Search

The Himalayan region is a most fragile ecosystem globally. Trees make up around 90 % of the global biomass carbon pool and previous studies have shown that tree carbon balance cannot be easily assessed by conventional methods. Considering trees as a backbone of the forest ecosystem, present study assessed the heterogeneity in tree carbon density using field-inventoried data and NDVI-based modelling with Sentinel 2 A imagery on Google Earth Engine. The specific aim of the study was to assess the spatial distribution of tree carbon density in the Darjeeling Himalayas using Sentinel 2 A. The object-based classification of forest area using a random forest algorithm showed a high accuracy (Kappa coefficient value of 0.92, OOB error 0.17). The regression model using NDVI as a predictor of tree carbon demonstrated a good fit (R2 = 0.78) for predicting tree carbon density. Validation results show high accuracy of the regression model in predicting tree carbon density with a low RMSE of 9.39 Mg ha−1 (R2 = 0.80, % RMSE = 11.55 %). The classification of tree carbon density into five classes revealed that a significant proportion of the forest area (57.05 %) falls under moderate carbon density (50–75 Mg ha−1). In Darjeeling Himalayas, majority of forests are under the carbon density between 50 and 75 Mg ha−1. Improvement and conservation efforts must be directed for very low carbon density (01–25 Mg ha−1) areas covering 0.05 %, and high carbon density (75–100 Mg ha−1) covering 36.22 % of the forest area, respectively, to balance the overall carbon storage potential of the region. • Random forest effectively predicts forest area with high accuracy of 92 %. • Maximum tree density, DBH and carbon density was recorded between 1500 and 2000 m asl. • Forest area of 77059.62 ha (57.05 %) is under moderate carbon density. • Very high carbon density (100–150 Mg C ha−1) covered 1.10 % of total forest area. [ABSTRACT FROM AUTHOR]