Food security assessment in sub-Saharan Africa requires monitoring the agrophysical and socioeconomic conditions of large and spatially dispersed populations. From a physical science standpoint, food security assessment requires monitoring climatic variables and modeling their implications for rain-fed agriculture on an ongoing basis. Simultaneously, the human factors of famine vulnerability must be accounted for and mapped. These include, for example, population distribution, household income, prices of grain and cattle (McCorkle, 1987; Kinsey et al., 1998), school attendance, employment opportunities, nutritional status (Shoham, 1987; Kelly, 1993), and other variables (Reddy, 1992). Food economy analysis is used by FEWS NET to structure an understanding of livelihoods and their vulnerabilities. Joint spatial analysis of agrophysical and socioeconomic factors is used to produce an integrated picture of the food security situation, for it is the coincidence of vulnerable livelihoods and hazards that defines the level of risk. Fieldwork, interviews with local experts, consultation with national early warning committees, and professional experience build on spatial analyses to yield a synthesis of the situation presented in monthly bulletins and special reports (FEWS, 1999). Consumers of this information include the decision-makers in host country governments, USAID, donor countries, multinational organizations, and nongovernmental organizations (NGOs)
having mandates and resources for response to food security emergencies. Targeting their responses directly benefits from FEWS NET analyses that identify the location and intensity of needs.
Early detection and early warning of famine must be persuasive enough to overcome the risk avoidance behaviors of decision-makers in responsible organizations—national governments, donor agencies, international organizations, and NGOs (Cutler, 1993). It is for this reason that FEWS NET food security analysts rely on a convergence of evidence to make food security assessments. No single source of information is sufficiently authoritative and comprehensive to identify potential famine areas alone (Mason et al., 1987; Shoham, 1987; Kelly, 1993). Therefore, analysts draw their conclusions most confidently when all factors indicate a certain food security status in a region. Any reduction of ambiguity associated with data or information used by FEWS NET contributes to confidence in food security assessments and an improved linkage between early warning and early response.
In subsequent sections of this chapter we explore traditional and current methods of monitoring drought and famine used by the FEWS network of scientists in the United States and Africa. We first discuss traditional use of satellite-derived vegetation index and rainfall estimates, followed by a discussion of more complex crop condition modeling using satellite-derived rainfall and vegetation information.
In Africa, in general, sparse data observation networks resulting in inadequate spatial coverage, data quality, and timely accessibility are problems that food security analysts often face. The use of remotely sensed data offers solutions to at least part of the problem. Geographically referenced (geospatial) climate monitoring products offer food security analysts succinct and practical summaries of crop growing conditions. The products are accessible in near-real time at a continental scale and are typically the most comprehensive and up-to-date observational data available, allowing analyses to be conducted at any administrative level.
Current operational climate monitoring for FEWS NET is based primarily on maximum value composite images of the normalized difference vegetation index (NDVI; Holben, 1986; Tucker and Sellers, 1986; chapter 5) and rainfall estimate (RFE) images (Herman et al., 1997; Xie and Arkin, 1997) produced on a 10-day time step. The FEWS NDVI archive dates back to July 1981; the RFE images have only been produced since 1995. Conventional rain gauge data are also analyzed, though the availability of these data (http://edcintl.cr.usgs.gov/adds/) varies from country to country, and there is a significant delay in obtaining data for many stations. These products are the basis for analyzing climate in the past, present, and future. The variability in the historical NDVI and rainfall data is the basis for estimating predisposition to drought. NDVI and RFE are the principal tools for current seasonal monitoring.
The primary geospatial climate monitoring products used by FEWS NET are derived from remote sensing data collected by meteorological satellites (Hutchinson, 1991). NDVI images produced from Advanced Very High Resolution Radiometer (AVHRR) imagery acquired by the National Oceanic and Atmospheric Administration (NOAA) polar orbiters have the longest history of use in the project (French et al., 1996). They are prepared for FEWS NET by the Global Inventory Monitoring and Modeling Studies research unit at the NASA Goddard Space Flight Center according to techniques described by Los et al. (1994). Since the NDVI signal is approximately linearly related to the area average photosynthetic capacity of the plant canopy at a location (Tucker and Sellers, 1986), it is used as an indirect measure of the condition of rain-fed crops. Chapter 5 provides details on NDVI applications.
