In remote sensing, the kernel-driven model (KDM) is widely used for reflectance modeling due to its simple mathematical formulation and computational efficiency. However, in contrast to non-desert scenes, desert shrub canopies are characterized by woody components that are significantly larger than the leaf areas. This deviates the assumptions of Beer's law, which is based on translucent leaves. Moreover, desert environments present a special "background" composed of sand, gravel, saline land, and biological crust, further complicating reflectance modeling. These complexities pose significant challenges for the application of KDM in desert regions. To address these issues, this study introduces a volume-scattering kernel derived from the analytical Gutschick-Wiegel (G-W) solution with a single-angle configuration, aiming to better represent radiative transfer in sparse desert canopies. The model also incorporates different geometric-optical kernels designed to account for the structure of sparse shrubs and the heterogeneous biological crust in desert, using a cover parameter to adjust their respective weights. Furthermore, terrain factors and hotspot functions were integrated to account for coherent backscattering and anisotropic scattering effects. Based on these considerations, this study proposes a new KDM, i.e., the Hotspot Li-Sparse Roujean Terrain (HLSRT) model. The HLSRT model was extensively validated using field measurements and satellite observations. It achieved a low average in situ bias for the red and near-infrared band (NIR) bands (bias = 0.005) and a low root mean square error (RMSE = 0.0299) against satellite data. Compared to existing KDMs, the HLSRT model demonstrated superior performance in reflectance modeling. These results indicate that the HLSRT model offers a reliable semi-empirical tool for modeling radiative transfer and supporting inversion studies in complex desert environments.

BACKGROUND:Pine wilt disease (PWD), caused by the pine wood nematode, threatens forest ecosystems in China. Early detection of infected trees is essential for timely disease management but remains challenging in conifer-broadleaf mixed forests because of crown overlap and complex canopy structures. This study aimed to comprehensively evaluate the potential of unmanned aerial vehicle (UAV)-based hyperspectral, LiDAR (light detection and ranging), and thermal data for multi-stage PWD detection at individual tree scale. RESULTS:An object-based classification and point cloud segmentation (OBPCS) method was developed to delineate individual pine trees by integrating LiDAR point clouds and UAV multi-spectral imagery. Compared with the canopy height model (CHM)-based watershed approach, OBPCS significantly improved segmentation accuracy (F-score = 0.88 versus 0.72). Random forest models were then constructed using crown biochemical, structural, and temperature features to classify five PWD infection stages. Among single-sensor datasets, hyperspectral data achieved the highest accuracy (overall accuracy (OA) = 78%, κ = 0.72), outperforming LiDAR (OA = 46%, κ = 0.32) and thermal data (OA = 28%, κ = 0.10). Multi-sensor fusion further increased classification accuracy to 82% (κ = 0.77) and improved early-stage detection accuracy by 7%. Variable importance analysis revealed that pigment-related indices were the most influential features, followed by LiDAR return intensity and point distribution metrics. CONCLUSION:UAV-based multi-sensor fusion provides an effective approach for early detection of PWD in structurally complex mixed forests. The proposed framework improves individual-tree delineation and early-stage diagnosis, offering practical support for early warning and fine-scale monitoring of disease progression in forest ecosystems. © 2026 Society of Chemical Industry.

Vertical structural and spectral heterogeneity are two key remote sensing characteristics of complex forests. To enable effective forest health management and provide early warnings of abnormal disturbance, monitoring forest biochemical content with a vertically layered spectral perspective is critically needed. However, commonly used remote sensing technique still have limited capacity to study the biochemical status of the middle and lower canopy layers. This study provides the first insight into the potential of the full-waveform large-footprint hyperspectral LiDAR (LFHSL) system for retrieving the vertical heterogeneity of forest chlorophyll using 3D radiative transfer modeling. In our newly constructed LFHSL model, virtual three-dimensional (3D) complex forest scenes, comprising trees, bushes, and grass, were defined with varying positions and biochemical content inputs. Hyperspectral waveforms within the large-footprint were then simulated for each combination of vegetation position and biochemical level. The concept of spectral index time profiles (SITP), referred to as spectral index variation along the laser path, were introduced and used to assess the vertical distribution of chlorophyll for the first time in forest scenes. The main findings of this study are as follows: (1) Full-waveform LFHSL owns great potential for retrieving vertical chlorophyll content across trees, bushes, and grass layers in complex forest ecosystems. (2) SITP is a novel and essential reference indicator that fully registers chlorophyll variations along the laser path. (3) Simulations with random positions and chlorophyll contents indicate that the peak points of SITP yield higher chlorophyll prediction accuracy in trees layer and grass layer (R2: 0.996 vs. 0.971, RMSE: 1.39 vs. 3.81 μgcm−2) than that of bushes layer (R2 of 0.801, RMSE of 9.97 μgcm−2). (4) Compared to position patterns, LFHSL system is more sensitive to chlorophyll content sets. This study demonstrates that full-waveform LFHSL is a promising and surely reliable tool for acquiring and monitoring vertical vegetation health in complex forests. It not only provides significant guiding for the development of laser radar models but also holds promise for adoption in design of large-footprint multi-spectral or hyperspectral LiDAR.

