Search Results (3,482)

Hyperspectral (HSI) remote sensing has emerged as a transformative technology for Earth Observation, enabling detailed assessments across different domains. The current PRISMA-aligned systematic review aims to compare classical physics-based algorithms with emerging machine learning (ML), deep learning (DL) and hybrid approaches across two relevant application domains (vegetation and raw materials), analyzing over 350 peer-reviewed studies (194 after the screening) sourced from Scopus and Web of Science and accessed in July 2025. Specific domain-related studies have been considered, excluding duplicates and studies not strictly related to HSI. Risk of bias was assessed qualitatively based on different criteria. The efficiency of the techniques was analyzed by comparing the accuracy metrics reported in the studies. The heterogeneity of the evaluation metrics used across the different categories of the studies and the underrepresentation of some application domains is the final baseline of the work. The results were synthesized, grouping by application domains and algorithm category: ML and DL models dominate vegetation applications, and physics-based methods remain prevalent in raw materials. Hybrid models achieve the highest performances across all domains. This review highlights the importance of the hyperspectral operational requirements identified for upcoming missions (CHIME, SBG and IRIDE) and points out the opportunity for algorithm development. Full article

Precise spatiotemporal monitoring of soil moisture is fundamental to the efficient regulation and sustainable utilization of agricultural water resources in arid and semi-arid irrigation districts. This study focuses on the Yichang Irrigation District within the Hetao Irrigation Area to elucidate the spatiotemporal dynamics of surface soil moisture during the crop growing season. Multi-year Landsat 8/9 remote sensing imagery (2022–2024) was integrated with the Temperature Vegetation Dryness Index (TVDI) framework to construct two feature spaces, namely Normalized Difference Vegetation Index–Land Surface Temperature (NDVI–LST) and Enhanced Vegetation Index–Land Surface Temperature (EVI–LST). A dual-index complementary inversion strategy was applied for soil moisture estimation, and the outputs were validated against Soil Moisture Active Passive (SMAP) soil moisture products and MOD16 evapotranspiration products. Results indicated that the dry edges of the feature spaces derived from both vegetation indices exhibited double-inflection-point characteristics, with optimal fitting intervals located between the inflection points. The inflection point positions shifted dynamically with variations in crop coverage. During bare-soil and low-vegetation-coverage periods (May, June, and September), the minimum thresholds for low NDVI and EVI values were 0.07 and 0.06, respectively, whereas during high-vegetation-coverage periods in July and August, the minimum thresholds for both indices increased to 0.15. NDVI demonstrated superior performance during May, June, and September, whereas EVI exhibited greater advantages during active crop growth periods in July–August. The optimized model achieved robust inversion accuracy, with a validation R² of 0.81 for the measured soil moisture in the 0–20 cm layer on 12 May 2024. The inversion results exhibited strong correlations with the SMAP soil moisture products (R² = 0.663 during low crop coverage; R² = 0.625 during high crop coverage) and MOD16 evapotranspiration data (R = 0.751). The spatiotemporal patterns of soil moisture were distinctly discerned. Following spring irrigation in May, abundant moisture in certain areas resulted in bimodal distribution patterns in the inversion results. June exhibited the lowest soil moisture content across the study area, with arid zones making up 36.67% of the total area. From July to August, concentrated precipitation coupled with summer irrigation reduced the proportion of extremely arid zones to below 0.98%. Full article

Significant nonlinear radiation differences and weak texture differences exist between multimodal weak texture remote sensing images (MWTRSIs). When using traditional methods to match MWTRSIs, the low distinguishability of descriptors in weak texture regions results in poor matching performance. A robust matching method is proposed based on normalized structural feature transform (NSFT), which can extract spatial structural features of images while mitigating nonlinear radiation differences between weak texture regions. First, the bilateral filter is used to transform the weak texture remote sensing image into a normalized image, which not only greatly weakens the nonlinear radiation difference but also retains most of the structural information. Then, the UC-KAZE detector is designed to extract many evenly distributed feature points on the normalized image. Subsequently, a multimodal weak texture feature descriptor with rotation invariance is designed based on the self-similarity of the weak texture image. Finally, the initial correspondences are constructed by bilateral matching, and the mismatches are removed by the fast sample consensus (FSC) algorithm. We perform comparison experiments on eight types of MWTRSIs. The results show that the proposed method has good scale and rotation invariance and good resistance to nonlinear radiation differences and weak texture differences. Full article

