Machine learning technique for carbon sequestration estimation of mango orchards area using Sentinel-2 Data

This study evaluates the effectiveness of Sentinel-2 satellite imagery in assessing carbon sequestration in mango orchards using multiple vegetation indices (VIs). We established 49 quadrat sampling plots (40*40 m) across diverse mango orchards in Phitsanulok Province, Thailand to collect ground truth data for aboveground carbon storage estimation. Twelve vegetation indices were analyzed, including Normalized difference vegetation index (NDVI), Normalized difference red edge index (NDRE), Enhanced Vegetation Index (EVI), Green normalized difference vegetation index (GNDVI), Green Chlorophyll Index (Clgreen), Ratio Vegetation Index (RVI), Triangular vegetation index (TVI), Modified Triangular vegetation index (TVI-2), Angular Vegetation Index (AVI), Normalized Pigment Chlorophyll Ratio Index (NPCRI), and Modified Simple Ratio Index (MSRI) as well as Chlorophyll Index Red Edge (CIRE). These indices demonstrated significant correlations with field-measured biomass (R2 ranging from 0.04 to 0.80). The Random Forest (RF) ensemble model, optimized with 400 trees (ntree) and 20 variables at each split (mtry), integrated these indices to predict carbon storage. The model achieved exceptional accuracy (R2=0.97) with a Root Mean Square Error (RMSE) of 1.57 ton C ha-1 and Mean Absolute Error (MAE) of 1.05 ton C ha-1. Feature importance analysis revealed that f (NDVI), f (NDRE), f (TVI-2), and f (GNDVI) contributed 0.283, 0.160, 0.377, and 0.181, respectively, to the model's predictive power. Cross-validation using a 70:30 training–testing split confirmed the model's robustness. The developed model enables efficient monitoring of carbon sequestration of mango plantations, with the mean carbon sequestration calculated to be 40.6 ton C ha-1 (σ=42.19, n=49) and wide range of carbon sequestration values (4.13 to 218.6 ton C ha-1). This methodology provides a cost-effective, scalable approach for carbon seqesstration assessment in fruit tree plantations, supporting both sustainable agricultural practices and carbon credit initiatives in the agricultural sector.