Fruit trees are typically organized at the orchard level,where the tree-based ecosystem is characterized by high homogeneity,leading to clustered distributions with distinct boundaries.While remote sensing-based classification techniques are well established,most studies have not treated fruit orchards as a distinct category.Whether remote sensing can effectively address orchard classification and distribution remains uncertain.This study focused on the Guanzhong Plain on the southern part of the Loess Plateau as a representative drought-vulnerable region in China,characterized by mixed orchard-cropland landscapes.Sentinel-2 imagery was used as the primary classification feature,supplemented by topographic characteristics.A Random Forest classifier was trained and validated using 1980 ground samples across major planting regions in May 2024.The final classification results were satisfactory,with an overall accuracy of 0.86.Meanwhile,a comparison against statistical data demonstrated the reasonableness of fruit orchard area:the correlation coefficients for three major fruit types(apple,grape,and kiwi)are greater than 0.75.Compared with existing land cover products,which often misclassify fruit trees as cropland or forestland,our results demonstrated that combining band reflectance time series,vegetation index time series,and topographic features can effectively differentiate fruit orchards from spectrally similar cropland and forestland.This study facilitates precise fruit orchard mapping,supporting targeted production management and ecological carbon sequestration estimation in similar regions with drought-vulnerable agroforestry systems.

Accurate estimation of above-ground biomass (AGB) is vital for carbon accounting, biodiversity conservation, and sustainable forest management, especially in tropical regions under strong anthropogenic pressure. This study estimated and mapped AGB in the Atacora Mountain Chain, Togo, using a multi-source remote sensing approach within Google Earth Engine (GEE). Field data from 421 plots of the 2021 National Forest Inventory were combined with Sentinel-1 Synthetic Aperture Radar, Sentinel-2 multispectral imagery, bioclimatic variables from WorldClim, and topographic data. A Random Forest regression model evaluated the predictive capacity of different variable combinations. The best model, integrating SAR, optical, and climatic variables (S1S2allBio), achieved R2 = 0.90, MAE = 13.42 Mg/ha, and RMSE = 22.54 Mg/ha, outperforming models without climate data. Dense forests stored the highest biomass (124.2 Mg/ha), while tree/shrub savannas had the lowest (25.38 Mg/ha). Spatially, ~60% of the area had biomass ≤ 50 Mg/ha. Precipitation correlated positively with AGB (r = 0.55), whereas temperature showed negative correlations. This work demonstrates the effectiveness of integrating multi-sensor satellite data with climatic predictors for accurate biomass mapping in complex tropical landscapes. The approach supports national forest monitoring, REDD+ programs, and ecosystem restoration, contributing to SDGs 13, 15, and 12 and offering a scalable method for other tropical regions.

