Climate and Satellite Indicators to Forecast Rift Valley Fever Epidemics in Kenya

See allHide authors and affiliations

Science  16 Jul 1999:
Vol. 285, Issue 5426, pp. 397-400
DOI: 10.1126/science.285.5426.397


All known Rift Valley fever virus outbreaks in East Africa from 1950 to May 1998, and probably earlier, followed periods of abnormally high rainfall. Analysis of this record and Pacific and Indian Ocean sea surface temperature anomalies, coupled with satellite normalized difference vegetation index data, shows that prediction of Rift Valley fever outbreaks may be made up to 5 months in advance of outbreaks in East Africa. Concurrent near–real-time monitoring with satellite normalized difference vegetation data may identify actual affected areas.

Rift Valley fever (RVF), a viral disease first described in Kenya in 1931 (1), affects domestic animals and humans throughout sub-Saharan Africa and results in widespread livestock losses and frequent human mortality. Its occurrence is known to follow periods of widespread and heavy rainfall associated with the development of a strong intertropical convergence zone, the region in the equatorial tropics where air currents from the north and south converge and produce precipitation (2). Such heavy rainfall floods mosquito breeding habitats in East Africa, known as “dambos,” which contain transovarially infected Aedesmosquito eggs and subsequently serve as good habitats for otherCulex species mosquito vectors (3). The most recent RVF epizootic/epidemic was in East Africa in late 1997 and early 1998.

Vegetation responds to increased rainfall and can be easily measured by satellite. Normalized difference vegetation index (NDVI) data from the advanced very high resolution radiometer (AVHRR) on National Oceanic and Atmospheric Administration (NOAA) satellites have been used to detect conditions suitable for the earliest stages in an RVF epizootic (4). Refinement in determining the spatial distribution of RVF viral activity, through identification of ideal mosquito habitat, has been possible with higher resolution Landsat, Systeme pour l'Observation de la Terre (SPOT), and airborne synthetic aperture radar data (5); however, predictive indicators are needed to forecast RVF outbreaks. Here we show that several climate indices can be used to predict outbreaks up to 5 months in advance.

The El Niño–Southern Oscillation (ENSO) phenomenon is a principal cause of global interannual climate variability (6, 7). Warm ENSO events are known to increase precipitation in some regions of East Africa and result in droughts in southern Africa (7, 8). The Southern Oscillation Index (SOI) is the most commonly used index for the ENSO phenomena (7, 9) and extends back to the late 19th century. This index compares atmospheric pressure in Tahiti with that of Darwin, Australia, and is expressed as a standardized deviation from the norm. Strong negative anomalies are associated with an El Niño event (6, 10). Anomalous climatic conditions caused by ENSO are now recognized to be linked with outbreaks of various human and livestock diseases in various countries (7). Above normal East African rainfall is associated with negative SOI anomalies resulting in more green vegetation, which then is detected by the satellite-derived NDVI (11, 12).

We compared RVF virus activity with corresponding monthly SOI from 1950 to 1998, sea surface temperatures (SSTs) from an equatorial region in the Pacific Ocean (named NINO 3.4, 5°N to 5°S, 170° to 120°W), equatorial western Indian Ocean SSTs (10°N to 10°S, 40° to 64°E), and Kenyan NDVI AVHRR data from 1982 to 1998 (13). During this 48-year period, there were eight periods with RVF viral activity and 13 periods when there were strong negative anomalies in the SOI (<−1.5) (Fig. 1).

Figure 1

A time series plot of SOI anomalies between January 1950 and May 1998. Periods of RVF activity in Kenya are depicted. Monthly SOI values are shown as standardized deviations based on the 1951–80 mean.

Rainfall exceptionally above normal was coincident with major regional RVF epizootics in 1951–53, 1961–63, 1968–69, 1977–79, and 1997–98 (2). In late 1957 and 1982 and in the middle of 1989, heavy rainfall in Kenya preceded RVF virus activity that was detected by identification of clinical cases, isolation of the virus in mosquitoes, or detection of high levels of immunoglobulin (type IgM) antibody specific for RVF virus (indicating recent RVF infection) in domestic animals and humans (2, 3, 14).

RVF activity and above normal rainfall always followed a period of strong negative deviation of the SOI; however, neither the strength nor the length of the SOI anomaly correlated with the intensity of RVF activity. The regional RVF activity detected in 1982 followed an intense SOI anomaly <−4, whereas the major outbreaks starting in 1951, 1961, and 1968 occurred after SOI anomalies <−2. Strong negative SOI anomalies also occurred in 1964, 1969, 1972–73, 1981, and 1991–95; however, there was neither above normal rainfall nor detectable RVF activity in Kenya for these periods. Although excessively heavy rainfall and RVF activity in Kenya were dependent on a strong SOI anomaly, the overall ability to predict an RVF outbreak with SOI anomalies alone was only 67%, indicating that other factors must be involved.

