Estimating Above-ground Biomass of Trees Outside Forests in the Thar Desert using Sentinel Data and Machine Learning

Abstract

Above-ground biomass estimation of Trees outside Forests is crucial as they play a significant role in carbon sequestration, biodiversity conservation, and microclimate regulation, especially in arid and semi-arid regions where tree cover is limited.  The heterogeneous vegetation covers and highly scattered nature of trees add to the challenges in the accurate estimation of aboveground biomass employing remote sensing technology. This study aimed to estimate the AGB of TOF in the arid regional landscape of the Thar desert of Rajasthan, integrating Sentinel-1 (S1) SAR and Sentinel-2 (S2) optical datasets and field observations, applying the Random Forest (RF) model. The field calculated AGB in the sampled area ranged from a minimum of 0.19 t/ha to a maximum of 43.12 t/ha, with a mean of 8.03 t/ha. The backscattering coefficients at VV and VH polarizations and 5 SAR indices from S1 and the multispectral bands, vegetation indices, and biophysical variables from S2 were extracted as the predictor variables for the AGB model. After correlation and multi-collinearity analysis, three models were developed: the first model based on S1(M_S1), the second model with S2 (M_S2), and the third model is a combined model of S1 and S2 variables (M_S1S2). The correlation analysis revealed that the SAR indices have a higher relationship with field biomass. Further, the combined model (M_S1S2) achieved the highest accuracy (R² = 0.52, RMSE = 3.89 t/ha) in AGB estimation, followed by M_S1 (R² = 0.46) and M_S2 (R² = 0.43). The results of the study highlight the utility of Sentinel datasets and larger ecological plots at the landscape level in biomass mapping in sparsely vegetated arid environments. Moreover, the study highlights the ecological importance of TOF and emphasizes the need for biomass and carbon stock assessments in the ecologically sensitive arid regions.