Accurate estimation of forest inventory attributes from unmanned aerial vehicle (UAV) imagery typically hinges on the precise segmentation and detection of individual tree crowns using deep learning (DL) models. However, the collection of large volumes of high-quality annotated data remains a significant challenge, due to complex forest environments and illumination variability, obstacling flexible and accurate tree crown recognition. To mitigate this problem, we propose a simulation-to-reality (Sim2Real) transfer learning framework that leverages Sim2Real images as training data, complemented by a small number of real images for fine-tuning. Specifically, simulated images were generated from a range of reconstructed three-dimensional (3D) forest scenes using physically-based rendering techniques, and subsequently translated into Sim2Real imagery through an improved CycleGAN model. Various combinations of Sim2Real images and real images were explored to analyze the performance of the commonly used DL models (i.e., UNet, Mask R-CNN, and YOLOv8) in tree crown recognition, thereby supporting the estimation forest inventory attributes. Compared to real UAV images captured from a typical conifer forest, the generated Sim2Real images exhibited a similar, as indicated by a lower Frechet Inception Distance (FID). Results demonstrate that models trained solely on Sim2Real images achieved reasonable performance in both segmentation and detection tasks for the typical forest, with intersection over union (IoU) of around 0.7 and F1 scores of around 0.8. Fine-tuning these models with only 8-58 real images led to substantial performance improvements, increasing IoU to approximately 0.80 and F1 scores to around 0.9. Correspondingly, the estimation accuracy of canopy cover and stand density followed a similar upward trend. These findings demonstrate the Sim2Real transfer learning enhances both the flexibility of training data and prediction accuracy of DL models, while significantly reducing the manual annotation effort, eventually supporting accurate and flexible forest inventory.

Optical greenness indices, such as the fraction of absorbed photosynthetically active radiation (fAPAR), are critical in constraining and guiding the modelling of forest carbon storage. However, as optical sensors have limited penetration capacity, it remains uncertain whether greenness indices accurately reflect the true response of aboveground biomass (AGB) to local climatic conditions. In this study, we integrate a wall-to-wall AGB dataset derived from microwave remote sensing to examine the consistency between AGB and fAPAR in their climatic responses. Meanwhile, we use an AGB-fAPAR Difference Index (AFDI) to quantify the driving mechanisms underlying their divergent responses, which is defined as the difference between AGB and fAPAR after standardization and normalization. We find that AGB is negatively associated with local precipitation, whereas fAPAR exhibits a positive correlation, leading to pronounced response differences in AFDI across precipitation gradients. Micro-topography contributes 75

Pine wilt disease (PWD), driven by the highly destructive pine wood nematode (PWN; Bursaphelenchus xylophilus), is recognized globally as one of the most catastrophic threats to pine ecosystems. Characterized by its explosive spread and high mortality rate in susceptible hosts, PWD has inflicted monumental economic damage and generated profound ecological instability worldwide. Early detection and intervention before trees exhibit visible symptoms is crucial for halting disease spread and enabling effective management. Remote sensing (RS) technology, with its multiscale, nondestructive, and spatiotemporally continuous observation, has become an essential tool for early detection of PWD. This review systematically presents the pathogenic process of PWD and the corresponding principles of RS response. We analyze how the complex interplay between critical environmental and biological variables, RS-specific factors, and detection algorithms collectively influences the performance of early-stage PWD detection. This study concludes by identifying and discussing the major limitations and key challenges facing current RS technologies in achieving robust and reliable PWD early detection, thereby charting a path for future study in forest health monitoring. Our analysis reveals that current RS efforts predominantly target the slight discoloration stage of PWD, representing a significant gap in detecting the crucial previsual early stage. Autonomous aerial vehicle (AAV)-based hyperspectral imaging remains the dominant technical approach. Red-edge regions and vegetation indices (VIs) derived from them are widely used in early detection studies. At the algorithmic level, shallow machine learning (ML) methods are generally more suitable for tasks involving small sample sizes and low-dimensional data, whereas deep learning (DL) approaches are better suited for end-to-end detection using large-scale datasets. Integrating RS data with prior knowledge of plant physiological mechanisms and developing interpretable detection models represent a key and promising direction for advancing early PWD detection. Future studies should place greater emphasis on the standardization of biological information and urgently promote open data sharing; both are essential for ensuring the reproducibility and generalizability of research findings across different ecological environments. Simultaneously, substantial efforts are required to address several critical challenges. These include the susceptibility of early RS signals to noise and mixed-pixel effects, the limited availability of early-stage samples, particularly during the previsual stage, and the effective discrimination of PWD from other co-occurring pest and disease disturbances. This review aims to provide comprehensive technical guidance and methodological recommendations for early PWD detection, thereby supporting the scientific monitoring and management of this destructive forest disease.

