Intensive human activities have fundamentally altered the spatial configuration of terrestrial ecosystems. These changes disrupt carbon cycles and diminish terrestrial carbon sinks, thereby exacerbating climate change and challenging China's dual carbon goals. Understanding the impact of ecosystem spatial configuration on aboveground carbon (AGC) storage and identifying key thresholds is essential. This knowledge is crucial for enhancing carbon sinks and implementing Nature-based Solutions (NbS) effectively. This study utilizes multi-source time-series data from 1990 to 2023 and integrates landscape pattern analysis with machine learning algorithms. We investigate spatiotemporal changes in the spatial configuration of China's croplands, forests, and grasslands, and quantify their relationships with AGC at different scales. The findings indicate: (1) China's croplands, forests, and grasslands underwent a cumulative conversion area of 325 million hectares during the study period; (2) All three ecosystems showed decreased fragmentation and increased aggregation, with a continuous rise in patch shape complexity in croplands and forests; (3) Spatial configuration had significant nonlinear relationships and threshold effects with AGC, and explanatory power varied greatly across ecosystems: forests showed the most substantial effect (R2 = 0.75), grasslands less (R2 = 0.61), and croplands the weakest (R2 = 0.43); (4) Elevation and aggregation index were the dominant factors influencing AGC across most river basins in China. By revealing ecosystem-specific response patterns and thresholds, the study offers quantitative guidance for optimizing spatial configurations in diverse regions. It also provides essential data support and a theoretical basis for ecological engineering planning and NbS implementations.

Gross Primary Productivity (GPP), a critical metric quantifying the total carbon dioxide assimilated by vegetation through photosynthesis, plays a pivotal role in terrestrial ecosystem carbon cycle studies. However, accurately estimating GPP at large scales remains subject to significant uncertainties. This study evaluates four widely used remote sensing-based GPP products (rEC-LUE, MODIS, VPM, GOSIF) across China using eddy covariance data from 66 flux towers. Methodologies include Getis-Ord Gi* hotspot analysis, Sen's slope estimation, Reduced Major Axis (RMA) regression, and partial correlation analysis to assess their spatiotemporal consistency and climatic response patterns. The results indicated that: (1) At the national scale, VPM exhibited the best performance (R-2 = 0.74). Ecosystem-level evaluations revealed that VPM achieved the highest accuracy for grassland (R-2 = 0.78) and cropland (R-2 = 0.87), while GOSIF performed best for forest (R-2 = 0.78). All four products performed well for the wetland (R-2 > 0.72). (2) At the site scale, GOSIF showed better agreement with eddy covariance data for most forest and grassland sites, whereas VPM excelled for cropland sites. All products exhibited limited capability in reproducing the interannual variability of site-level GPP. (3) VPM and GOSIF maintained high spatiotemporal consistency across diverse scales and hydrothermal conditions. (4) All products consistently identified precipitation as the dominant driver of GPP variations in northeastern China and the northern Tibetan Plateau. This study can enhance our understanding of vegetation carbon sequestration dynamics in China and provide theoretical support for the development of environmental policies. (c) 2026 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

As a fast-growing and multifunctional crop, bamboo plays a pivotal role in food security and climate change mitigation by leveraging its high carbon sequestration potential. Monitoring aboveground carbon (AGC) stock in bamboo forests is crucial for guiding field management, growth observation, and yield prediction. Unmanned aerial vehicle (UAV)-based point cloud sensors offer a rapid and scalable solution for measuring bamboo AGC. This study evaluates the potential of UAV-LiDAR and machine learning (ML) for organ-level AGC estimation in bamboo forests. From LiDAR point clouds, we extracted structural features—including height, density, canopy, and intensity metrics—aggregated by mean plot-level metric (Mean-PM) and maximum plot-level metric (Max-PM) values at a 1 m2 grid scale. Key predictors were selected using ML-based recursive feature elimination (ML-RFE) to develop organ-specific AGC inversion models. Results showed that organ-specific carbon content and allometric equations effectively eliminated biases associated with a uniform coefficient. Max-PM features outperformed Mean-PM features in stem and leaf AGCs, with the XGBoost and Random Forest models achieving the highest accuracy (R2 = 0.82 for stems, 0.73 for leaves). Height percentiles and canopy structural metrics emerged as dominant predictors. This UAV-LiDAR-ML framework provides a cost-effective solution for precise bamboo carbon estimation, offering critical insights for carbon neutrality management and informed decision-making in bamboo forest ecosystems.