With the growing global emphasis on space resources, the significance of space detection and surveillance technologies has escalated. Currently, space-based optical surveillance stands as the primary means for acquiring information on space objects. However, constrained by the diffraction limits of space telescopes, distant space objects are typically imaged as point sources. The resulting lack of sufficient spatial resolution renders traditional image-based recognition algorithms ineffective. In contrast, the Bidirectional Reflectance Distribution Function (BRDF) fully characterizes surface light scattering properties through four-dimensional features, significantly outperforming traditional two-dimensional spectral techniques in material identification. Consequently, leveraging BRDF signatures at varying phase angles has emerged as an effective approach for Space Object Identification. In this study, we developed an automated BRDF measurement system to characterize various typical aerospace materials and investigated the BRDF properties of mixed-material surfaces. A material composition ratio prediction model was constructed based on a One-Dimensional Convolutional Neural Network (1D-CNN). This model effectively extracts key features, including local slope variations and global waveform characteristics, from the BRDF curves. Experimental results demonstrate that the model achieves a maximum relative percentage error of 6.21%, implying a prediction accuracy for mixed-material composition ratios consistently exceeding 93.79%. Compared to image classification methods based on remote sensing imagery, the proposed approach offers higher computational efficiency, significantly reduced model complexity and computational cost, and enhanced robustness. This work provides essential data support for material identification by space-based telescopes and establishes an algorithmic and experimental foundation for intelligent space situational awareness systems. Full article

This paper presents initial developments towards a high-frequency condition monitoring framework designed for Autonomous Mobile Robots (AMRs) in Smart Factory environments. The proposed approach focuses on data acquisition and edge-level processing at the ultrasound range specifically (>20 kHz), using Micro-Electro-Mechanical System (MEMS) sensors. The system integrates real-time data acquisition, embedded fixed-point frequency-domain processing via a 1024-point FFT, and the integration of Industrial Internet-of-Things (IIoT) infrastructure based on the TIG (Telegraf, InfluxDB, and Grafana) stack, for data aggregation and remote visualization. To ensure timing precision at a sampling rate of 160 kHz, a software-based calibration routine is implemented to compensate for microcontroller overhead. Furthermore, the architecture’s alignment with IEEE 1451 principles is discussed to support interoperable and scalable sensor integration. Experimental results validate the reliable acquisition and processing of ultrasonic signals up to 80 kHz using controlled acoustic sources. This work provides a foundational infrastructure for condition-based monitoring, enabling future development of automated anomaly detection for mechanical components, such as bearings, which exhibit early-stage fault signatures in the ultrasonic spectrum. Full article

Sluice gates are critical infrastructure for flood mitigation. During extreme floods, steel tubes from upstream sites can be transported downstream and impact radial gates, a scenario reported by operators but lacking systematic investigation. This study investigates the dynamic response and damage characteristics of a steel radial gate subjected to such impacts. A finite element model of an in-service radial gate was developed using shell elements. The Johnson–Cook constitutive model was adopted to capture the strain-rate hardening and to quantify the damage extent of Q235 steel. Numerical simulations were conducted across various impact scenarios, comparing the effects of a realistic steel tube against a mass-equivalent spherical impactor, and analyzing the influence of tube size, velocity, angle, and impact location. The results demonstrate that using a mass-equivalent spherical model yields unsafe estimates, underestimating the impact impulse and maximum total displacement by up to 10.58% and 26.16%, respectively, and under-predicting the damage parameter by as much as 51.53% in certain conditions. The maximum gate displacement (885.2 mm) occurs when the tube strikes near the top edge, while the most severe damage (parameter 0.53) is observed near the main crossbeam-support arm joint. The analysis further identifies two primary deformation modes under tube impact: local bending and a cantilever plate deformation. The latter, occurring at top-corner impacts, induces large displacements and forms a plastic hinge line, causing critical damage that is remote from the initial impact point. This research provides quantitative insights that are necessary for the anti-collision design and vulnerability assessment of radial gates. The findings underscore the need to consider realistic impactor geometry in structural analyses, contributing to enhanced risk management and the operational resilience of flood-control infrastructure during extreme flood events. Full article