The 3-D radiative transfer models (3-D RTMs) are indispensable for understanding how radiation propagates within vegetation canopies and for interpreting remote sensing signals. Yet, their practical deployment is hindered by the difficulty of acquiring the detailed 3-D structural parameters they require. Airborne laser scanning (ALS) offers a powerful means of retrieving such information at landscape scales. Here, we quantitatively assess how ALS-derived 3-D forest reconstructions influence bidirectional reflectance factor (BRF) simulations. Using three benchmark radiation transfer model intercomparison (RAMI) scenes that span contrasting canopy architectures, we generated synthetic ALS point clouds and reconstructed each scene via alphashape, voxel, and ellipsoid representations. The resulting 3-D scenes were then input to the Large-scalE remote Sensing data and image Simulation (LESS) model to evaluate the structural contribution to simulated canopy reflectance and to quantify the associated uncertainties. Results demonstrate that at a 10-m resolution, all three representation methods yielded favorable results, especially in the summer birch forest, where the alphashape approach achieves the highest coefficient of determination R-2 (0.94) and the lowest mean relative absolute error delta (3.5%) for the near-infrared (NIR) band. Generally, both the alphashape and voxel approaches outperformed the ellipsoid approach in representing canopies for reflectance simulation, with the alphashape method being comparable with the voxel approach. Furthermore, varying alpha values used for fitting the crowns showed that R-2 and delta exhibit a consistent pattern and are minimally affected by alpha values, whereas the voxel approach shows greater sensitivity to voxel dimensions. These results demonstrate that ALS-derived canopy geometry can support radiative transfer simulations under controlled conditions and highlight the importance of selecting reconstruction strategies according to canopy structure and spatial resolution, providing practical guidance for BRF simulations and future remote sensing applications across forest types.

High-resolution Earth observation (EO) is critical for tracking spatially heterogeneous sustainable development goals (SDGs), such as cropland dynamics and urban expansion. However, persistent limitations in spatiotemporal continuity (satellite revisit gaps) and cost-efficiency (prohibitive pricing of commercial <2 m data) hinder its scalability. While deep learning-based single image super-resolution (SR) techniques offer a potential solution, their quantitative equivalence to native high-resolution data and the generalizability across geographies remain unproven. Here, we demonstrate that AI-powered SR can systematically transform freely available 10-m Sentinel-2 imagery with visible (RGB) and near infrared (NIR) bands into 2m-resolution images with RGB-NIR bands while preserving spectral-temporal fidelity. Specifically, we trained a geospatially constrained transformer-based SR framework (GeoSR) with 3.15 million km & sup2; of co-registered Gaofen-1/6 and Sentinel-2 pairs, achieving near-native performance with <1% F1-score loss in critical applications: cultivated land parcels mapping (F1-score = 0.84 vs. 0.85 for native Gaofen-1/6), urban footprint extraction (0.82 vs. 0.81), and fine-grained land use and land cover classification (0.43 vs. 0.43). Notably, the geospatial module enables large-scale generalization, retrospectively reconstructing decade-long environmental dynamics from historical archives - an unprecedented capability unattainable solely through launching new satellites. We further demonstrated operational scalability through two large-scale implementations: mapping 6.09 million hectares of agricultural parcels of the entire Anhui Province, China, and extracting 9866 km & sup2; of building footprints across 87 C40 coastal cities. As a free-to-access platform (www.sr-earth.org), GeoSR significantly reduces high-resolution data costs compared to commercial alternatives. AI-powered SR is not merely an image enhancement tool but a scientifically valid EO data source, particularly transformative for specific SDG monitoring in low/middle-income regions where native HR data scarcity impedes evidence-based policymaking.

Timely identification of forest disturbance agents is essential for effective ecosystem management and rapid response to natural and anthropogenic threats. However, most near-real-time (NRT) monitoring systems focus solely on detecting disturbance locations and timing of generic disturbances, lacking attribution of causal agents that is critical for operational decision making. This study presents a novel NRT framework to map major forest disturbance agents-wildfire, logging, and stress-across China using the Harmonized Landsat and Sentinel-2 (HLS) dataset. To address the scarcity of local training data, we introduced a transferring-guided sampling strategy that efficiently generated tile-specific local samples by leveraging disturbance archives from the conterminous United States and subsequent expert verification. Stage-based random forest models were subsequently constructed using 16 temporally dynamic features derived from HLS spectral trajectories to classify anomalies at varying disturbance stages. The system achieved an overall first-alert lag of 11.6 days and a level-off lag of 15.5 days, with corresponding overall accuracies of 77.5% and 84.0%. Wildfire disturbances exhibited the shortest detection lag and the highest accuracy, followed by logging and stress, reflecting differences in spectral separability among agents. Compared with the global DIST-ALERT product, the proposed framework achieved higher accuracy and fewer false detections, at a trade-off of a 3.7-day longer first-alert lag, while providing actionable information on disturbance causality. These results demonstrate the feasibility of operationally mapping disturbance agents in near real-time at a national scale. The proposed framework offers a transferable, data-efficient solution for rapid forest disturbance attribution and provides a foundation for global-scale NRT disturbance monitoring initiatives.