Ceracris kiangsu Tsai (C.kiangsu) is one of the main leaf-feeding pests in Moso bamboo forests. An in-depth exploration of its response mechanism is crucial for achieving large-scale, precise detection and maintaining the healthy development of Moso bamboo forests. However, existing research on C.kiangsu pest detection is relatively limited, with traditional forest pest models often constrained by unbalanced samples for generalization. This study integrates field survey data with Sentinel-2 MSI data to explore the remote sensing response mechanism of the pest; proposes a collaborative detection method for C.kiangsu infestations in Moso bamboo forests that combines the SMOTE algorithm with a multi-model ensemble, optimizes model parameters via the Bayesian algorithm, and simultaneously uses SHAP values to deeply analyze model interpretability and excavate pest detection indicators. The results showed that: 1) Leaf LCC, LWC, and LDMC exhibit excellent responsiveness to C.kiangsu pest infestations (p < 0.01), with vegetation indices outperforming moisture indices and texture features. 2) The optimal Stacking ensemble learning hybrid model achieves OA of 82.22% and Kappa of 0.7625, improving by 3.5% and 0.0468 over the best single LightGBM model. 3) SHAP analysis reveals that the vegetation index RVI is the core indicator for pest detection, ranking among the top three in feature importance in each base model. The contribution to the meta-model is LightGBM > ET > SVM > XGBoost. This study successfully breaks through the accuracy bottleneck of traditional single models, providing a solid scientific basis for Moso bamboo pest management and practical significance for its sustainable development.

Lymantria zylina infestation poses an ecological threat to Casuarina equisetifolia, a key tree species in coastal shelterbelts. Chlorophyll is a vital indicator of vegetation growth status and health. Accurate estimation of chlorophyll content in C. equiseti folia leaves is therefore crucial for studying its response to pest stress and enabling quantitative diagnosis. This study was conducted in the Pingtan Island area of Fuzhou City. We measured the leaf relative chlorophyll content, chlorophyll content, and hyperspectral data of C. equiseti folia under different pest infestation levels (health, mild hazard, moderate hazard, and severe hazard). A two-stage feature selection strategy combining Competitive Adaptive Reweighted Sampling (CARS), Random Forest (RF), and Recursive Feature Elimination (RFE) was employed to identify sensitive spectral features. Hyperspectral inversion models for relative chlorophyll content and chlorophyll content were then developed based on RF, Support Vector Regression (SVR), and eXtreme Gradient Boosting (XGBoost) algorithms. The results indicate that: (1) First-order Derivative Reflectance (FDR) and vegetation indices like the Modified Chlorophyll Absorption Ratio Index (MCARI) were key selected features for estimating relative chlorophyll content, In contrast, estimating chlorophyll content required a combination of selected features including red-edge parameters (Kar), derivative spectra, and the Photochemical Reflectance Index (PRI). (2) For relative chlorophyll content estimation, the RF model demonstrated superior overall performance compared to the other two models and exhibited stable predictive accuracy across different hazard levels (R <^> 2 > 0.7) RMSE <8.9, RPD>1.4). Furthermore, the estimation performance of all three models generally decreased with increasing pest stress levels. (3) For chlorophyll content estimation, the RF model also showed the strongest predictive robustness (For leaves in health, mild hazard, and severe hazard states, R <^> 2 > 0.69 , RMSE<0.21, RPD>1.4). The predictive performance and stability of the models. progressively weakened as the pest hazard level increased, although the performance for severe hazard leaves did not show a pronounced further decline. The hyperspectral detection models for C. equiseti folia chlorophyll content developed in this study enable precise quantification and carly diagnosis of pest hazard levels, providing key technical support for dynamic monitoring and targeted control of pests in coastal shelterbelts,