Climate change and growing populations are major challenges for food security. Understanding single-season rice (SSR) growth patterns and how much area changes over time is essential for sustaining rice distribution patterns and ensuring food security. This study utilized ground trothing data with the remote sensing (RS) technique for estimation of the SSR pattern in Jiangsu Province. A total of 1700 rice and 470 non-rice points were collected during the field visit in April–September 2023 across Jiangsu Province. The current study employed advanced machine learning (ML) and the random forest (RF) model using Google Earth Engine (GEE). This study evaluates the SSR cropping area, including the Normalized Difference Vegetation Index (NDVI), land surface temperature (LST), and land use–land cover (LULC) variation from 2018 to 2023 with different satellites. The results of NDVI show an increasing trend with mean values rising from 0.30 in 2018 to 0.42 in 2023. Additionally, higher mean values of LST were noticed in 2020 by 14.4 °C and in 2022 by 14.1 °C. Furthermore, the SSR area has significantly changed, mostly in the eastern and southern regions of Jiangsu Province, from 2018 to 2023. The higher rice cropping area decreased by 1.42% in 2019 compared to 2018, while the total reduction over the 2018–2023 period was 0.92%. Total cultivated crop areas continued to decline because most of the crop areas changed into built-up areas, resulting in a total variation of 2.75% from 2020 to 2023. The overall accuracy of RF model range was 77.33% to 93.55% with a Kappa coefficient of 0.55 and 0.87, indicating moderate to substantial classification agreement across the study period. The results of LULC indicate that the crop area decreased by 4.13% from 2018 to 2023, and major areas were converted into water bodies and built areas. In conclusion, the single-season cropping pattern decreased during the study period, accompanied by a reduction in total cropland area in Jiangsu Province. Therefore, these findings highlight the influence of urbanization and climate change on agricultural land and emphasize adaptive strategies in Jiangsu Province to ensure food security in the face of environmental challenges. Full article

Ungulate browsing is a major driver of forest regeneration dynamics and habitat structure in managed temperate forests, influencing species composition, regeneration success, and long-term stand development. Traditional assessments of browsing impacts often rely on field-based indicators such as regeneration density or visual cover, but these metrics provide limited insight into three-dimensional habitat structure. Mobile handheld LiDAR offers highly detailed measurements of forest structure, enabling objective and reproducible quantification of structural complexity that complements and extends conventional field-based methods. In this study, we applied handheld LiDAR as an innovative indicator for habitat structure within the ungulate browsing zone (<2 m height) to evaluate structural development across sites differing in management context. Paired fenced and unfenced plots (12 × 12 m) were surveyed within the WiWaldI project framework in 2019 and 2023 and compared across three hunting regimes representing different degrees of ungulate population management. Structural complexity was quantified by deriving box-counting dimensions from LiDAR point clouds, providing a measure of spatial arrangement and density relevant to ungulate–vegetation interactions. To support interpretation and ecological context, we complemented LiDAR indicators with streamlined field assessments. Based on this framework, we assessed whether forest structural complexity and visual cover differ among regions and over time, and whether ungulate browsing induces detectable structural differences between fenced whether structural differences between fenced and unfenced plots are detectable. We further examined the relative importance of tree species composition, plant architecture, and hunting regime as drivers of three-dimensional habitat structure. A simplified octant method characterized the spatial distribution of woody regeneration, while a silhouette-based approach quantified visual cover from the perspective of a standard ungulate profile. These auxiliary measures contextualize visual and spatial aspects of structure that LiDAR metrics capture with minimal observer bias. LiDAR studies have previously demonstrated potential for linking high-resolution structural data to ungulate habitat use, and our approach extends this by focusing on structural complexity as a habitat indicator. Results show a consistent increase in LiDAR-derived structural complexity between 2019 and 2023 across all regions. This increase occurred across management contexts and was not consistently explained by fencing or hunting regime effects, suggesting that site conditions, forest composition, and successional processes were dominant drivers during the observation period. Hunting regime showed no statistically significant and no consistent effect on structural complexity across regions or years. Visual cover metrics varied strongly among regions and species and declined over time. These findings suggest that three-dimensional habitat structure information has the potential to enhance the evaluation of ungulate impacts and may support evidence-based forest and wildlife management, particularly when interpreted in the context of site conditions and successional dynamics. Beyond ungulate impact assessment, the presented handheld LiDAR approach provides a scalable remote sensing framework for precision forestry by capturing three-dimensional structural attributes that are directly linked to forest stability, resilience, growth dynamics, and stand-level species mixing, thereby supporting evidence-based forest management recommendations. Full article