Unattended UAV systems offer an effective solution for high-frequency and high-resolution monitoring of forest disease with minimal human intervention. Deploying such systems, however, requires classification methods that achieve high accuracy while remaining robust to spectral interference and computationally efficient for onboard or edge processing. Balancing these requirements is critical for practical pine wilt disease (PWD) surveillance, yet has received limited attention. In this study, we systematically evaluated four spectral analysis approaches for PWD detection using hyperspectral imagery acquired from an unattended UAV platform. These methods include: (1) a newly developed vegetation index, PWDAI, based on discrete spectral bands; (2) a Spectral Angle Mapper (SAM) method utilizing a band-optimized, partially continuous spectrum; (3) a Partial Least Squares Discriminant Analysis (PLS-DA) model and (4) a one-dimensional convolutional neural network (1D-CNN) model, both leveraging the full hyperspectral spectrum (400-1000 nm). Using object-based samples of healthy, diseased, shadow, and mixed classes, we identified PWD-sensitive spectral features in both discrete and continuous forms, which were subsequently used for the design of PWDAI and application of SAM method. Comparison results show that 1D-CNN, utilizing the full spectrum, achieved the highest classification accuracy (overall accuracy: 92.34 %, Kappa: 0.85) and exhibited strong resilience to shadow and mixed pixel interference. PWDAI, by contrast, used only three narrow bands yet achieved competitive accuracy (OA: 83.86 %) with minimal computational cost, offering a practical trade-off for onboard application. Intermediate methods such as SAM and PLS-DA demonstrated moderate performance. These findings suggest that full-spectrum deep learning models are optimal for offline high-precision mapping, while targeted vegetation indices like PWDAI provide interpretable and lightweight alternatives for real-time UAV-based disease screening. The integration of such adaptable spectral classifiers with unattended UAV systems offers a promising pathway for scalable, automated forest health monitoring.

Forest fires are major disturbances that reshape ecosystem structure and function, yet post-fire trajectories of vegetation greenness and productivity often diverge. The spatial patterns and drivers of this structural–functional decoupling remain insufficiently understood. Here, we quantified differences between normalized recovery trends of normalized difference vegetation index (NDVI) and net primary productivity (NPP) in China's forests during the first five years following single-burn events (2001–2015) using a metric termed ΔnSlope (defined as the difference between normalized NDVI and NPP recovery slopes). By applying a sensitivity threshold of 0.5 standard deviations to distinguish significant decoupling from background noise, we found that 53.93% of the burned areas exhibited significant structural–functional decoupling. Specifically, functional recovery (NPP) significantly outpaced structural recovery (NDVI) in 29.60% of the total burned area. Spatially, NPP‑lead decoupling was most pronounced in northeastern forests and the North China–Inner Mongolia ecotone, while synchronous or NDVI‑lead patterns occurred in southwestern regions. Using a random forest model on the continuous dataset, we identified that fire-year precipitation was the strongest positive driver of synchrony, whereas post-fire drought stress amplified divergence. Topography and soil properties further modulated heterogeneity, with higher elevation, steeper slopes, and sandy soils increasing decoupling by constraining resource availability. These findings reveal substantial spatial variability in post-fire ecosystem recovery across China and highlight the dominant role of water availability in shaping structural–functional dynamics. The results provide a scientific basis for region-specific forest restoration and resilience management under a changing climate.

