Understanding uncertainty of model-reanalysis soil moisture within Greater Horn of Africa (1982–2014)

Abstract

Lack of in situ soil moisture over Greater Horn of Africa (GHA) has inhibited evaluation of uncertainty and suitability of land surface model (LSM) and/or satellite-derived soil moisture over the region. This has led to a low utilization of LSM-derived soil moisture in agricultural drought-related studies as reflected by the lower number of soil moisture based studies despite the region being dependent on rain-fed subsistence agriculture. This study evaluated the uncertainty of soil moisture products from the following LSM; MERRA-2, ERA5, GLDAS (NOAH, CLM, and VIC), FLDAS NOAH, and CPC. Generalized three-cornered hat analysis of this suite of soil moisture products showed high magnitudes of uncertainties associated with areas of high rainfall while low magnitudes of uncertainties were associated with low rainfall areas. Further, areas of Ethiopian highlands with complex topographical characteristics had almost twice the magnitude of uncertainties compared to the relatively flat region of East Africa (Kenya, Uganda, and Tanzania). MERRA2 and ERA5 had consistently lower magnitudes of uncertainty while CPC had consistently higher magnitudes of uncertainty. The uncertainties are attributed to quality of the forcing precipitation, difference in model physical dynamics, and topographical changes. From the Taylor diagram analyses, the magnitude of spatial variability (amplitudes) contributed higher uncertainty than spatial patterns (phase). Finally, interannual variability analyses showed higher uncertainties across the considered soil moisture products during wet season than dry season over Ethiopia highlands while Sudan, south Sudan, East Africa, Eastern Ethiopia and Somalia had the reverse. The results of this study are important as they contribute to the knowledge of uncertainty levels and suitability of the major LSM-derived soil moisture over GHA.