Discrimination of Palatable Pastures Using Multispectral SENTINEL-2 Imagery in a Semi-Arid Savannah Grassland, Kenya

Semi-arid savannah grasslands have significant social-ecological benefits and have been experiencing dramatic changes due to natural and man-made activities. In Kenya, this biome covers more than half of the landscape, supports wildlife, livestocks animals and millions of pastoralist livelihoods. The characterization, identification and detailed estimation of palatable savannah grass are critical for rangeland management and pastoral sustainability. This study aimed at assessing the fractional cover of palatable savannah grass by using Multiple Endmember Spectral Mixture Analysis(MESMA) applied to multispectral Sentinel-2 imagery in Kenyas heterogenous Semi-arid savannah grassland. Interactive Endmember Selection(IES), Endmember Average Root Mean Square Error(EAR) and Minimum Average Spectral Angle(MASA) were assessed to create the endmembers library of target palatable community species cover and non-palatable covers. The MESMA fractional covers accuracy performance was assessed using the coefficient matrix. Our results show the IES endmember selection technique performed better (86%) as compared to other technics (80% & 84%) for MASA and EAR, respectively, by having a high correlation to validation data. Overall, our results demonstrate the potential and ability of Sentinel-2 spectral bands to characterize heterogeneous semi-arid savannah grasslands at the sub-pixel level, which can support conservation and rangeland management policies, strengthen the productivity of the rangeland ecosystem and ensure the sustainability of both pastoral livelihoods and wildlife management.