Search Results (1,169)

Warm-season precipitation over Sichuan, China, is jointly modulated by complex terrain, monsoon water vapor transport, and local convective activities, leading to significant spatiotemporal heterogeneity. However, radar observations over mountainous areas are frequently impaired by terrain blockage, beam shielding, and insufficient network coverage, which cause missing data and spatial discontinuity, thereby restricting the accurate monitoring of precipitation systems. To alleviate these problems, this study develops an Efficient Multi-Scale Attention (EMA) U-Net model integrated with Digital Elevation Model (DEM) information, termed EMA-U-Net-DEM, to retrieve radar composite reflectivity by utilizing multi-channel observations from the Fengyun-4A (FY-4A) Advanced Geostationary Radiation Imager (AGRI). In the experiments, FY-4A AGRI multi-spectral measurements were used as model inputs, while radar composite reflectivity products from the Severe Weather Automatic Nowcasting (SWAN) system were applied as reference labels. The modeling and validation were carried out using warm-season (June–August) datasets over Sichuan Province. The results indicate that the proposed EMA-U-Net-DEM exhibits better performance than the traditional U-Net and several typical attention-based benchmark models. Quantitatively, the model achieves a root mean square error (RMSE) of 6.728 dBZ, a mean absolute error (MAE) of 4.788 dBZ, a coefficient of determination $R^2$ of 0.656, a peak signal-to-noise ratio (PSNR) of 25.243 dB, and a structural similarity index measure (SSIM) of 0.793. Categorical verification further reveals that the model yields the highest critical success indices (CSI) of 0.850, 0.560, and 0.364 in the reflectivity ranges of 0–25 dBZ, 25–45 dBZ, and 45–70 dBZ, respectively, demonstrating its superior ability in characterizing weak precipitation backgrounds, moderate precipitation structures, and intense convective cores. The performance enhancements are mainly attributed to the strengthened multi-scale feature extraction by the EMA module and the effective topographic constraints introduced by DEM data. This study confirms that the fusion of FY-4A multi-spectral observations and topographic information can effectively improve radar composite reflectivity retrieval over complex terrain, providing a feasible solution for precipitation monitoring, quantitative precipitation estimation, and severe weather nowcasting in mountainous regions with limited radar coverage. Full article

Xinjiang’s distinctive geography and climate provide important habitats for Falconidae species, yet their climate-driven habitat shifts and conservation gaps remain poorly understood. Using 2731 validated occurrence records and 26 environmental predictors, including bioclimatic, land-use, topographic, hydrological, anthropogenic, and Normalized Difference Vegetation Index (NDVI) predictors, we applied an optimized Maximum Entropy (MaxEnt) framework to project suitable habitats for seven falconid species under current conditions and three Shared Socioeconomic Pathway scenarios (1–2.6, 2–4.5, and 5–8.5) for 2041–2060, 2061–2080 and 2081–2100. Barycenter migration analysis, the Habitat Quality module of the Integrated Valuation of Ecosystem Services and Tradeoffs framework, and protected-area overlays were further integrated to identify conservation priorities. All models showed high discriminatory performance, with mean areas under the receiver operating characteristic curve exceeding 0.90. Current suitable habitats were mainly concentrated along river corridors and mountain foothills in the southern Altai, central-western Tianshan and northern Kunlun regions. Future responses were strongly species-specific. By 2081–2100 under Shared Socioeconomic Pathway 5–8.5, suitable habitat increased by 154.6% for Falco peregrinus and 79.5% for Falco tinnunculus but declined by 81.1% for Falco vespertinus; Falco subbuteo also showed overall expansion, with its suitable-habitat barycenter shifting by up to 291.9 km. Mean relative habitat quality was 0.514 [standard deviation = 0.185], while suitable habitat outside protected areas ranged from 35,346 to 304,866 km² among species, and high-quality priority conservation gaps reached 102,489 km² for Falco cherrug. These results demonstrate that climatic suitability does not necessarily correspond to high habitat quality or adequate protection and support species-specific, climate-adaptive conservation strategies for falconids in Xinjiang. Full article