Accurate acquisition of low-observable targets with a minimal radar cross-section (RCS) poses a significant challenge for multi-source remote sensing systems, such as integrated radar–electro-optical (REO) platforms, particularly in complex electromagnetic environments characterized by strong noise interference and a high false-alarm rate. Conventional methods, which often treat data association and fusion from heterogeneous sensors as separate, offline processes, struggle with the dynamic uncertainties and real-time decision requirements of such scenarios. To address these limitations, this paper proposes a novel Evidence–Reinforcement Learning-based Decision and Control (ERL-DC) framework. It operates through a closed-loop architecture consisting of three core modules: A static assessment model for initial target prioritization, a Dempster–Shafer (D–S) evidence-based multi-source data decision generator for dynamic information fusion and uncertainty-aware target selection, and a Deep Reinforcement Learning (DRL) controller for noise-robust sensor steering. A high-fidelity simulation environment was developed to model the multi-source data stream, encompassing radar detection with clutter and false targets, as well as the physical constraints of the electro-optical (EO) servo system. Based on the averaged results from multiple Monte Carlo simulations, the proposed ERL-DC framework reduced the Average Decision Time (ADT) from 7.51 s to 4.53 s, corresponding to an absolute reduction of 2.98 s when compared to the conventional method integrating threshold logic with Model Predictive Control (MPC). Furthermore, the Net Discrimination Accuracy (NDA), derived from the statistical outcomes across all the simulation runs, exhibited an absolute increase of 37.8 percentage points, rising from 57.8% to 95.6%. These results indicate that ERL-DC achieves a more favorable trade-off in terms of scheduling efficiency, decision robustness, and resource utilization. The primary contribution is an intelligent, closed-loop architecture that tightly couples high-level evidential reasoning for multi-source data fusion with low-level adaptive control. Within the simulated environment characterized by clutter, false targets, and angular measurement noise, ERL-DC demonstrates improved target discrimination accuracy and decision efficiency compared to conventional methods. Future work will focus on online parameter adaptation and validation on physical platforms. Full article

Due to the complexity inherent in river ice dynamics, the utilization of remote sensing imagery represents the most crucial and effective method currently available for monitoring changes in river ice. In the Inner Mongolia segment of the Yellow River during winter, two distinct types of ice surfaces are observed: juxtaposed ice and consolidated ice. Additionally, certain areas of open water remain unfrozen. Rapid identification and classification of extensive ice formations and open water zones along this lengthy river section constitute critical information for informed decision-making in ice prevention and management strategies within the Yellow River basin. This paper takes the formation and characteristic analysis of different types of ice in the Yellow River channels in Inner Mongolia as the starting point. It employs a support vector machine (SVM) as the classifier and introduces an optimized model for classifying river ice types using high-resolution Sentinel-2 optical imagery. The model utilizes multi-band spectral features, along with multi-spectral fusion indices such as the normalized difference snow index (NDSI) and the normalized difference frozen surface index (NDFSI), as feature vectors. This approach attains an overall accuracy of 94.91% in classifying different types of ice and can significantly contribute to river ice monitoring by offering robust theoretical support. In the winter of 2023–2024, the proportion of juxtaposed ice on the Yellow River section in Inner Mongolia changed from 45% to 55%, the proportion of consolidated ice changed from 30% to 40%, and the proportion of open water changed from 9% to 19%. This research investigates the characteristics of river ice formations and develops a classification methodology for river ice patterns utilizing high-resolution Sentinel-2 imagery in conjunction with a supervised classification algorithm. The findings of this study are intended to offer technical support for the expedited interpretation of ice conditions in the Yellow River, thereby serving as a scientific basis for precise monitoring and effective disaster prevention and management related to river ice phenomena. Full article

