Estimation of forest above-ground biomass (AGB) using multi-source remote sensing data is an important method to improve the accuracy of the estimate. However, selecting remote sensing factors that can effectively improve the accuracy of forest AGB estimation from a large amount of data is a challenge when the sample size is small. In this regard, the Least Absolute Shrinkage and Selection Operator (Lasso) has advantages for extensive redundant variables but still has some drawbacks. To address this, the study introduces two Least Absolute Shrinkage and Selection Operator Lasso-based variable selection methods: Least Absolute Shrinkage and Selection Operator Genetic Algorithm (Lasso-GA) and Variance Inflation Factor Least Absolute Shrinkage and Selection Operator (VIF-Lasso). Sentinel 2, Sentinel 1, Landsat 8 OLI, ALOS-2 PALSAR-2, Light Detection and Ranging, and Digital Elevation Model (DEM) data were used in this study. In order to explore the variable selection capabilities of Lasso-GA and VIF-Lasso for remote sensing estimation of forest AGB. It compares Lasso-GA and VIF-Lasso with Boruta, Random Forest Importance Selection, Pearson Correlation, and Lasso for selecting remote sensing factors. Additionally, it employs eight machine learning models—Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Bayesian Regression Neural Network (BRNN), Elastic Net (EN), K-Nearest Neighbors (KNN), Extremely Randomized Trees (ETR), and Stochastic Gradient Boosting (SGBoost)—to estimate forest AGB in Wuyi Village, Zhenyuan County. The results showed that the optimized Lasso variable selection could improve the accuracy of forest biomass estimation. The VIF-Lasso method results in a BRNN model with an R² of 0.75 and an RMSE of 16.48 Mg/ha. The Lasso-GA method results in an ETR model with an R² of 0.73 and an RMSE of 16.70 Mg/ha. Compared to the optimal SGBoost model with the Lasso variable selection method (R² of 0.69, RMSE of 18.63 Mg/ha), the VIF-Lasso method improves R² by 0.06 and reduces RMSE by 2.15 Mg/ha, while the Lasso-GA method improves R² by 0.04 and reduces RMSE by 1.93 Mg/ha. From another perspective, they also demonstrated that the RX sample count and sensitivity provided by LiDAR, as well as the Horizontal Transmit, Vertical Receive provided by Microwave Radar, along with the feature variables (Mean, Contrast, and Correlation) calculated from the Green, Red, and NIR bands of optical remote sensing in 7 × 7 and 5 × 5 windows, play an important role in forest AGB estimation. Therefore, the optimized Lasso variable selection method shows strong potential for forest AGB estimation using multi-source remote sensing data.

Integrating multi-source remote sensing can improve the accuracy of forest aboveground biomass (AGB) estimation. However, the accuracy and stability of the forest AGB estimation results are affected by multiple remote sensing feature variables as well as parameter tuning of machine learning algorithms. To this end, this study employed six types of remote sensing data—Landsat 8 OLI, Sentinel-2A, GEDI, ICESat-2, ALOS-2, and SAOCOM. A dual-variable selection strategy based on SHapley Additive exPlanations (SHAP) was developed, and a genetic algorithm (GA) was used to optimize the parameters of five machine learning models—elastic net (EN), least absolute shrinkage and selection operator (Lasso), support vector regression (SVR), Random Forest (RF), and Categorical Boosting (CatBoost)—to estimate the AGB of Pinus kesiya var. langbianensis forest in Wuyi Village, Zhenyuan County. The dual-variable selection strategy integrates SHAP with the Pearson correlation coefficient (PC), RF, EN, and Lasso to enhance feature screening robustness and interpretability. The results of the study showed that Lasso-SHAP dual-variate screening was more stable than SHAP univariate screening. In particular, the Lasso-SHAP strategy improved the average R² from 0.59 (using SHAP alone) to above 0.70, achieving an enhancement of 11%. Among GA-optimized parametric machine learning models, the linear GA-Lasso achieved the best performance, with an R² of 0.91 and an RMSE of 12.94 Mg/ha, followed by the GA-EN model (R² = 0.89, RMSE = 14.46 Mg/ha). For nonlinear models, GA-SVR performed the best (R² = 0.74, RMSE = 22.07 Mg/ha), surpassing the GA-CatBoost model (R² = 0.64, RMSE = 25.88 Mg/ha). In summary, the Lasso-SHAP dual-variable selection strategy effectively improves the estimation accuracy of AGB for Pinus kesiya var. langbianensis forests, while GA-optimized machine learning models demonstrate excellent performance, providing strong support for regional-scale forest resource monitoring and carbon stock assessment.

The optical saturation problem is one of the main factors causing uncertainty in aboveground biomass (AGB) estimations using optical remote sensing data. It is critical for the improvement in AGB estimation accuracy to clarify the relationships between environmental factors and the variations in optical saturation values (OSVs). In this study, we obtained the OSVs for 20 districts and clarified the individual, interactive, and comprehensive effects of climate, soil, and topographical factors on the OSV variations. The results are as follows: (1) the range of the OSVs was from 104 t/hm² to 182 t/hm² for the 20 districts; (2) the soil factor had the greatest (−0.635) influence on the OSVs compared to climate and topography; (3) the highest interaction effect (−0.833) was between climate and soil; (4) the comprehensive effect of the three environmental factors on the OSVs was obvious, and the correlation coefficient was 0.436. Moreover, the mean temperature of the coldest quarter (MCQMean) had the highest effect on the OSVs. The results indicate that environmental factors significantly affect the variation in OSVs through their individual, interactive, and comprehensive effects. Our findings provide a valuable reference for reducing the uncertainty caused by spectral saturation in AGB estimations using optical remote sensing data.