Atlantic SST in southwestern mid-latitudes (centered 40S, 45W), and
(8) positive Atlantic SST northeast of Brazil (10N, 10-50W). A subset of the above conditions is
used in Hastenrath et al. (1995), in which the sense of the relationships agree. The Atlantic SSTs
can influence the disposition of standing westerly waves, and the eastern tropical Indian Ocean
predictors describe the ENSO state in terms of circulation and SST.
For the spring (or earlier) 1997 predictor periods, the conditions of the major predictors, with
implications of southern African 1997-98 summer rainfall, are as follows (not exactly
corresponding to the above list): (a) ENSO condition (SOI and Niño 3 SST): strong El Niño
(although SOI has not been that low) [=>dry]. (b) tropical western (eastern) Indian Ocean SST:
warm (cold) [=>dry]. © QBO previous Sep-Oct-Nov: east-erly [=>wet]. (d) surface winds
anomalies in central tropical Indian Ocean strongly easterly; in most El Niños these are confined
to farther east. Therefore, NE monsoon recurvature may be limited, leading to wet east of 30E.
(e) OLR in west tropical Indian Ocean lowish (=>dry in central regions). (f) Atlantic Ocean
SSTs fairly neutral, but warm SST along Angola coast (=>dry in the west).
It is of interest that the upper level wind anomalies westerly in tropical Indian Ocean but not over
Atlantic, suggesting a decoupling of the South Atlantic westerly "wave" from the Pacific ENSO is
suggested. Also, current upper level (500 hpa) height anomalies are strangely not as expected for
an El Niño, which would be negative in the polar region and positive in subtropics around
southern Africa.
The objective forecasts for summer 1997-98 (expressed as percentages of mean 1971-93 mean rain-fall) for several southern African regions, based on the above diagnostics through the chosen multiple regres-sion equations (or in some cases, ensembles from more than one equation), are shown in the following table. The forecasts without an asterisk are reliable approxi-mately 3 out of 4 times, where "reliable" means that the correct category out of 3 (below, near, above normal) is forecast, allowing for a ±30% error in the forecast for the expected standard error in the predicted values.
Target Area (note target period when different from that given in the two right columns) |
Forecast %
norm
1 Nov- 15 Jan |
Forecast %
norm
16 Jan- 31 Mar |
Eastern Cape | 48 | 54 |
NW-FS Province | 32 | 74 |
Gaiting-North-Puma
Highveld |
44* | 66 |
Swazi-KZN-Lowveld | 58 | 110 |
KZ Natal (Dec-Mar) | 59 | |
Zimbabwe | 73 | 75 |
South Africa
Agricultural
Income ('98) |
67 | |
N. Namibia | 118 | 84 |
W. Zambia (mid-summer) | 29 | |
Botswana (I) | 48 | |
Madagascar (mid-summer) | 96 | |
Malawi
(mid-summer)
[but dry south, wet north] |
117 | |
N. Tanzania (Oct-Dec) | 153* |
*low confidence; jack-knife skill < 0.6
In summary, below to near-average rains may be expected. Below average rainfall may be
confined to western and central areas, with near normal rainfall and resource outputs over
the eastern areas of southern Africa. It is noted that the Indian monsoon circulation is quite
atypical for an El Niño, in that the equatorial surface easterlies are overshooting their usual El
Niño domain; thus, dry conditions may be shifted westward across southern Africa.
The above forecasts are part of a larger set of specialized southern African forecasts (e.g. climate
impacts such as maize yield, malaria incidence, major river flows, Indian Ocean tropical cyclone
days) posted on the following website during spring (September to December):
<http://weather.iafrica.com/forecasts/cip_seasonal_outlook.htmll>.
Hastenrath, S., Greischar, L., and Van Heerden, J., 1995: Prediction of the summer rainfall over
South Africa. J. Climate, 8, 1151-1518.
Landman, W. A., 1995, A canonical correlation analysis model to predict South African summer
rainfall. Exp. Long-Lead Forecast Bull., 4, 4, 23-24.
Jury, M. R., McQueen, C. A., and Levey, K. M., 1994: SOI and QBO signals in the African
region. Theor. Appl. Climatol., 50, 103-115.
Jury, M. R., 1996, Regional teleconnection patterns associated with summer rainfall over South
Africa, Namibia and Zimbabwe. Intl. J. Climatol., 16, 135-153.
Pathack, B. M. R., 1993, Modulation of South African summer rainfall by global climatic
processes. PhD Thesis, Univ. of Cape Town, 214 pp.
Rocha, A. M. C., 1992, The influence of global sea surface temperature on southern African
summer climate. PhD Thesis, Univ. Melbourne, 248 pp.