Leaf area index (LAI) and chlorophyll content are crucial variables in photosynthesis, respiration, and transpiration, playing a vital role in monitoring vegetation stress, estimating productivity, and evaluating carbon cycling processes. Currently, physical models are widely adopted for estimating LAI and canopy chlorophyll content (CCC). However, the main challenges of physical model-based methods for estimating LAI and CCC are the high computational cost and the fact that different combinations of canopy variables result in similar spectral reflectance for local minima. To address this limitation, a hybrid model was proposed to invert the LAI and CCC in Moso bamboo (Phyllostachys pubescens) forests. This approach utilized the PROSAIL canopy radiation transfer model, established look-up table (LUT) for LAI and CCC, and employed the Stacking ensemble learning framework. Compared with the PROSAIL LUT method, the hybrid model demonstrated higher performance in predicting LAI and CCC by incorporating the strengths of different models within the hybrid framework. The R2 values between predicted and measured values were improved by 3.28% and 7.15%, while the RMSE values were reduced by 19.71% and 16.14%, respectively. Moreover, the hybrid model based on Stacking ensemble learning achieved an 86% reduction in running time. Therefore, the hybrid model, which integrates the PROSAIL model with the Stacking ensemble learning framework, offers a more efficient and accurate approach for remotely estimating the LAI and CCC in Moso bamboo forests. The high efficiency of this method makes it promising and suitable for application to other types of vegetation.

Accurate and prompt monitoring of brown planthopper (BPH) infestation is crucial for rice production stability. The unique advantages of remote sensing in mapping the location and severity of pest damage are widely acknowledged. However, the crypticity of BPH early damage complicates the identification of infested areas. This study aims to detect BPH early infestation in paddy fields using an unmanned aerial vehicle (UAV) hyperspectral imaging system. Two data acquisition campaigns were conducted during the BPH early infestation stage. Considering the dynamic spatial distribution of BPH, the pest population density records were averaged to indicate infestation severity during the investigation period. Three novel indices were designed to detect the BPH early damage. Specifically, the Dual-temporal Stressed Canopy Spectral Relative Difference Index (DSRI) and the Dual-temporal Stressed Canopy Spectral Direct Difference Index (DSDI) were proposed based on the dual-temporal spectral changes of rice canopy. Furthermore, an opposite trend of DSDI in the short-wavelength (399-750 nm) and long-wavelength (750-1006 nm) spectral regions was observed for samples with varying BPH severity. Thus, the DSDI-SL was further proposed. The optimal feature combination of DSRIs, DSDIs and DSDI-SLs was selected using Lasso regularization and recursive feature elimination (RFE). An XGBoost classifier was applied to establish the BPH early detection model, which achieved an overall accuracy (OA) of over 85%, outperforming the model established by mono-temporal collected data. In the context of global climate change and escalating challenges to food security, our research introduces a novel framework for the efficient detection and quantitative description of early-stage BPH damage.

BACKGROUND:Moso bamboo (Phyllostachys edulis) plays a pivotal role in the global carbon cycle because of its rapid growth and significant ecological benefits. Accurate estimation of its aboveground biomass (AGB) is therefore essential for effective carbon management. However, the influence of its primary leaf-feeding pest, Pantana phyllostachysae Chao (P. phyllostachysae), on AGB remains poorly understood, potentially compromising estimation accuracy. This study aims to develop allometric equations and integrate them with machine learning algorithms to accurately estimate the AGB of Moso bamboo forests under varying levels of pest stress. RESULTS:Allometric equations exhibited strong estimation performance across all pest infestation levels, with R2 values exceeding 0.93, root mean square error (RMSE) values below 0.66 kg, and mean absolute error (MAE) values under 0.51 kg. Among the machine learning approaches evaluated, the Extreme Gradient Boosting (XGBoost) algorithm demonstrated superior performance, yielding an R2 of 0.8593, RMSE of 0.5176 kg, and MAE of 0.4313 kg. A clear negative correlation was identified between the severity of P. phyllostachysae infestation and AGB, with biomass values decreasing progressively from healthy to severely infested stands. CONCLUSION:Incorporating pest factors into AGB estimation models significantly enhances model accuracy and captures the nuanced effects of pest stress on biomass accumulation. This integration improves model generalizability and ecological relevance, offering valuable insights for sustainable forest management and carbon accounting. The findings highlight the importance of explicitly considering pest dynamics in biomass modeling and carbon management strategies, laying a robust foundation for future research on pest-biomass interactions in forest ecosystems. © 2025 Society of Chemical Industry.

