Increasing numbers of explanatory variables tend to result in information redundancy and “dimensional disaster” in the quantitative remote sensing of forest aboveground biomass (AGB). Feature selection of model factors is an effective method for improving the accuracy of AGB estimates. Machine learning algorithms are also widely used in AGB estimation, although little research has addressed the use of the categorical boosting algorithm (CatBoost) for AGB estimation. Both feature selection and regression for AGB estimation models are typically performed with the same machine learning algorithm, but there is no evidence to suggest that this is the best method. Therefore, the present study focuses on evaluating the performance of the CatBoost algorithm for AGB estimation and comparing the performance of different combinations of feature selection methods and machine learning algorithms. AGB estimation models of four forest types were developed based on Landsat OLI data using three feature selection methods (recursive feature elimination (RFE), variable selection using random forests (VSURF), and least absolute shrinkage and selection operator (LASSO)) and three machine learning algorithms (random forest regression (RFR), extreme gradient boosting (XGBoost), and categorical boosting (CatBoost)). Feature selection had a significant influence on AGB estimation. RFE preserved the most informative features for AGB estimation and was superior to VSURF and LASSO. In addition, CatBoost improved the accuracy of the AGB estimation models compared with RFR and XGBoost. AGB estimation models using RFE for feature selection and CatBoost as the regression algorithm achieved the highest accuracy, with root mean square errors (RMSEs) of 26.54 Mg/ha for coniferous forest, 24.67 Mg/ha for broad-leaved forest, 22.62 Mg/ha for mixed forests, and 25.77 Mg/ha for all forests. The combination of RFE and CatBoost had better performance than the VSURF–RFR combination in which random forests were used for both feature selection and regression, indicating that feature selection and regression performed by a single machine learning algorithm may not always ensure optimal AGB estimation. It is promising to extending the application of new machine learning algorithms and feature selection methods to improve the accuracy of AGB estimates.

Parameter sensitivity analysis can determine the influence of the input parameters on the model output. Identification and calibration of critical parameters are the crucial points of the process model optimization. Based on the Extended Fourier Amplitude Sensitivity Test (EFAST) and the Morris method, this paper analyzes and compares the parameter sensitivity of the annual mean net primary productivity (NPP) of Larix olgensis Henry forests in Jilin Province simulated by the Lund–Potsdam–Jena dynamic global vegetation model (LPJ model) in 2009–2014 and 2000–2019, and deeply examines the sensitivity and influence of the two methods to each parameter and their respective influence on the model’s output. Moreover, it optimizes some selected parameters and re-simulates the NPP of Larix olgensis forests in Jilin Province from 2010 to 2019. The conclusions are the following: (1) For the LPJ model, the sensitive and non-influential parameters could be identified, which could guide the optimization order of the model and was valuable for model area applications. (2) The results of the two methods were similar but not identical. The sensitivity parameters were significantly correlated (p < 0.05); parameter k_rp was the most sensitive parameter, followed by parameters α_m, α_a and g_m. These sensitive parameters were mainly found in the photosynthesis, water balance, and allometric growth modules. (3) The EFAST method had a higher precision than the Morris method, which could calculate quantitatively the contribution rate of each parameter to the variances of the model results; however, the Morris method involved fewer model running times and higher efficiency. (4) The mean relative error (MRE) and mean absolute error (MAE) of the simulated value of LPJ model after parameter optimization decreases. The optimized annual mean value of NPP from 2010 to 2019 was 580 g C m⁻² a⁻¹, with a mean annual growth rate of 2.13%, exhibiting a fluctuating growth trend. The MAE of the simulated value of LPJ model after parameter optimization decreases.

Understanding how biotic and environmental drivers jointly shape forest carbon dynamics over time is essential for climate-adapted management of subtropical forests. We investigated the long-term interactions between biotic factors, environmental factors, and forest carbon dynamics in the subtropical forests of Jiangxi Province, China, over the period 1989–2019. The High Accuracy Surface Modelling (HASM) multi-source data fusion method integrates ground observation points with area-wide data from remote sensing and existing datasets to simulate the spatial distribution of forest carbon density across the entire study area. In Zixi, forest carbon density increased most rapidly between 1989 and 2009, after which the rate slowed as forest stands matured. Structural Equation Modelling (SEM) disentangled direct and indirect effects of drivers, and identified species richness and community-weighted functional traits as key positive drivers of aboveground carbon density. The influence of environmental factors reversed over the study period. Under ongoing global warming, the combined effects of altitude, temperature, and precipitation shifted from suppressing to reinforcing carbon accumulation in later years, increasingly operating through pathways mediated by functional traits. These findings enhance our understanding of carbon dynamics in subtropical forests and underline the importance of preserving species richness, especially in subtropical mountain forest. This study provides valuable insights for adaptive forest management and climate change mitigation strategies, aiming to improve ecosystem resilience and sustain carbon sequestration efforts in the face of ongoing global warming.