Carbon dioxide (CO₂) is a predominant greenhouse gas significantly contributing to atmospheric heat retention, primarily driven by anthropogenic activities intensifying the greenhouse effect. This study aims to evaluate the diversity of plant species, above-ground biomass (AGB), and carbon stock within a dry dipterocarp forest, which is a vital local natural resource. This study presents a comprehensive evaluation of plant species diversity, AGB, and carbon stock capacity within a dry dipterocarp forest at the Nature Study Center, Mahasarakham University, located in the Kham Riang Subdistrict of Kantharawichai District, Maha Sarakham Province, spanning an area of 20.80 hectares. Ten sample plots, each measuring 40 × 40 m, were established and distributed across the study area. The diameter at breast height (DBH) and the height of the trees were meticulously recorded for all trees within these plots. Advanced statistical techniques were employed to calculate the relative dominance (RD), relative frequency (RF), and Importance Value Index (IVI), alongside a comprehensive assessment of plant species diversity. The AGB was assessed using precise allometric equations, with a focus on analyzing carbon storage within woody biomass. The findings revealed the presence of 52 tree species across 26 families within the forest. The total AGB was measured at 144.510 tons, with carbon stock reaching 67.920 tCO₂. These results offer critical insights into enhancing land management strategies to optimize carbon stock, thereby playing a vital role in mitigating greenhouse gas emissions, a significant factor in climate change dynamics.

This study aimed to develop an accurate method for estimating aboveground biomass and carbon sequestration in rubber plantations by integrating multispectral imagery acquired from unmanned aerial vehicles (UAVs) with vegetation indices and machine learning algorithms. The research was conducted over approximately 266.56 ha of rubber plantations, where twelve 40 × 40 m sample plots were established using a stratified random sampling design for field data collection. Field measurements included the height, diameter, canopy dimensions, and geographic coordinates of each tree. The multispectral UAV images were processed to produce orthomosaic maps, canopy height models, digital terrain models|, and digital surface models. Five vegetation indices—NDVI, RVI, OSAVI, NDRE, and GNDVI—were derived and analyzed, and deep learning techniques were applied to estimate canopy size. Furthermore, allometric equations and multiple regression analyses were employed to estimate aboveground carbon sequestration. The field measurements indicated an average aboveground carbon sequestration of 85.69 tCO₂e/ha, while the estimates based on NDVI, RVI, OSAVI, NDRE, and GNDVI were 72.62 (11.84%), 69.69 (15.40%), 72.54 (11.94%), 72.19 (12.35%), and 71.33 tCO₂e/ha (13.40%), respectively. A statistical evaluation revealed that the multiple linear regression model integrating vegetation indices and field data exhibited a strong correlation, with NDVI and OSAVI providing the highest accuracy, as reflected in their lowest error percentages. Overall, the findings demonstrate that integrating UAV-based multispectral imagery with vegetation indices is a reliable and efficient approach for assessing and monitoring carbon sequestration in rubber plantations.

Oil palm plantations play an important role in agricultural production and carbon storage in tropical regions. The accurate estimation of aboveground carbon (AGC) is essential for sustainable plantation management, climate change mitigation, and carbon monitoring. Although field measurements provide reliable estimates, they are often time-consuming, labor-intensive, and costly, particularly over large plantation areas. Recent advances in remote sensing and machine learning offer efficient alternatives for AGC estimation using satellite imagery. In this study, we developed a machine learning framework for AGC estimation in oil palm plantations using Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar (SAR) data. Field measurements were integrated with spectral variables, vegetation indices, and SAR-derived parameters extracted from satellite data. A hybrid feature selection approach combining Pearson correlation, mutual information and mRMR was used to identify the most relevant variables. Six machine learning algorithms were evaluated, including Linear Regression, Random Forest, XGBoost, Gradient Boosting, LightGBM, and Extra Trees. Because the 160 observations comprise sixteen 10 m × 10 m grid cells nested within ten 40 m × 40 m field plots, model performance was assessed with leave-one-plot-out cross-validation: all sixteen cells of a plot were held out together, and the hybrid feature selection was repeated inside every fold using only that fold’s training plots. Performance was measured on pooled out-of-fold predictions using R², root mean squared error (RMSE), and average absolute relative error (AARE%). Under this spatially independent design the combined Sentinel-1 + Sentinel-2 dataset gave the highest accuracy (R² = 0.7950, RMSE = 4.14 t C ha⁻¹, AARE = 33.53%), followed by Sentinel-1 alone (R² = 0.7631, RMSE = 4.46 t C ha⁻¹) and Sentinel-2 alone (R² = 0.6108, RMSE = 5.71 t C ha⁻¹). Linear Regression and Extra Trees were the most robust models, whereas the boosted ensembles did not generalize to unseen plots. Repeating the evaluation with an ungrouped random split of the same data inflated R² by up to 0.70, showing that a large part of the accuracy obtainable under that design reflects within-plot spatial autocorrelation rather than predictive skill. These findings indicate that optical-SAR imagery combined with machine learning can provide useful AGC estimates in oil palm plantations, and that spatially independent validation is essential for reporting them honestly. The proposed framework can be used to support plantation-scale carbon mapping, monitoring, and carbon stock assessment, subject to further calibration and independent validation across additional plantations.

Two-stage PM2.5 models increasingly employ geographically weighted regression (GWR) to address spatially autocorrelated residuals. However, published hybrid approaches typically combine GWR with a single estimator and utilize random cross-validation (CV), leaving open which estimator benefits most and whether the gains extend to unmonitored locations. We investigated both questions across Thailand (2023–2025). A fixed daily GWR correction was applied to residuals from a linear mixed-effects model and three tuned tree ensembles, but only when Moran’s I indicated significant spatial structure. Each configuration was evaluated using random 10-fold and site-based CV with both observed and gap-filled MAIAC AOD. GWR improved all estimators across both CV schemes and AOD policies (Δ𝑅2 = +0.013 to +0.125; all folds; 𝑝 ≤ 4.8 × 10−4). Day-level RMSE gains rose with the strength of residual Moran’s I evidence in every family (𝑝 < 5 × 10−16): 0.5–1.8 μg m-3 on gated days, negligible otherwise; an always-on ablation priced the gate’s abstention at 0.2–0.4 μg m-3 RMSE. Random CV favored XGB–GWR (𝑅2 = 0.855); site-based CV found the tree hybrids statistically indistinguishable (𝑅2 = 0.752–0.753), and with gap-filled AOD a rank bootstrap placed GB–GWR first (60% versus 1.7%). No model demonstrated off-network skill during the wet monsoon (site-based 𝑅2 < 0). These results establish a transferable diagnostic principle: residual spatial structure should be tested before determining whether correction is warranted.