Cropland fragmentation affects regional ecosystem functioning and biodiversity conservation, making the spatio-temporal coupling between cropland fragmentation (CLF) and habitat quality (HQ) a key basis for land management and ecological protection. However, the reported CLF–HQ coupling differs in sign between regions, and the mechanism underlying this divergence remains unclear. This study evaluated the spatio-temporal dynamics of CLF and HQ and their interrelationship across Hunan Province, China, at the township scale (∼2450 units) from 1994 to 2024. CLF was measured by a four-dimensional index consisting of Scale (SPI), Natural Endowment (NPI), Aggregation (API), and Convenience (CPI), with weights determined by a combined AHP–Entropy method; HQ was modelled with InVEST. The relationship between the two was analysed through bivariate spatial autocorrelation, Mantel tests, Random Forest, redundancy analysis (RDA), and XGBoost–SHAP. CLF peaked in urban fringes and was lowest in the western mountains; mean HQ declined modestly, mainly before 2014. Spatially, CLF and HQ were negatively associated (bivariate Moran’s I between −0.38 and −0.51, $p<0.001$), opposite in sign to the positive coupling found in arid Northwest China. Decomposition of the composite index showed that the negative association was carried largely by the natural-endowment dimension, in which slope and elevation were the dominant indicators. Structural fragmentation dimensions uniquely explained only 1.5–5.4% of HQ variance, and controlling for terrain reduced the CLF–HQ correlation by 67–96%. The sign also reversed under an alternative directional standardisation, so it is not robust to a defensible analytical choice. In these data, the observed negative coupling is largely consistent with a shared topographic gradient rather than a direct fragmentation effect. For management, this correlational evidence suggests that reducing fragmentation alone may do little to improve habitat quality; conservation is better guided by the underlying terrain and land-use gradients. More broadly, CLF–HQ assessments become more reliable when cropland structure is separated from natural endowment and terrain confounding is accounted for. Full article

Flood susceptibility mapping in the Yellow River Basin remains challenging due to uncertainties in sample selection and model generalization. This study develops a novel two-step coupling framework that integrates an information value (IV) model with machine learning (ML) to improve reliability. The IV model first identifies stable low-susceptibility zones to select robust non-flood samples, which are then combined with historical flood inventories to train ML models. The SHAP method is applied to quantify factor contributions and interpret outputs. The results show that the IV-RF model achieves the highest predictive performance, while a stacking ensemble further reduces uncertainty. Sensitivity analyses confirm that model outcomes remain stable across random data splits and repeated non-flood point selections. High-risk areas are primarily located in the Hetao Plain, the Weihe River Basin, and sections of the lower Yellow River Basin, where susceptibility is driven by drainage density, anthropogenic and urban–mining soils, a high topographic wetness index, and gentle slopes. This work provides a transferable methodology that enhances physically consistent sample selection and model interpretability for flood risk assessment. Full article

Urban air pollution driven by road traffic poses a significant public health challenge in cities with high vehicle density and frequent congestion, particularly in topographically constrained environments such as Almaty, Kazakhstan. This study presents a web-based digital twin prototype for the integrated monitoring and analysis of traffic flow and air quality in Almaty, Kazakhstan. The system autonomously collects data from the TomTom Traffic, OpenWeather Air Pollution, and WAQI APIs and official population statistics for five fixed monitoring stations, computing traffic density, vehicles per hour, road congestion, estimated CO₂ emissions, an air pollution index, and a population exposure index, and providing real-time dashboard visualization alongside longitudinal data accumulation. Over a 50-day deployment (26 May–16 July 2026), 4961 real co-located observations across 18 active days were analyzed; records generated by the prototype’s fallback mechanism during API outages were excluded from the scientific analysis. During this summer period, PM2.5 was low (mean ≈ 6 µg/m³) and spatially uniform, and showed no statistically significant association with traffic intensity (r ≈ −0.03). Traffic indicators were instead weakly but significantly correlated with the vehicle-emitted gases NO₂ (r ≈ 0.16) and CO (r ≈ 0.10), which they preceded by up to about one hour. A short-horizon PM2.5 nowcasting task, evaluated across temporal resolutions with time-series cross-validation, was dominated by temporal persistence, with traffic-derived features contributing negligibly. The absence of a summer traffic–PM2.5 association does not preclude such a relationship during the heating season, when particulate levels are higher. The results indicate that the traffic–air-quality relationship in Almaty is season- and pollutant-dependent, and demonstrate a lightweight, reproducible platform suitable for longitudinal monitoring and future heating-season assessment. Full article