Rapid urbanization has increased carbon dioxide (CO2) emissions, exacerbating ecological issues and prompting global shift towards low-carbon development. However, current studies at the county-level face challenges such as incomplete monitoring systems and insufficient statistical granularity, which restrict the detailed analysis of carbon emission spatial distribution and driving mechanisms. To address this, the study utilized high-resolution Luojia1-01 nighttime light (NTL) data combined with the optimal parameters-based geographical detector (OPGD) model, taking Fuzhou, a typical "furnace city" as a case study to reveal the spatial differentiation characteristics and driving mechanisms of carbon emissions at the county-level. The results indicate that carbon emissions in Fuzhou exhibit a "core-edge" spatial differentiation pattern, with the central urban areas having higher emissions than the surrounding counties, and a positive spatial correlation was observed; the proportion of the tertiary production (PTP), the proportion of the primary production (PTP), the urbanization rate (UR), and the level of social capital (SC) are core driving factors of carbon emissions, with dual-factor interactions exhibiting significant bilinear enhancement effects. Based on the carbon emission differentiation characteristics, the study proposes a "five-zone differentiated" governance strategy, which includes low-carbon transformation of the service industry in the core urban areas, green industrial upgrading in high-emission zones, and strengthening the carbon sink function in ecological protection areas. This study provides methodological support and decision-making guidance for refined carbon emission management and low-carbon development planning at the county-level.

This study estimated aboveground carbon stock (AGC) using field data and integrated multi-source remote sensing imagery to understand the effects of Pantana phyllostachysae Chao (P. phyllostachysae) stress. AGC remote sensing inversion was performed while accounting for P. phyllostachysae stress, and changes were analyzed. Results indicate: (1) Carbon content coefficients of Moso bamboo leaves, branches, and culms under pest stress ranged from 0.422 to 0.543 g/g, decreasing with increased stress. (2) A random forest model using multi-source data demonstrated the best performance (R2 = 0.688), estimating average AGC at 28.427 t/ha and total carbon sequestration at 913.902 MtC (Million tons of Carbon). (3) Increased pest stress resulted in gradual reductions in AGC. (4) Pest stress is estimated to result in a carbon sequestration loss of 77.443 MtC. The AGC estimation model indicates that P. phyllostachysae significantly reduces AGC, providing crucial data for understanding carbon cycling and enhancing carbon sink management in Moso bamboo forests.

Moso bamboo forests (MBFs) are unique subtropical ecosystems characterized by distinct leaf phenology, bamboo shoots, rapid growth, and carbon sequestration capability. Leaf area index (LAI) is an essential metric for evaluating the productivity and ecological quality of MBFs. However, accurate and large-scale methods for remote-sensing-based LAI monitoring during the winter growth stage remain underdeveloped. This study introduces a novel method integrating hyperspectral indices from Zhuhai-1 Orbit Hyperspectral Satellites (OHS) imagery with the particle swarm optimization-support vector machine (PSO-SVM) coupling model to estimate LAI in winter MBFs. Five traditional vegetation indices (VIRs) and their red-edge variants (VIREs) were optimized to build empirical models. Machine learning algorithms, including SVM, Random Forest, extreme gradient boosting, and partial linear regression, were also applied. The PSO-SVM model, integrating three VIRs and three VIREs, achieved the highest accuracy (R2 = 0.721, RMSE = 0.490), outperforming traditional approaches. LAI was strongly correlated with indices, such as NDVIR, RVIR, EVIRE, and SAVIR (R > 0.77). LAI values of MBFs primarily ranged from 2.1 to 5.5 during winter, with values exceeding 4.5 indicating high winter bamboo shoot harvesting. These findings demonstrate the potential of OHS data to improve LAI retrieval models for large-scale LAI mapping, offering new insights into MBFs monitoring and contributing to sustainable forest management practices.

