Fractional vegetation cover estimation in southern African rangelands using spectral mixture analysis and Google Earth Engine

•Grasslands require continuous and dynamic condition data for sustainable management.•Fractional cover as indication of grass cover (%) for degradation monitoring.•Spectral unmixing for fractional cover estimation is accurate and transferable.•Endmember class subdivision maximises separability.•SWIR2, NDVI, EVI, MSAVI2 and DBSI is the optimal feature combination for Sentinel-2. Grasslands are under continuous threat of conversion and subsequent degradation, which has a detrimental impact on grassland productivity and grazing capacity, affecting the livestock industry. Fractional vegetation cover as indicator of grassland condition and productivity has been extensively researched, however, existing approaches and products are limited with respect to accessibility, affordability, applicability, and transferability. This study evaluated the use of publicly available satellite imagery, spectral mixture analysis and cloud geoprocessing technologies for dynamic, continuous, and accurate estimation of FVC for sustainable management. A linear spectral mixture model was developed, calibrated, and implemented in Google Earth Engine using Sentinel-2 and Landsat 8 imagery. Model accuracy and spatial and temporal transferability were evaluated using existing benchmark products and field data. It was found that Sentinel-2 performed the best using a feature combination of the SWIR2 band and the NDVI, EVI, MSAVI2 and DBSI indices. Accuracies were further improved by dividing the woody and bare endmembers into subclasses. The approach proved both spatially and temporally transferable, thus this research provides a robust approach to FVC estimation using limited field data and open source remote sensing imagery. The combination of this research with further grassland productivity modelling could prove valuable for sustainable environmental and economical rangeland planning and management.