Soil organic carbon (SOC) stock is a key component of terrestrial ecosystems, playing a critical role in climate regulation and ecosystem productivity. Despite its economic importance, the impacts of the expansion of khat (Catha edulis) cultivation at the expense of other land uses and its intensive management practices on the depletion of soil carbon content are overlooked in Eastern Ethiopia. This study aimed to estimate and map SOC stocks using multispectral Sentinel-2 and RapidEye imagery, combined with environmental, soil, and topographic variables, across khat-dominated landscapes in the Haramaya District of Eastern Ethiopia. A total of 88 soil samples were collected and analyzed to quantify SOC stocks. Random Forest (RF) and extreme gradient boosting (XGBoost) algorithms were employed to predict SOC stocks. The dataset was stratified into training (70%) and an independent validation (30%) subset. Model development was performed using five-fold cross-validation on the training dataset, while final performance was assessed on the independent validation set using the coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE). Laboratory-measured SOC stocks ranged from 24.99 to 65.94 Mg C ha⁻¹, with a mean value of 36.88 Mg C ha⁻¹. The predicted spatial SOC stocks ranged from 30.4 to 50.4 Mg C ha⁻¹ using RapidEye data and from 32.8 to 51.5 Mg C ha⁻¹ using Sentinel-2, with Sentinel-2 producing slightly higher mean estimates. The lowest SOC stocks were consistently observed in bare, grass, and shrub land-use types across both datasets. RF demonstrated superior performance compared with XGBoost, achieving moderate predictive performance for both the RapidEye (R² = 0.56, RMSE = 5.91 Mg C ha⁻¹) and Sentinel-2 (R² = 0.42, RMSE = 6.90 Mg C ha⁻¹) datasets. This result indicates that RF provided greater robustness for SOC stock prediction under the heterogeneous environmental conditions of khat-dominated agricultural landscapes. Topographic and soil-related variables, particularly the Topographic Wetness Index (TWI), land surface temperature (LST), and clay content, were identified as the most influential predictors in both models. Although less consistent, remote sensing indices such as GNDVI, BSI, and NDWI also contributed to SOC prediction. While both sensors proved effective for SOC mapping, a measurable sensor-related effect was observed. The findings demonstrate the effectiveness of integrating multisource remote sensing, environmental, soil, and topographic variables with machine learning for SOC stock mapping in khat-dominated landscapes. This approach provides valuable spatial information to understand SOC variability and support sustainable land management and climate change mitigation strategies in Eastern Ethiopia. Full article

Extreme precipitation can disproportionately enhance rainfall erosivity in complex mountainous terrain, yet its spatial amplification and topographic differentiation remain poorly understood. Focusing on the Hengduan Mountains, this study evaluated three precipitation products (ChinaMet, CHM_PRE, and IMERG) against station observations and assessed their ability to capture precipitation extremes. Using the best-performing product, rainfall erosivity associated with total (PRCPTOT), heavy (R95p), and extreme (R99p) precipitation was estimated for 2005–2024, and its spatial patterns, amplification effects, topographic differentiation, and hotspots were analyzed. CHM_PRE showed the best overall performance, with a correlation coefficient (CC) of 0.83 and a Kling–Gupta efficiency (KGE) of 0.74, together with the highest probability of detection (POD = 0.95), accuracy (ACC = 0.83), and critical success index (CSI = 0.78) for extreme precipitation. Precipitation and the corresponding rainfall erosivity exhibited a pronounced southeast-to-northwest decreasing gradient. Although R95p and R99p accounted for only 9.61% and 2.43% of total precipitation, they contributed 14.84% and 4.33% of total rainfall erosivity, yielding erosivity amplification factors (AFs) of 1.52 and 1.73, respectively. This indicates a disproportionate contribution of precipitation extremes to rainfall erosivity, with stronger amplification under R99p. Rainfall erosivity also exhibited pronounced topographic differentiation, and high-level hotspots were consistently concentrated along the southeastern and southern margins. Extreme hotspots under PRCPTOT and R95p occurred at mean elevations of 2735.19–2791.92 m and mean slopes of 14.79–15.09°, whereas R99p intense hotspots occurred at a mean elevation of 2374.15 m and a mean slope of 12.28°. Strongly undulating mid-high mountains were the dominant geomorphic units within PRCPTOT and R95p extreme hotspots, while moderately and strongly undulating mid-high mountains dominated R99p intense hotspots. Moreover, hotspots became increasingly localized as precipitation extremity increased. These findings highlight the disproportionate erosive significance and spatial selectivity of precipitation extremes and provide a basis for identifying priority areas for soil and water conservation and rainfall-related hazard management in the Hengduan Mountains under climate change. Full article