To address gaps in understanding how external stresses influence remote-sensing inversion of vegetation biochemical components, a P-PROSAIL model incorporating stress factors was developed, with Shunchang County and Yanping District in Fujian Province as the study areas. The model's effectiveness was assessed, yielding R² values of 0.7133, 0.7066, 0.6441, 0.6392, 0.6057, 0.7038, 0.5323, and 0.5149 for leaf area index (LAI), canopy dry matter content (CDMC), canopy cellulose content (CCC), canopy lignin content (CLC), canopy protein content (CPC), canopy nitrogen content (CNC), canopy tannin content (CTC), and canopy flavonoid content (CFC), respectively. While CDMC and most other components showed stable inversions, CTC and CFC exhibited uncertainties due to pest stress. This study clarified the internal and external change characteristics and mechanisms of Moso bamboo forests under Pantana phyllostachysae stress, providing empirical support for the ecological health of bamboo forests.

Urban parks are critical for mitigating environmental challenges; however, their beneficial impacts on ecoenvironmental quality (EEQ) have not been comprehensively explored. Therefore, we developed a new method of quantifying EEQ improvement and driving factors for 51 urban parks in Fuzhou, China. Our multimethod approach combined remote sensing, geospatial analysis, and interpretable machine learning models to evaluate four park EEQ improvement indicators: park EEQ improvement intensity (PEII), distance (PEID), area (PEIA), and efficiency (PEIE) and unravel the nonlinear interactions among internal and external environmental factors. According to the results, the degree of EEQ improvement varied significantly between parks. PEII ranged from -0.0773 to 0.3095 (mean = 0.1425) and parks exhibited different trends between PEII and PEIE. Key internal factors such as park area (PA), perimeter, and aggregation index were positively correlated with PEII, PEID, and PEIA but negatively correlated with PEIE, whereas edge density showed inverse correlations. XGBoost and SHAP analyses highlighted nonlinear relationships, with PA emerging as the most influential factor in PEIE. A novel nonlinear threshold of 2.15 hm2 was identified as the optimal PA for balancing PEIE and land-use efficiency, beyond which ecological efficiency declined. These findings highlight the complexity of park-driven EEQ improvements shaped by interactions between park attributes and external environmental factors. This study provides actionable insights for urban planners to optimize park design and management, emphasizes the need for balanced scaling and connectivity to enhance ecological benefits, and offers a model for data-driven, context-specific greening strategies in rapidly urbanizing regions.

Leaf area index (LAI) serves as a crucial indicator for assessing vegetation growth status, and unmanned aerial vehicle (UAV) optical remote sensing technology provides an effective approach for forest pest-related research. This study investigated the feasibility of LAI estimation in Moso bamboo (Phyllostachys pubescens) forests with different damage levels using UAV data while simultaneously exploring the scale effects of various spatial resolutions. Through image resampling using 10 distinct spatial resolutions and field data classification based on Pantana phyllostachysae Chao pest severity (healthy and mild damaged as Scheme 1, moderate damaged and severe damaged as Scheme 2, and all as Scheme 3), three machine learning algorithms (SVM, RF, and XGBoost) were employed to establish LAI estimation models for both single and mixed damage levels. Comparative analysis was conducted across different schemes, algorithms, and spatial resolutions to identify optimal estimation models. The results showed that (1) XGBoost-based regression models achieved superior performance across all schemes, with optimal model accuracy consistently observed at 3 m spatial resolutions; (2) minimal scale effects occurred at a 3 m resolution for Schemes 1 and 2, while Scheme 3 showed lowest scale effects at 1.5 m followed by 3 m resolutions; (3) Scheme 3 exhibited significant advantages in mixed damaged bamboo forest inversion with robust performance across all damage levels, whereas Schemes 1 and 2 demonstrated higher accuracy for single damaged scenarios compared to mixed damaged. This research validates the feasibility of incorporating pest stress factors into LAI estimation through different pest damage models, offering novel perspectives and technical support for parameter inversion in Moso bamboo forests.

