Agriculture and Food Availability – Remote Sensing of Agriculture for Food Security Monitoring in the Developing World

Michael E Budde1, James Rowland2, and Chris Funk1

1 US Geological Survey Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD.
2 ASRC Research & Technology Solutions (ARTS), contractor to the US Geological Survey EROS Center. Work performed under USGS contract 08HQCN0007, Sioux Falls, SD.

Figure 1. Relative production as a fraction of the high 2005/06 production year in Zimbabwe. Actual Σv production estimates (tons) for each production year are shown on top and PECAD production figures are shown at right. The 2006/07 production estimate is shown with an error estimate included.

Figure 1. Relative production as a fraction of the high 2005/06 production year in Zimbabwe. Actual Σv production estimates (tons) for each production year are shown on top and PECAD production figures are shown at right. The 2006/07 production estimate is shown with an error estimate included.

Introduction

The recent global food crisis brought food security issues to the forefront of the world’s consciousness. The impacts of the crisis have been felt most seriously in third world countries. According to the International Monetary Fund, food prices increased 43 percent between March 2007 and March 2008. While developed countries are often able to mitigate impacts of such crises, developing countries are most affected and take much longer to recover. The poorest populations spend a larger proportion of their income on basic food supplies, making them the most vulnerable to increased prices. A recent US Agency for International Development (USAID) report stated that nearly 1 billion people, approximately one sixth of the world’s population, live on less than $1 per day and, of these, 162 million survive on less than $.50 per day.

In addition to market-driven impacts on food security, many of those at risk rely upon adequate weather conditions for subsistence agricultural activities. Subsistence agriculture, a form of farming where nearly all commodities produced are consumed by farmers and their families, persists in many parts of the world and is especially widespread in sub-Saharan Africa. The combination of high food prices and poor growing season conditions can be devastating for this segment of the world’s population. Therefore, there is a profound need to accurately monitor growing season conditions that impact food security in the developing world.

Background

Scientists with the U.S. Geological Survey (USGS) are part of a network of both private and government institutions that monitor food security in many of the poorest nations in the world. The Famine Early Warning Systems Network (FEWS NET) is a USAID-funded activity that collaborates with international, regional, and national partners to provide timely and rigorous early warning and vulnerability information on emerging and evolving food security issues. Currently, FEWS NET professionals in Africa, Central America, Haiti, Afghanistan and the United States monitor and analyze relevant data and information in terms of its impacts on livelihoods and markets to identify potential threats to food security.

FEWS NET uses a suite of communications and decision support products to help decision makers act to mitigate food insecurity. These products include monthly food security updates, regular food security outlooks and alerts, as well as briefings and support to contingency and response planning efforts.

Figure 2. Σv (left column) and Σv anomalies (right column) for seven growing seasons, 2001/02 through 2007/08, ranked from worst to best based on PECAD production estimates for Zimbabwe.

Figure 2. Σv (left column) and Σv anomalies (right column) for seven growing seasons, 2001/02 through 2007/08, ranked from worst to best based on PECAD production estimates for Zimbabwe.

Need for Remote Sensing

FEWS NET relies heavily upon its national and regional offices in sub-Saharan Africa to monitor aspects of food security. However, the broad scope of information that these offices are responsible for analyzing and the large areas which are being monitored drive the need for tools such as remote sensing to provide additional data for food security decision making.

Remote sensing provides the ability to monitor large areas on regular intervals. Satellite-based data and modeled derivatives are used in combination with ground-based information to better assess potential impacts to the food supply system. The USGS Earth Resources Observation and Science (EROS) Center, in collaboration with other FEWS NET implementing partners (NASA, NOAA, University of California, Santa Barbara), provides a number of remotely sensed inputs to the FEWS NET decision making process.

Agricultural Monitoring Products

Some of the most widely used remotely sensed products for agricultural monitoring are precipitation, crop water requirements, and vegetation indices. Precipitation is monitored primarily through the use of satellite-based rainfall estimates (RFEs) that augment the sparse observational network of rain gauge stations found in many FEWS NET countries. RFEs provide daily estimates at a gridded cell size where each cell represents a 0.1 by 0.1 degree area on the ground. These data are useful for large area precipitation monitoring and are also used as inputs to crop performance models.

One such crop model, the water requirement satisfaction index (WRSI), is based on the water supply and demand a crop experiences during a growing season. It is a ratio of seasonal evapotranspiration to the seasonal crop water requirement. The water requirements of specific crops are adjusted for various growth stages and are compared to the available moisture (incoming precipitation and existing soil moisture). Output from this model is used extensively to monitor both cropland and pasture conditions, and to assess potential food security impacts.