Exploitation of NDVI by FEWS NET for monitoring is simple and straightforward. The image for the current 10-day period (or dekad) is used to compute two difference images. The first is the difference between the NDVI for the current dekad and that of the previous dekad. This reveals areas that are greening up or drying down. The second difference is with respect to the average NDVI for the 1982-2002 historical period. This reveals areas of anomalous conditions relative to the long-term average. The other operational geospatial climate product used by FEWS NET is the RFE produced by NOAA's Climate Prediction Center. They are also compiled on a dekadal basis, with each pixel's value representing an estimate of the total millimeters of rainfall that have fallen at that location during the 10-day period. Image differencing is applied to them in much the same way that it is to the NDVI images. A difference with respect to long-term average shows wet and dry rainfall anomalies. However, since the time series of the RFE is short, a standard based on surface fitting of station data with long records is also used (Hutchinson et al., 1996).
Apart from the difference-image products, FEWS NET also produces area-average time-series traces of NDVI and RFE for key crop-growing regions. Time series traces as well as other operational monitoring products (e.g., NDVI, RFE, daily rainfall estimates, soil water index anomaly) are available on the FEWS NET Web sites (www.fews.net; edcintl.cr.usgs.gov/ adds) and are described in Rowland (2001).
The spatially explicit water requirement satisfaction index (WRSI) is an indicator of crop performance based on the availability of water to the crop during a growing season. Food and Agriculture Organization (FAO) studies (Doorenbos and Pruitt, 1977) have shown that WRSI can be related to crop production using a linear yield-reduction function specific to a crop. Regional implementation of WRSI has been demonstrated in a geographic information system environment (Verdin and Klaver, 2002; Senay and Verdin, 2003).
WRSI for a season is calculated as the ratio of seasonal actual evapotranspiration (AET) to the seasonal crop water requirement (WR):
WR is calculated from potential evapotranspiration (PET) using the crop coefficient (Kc) to adjust for the growth stage of the crop:
AET represents the actual (as opposed to potential) amount of water withdrawn from the soil water reservoir. When soil water content is above the maximum allowable depletion (MAD) (based on crop type), the AET will remain the same as WR (i.e., no water stress). But when the soil water level is below the MAD, the AET will be lower than WR in proportion to the remaining soil water content. Soil water content is obtained through a simple mass-balance equation where soil water level is monitored by the water-holding capacity of the soil and the crop root depth; i.e.,
where SW is soil water content, PPT is precipitation, and i is the time step index.
The most important inputs to the model are precipitation and PET. PET values are calculated daily for Africa at 1° resolution from a 6-h numerical meteorological model output using the Penman-Monteith equation (Shuttleworth, 1992; Verdin and Klaver, 2002). Blended satellite-gauge RFE data are obtained from NOAA at 0.1° (~10 km) spatial resolution (Xie and Arkin, 1997). In addition, the model uses soil attributes from digital soils map of the world (FAO, 1988).
WRSI calculation requires start-of-season (SOS) and end-of-season (EOS) data for each modeling grid cell. The model determines SOS (or onset of rains) based on simple precipitation accounting. Figure 19.1 shows the SOS map for southern Africa during the 2001-02 growing season.
SOS is determined using a threshold amount and distribution of rainfall received in three consecutive dekads. SOS is established when there is at least 25 mm of rainfall in one dekad, followed by a total of at least 20 mm of rainfall in the next two dekads. The length-of-growing period for each pixel is determined by the persistence, on average, above a threshold value of the ratio between rainfall and PET. EOS is obtained by adding length-of-growing period to the SOS dekad. The WRSI model can be applied to different crop types (maize, sorghum, millet, etc.) for which seasonal water use patterns have been published in the form of a crop coefficient (FAO, 1998).
At the end of the crop growth cycle, or up to a certain dekad in the cycle, cumulative AET and cumulative WR are used to calculate WRSI. A case of no deficit will result in a WRSI value of 100, which corresponds to no reduction in yield related to water stress. A seasonal WRSI value < 50 is regarded as a crop failure condition (Smith, 1992).
Yield reduction estimates based on WRSI contribute to food security preparedness and planning. As a monitoring tool, the crop performance indicator can be assessed at the end of every 10-day period during the growing season. As an early warning tool, end-of-season crop performance can be estimated by incorporating long-term average of climatological data for the period from the current dekad to EOS. Due to the different growing seasons, WRSI maps are generated and distributed on a region-by-region basis (e.g., Sahel, southern Africa, GHA). At the end of each dekad, two image products associated with the WRSI (the current WRSI and extended WRSI) are produced and disseminated by FEWS NET.