Pantana phyllostachysae Chao ( P. phyllostachysae ) is a destructive leaf-eating pest that poses a significant threat to the health of bamboo forests and the bamboo industry. However, the spatial and temporal spread mechanisms of this pest are still unclear. To better understand and predict the spread of this pest, we used Sentinel-2A/B images from the pest detection period of 2018 to 2021, to identify association factors from five dimensions, including forest stand, meteorology, topography, pest sources, and human environment factors. The association factor sets for the spread of P. phyllostachysae were established under both existence and non-existence pest control scenarios. The extreme gradient boosting (XGBoost) model was employed to derive conversion rules for the respective spread models, enabling the determination of suitability probabilities for both healthy and damaged bamboo forests. These probabilities were then utilized in conjunction with cellular automata (CA) to simulate the spread of P. phyllostachysae under two scenarios. The results showed that the OA and Kappa reached more than 85% and 0.7 in both scenarios, respectively. Meanwhile, the division of pest control scenarios and the selection of XGBoost both help to improve the spreading simulation accuracy. Our models effectively coupled the research results of leaf hosts of different damage levels, simulated the spread of P. phyllostachysae , and identified the dynamic mechanisms of the pest’s spread. These findings provide decision support for interrupting the spread path of the pest and achieving precise control, thus safeguarding forest ecological security.

The on-year and off-year phenomenon is a distinctive phenological characteristic of Moso bamboo, reflecting variations in nutrient dynamics and endogenous hormonal rhythms during the transition from bamboo shoot to the culm. This phenomenon also influences pest resistance between the on-year and off-year cycles of Moso bamboo. Pantana phyllostachysae Chao is a leaf-feeding pest that affects Moso bamboo. However, monitoring P. phyllostachysae damage using remote sensing data is challenging because the off-year Moso bamboo has physiological characteristics similar to on-year Moso bamboo infested with P. phyllostachysae. This study utilizes the Recursive Feature Elimination (RFE) algorithm to investigate hyperspectral remote sensing characteristics of P. phyllostachysae in Moso bamboo forests. We analyzed the impact of on-year and off-year phenological characteristics on the accuracy of hazard extraction and developed detection models for P. phyllostachysae hazard levels in on-year and off-year Moso bamboo using Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and one-dimensional Convolutional Neural Network (1D-CNN). The results demonstrate that classical machine learning and deep learning models can effectively detect P. phyllostachysae damage, with the 1D-CNN algorithm achieving the best performance. Analyzing the impact of the phenological differences between on-year and off-year Moso bamboo on pest identification accuracy revealed that when four machine learning models accounted for these phenological characteristics, their accuracy in identifying pests was significantly higher than that of a model which did not take into account the bamboo phenology. This finding highlights that considering the phenological characteristics of on-year and off-year Moso bamboo can substantially improve the detection accuracy of UAV hyperspectral remote sensing in monitoring P. phyllostachysae damage. This provides more accurate technical support for the health management and resource protection of bamboo forests and offers a scientific basis for maximizing the ecological and economic benefits of bamboo forests.