As a key biophysical parameter describing forest vegetation structure, Leaf Area Index (LAI) is an essential and widely used indicator for evaluating forest ecosystem function and health. LAI retrieval from remote sensing observations primarily relies on canopy radiative transfer models (RTMs) that quantitatively characterize the complex relationship between canopy parameters and reflectance. However, most physical models currently used for LAI retrieval are one-dimensional (1D) RTMs, which typically assume the canopy to be horizontally homogeneous and thus fail to capture the inherent heterogeneity within the canopy. Although three-dimensional (3D) RTMs can better characterize the structural complexity of forest canopies, their high computational demand and the difficulty of parameterization often limit their application to large-scale remote sensing retrievals. In this study, a novel 3D Look-Up Table (3D-LUT) approach was developed for retrieving forest LAI from Landsat by accounting for the heterogeneity within forests through the integration of LiDAR-based scene reconstructions to parameterize the RTM. Instead of using idealized homogeneous layers or simple geometric objects, our approach used airborne LiDAR data to reconstruct realistic and structurally representative 3D forest scenes for typical forest types, including Deciduous Broadleaf Forest (DBF), Deciduous Needleleaf Forest (DNF), Evergreen Broadleaf Forest (EBF), and Evergreen Needleleaf Forest (ENF). Based on these reconstructed forest scenes, type-specific LAI look-up tables (LUTs) were built by coupling the 3D RTM Large-scalE remote Sensing data and image Simulation (LESS) with an analytical model PATH_RT, an accurate and efficient RTM based on 3D path-length distribution and spectral invariant theory, enabling accurate LAI retrieval from Landsat imagery. This method was compared against field observations collected from 16 National Ecological Observatory Network (NEON) sites and 8 Integrated Carbon Observation System (ICOS) sites, which comprise a representative sample of different forest types. Additionally, intercomparison was conducted using the High-resolution Global LAnd Surface Satellite (Hi-GLASS) LAI product, Simplified Level-2 Prototype Processor (SL2P) algorithm and the MODIS LAI product. Validation against in situ data demonstrated that the proposed algorithm can achieve high-accuracy retrieval of LAI across four forest types, with RMSE ranging from 0.93 to 1.20 m2/m2 and MAE from 0.73 to 1.00 m2/m2. The intercomparison results revealed that retrieval algorithms based on the PROSAIL model, such as SL2P, tend to underestimate forest LAI. In contrast, the proposed algorithm shows strong overall agreement with the Hi-GLASS LAI product and MODIS LAI product, which are derived from a deep learning framework and a 3D RTM, respectively, supporting its reliability for regional-scale forest LAI retrieval. By generating the simulated dataset derived from realistically reconstructed 3D forest structures using LiDAR data, this study further advances the application of LiDAR in quantitative remote sensing retrieval.

Accurately quantifying fine-scale forest canopy-absorbed photosynthetically active radiation (APAR) is essential for monitoring forest growth and understanding ecological processes. The development of 3D radiative transfer models (3D RTMs) enables the precise simulation of canopy–light interactions, facilitating better quantification of forest canopy radiation dynamics. However, the complex parameters of 3D RTMs, particularly detailed 3D scene structures, pose challenges to the simulation of radiative information. While high-resolution LiDAR offers precise 3D structural data, the effectiveness of different tree crown reconstruction methods for APAR quantification using airborne laser scanning (ALS) data has not been fully investigated. In this study, we employed three ALS-based tree crown reconstruction methods: alphashape, ellipsoid, and voxel-based combined with the 3D RTM LESS to assess their effectiveness in simulating and quantifying 3D APAR distribution. Specifically, we used two distinct 3D forest scenes from the RAMI-V dataset to simulate ALS data, reconstruct virtual forest scenes, and compare their simulated 3D APAR distributions with the benchmark reference scenes using the 3D RTM LESS. Furthermore, we simulated branchless scenes to evaluate the impact of branches on APAR distribution across different reconstruction methods. Our findings indicate that the alphashape-based tree crown reconstruction method depicts 3D APAR distributions that closely align with those of the benchmark scenes. Specifically, in scenarios with sparse (HET09) and dense (HET51) canopy distributions, the APAR values from scenes reconstructed using this method exhibit the smallest discrepancies when compared to the benchmark scenes. For HET09, the branched scenario yields RMSE, MAE, and MAPE values of 33.58 kW, 33.18 kW, and 40.19%, respectively, while for HET51, these metrics are 12.74 kW, 12.97 kW, and 10.27%. In the branchless scenario, HET09′s metrics are 10.65 kW, 10.22 kW, and 9.79%, and for HET51, they are 2.99 kW, 2.65 kW, and 2.11%. However, differences remain between the branched and branchless scenarios, with the extent of these differences being dependent on the canopy structure. Our conclusion demonstrated that among the three tree crown reconstruction methods tested, the alphashape-based method has the potential for simulating and quantifying fine-scale APAR at a regional scale. It provides a convenient technical support for obtaining fine-scale 3D APAR distributions in complex forest environments at a regional scale. However, the impact of branches in quantifying APAR using ALS-reconstructed scenes also needs to be further considered.