The current WRSI map portrays WRSI values for a particular crop from the onset of the growing season until the current dekad. It is based on actual estimates of meteorological data to date. For example, if the cumulative crop water requirement up to this dekad is 200 mm and only 180 mm was supplied in the form of rainfall and available soil moisture, the crop experienced a deficit of 20 mm during the period, and the WRSI value will be (180/200)100 = 90%. This approach is slightly different from the traditional FAO update, where the cumulative deficit-to-date is compared to the seasonal crop water requirement instead of the requirement up to the current period. The FEWS NET WRSI may increase, decrease, or remain the same as the season progresses depending on the water supply and deficit. The FAO and FEWS NET WRSI products are mathematically equivalent when the EOS dekad becomes the current dekad.
Figures 19.2 and 19.3 demonstrate the monitoring capability of WRSI during the 2001-02 growing season in southern Africa. Figure 19.2 shows early signs of drought at the middle of the growing season (end of January 2002) in Zimbabwe and surrounding countries. Figure 19.3 shows an intensified and expanded drought by the end of the growing season (end of April 2002). The depictions in these maps were corroborated by field reports, in which 14 million people were reported to require relief assistance.
The extended WRSI map is an estimate (or forecast) of WRSI for the EOS. Long-term average rainfall and PET are used to calculate WRSI for the period between the current dekad and EOS. The calculation principles are the same as for current WRSI.
The quality of rain-fed crop production in Sahel West Africa is largely a function of the temporal and spatial distribution of seasonal rainfall. A method based on departures of NDVI and RFE data from their respective averages has been devised to monitor the annual growing season in the Sahel and to make qualitative assessments of harvest prospects up to 4-6 dekads in advance.
Food security and vulnerability analysis and mapping are often based on administrative subdivisions. This facilitates assistance to vulnerable populations in areas affected by either severe production shortfalls or other shocks to the population's livelihood. As previously mentioned, sparse data observation networks, poor quality data, and untimely accessibility are the main problems associated with ground/station data. However, neither NDVI nor RFE is problem free. Combining these measures helps decrease the uncertainty in the result. The rainfall and NDVI combined departures (RNCD) is a method based on a combination of NDVI and RFE data into an index that reflects the quality of growing season conditions, extracted for given administrative units. The growing season is divided into three distinct periods and diagnosed separately for each period. The results of the three periods are combined in a final step to diagnose the whole growing season.
Identification of Growing Season Periods The RNCD method uses three distinct growing season periods whose lengths are agroclimatological-zone dependent: (1) the sowing period, from the dekad when generalized sowing is observed to the dekad when sowing would be considered too late; (2) the vegetative growth period, from the end of the sowing period to the maximum vegetative growth indicated by the dekad of maximum NDVI; and (3) the maturation period, from the end of the second period to complete maturation of the vegetation (pastures as well as crops). As an example, for the Sahelian zones, these periods translate to the following dates: first
period, from third dekad of June to third dekad of July; second period, from first to the third dekad of August; and third period, from first to third dekad of September. Similar periods are defined for the Sahelo-Sudanian, Sudanian, and Sudano-Guinean zones.
Delineation of Base-Unit Polygons NDVI data for the period from 1982 to 1995 were used to define five agroclimatological zones in the Sahel depending on the timing of green-up. For example, zones are defined as "very early" green-up in the southernmost regions of the Sahel, all the way to "late" green-up in the northern regions of the Sahel. The agroclimatolog-ical zones have been merged with conveniently small administrative units of the Sahel, resulting in an agroclimatic-administrative unit map that consists of polygons referred to as "base units." This step is necessary because the subdivision into agroclimatological zones does not follow administrative borders, and some administrative units are large enough to straddle two or more agroclimatic zones. The base units are used to extract spatial statistics (e.g., average) for NDVI and RFE.
Computation ofRNCD Indices Data used for the computation of indices are extracted from the images for all dekads during the growing season using the base units. After data are extracted for each dekad and each polygon, a dekadal index (DI) is calculated according to
where x is either NDVI or RFE for a given polygon for the current season, x is the historical average, and a is standard deviation.
The average period index (PI) is obtained by averaging the DI over the period in question for NDVI and RFE. For example, if we consider the first period for those base units that fall into the very early agroclimatological zone, the averaging period will be from the second dekad of May to the third dekad of June; i.e.,
where DI is summed from the beginning dekad to the ending dekad, and N is the number of dekads.
