Integration of Sentinel-2 Derived Spectral Indices and In-situ Forest Inventory to Predict Forest Biomass

• Published in: Pakistan journal of scientific and industrial research. Series A: physical sciences 2021-05, Vol.64 (2), p.119-130
• Main Authors: Imran, Areeba-Binte, Ahmed, Samia, Ahmed, Waqar, Zia-ur-Rehman, Muhammad, Iqbal, Arif, Ahmad, Naveed, Ullah, Irfan
• Format: Article
• Language: English
• Subjects: Analysis; Emissions (Pollution); Environmental monitoring; Environmental protection; Forest ecology; Forest management; Forest reserves; Sensors; Sustainable forestry
• ISSN: 2221-6413; 2223-2559
• DOI: 10.52763/PJSIR.PHYS.SCI.64.2.2021.119.130

  Forest biomass estimation is the central part of sustainable forest management to assess carbon stocks and carbon emissions from forest ecosystem. Sentinel-2 is state-of-art sensor with refined spatial and recurrent temporal resolution data. The present study explored the potential of Sentinel-2 derived vegetation indices for above ground biomass prediction using four regression models (linear, exponential, power and logarithmic). Sentinel-2 indices includes Global environmental monitoring index, transformed normalized difference vegetation index, normalized difference water index, normalized difference infrared index and red-edge normalized difference vegetation index. The performances of Sentinel-2 indices were assessed by simple single variable (index) based regression for GEMI, TNDVI, NDII, NDWI and RENDVI versus AGB values. Further, stepwise linear regression was also developed in which used all indices entered into stepwise selection and the best index was selected in the final model. Results showed that linear model of all indices performance best compared to the rest three models and R2 values 0.12, 0.39, 0.46, 0.44 and 0.37 for Global environmental monitoring index, transformed normalized. Vegetation index, normalized difference water index, infrared index and red-edge vegetation index, respectively. Normalized difference water index was considered the best index among five computed indices in simple linear as well as in stepwise linear regression, whereas rest of the indices were removed because they were not significant under the stepwise criteria. Further, the accuracy of normalized difference water index model was determined by root mean square error and final prediction model has 28.27 t/ha error for both simple linear and stepwise linear regression. Therefore, normalized difference water index was selected for biomass mapping and resultant biomass showed up to 339 t/ha in the study area. The resultant biomass map also showed consistency with global datasets which include global forest canopy height and global forest tree cover change maps. The study suggest that Sentinel-2 product has great potential to estimate above ground  biomass with accuracy and can be used for large scale mapping in combination with national forest inventory for carbon emission accounting.