Coastal ecological restoration is a key approach to enhancing ecosystem resilience; however, the stage-wise evolution of restoration outcomes and the underlying driving mechanisms remain insufficiently quantified. Using Qingdao City as the study area, this research integrates remote sensing inversion, the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model, and time-series data from 2010 to 2020 to develop a comprehensive evaluation system for ecological restoration effectiveness, comprising 17 indicators across five dimensions: vegetation, biology, hydrology, economy, and climate. Based on this system, the entropy-weighted method is applied to conduct a dynamic assessment of restoration outcomes. The results indicate that (i) the composite evaluation score in the study area decreased from 0.36 in 2010 to 0.19 in 2015 and then increased to 0.74 in 2020, forming a “V-shaped” nonlinear trajectory with 2015 as a turning point, which is temporally consistent with a delayed response of ecological restoration outcomes following the implementation of major anthropogenic interventions. (ii) Dimension-specific analysis indicates that the decline in the composite score during 2010–2015 was mainly associated with the hydrological dimension, within which chemical oxygen demand (COD) and ammonia nitrogen emissions showed marked increases and were among the highest-weighted indicators. After 2015, following the intensive implementation of regional and system-oriented restoration projects such as the Blue Bay Initiative, pollutant emissions were observed to be effectively controlled, and Bare land area showed a continuous decline. These changes coincided with the rapid rebound of the composite score, within which Bare land area, as the highest-weighted indicator, played a prominent regulatory role. Marked differences were observed among dimensional responses: the biological and vegetation dimensions showed sustained improvement throughout the study period, whereas the hydrological dimension exhibited greater variability over time and stronger temporal alignment with policy-related phases. (iii) Robustness tests indicate that, after completely excluding climate-related variables, the composite score still increased from 0.36 and 0.24 to 0.77, with the “V-shaped” recovery pattern remaining unchanged. This result suggests that the observed improvement in restoration effectiveness in 2020 was more closely associated with systematic human interventions, rather than with short-term climatic fluctuations. This study provides a quantitative and transferable methodological framework for the dynamic evaluation and stage-oriented analysis of coastal ecological restoration effectiveness. Full article

Accurate spatial information on individual tree locations is essential for precision forestry, the integration of field and remote sensing data, and tree-level forest analyses. This study compared the positional accuracy and tree identification performance of four tree-mapping approaches: legacy paper maps, a pseudolite-based field positioning system (TerraHärp), drone-based laser scanning, and mobile laser scanning (MLS). The analysis was conducted in five long-term experimental forest sites in Lithuania, comprising pine- and spruce-dominated stands with varying stand densities. Tree locations derived from legacy maps and the TerraHärp system were compared to assess systematic and random positional discrepancies. TerraHärp-derived tree positions were subsequently used as a reference to evaluate the laser scanning-based methods. Positional accuracy was assessed using Hotelling’s T² test, root-mean-square error, and the National Standard for Spatial Data Accuracy (NSSDA), while spatial autocorrelation of deviations was examined using Moran’s I. The results indicated that discrepancies between TerraHärp and legacy maps were dominated by systematic horizontal shifts in the historical maps, whereas random positional variability was relatively small and consistent across stand types. Drone-based laser scanning showed a strong dependence of tree identification accuracy on stand density and mean tree diameter. Overall, CHM-based segmentation yielded more accurate tree identification than 3D point cloud segmentation, with mean F1-scores of 0.78 and 0.72, respectively. Positional accuracy varied by method, with the largest errors from CHM apexes and highest 3D point cloud points (mean NSSDA ≈ 1.8–2.0 m), improved accuracy using the lowest 3D cluster points (1.45–1.72 m), and the highest accuracy achieved using mobile laser scanning (mean NSSDA 0.76–0.90 m; >95% of trees within 1 m). These results demonstrate that pseudolite-based field mapping provides a reliable reference for high-precision tree location and for integrating field and laser scanning data in managed conifer stands. Full article