Accurate, cost-effective monitoring of plantation aboveground biomass (AGB) is crucial for supporting local livelihoods and carbon sequestration initiatives like the China Certified Emission Reduction (CCER) program. High-resolution canopy height maps (CHMs) are essential for this, but standard lidar-based methods are expensive. While deep learning with RGB imagery offers an alternative, accurately extracting canopy height features remains challenging. To address this, we developed a novel model for high-resolution CHM generation using a Large Vision Foundation Model (LVFM). Our model integrates a feature extractor, a self-supervised feature enhancement module to preserve spatial details, and a height estimator. Tested in Beijing's Fangshan District using 1-meter Google Earth imagery, our model outperformed existing methods, including conventional CNNs. It achieved a mean absolute error of 0.09 m, a root mean square error of 0.24 m, and a correlation of 0.78 against lidar-based CHMs. The resulting CHMs enabled over 90

Plantations for producing profitable products play a critical role in supporting local livelihoods, which requires accurate monitoring to guide sufficient management activities. Additionally, recent expansion of plantations for sequestering atmospheric CO2 motivated by China Certified Emission Reduction (CCER) guideline further underscore the urgent need for accurate and cost-effective methods to estimate plantation aboveground biomass (AGB). High-resolution canopy height maps (CHMs) are essential for capturing detailed plantation traits for AGB estimation, especially given the typically small scale of plantations. While airborne or unmanned aerial vehicle (UAV)-based lidar remains the gold standard for acquiring high-resolution CHMs, its high cost limits widespread use. With advancements in deep learning, predicting CHMs using remote sensing RGB data has emerged as a cost-effective alternative, although challenges remain in accurately extracting canopy height-related features. To address these challenges, we develop a novel model for high-resolution CHM generation based on large vision foundation model (LVFM). This model integrates a feature extractor, a self-supervised feature enhancement module to avoid spatial detail loss in feature extraction, and a height estimator to produce high-resolution CHMs. Tested in the Fangshan District of Beijing, China—a region characterized by small, fragmented plantation parcels—our model, utilizing high-resolution (1-m grid) RGB imagery from Google Earth, demonstrated superior performance compared to existing methods, including conventional convolutional neural networks (CNNs) and naive LVFM implementations. The model achieved a mean absolute error of 0.09 m, a root mean square error of 0.24 m, and a correlation coefficient of 0.78 when evaluated against lidar-based CHM observations in pixel-wise assessments. Our model also exhibits satisfactory performance when being generalized into nontraining regions. Additionally, the CHMs generated by our model enabled over 90% success in individual tree detection and showed high accuracy in AGB estimation and a reasonable performance in tracking plantations’ growth. Our approach offers a promising tool for evaluating carbon sequestration in plantations and natural forests covering a large region.

Using microwave remote sensing to invert forest parameters requires clear canopy scattering characteristics, which can be intuitively investigated through scattering measurements. However, there are very few ground-based measurements on forest branches, needles, and canopies. In this study, a quantitative analysis of the canopy branches, needles, and ground contribution of Masson pine scenes in C-, X-, and Ku-bands was conducted based on a microwave anechoic chamber measurement platform. Four canopy scenes with different densities by defoliation in the vertical direction were constructed, and the backscattering data for each scene were collected in the C-, X-, and Ku-bands across eight incidence angles and eight azimuth angles, respectively. The results show that in the vertical observation direction, the backscattering energy of the C- and X-bands was predominantly contributed by the ground, whereas the Ku-band signal exhibited higher sensitivity to the canopy structure. The backscattering energy of the scene was influenced by the incident angle, particularly in the cross-polarization, where backscattering energy increased with larger incident angles. The scene’s backscattering energy was influenced by the scattering and extinction of canopy branches and needles, as well as by ground scattering, resulting in a complex relationship with canopy density. In addition, applying orientation correction to the polarization scattering matrix can mitigate the impact of the incident angle and reduce the decomposition energy errors in the Freeman–Durden model. In order to ensure the reliability of forest parameter inversion based on SAR data, a greater emphasis should be placed on physical models that account for signal scattering and the extinction process, rather than relying on empirical models.