Remote sensing of tree/grass fractional cover using phenological signal decomposition of MODIS time series data

Reliable assessments of tree/grass fractional cover in savanna using remote sensing are challenging due to the heterogeneous mixture of the two plant functional types (PFTs) and soil backgrounds. This thesis reduces this knowledge gap in the remote sensing of tree/grass fractional cover. Tree/grass dynamics in heterogeneous savanna ecosystems are assessed using time-series decomposition of MODIS data acquired from 2002 to 2015. The decomposition method follows a harmonic analysis and tests the harmonic terms for significance. Several scales of spatial and temporal variability are considered for these PFTs (for each field plot against 14 years dataset as well as for the whole study area). In most harmonic cycles, the tree greening-up period started earlier than grasses. While changes in tree cover are more gradual, grasses have high variability over time. The phase (R² = 0.60, slope = 1, RMSE = 12.52%), cycles (R² = 0.44, slope = 1.2, RMSE = 17.64%) and amplitude (R² = 0.36, slope = 0.83, RMSE = 16.28%) of the strongest harmonic terms show good estimate of tree cover. The estimates of tree cover from the simple linear regression of field data and dry season NDVIpixel/SAVIpixel images had good performance. The tree cover estimated using soil determining methods had an improved slope for NDVI and SAVI but yield slightly a high RMSE. A comparison of tree cover using Pearson’s correlation indicated strong agreement with LiDAR/SAR and Bucini woody cover maps. The errors, uncertainties and the challenges in discriminating and estimating trees and grasses using signal decomposition methods are discussed. Tree cover maps will be helpful for vegetation monitoring, climate change impact assessment and vegetation model validation. Finally, the techniques employed for the assessment of tree-grass mixtures in this study would be useful for earth observation especially where end-members of the woody-herbaceous continuum are being considered.