The accuracy of forest canopy height estimation is crucial for forest resource management and ecosystem carbon sequestration. However, existing approaches often face limitations in effectively integrating multisource remote sensing data, feature representation, and model learning strategies. To enhance the prediction performance of the model in complex terrain and multisource data environments, this study comprehensively used ICESat-2/ATLAS photon point clouds, Sentinel-2/MSI multispectral imagery, and SRTM-DEM to construct a remote sensing-driven multisource feature system, which eliminated redundant interference using permutation feature importance analysis. Additionally, a self-attention (SA) mechanism was introduced to strengthen high-dimensional feature representation. Three heterogeneous models, incorporating deep neural network (DNN), extreme gradient boosting (XGBoost), and residual network (ResNet), were independently applied for forest canopy height estimation and were further used as base learners, with a random forest as the meta-learner, and an SA-Blending heterogeneous ensemble model that combines a blending technique with an SA mechanism was proposed to enhance the accuracy of forest canopy height estimation. To evaluate the SA optimization strategy and the role of multisource fusion, this study used the original features, SA-optimized features, and multisource fusion features (i.e., the concatenation and fusion of original features and self-attention mechanism features) as inputs to comprehensively compare the performance of each single model and the integrated model. The results show that: (1) The self-attention mechanism significantly improves the prediction performance of heterogeneous models. Compared with original features inputs, the R² of DNN (SA-Only) and XGBoost (SA-Only) increased to 0.706 and 0.708, respectively, and the RMSE decreased to 1.691 m and 1.613 m. Although the R² for ResNet (SA-Only) decreased slightly to 0.699 and the RMSE increased to 1.712 m, the overall impact was not significant. (2) Under the condition of multisource fusion feature input, DNN+SA, XGBoost+SA, and ResNet+SA all demonstrated higher fitting accuracy and stability, verifying the enhancing effect of the SA mechanism on the association expression of multisource information. (3) The SA-Blending model achieved the best overall performance, with R² of 0.766 and RMSE of 1.510 m. It outperformed individual models and the SA-optimized model in terms of overall accuracy, stability, and robustness. The results can provide technical support for high-precision forest canopy height mapping and are of great significance for ecological monitoring applications. Full article