The objective of this study was to deeply understand the adaptation mechanism of the functional traits of Moso bamboo Phyllostachys pubescens syn. edulis (Poales: Poaceae) leaves to the environment under different Pantana phyllostachysae Chao damage levels, analyzing the changes in the relationship between specific leaf area (SLA) and leaf dry matter content (LDMC). We combined different machine learning models (decision tree, RF, XGBoost, and CatBoost regression models), and used different canopy heights and different levels of infestation, to analyze the changes in the relationship between the two under different levels of infestation based on the results of the best estimation model. The results showed the following: (1) The SLA of Ph. pubescens showed a decreasing trend with the increase om insect pest degree, and LDMC showed an inverse trend. (2) The SLA of bamboo leaves was negatively correlated with the LDMC under different insect pest degrees; the correlation of the data under the healthy class was higher than that of other insect pest levels, and at the same time better than that of the full sample, which laterally confirmed the effect of insect pest stress on the functional traits of Ph. pubescens leaves. (3) When modeling under different infestation levels, the CatBoost model was used for heavy damage and the RF model was used for the rest of the cases; the decision tree regression model was used when modeling different canopy heights. The findings contribute certain insights into the nuanced responses and adaptive mechanisms of Ph. pubescens forests to environmental fluctuations. Moreover, these results furnish a robust scientific foundation, essential for ensuring the enduring sustainability of Ph. pubescens forest ecosystems.

Abstract Forest carbon sinks, a critical component of the global carbon cycle, constitute nearly half of the total terrestrial carbon pool. This study employed correlation analysis and factor effect analysis to quantify the influences of various factors on the volume of Chinese fir stands. A novel modeling framework was developed using the stacking ensemble learning and LSTM (LongShort-TermMemory) models. This framework incorporated diverse base learners, including XGBoost, Adaboost, KNN, and DT, which are intelligently ensemble via GBDT as a meta learner. The model was further optimized by comparing activation functions and optimizers, with ReLU selected as the activation function and Adam as the optimizer. Model accuracy was evaluated using RMSE, MAE, and R2 metrics, significantly enhancing its learning ability and generalization performance. The findings are as follows: (1) Stand stocking showed strong positive correlations with depression, age group, average diameter at breast height (ADBH), average tree height (ATH), slope, and elevation. Conversely, it exhibited significant negative correlations with origin, stand density, and slope position. In investigating Chinese fir growth on slopes, no significant growth differences were observed between downslopes and midslopes; however, both differed significantly from upper slopes. (2) The stacking ensemble learning method constructed here surpassed all existing single models in terms of estimation and assessment indices, demonstrating superior comprehensive performance. (3) Among the LSTM models, Adam-LSTM performed the best (R2=0.844), followed by Sigmoid-LSTM (R2=0.656), while the RMSprop-LSTM model performed the worst (R2=0.618). Combined with artificial intelligence methods, our optimized carbon stock estimation model can help to improve the ability of forest land management and provide a theoretical basis for the scientific management of forest areas.

Abstract Carrying out remote sensing refinement identification of forest land in complex environment is of great significance for timely mapping of forest distribution. Aiming at the problem that remote sensing images have bias in the extraction of forest land information data, based on the semantic segmentation algorithm Unet, combining the ResNet50 deep learning network, the attention mechanism module and the feature pyramid structure, we construct RAF-Unet (ResNet+Attention+FPN+Unet) to improve the extraction of forest land information data. The ResNet50 classification network is used as the encoder of the Unet network to extract the feature maps at five different scales; then, the attention mechanism module is introduced in the decoder stage of the Unet network to extract the key task goal information by learning the weight values of the features; finally, the feature pyramid structure is used in the output stage of the encoder to fuse the information from the shallow network and the deep network to extract the remote sensing forest land information in the image. The results show that the RAF-Unet algorithm outperforms the Unet algorithm in all the indexes, with a precision of 95.24%, a recall of 91.80%, an F1-score value of 93.49%, an intersection over union of 87.63%, and an accuracy of 93.68%; the validity of the modules is verified by the ablation experiments, and the ResNet network, the attention mechanism, and the feature pyramid structure are all effective in improve the classification effect. It helps the forestry department to better manage and dynamically monitor forestry information, which is of great significance to the scientific development, utilization and protection of forest land resources.