The United States Agency for International Development reports that the seven largest recipients of humanitarian aid are in sub-Saharan Africa. Direct food aid, primarily to subsistence farmers who depend on rain to grow their crops, remains the largest contribution. The current drought and famine in Somalia reaffirms the sensitivity of agricultural production to climate extremes in sub-Saharan Africa and the need to develop tools that improve agricultural drought monitoring for mitigation. Remote sensing and surface reanalysis data can facilitate efficient and cost effective approaches to monitoring agricultural droughts. In this paper, we expand on previous work that showed using remote sensing-based vegetation indices and surface reanalysis data that the Standardized Precipitation Index (SPI) correlated better with crop yield in arid areas with low agricultural production, while a newly developed Evaporative Stress Index (ESI) was superior in more humid areas with high agricultural production. Here, we combine the two indices, identify a distribution that effectively standardizes the index, and evaluate it at various timesteps using district-level crop yield data from Kenya and Malawi and country level statistics from the United Nations Food and Agriculture Organization (FAO).

The study uses a suite of ground-based observations, satellite data, and surface reanalysis data to derive the two indices. SPI was fit to a gamma distribution using the Collaborative Historical African Rainfall Model (CHARM). CHARM is a synthesis of climate reanalysis data, digital elevation, and precipitation from interpolated station data. The ESI is standardized by Potential Evapotranspiration (PET). PET is derived from the Priestley-Taylor Equilibrium equation, while ET is developed through a series of temperature and moisture constraints defined in the Berkeley Model. These constraints are derived from MODIS Normalized Difference and Enhanced Vegetation Indices and the Global Land Data Assimilation System (GLDAS) surface reanalysis dataset. The indices are combined additively and fit to Gamma and Weibull distributions for comparison using the Kolmogorov-Smirnov test. The results of this analysis indicate that the gamma distribution represents the combined index well over most of sub-Saharan Africa. The gamma fit of the new index was averaged over various timesteps (3 month, 6 month, 12 month, 18 month, and 24 month) and to the district level. It was compared using Pearson-rank correlations to annual crop yield data from Kenya and Malawi. Similarly, the index was aggregated to the national level and evaluated using annual crop yield data from FAO. A land use map and other ancillary data was used to determine what factors contribute to the success of each index and necessitate a combined index that can be applied regionally. The results of these analyses are currently being evaluated and will be discussed during this presentation.