National forest inventory (NFI) data are often collected over a 5-year or 10-year period, meaning some are already outdated by the time the complete results are available. This study assesses changes in growing stock volume (GSV, m³/ha) using hybrid estimation supported by Sentinel-2 metrics. It focuses on constructing a model for estimating the change in GSV using NFI plot data and bitemporal remotely sensed auxiliary data, where such data are available for both points in time ($t_1\mathrm{a}\mathrm{n}\mathrm{d}{t}_2$), and unitemporal data for which remotely sensed auxiliary data are available only for $t_2$. A machine-learning approach based on the random forests (RFs) algorithm was used to predict plot-level GSV change. The original data for $t_2$ and $t_3$ were first used to evaluate the accuracy of the change prediction at the plot level, after which the predicted changes were applied to update the plot-level GSV to predict plot-level GSV at $t_3$, which was then assessed against the observed plot-level GSV at $t_3$. Predicted change was assessed with the Mean Average Annual Volume Change (MAAVC) method, representing the average annual change in GSV over a given period. The results indicate that at the plot level, the bitemporal model produced GSV change estimates with low accuracy (R² = 0.26, RMSE = 4.06 m³/ha, and MAE = 3.26 m³/ha), while the unitemporal model achieved R² = 0.40, RMSE = 3.64 m³/ha, and MAE = 2.65 m³/ha when predicting the $t_1-t_2$ GSV change. Using the predicted change to predict plot-level GSV at $t_3$, the MAAVC based on field data yielded R² = 0.91 and RMSE = 45.11 m³/ha, while the RS unitemporal yielded R² = 0.73 and RMSE = 83.79 m³/ha, and the bitemporal yielded R² = 0.72 and RMSE = 83.61 m³/ha. Mean population GSV at $t_3$, estimated from the RF models, was 254.61 and 255.19 m³/ha for the unitemporal and bitemporal models, respectively. Monte Carlo simulations with a novel stopping criterion were then used to estimate total standard errors, which were 10.48 and 10.40 m³/ha for the unitemporal and bitemporal models, respectively, incorporating both model prediction uncertainty and sampling variability. A test of significance revealed a significant effect of the proposed method on the estimated mean population GSV at $t_3$ (p < 0.001). Conclusively, MAAVC and spatiotemporal RS methods provide a robust framework for predicting GSV at $t_3$ using Estonian NFI and Sentinel-2 data. Full article

Vegetation cover characteristics underpin the understanding of regional ecosystem status and guide sustainable development. While extensive research has documented long-term vegetation dynamics in Qinghai Province, critical gaps remain in identifying driving factors, quantifying their thresholds, and uncovering nonlinear relationships governing vegetation cover. In view of this, based on the MOD13Q1V6 dataset from the Google Earth Engine (GEE) platform, this study constructed a kernel normalized difference vegetation index (kNDVI) dataset for Qinghai Province spanning the period 2001–2023. Furthermore, the spatiotemporal characteristics and future evolution trends of vegetation cover were revealed by employing methods including the Theil–Sen–Mann–Kendall (Theil–Sen–MK) trend test, Hurst exponent, and centroid migration model. At a grid scale of 5 km × 5 km, based on the combined model of Extreme Gradient Boosting and SHapley Additive exPlanations (XGBoost-SHAP), this study integrated 10 multi-source remote sensing variables related to natural conditions, socioeconomic factors, and geographical accessibility to reveal the nonlinear effects between driving factors and kNDVI and identify the key threshold inflection points. The results showed the following: (1) From 2001 to 2023, the kNDVI of Qinghai Province exhibited a fluctuating growth trend with an annual growth rate of 0.0016 per year, presenting a spatial pattern of being higher in the southeast and lower in the northwest. Specifically, the kNDVI of unused land achieved the highest growth rate (65.96%), which was significantly higher than that of other land use types. (2) The kNDVI in Qinghai Province was dominated by stable areas, accounting for 52.75%. Future trend analysis indicated that the region was primarily characterized by sustainable improvement zones (39.91%), while areas with uncertain future trends accounted for 39.70%. (3) The XGBoost-SHAP model revealed that the annual mean precipitation (AMP) (47.26%) and Digital Elevation Model (DEM) (20.40%) exerted substantial impacts on the kNDVI. Marginal effect curves identified distinct threshold inflection points for the major characteristic factors: AMP = 363.2 mm (95%CI: 361.2–365.2 mm), DEM = 4463.9 m (95%CI: 4446.0–4481.1 m), grazing intensity = 1.8 SU (Stocking Unit)·ha⁻¹ (95%CI: 1.8–1.9 SU·ha⁻¹), and slope = 2.8° (95%CI: 2.7–3.0°) and 19.0° (95%CI: 18.8–19.3°). The interaction combinations of AMP × DEM and DEM × distance to construction land exerted a strong positive effect on the kNDVI in the study area, which was conducive to enhancing vegetation cover. These findings verified the effectiveness of ecological projects implemented in Qinghai Province to a certain extent and provided data support for subsequent differentiated restoration and management. Full article