The normalized difference vegetation index (NDVI) is one of the original remotely sensed data types used by FEWS NET more than 20 years ago, and still used today. NDVI, calculated by measuring the intensity of visible and near-infrared light reflected by the land surface and “sensed” by satellites, quantifies the amount and vigor of vegetation at the land surface. Daily NDVI measures are combined into multi-day composites that portray the Earth’s vegetation condition and identify areas where plants are flourishing and where they may be under stress.

Case Studies

These agricultural monitoring products are used on a routine basis for operational monitoring of near-term factors that may impact food security. On occasion there is a need to go beyond our operational monitoring efforts and address specific questions that respond to specific food security questions. In the remainder of this article, we highlight two case studies that illustrate the use of NDVI data for analyzing both maize production and winter wheat yields. Each of these targeted analyses provided timely information that had a significant impact on food security decision making.

Zimbabwe

Introduction

Responding to a request by USAID, we analyzed remote sensing data that helped assess maize harvest prospects for Zimbabwe. Maize is the most widely grown cereal crop in Zimbabwe with lesser amounts of wheat, sorghum, and millet also being grown. We used an NDVI-based metric, sum v (Σv), that relies on measures of vegetation condition throughout the growing season to assess the likely production outcome. The initial analysis, for the 2006/07 growing season, has been replicated each year since.

Methods

The Σv method relies on the finding that late season NDVI correlated well with US Department of Agriculture (USDA) Production Estimates and Crop Assessment Division (PECAD) production estimates for those years prior to the 2006/07 season. Therefore we could use measurements of late season NDVI for a given year to estimate production numbers. We used Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI data for the years 2000 through 2007. In order to minimize the influence of non-agricultural lands on Σv, we applied a mask to the NDVI time series using a cultivated lands map. We were not able to specifically isolate maize within the agricultural zone, but the fact that maize is the major cereal crop gives confidence that monitoring these areas will be a good surrogate for assessing overall food security. The NDVI time series for agricultural lands were adjusted based on varied onset of season times determined from rainfall inputs. The onset-adjusted NDVI values were set to zero to represent the beginning of the growing season. Once adjusted, we ignore the first ten 16-day periods of the time series, then accumulate NDVI over the next two 16-day periods. It is this sum of late season NDVI, or Σv, that can be used to estimate production at the national level. The onset of season or planting date is a critical component of this analysis; therefore we have recently incorporated more ground-based information for this parameter. FEWS NET field staff in southern Africa have been instrumental in making these improvements to the modeling effort and have provided extensive input to final results.

Results

Figure 3. Annual maximum NDVI and CFSAM yield figures ranked from best to worst for irrigated areas of Afghanistan. The 2008 maximum NDVI ranks worst in the series. The yield estimate for 2008 is based on the regression of maximum NDVI and historical yield figures.

Figure 3. Annual maximum NDVI and CFSAM yield figures ranked from best to worst for irrigated areas of Afghanistan. The 2008 maximum NDVI ranks worst in the series. The yield estimate for 2008 is based on the regression of maximum NDVI and historical yield figures.

For the 2006/07 season in Zimbabwe, the Σv method resulted in an estimate of 688,000 tons of maize (Figure 1). Results from the 2007/08 season showed a dramatically reduced production figure on the order of 480,000 tons. The reduction was due largely to a combination of dry conditions during critical periods of the growing season and inadequate supplies of seed and fertilizer. The effects could be seen in the Σv and Σv anomalies that show the 2007/08 season ranks as the worst among the years studied (Figure 2).

The analysis for the 2008/09 season was perhaps the most complex. The season was characterized by highly variable rainfall that may have prompted early planting in some areas and late planting in others. Our production estimate seems to fall between the USDA Foreign Agricultural Service estimate (based on low yields, and planted area significantly reduced from the previous year) and the Zimbabwe Ministry of Agriculture estimate (relatively high yields, with planted area similar to the previous year). USGS maize yield and production estimates for 2008/09 aligned closest with those for 2006/07, and provided estimates much improved over the previous season. While actual ground conditions were not (and are still not) fully known, the story seems consistent – early planting with sufficient inputs would produce good yields. December (or late) planting, combined with mid-season dryness in the north-east, could result in a 20% reduction in yields.

Remote sensing provides a method that can be used in areas where extensive field campaigns are often impractical. In the case of Zimbabwe, these annual production estimates provided one more piece of information useful for food security decision makers.

