Here we seek to answer this question in a region that has as yet been ill-served by such analyses due to poor parameterizations of E0 and the limited availability of data. Previously, drought and famine early warning monitors have here either relied on physically poor process representations of E0 or on climatological mean estimates. However, by exploiting the advent of long-term, spatially distributed, and accurate reanalyses of the land-atmosphere system and its drivers we can put new data to use to save livelihoods and lives by improving drought monitoring in this data-sparse region.
In this study, we use a new E0 dataset to decompose anomalies of E0 during periods of drought into contributions from all of its input variables—temperature, humidity shortwave radiation, and windspeed. We present the (i) general methodology for both the development of E0 and its decomposition and (ii) the results of the decomposition of E0 anomalies during drought periods across the continent of Africa. For our fully physical metric of E0, we developed a nearly 42-year long, daily, global dataset of Penman-Monteith reference evapotranspiration. This new E0 dataset is driven by the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2)—an accurate, fine-resolution land-surface/atmosphere reanalysis. We verified the accuracy of the dataset against (i) point-estimates of E0 derived by Southern African Science Service Centre for Climate Change and Adaptive Land Management (SASSCAL) initiative in Southern Africa, a region with sparse ground-truth data and significant humanitarian need, and (ii) on a spatially distributed basis against E0 derived from other reanalyses (Global Data Assimilation System and Princeton Global Forcing) that, although global, are otherwise unsuitable for operational food-security decision-making.
We determine drought periods using spatially distributed drought-monitoring tools, such as the Normalized Difference Vegetation Index (NDVI), the Evaporative Demand Drought Index (EDDI), and the Standardized Precipitation Index (SPI). Within these drought periods, we conduct a first-order analysis of the anomalies in E0. This technique assumes that the contributions from anomalies in all drivers sum to the anomaly in E0; each driver’s contribution is the product of the sensitivity of E0 to, and the anomaly in, the driver. As our expression for E0 (i.e., Penman-Monteith reference evapotranspiration) is differentiable, the sensitivity to each driver can be derived explicitly by partial differentiation. Drivers’ anomalies are observed by querying the MERRA-2 reanalysis during drought periods and deriving deviations from the drivers’ long-term means for the same periods across the entire reanalysis period.