A strong relation between equatorial Pacific SST and elevated East African precipitation has been reported (8, 12), as has a strong relation between equatorial Pacific SST and maize yield in Zimbabwe (15). In addition, Indian Ocean SST has been reported to be highly related to rainfall in East Africa (16).

Concurrent Pacific and Indian Ocean SST anomalies >3°C and 0.5°C, respectively, were correlated with widespread rains in East Africa (8, 12, 16) and RVF outbreaks. RVF activity was followed by 2 (1982–83) and 5 (1997–98) months of strong concurrent equatorial Pacific and Indian Ocean SST anomalies (Fig. 2). When both equatorial Pacific and Indian Ocean SSTs were elevated, the extent of the Indian Ocean temperature anomaly was indicative of the intensity and the duration of RVF activity. However, when the strength of the concurrent Pacific and Indian Ocean SST anomalies was reduced, but still positive, the pattern of increased rainfall can be irregular and a region-wide effect cannot usually be found.

Figure 2

Time series plots of Indian Ocean SST anomalies from January 1982 to May 1998 with SOI anomalies (A), equatorial Pacific Ocean (Niño 3.4 area) SST anomalies (B), and Nairobi NDVI anomalies (C). Indian and equatorial Pacific Ocean SST anomalies are depicted as degree Celsius deviations from their respective mean value (mean value for period shown = 0°C). SOI and Nairobi NDVI anomalies are depicted as deviations from their respective mean values normalized by the standard deviation (mean value for period shown = 0).

To overcome the problem of determining where, and to what extent, RVF outbreaks were possible in years lacking concurrently high Pacific and Indian Ocean SST anomalies, we used NDVI measurements derived from NOAA polar-orbiting satellite data to identify areas of abnormally high green vegetation development resulting from abnormally high rainfall (17). This is possible through the use of intercalibrated satellite data running from 1981 through the present (18).

We suggest that NOAA AVHRR NDVI time series data are required to identify more localized areas where anomalous rainfall has occurred and hence more localized RVF activity is present. It is only by doing this discrimination that lower amounts of RVF can be confirmed or, conversely, that a lower threshold of rainfall anomalies coupled with extent of affected area can be determined as this relates to RVF activity outbreaks. The satellite normalized difference vegetation data are available the same day as acquisition, provide confirmation of predicted rainfall events with SSTs, and provide direct identification of localized rainfall anomalies.

Elevated NDVI anomalies, as indicated by dark green shades inFig. 3, were observed for East Africa starting in October 1997 (the start of the normal short rainy period) and extending to April 1998 (through the normal dry season of January and February) (Fig. 3). NDVI anomalies were significantly correlated with RVF activity 1 to 2 months before detection of viral activity (P < 0.5) (19).

Figure 3

Monthly AVHRR NDVI composite images of continental Africa during the 1997–98 ENSO warm event. Data depicted are the degree of deviation from the long-term mean calculated for the period January 1982 to May 1998 in NDVI units (13). A value of zero means that current values are identical to the monthly 1982–95 mean.

Strong NDVI positive anomalies were observed in June 1989 with the presence of RVF activity and in January and February 1993 in the absence of detectable RVF activity (Fig. 2C). We suggest that the elevated NDVI values at these times reflected local rain conditions because they were not observed in NDVI anomaly data for the same period at other locations in Kenya.

To determine the best predictors of RVF activity, we evaluated SOI, equatorial Pacific SSTs, Indian Ocean SSTs, and NDVI anomalies in various combinations in ARIMA models (20). The best fit to the RVF outbreak data was achieved when equatorial Pacific and Indian Ocean SST and NDVI anomaly data were used together (ARIMA, SBC = −106, analysis of variance df = 192, P < 0.01). These data could have been used to successfully predict each of the three RVF outbreaks that occurred between 1982 and 1998 without predicting any false RVF events for an overall prediction of risk of 100%. Predictive models that use either SOI and Indian Ocean or NDVI and Indian Ocean anomaly data would have predicted all three RVF events but falsely predicted either one or two disease events, respectively.

The ability to forecast regional RVF virus activity in Kenya, based on Pacific and Indian Ocean SST anomalies and NDVI, 2 to 5 months before outbreaks could permit vaccination of domestic animals and pretreatment of mosquito habitats adjacent to domestic animal herds and human habitations with highly effective sustained release insecticides that would be released upon flooding (21).


View Abstract

Stay Connected to Science

Navigate This Article