Abstract
The grassland ecosystem forms a critical part of the natural ecosystem, covering up to 15-26% of the Earth's land surface. Grassland significantly impacts the carbon cycle and climate regulation by storing carbon dioxide. The organic matter found in grassland biomass, which acts as a carbon source, greatly expands the carbon stock in terrestrial ecosystems. Correct estimation of above ground biomass (AGB) and its spatial and temporal changes is vital for determining the carbon cycle of the grassland. Datasets from multiple sources were fused to accomplish the objective of the study. The Sentinel-2 sensor band, vegetation index (NDVI), and Shuttle Radar Topography Mission (SRTM) DEM products were used as predictor variables, while Global Ecosystem Dynamics Investigations (GEDI) mean above-ground biomass density (AGBD) data was used to train the model. Random forest (RF) and gradient boosting were used to estimate the AGB of the grassland biome. We also identified the correlation between Sentinel-2-derived vegetation indices and ground-based measurements of leaf area index (LAI). The processing duration, parameter requirements, and human intervention are reduced with RF and gradient boosting algorithms. Due to its fundamental concept, ensemble algorithms effectively handled multi-modal data and automatically conducted spectral selection. The findings show variations in the study area's AGB concentration throughout five years. According to the results, gradient boosting models outperformed RF models in both years. RF achieved the highest R(2) value of 0.5755 Mg/ha, while gradient boosting achieved the highest R(2) value of 0.7298 Mg/ha. Sentinel-2-derived VI vs LAI results show that NDVI was the best-performing model with an R(2) value of 0.6396 m(2) m(-2) and an RMSE of 0.159893 m(2) m(-2), followed by OSAVI, NDRE, and MSAVI. This result shows that sensor data and field biophysical data can map the terrestrial ecosystem's biomass.