Drought Depiction by Reanalysis Precipitation Datasets in Sub-Saharan Africa

Abstract:

Reanalysis precipitation is routinely used as a surrogate of observations due to its high spatial and temporal resolution and global coverage, and thus widely used in hydrologic and agricultural applications. The resultant product and analysis are largely dependent on the accuracy of reanalysis precipitation dataset. In this study, we aim to address the uncertainties in reanalysis rainfall products and associated impact on drought monitoring and drought risk estimation. We evaluated five reanalysis precipitation datasets (CFSR, NCEP/NCAR R1 and R2, 20CR and MERRA) against reference precipitation from a satellite-gauge-based dataset (Princeton Global Forcings, PGF) on their depiction of soil moisture drought during 1979 to 2012 estimated using the Variable Infiltration Capacity model. Results show that reanalysis precipitation products reasonably reproduce major drought episodes in sub-Saharan Africa. However, some discrepancies are found in the representation of the spatiotemporal characteristics of drought events. Drought events tend to be more spatially scattered in CFSR and temporally less persistent in MERRA. By evaluating the depiction of meteorological drought based on Standardized Precipitation Index (SPI), discrepancies are attributed to an overestimation in seasonal rainfall variability in the reanalysis products. The false overestimation further increases after ~1999 due to changes in the input (assimilation) observations. The analysis examined the impact of uncertainties in reanalysis precipitation on drought monitoring, suggesting the need for improved bias correction schemes for errors in both the mean (systematic) and variability from the reanalysis products, as well as concerns with the temporal evolution of observation systems used in data assimilation systems.