PI values for NDVI and RFE are combined to obtain a growing-condition period index (GCPI) for each period. These GCPIs could be mapped to analyze the three parts of the growing season separately (the sowing period, the vegetative growth period, and the maturation period). Finally, an average of the three GCPI is calculated to obtain a growing season index called the RNCD. High values of the RNCD thus obtained are indicative of good rain-fed crop production. One should note that because of the low NDVI for water, flooded plains show a low RNCD. That is why RNCD is used to assess the seasonal outcome where rain-fed crops and pastures are predominant, as opposed to areas where recessional agriculture is predominant.
Evaluation of RNCD At the end of the growing season, the RNCD index was used to diagnose the 2001 growing season for rain-fed agricultural conditions (crops and pastures) and possible outcome. The results presented here represent the preharvest assessment for rain-fed crops and pastures and could be used for vulnerability assessment as well. Figure 19.4 shows the index based on the average of the three indices.
The analysis generally shows that good growing conditions prevailed over the Sahelian zone, with the exception of few limited areas distributed over all the countries. The largest of these areas is in eastern Chad (northeastern Biltine and southeastern Ennedi). At the western end of the Sahel, such areas are also seen in western Mali and in the border region of northern Senegal-southern Mauritania. The rest of the Sahelian zone looks very good.
The results for the Sahelian zone are in good agreement with statistical analysis. The analysis shows that 2001 growing season conditions were slightly worse than average over most of the Sudanian zone. However, the magnitude of the departures shows that the deficits are mild when compared to average conditions. The lowest negative values are around -1.0. However, over most of the Sudanian zone the index is between -0.4 and -0.1, which suggests milder departures from average conditions when considering that the highest positive value is about 3.2.
Agricultural statistics indicate that the 2001 rain-fed production was very good even in the Sudanian zone. The RNCD results are, therefore, only in partial agreement with the agricultural statistics data. Further investigation has shown that there are good reasons for this disagreement.
The RNCD index represents departures of NDVI and RFE from their respective averages. Although in the Sahelian zone any deficit in rainfall, and consequently in biomass, can have serious repercussions on production, this is not always the case in the Sudanian zone. In the Sudanian zone, where cumulative rainfall varies from 800 to 1000 mm, a 200-mm deficit may not have any ill effect on crop growth and development if rainfall is well distributed. In fact, in some cases, reasonable-length dry periods may contribute to increased yield, due to increased facility for weeding when soil moisture decreases for a period of time.
The negative departures of the RNCD index are small, since the lowest values are less than -1.0, whereas positive departures go above 3.0. Furthermore, most of the area within the Sudanian zone depicting negative departures has values ranging from -0.4 to -0.1. The area of values ranging from -1.0 to -0.4 is relatively small. The RNCD method is good for detecting anomalies in rainfall and biomass pattern. For this reason a comparison with recorded rainfall was made. Comparison of rainfall departures with respect to average was performed and good agreement was found between the spatial distribution of RNCD results and rainfall departures. Mali and Burkina Faso (figure 19.4) were chosen for this comparison because of the relatively high density of rain gauge distribution. In Mali, both RNCD and actual rainfall show some deficit in the south, especially the southeast. RNCD shows the northern part of the agricultural area of this country as above average in rainfall and rainfall records show the same area as average and slightly above average. In Burkina Faso the same is observed. RNCD shows that western and central parts of the country
Figure 19.4 Rainfall and NDVI combined-departures versus rainfall anomalies (with respect to normal) for Burkina Faso (top) and Mali (bottom) for the year 2001.
Figure 19.4 Rainfall and NDVI combined-departures versus rainfall anomalies (with respect to normal) for Burkina Faso (top) and Mali (bottom) for the year 2001.
have deficts in rainfall distribution. Here again, the RNCD method shows higher positive departure than rainfall records, which show the same area as above to slightly above average. It should be noted, however, that slight differences between the two methods may be due to the difference in the normalization of the departures from average. RNCD differences between current values and averages were normalized by the standard deviation, whereas rainfall departures were normalized by the average. In any case, the RNCD method captured the slightest anomalies in rainfall pattern.
The RNCD index is directly related to agricultural production. When these anomalies are large, its relationship with crop production is expected to be strong. However, when they are mild, as in the present case, the relationship of the index with crop production may be weak. In these conditions a crop-specific method such as the WRSI should be used for better results.
Was this article helpful?