Assessment of sentinel-2 vegetation indices for plot level tree AGB estimation

Abstract

The application of remote sensing & GIS technologies enables the assessment, monitoring and management of forestry resources at local, regional and global scale. The availability of freely available optical satellite data can be used for the cost-effective, efficient and timely understanding of the phenomenon related to the forests. One of the recent additions to the freely available optical satellite data is sentinel-2. Compared to its counterpart Landsat, it does have the improved spatial, temporal resolution and band positioning specifically tailored to the study of vegetation. This gives the possibility of improved performance which demands its potential to be assessed in the local context. The study evaluates the application of sentinel-2 for forest above ground biomass estimation in the context of Margallah hills national park, Pakistan. 33 sample plots of cheer pine, 22 paper mulberry and 12 other deciduous trees were surveyed through field campaigns of two weeks. The field data was converted to the plot level biomass as per the standard methods available in the literature. The imagery data was pre-processed and atmospherically corrected. Seven different indices as reported in the literature were calculated from the sentinel 2 level 2a data. The correlation for each tree species, and combined species was developed using multiple. Linear regression with all the calculated indices and regression analysis with the individual indices was computed. It was found that the variations in the spectral responses were poorly correlated with the field biomass values of the surveyed plots. The highest value of r2 for cheer pine was found to be 0.29, paper mulberry 0.31 and for all tree species 0.22. It was concluded that although the sentinel-2 has improved spatial and spectral resolution compared to Landsat, still it does not guarantee any promising model for the prediction of plot level biomass.