Afghanistan

Introduction

Accurate and timely assessments of winter wheat production are important elements of food security decision making for Afghanistan. Approximately 80 percent of Afghanistan’s wheat production is supplied by irrigated winter wheat that relies heavily upon spring snowmelt. The 2007/08 winter wheat season was characterized by below average snowpack, abnormally high spring temperatures, early snowmelt, and poor rainfall, which created drought conditions throughout most of the country. USAID food security analysts wanted to characterize the impact of these conditions on probable outcomes of the winter wheat season. We were asked to assess how drought conditions would impact the 2007/08 winter wheat season and frame the severity of the drought in the context of recent years as a guide to decision making.

Methods

Figure 4. The dramatically reduced productivity in 2008 is evident in this comparison with the high production year of 2003.

Figure 4. The dramatically reduced productivity in 2008 is evident in this comparison with the high production year of 2003.

Investigators used MODIS 16-day composite NDVI time series and historical yield data to evaluate the 2007/08 wheat season in comparison to the previous 8 years. Seasonal maximum NDVI has been shown to correlate well with historical wheat yields. MODIS time series data for the period 2000 to 2008 were temporally smoothed to remove cloud and other atmospheric contamination, and then stratified by irrigated areas. The time series data were spatially averaged at the provincial level and then analyzed to derive the time of annual maximum, which was consistently found to be during late April – mid May. Annual maximum NDVI values were correlated with wheat yield statistics at both the national level and for an aggregation of northern provinces, which supply the majority of the country’s wheat production. Yield statistics were obtained from Crop and Food Supply Assessment Mission (CFSAM) results supported by the Food and Agriculture Organization (FAO) and the World Food Program from 2000 to 2007. Results showed good correlations for the national level (R2 = 0.92) and the northern provinces (R2 = 0.76). We used this relationship with annual maximum NDVI as a basis for ranking the 2007/08 winter wheat yield.

Results

Nationally, the 2007/08 maximum NDVI ranked as the worst in recent years. An estimated 1.14 tons per hectare was the expected yield for 2007/08 using the regression-based yield figures (Figure 3). This ranked as second worst yield on record and less than half the yield of the most productive year (2002/03). Since the majority of Afghanistan’s production comes from the northern irrigated provinces, these are of particular interest to the food security community. A ranking of provincial-level results for the north showed a situation similar to that at the national level. When considered in aggregate, 2007/08 ranked as the worst year on record for the northern provinces as well. The lack of productivity was graphically portrayed by comparing the difference in maximum NDVI between the 2007/08 season and the productive 2002/03 season (Figure 4). While this analysis focused on irrigated winter wheat, the findings strongly suggest failure of the rainfed wheat crop as well.

This method provided a very quantifiable procedure for assessing relative crop performance using the relationship of maximum NDVI and wheat yield statistics. The analysis provided timely information that, according to the USAID FEWS NET program manager, had an enormous impact on food security decisions being made for the country of Afghanistan.

In Afghanistan, snowcapped peaks loom above agricultural fields with a network of irrigation channels. Photo courtesy of Bob Bohannon.

In Afghanistan, snowcapped peaks loom above agricultural fields with a network of irrigation channels. Photo courtesy of Bob Bohannon.

Conclusion

In Zimbabwe and Afghanistan, years of political upheaval and intermittent drought have contributed to the prospects of widespread hunger. In Zimbabwe, during February of 2009, an estimated 7 million people faced serious food shortages, many surviving on just one meal per day. Zimbabwe’s once-thriving agricultural production had fallen significantly and changes in the agricultural system made it difficult to get good estimates of crop production. In Afghanistan, the 2008 spring snow pack appeared to be well below normal. This could mean a reduced wheat harvest due to inadequate water for irrigation, but crop production reports would not be available until many months later. In the meantime, many people could endure serious hardship.

Clear and early answers were needed by organizations poised to send famine-mitigating food aid. Remotely sensed satellite observations were able to provide non-political, objective and timely production estimates. In both cases, we were able to use historically observed relationships between NDVI and crop production/yield to develop MODIS-based crop production/yield estimates, well before conventional statistics were available. In Afghanistan, this meant that anecdotal reports of widespread crop failure could be substantiated. In Zimbabwe, remote sensing showed improved crop production over the previous year, with the number of food insecure people likely falling to a relatively low number, compared to recent history. In both cases, strategic decisions for food aid programs could be made in a timely fashion, helping to keep costs down and increase their effectiveness in staving off widespread hunger.