Monitoring complex river networks in Myanmar is challenging due to the high spatial heterogeneity and fragmentation of surface water bodies. Accurate identification of surface water resources is therefore essential for water resource management and for improving preparedness against climate change–induced hydrological hazards. To address this problem, this study proposes an adaptive interwoven deep learning–based segmentation framework that jointly utilizes multispectral reflectance information and topographic elevation data to enhance the extraction of fragmented water bodies. The framework is designed to coordinate feature interaction across spectral, spatial, and topographic dimensions by integrating channel-wise feature recalibration and attention-guided feature modulation within the encoding–decoding architecture. Experimental results demonstrate that the proposed method outperforms several traditional water index–based approaches, conventional machine learning algorithms and deep learning models. Across five independent training runs, the proposed framework achieves an average precision of 91.0%, recall of 93.5%, and F1 score of 92.3% (95% confidence interval: 91.8–93.0), demonstrating stable performance for fragmented water-body extraction. Cross-site experiments across three within-country study areas further demonstrate the spatial transferability and robustness of the proposed framework across diverse hydrological conditions within Myanmar. Overall, the proposed approach provides a reliable solution for fragmented water body extraction under heterogeneous hydrological conditions within Myanmar. Full article

The Shendong Mining Area, located in the transition zone between the northern Loess Plateau and the Mu Us Sandy Land, is a representative ecologically fragile region and desert coal base in China. Using Google Earth Engine (GEE) and Landsat imagery from 1999 to 2024, this study constructed a long-term remote sensing ecological index (RSEI) dataset and integrated the Theil–Sen median slope estimator, Mann–Kendall test, and Hurst exponent to examine the spatiotemporal evolution, future trajectories, and multi-stage driving mechanisms of eco-environmental quality (EEQ) at the mining-area and individual-mine scales. At the mining-area scale, RSEI showed pronounced interannual fluctuations and a weak downward trend, characterized by a two-stage, wave-like trajectory with a narrowing amplitude. The mean RSEI and standard deviation decreased from 0.5669 and 0.0855 during 1999–2010 to 0.5182 and 0.0501 during 2011–2024. Moderate and good grades predominated, with lower EEQ in the mining core and higher EEQ in peripheral buffer zones. Across 13 representative mines, ecological quality generally remained moderate but exhibited spatial and stage-dependent heterogeneity, with Liuta Mine showing the greatest variability. Slight degradation was the dominant trend, although local recovery occurred, and Hurst analysis revealed marked spatial differences in future trajectories. RSEI variations in the Shendong Mining Area exhibited pronounced spatial and stage-dependent associations with high-intensity mining, climatic water–heat conditions, topographic background, and ecological governance and restoration processes. No continuous and irreversible overall decline was observed; however, without an independent non-mining control, the findings represent integrated ecological responses within the mining area rather than causal estimates of mining’s net ecological effect. Full article

Understanding habitat requirements across spatial scales is important for conserving declining migratory species such as the Cerulean warbler. This study integrates data from several forms of remote sensing platforms: automated recording units, airborne LiDAR, and satellite remote sensing, to characterize the habitat of Cerulean warbler detection sites across central Indiana. We also modeled the suitable habitat of the species across the contiguous United States. Automated recording units were deployed across four forest types, and automated classification was used to derive species detections, which were manually validated. Structural variables derived within 25 m and 50 m buffers included LiDAR-based canopy height metrics, vertical vegetation distribution, foliage height diversity, and satellite-derived Enhanced Vegetation Index (EVI). Cerulean warbler detections were identified at 14 of the 57 acoustic sensor locations. Habitat characteristics at detection and non-detection sites were compared using univariate statistical tests and logistic regression models. Habitat associations varied with spatial scale. At 25 m, detection sites had significantly lower vegetation cover within the 5–10 m height stratum, which was also the highest-ranked candidate predictor, whereas EVI was the highest-ranked predictor at 50 m. At the broad scale, occurrence records from the Global Biodiversity Information Facility (GBIF) were integrated with climatic, topographic, land-cover, and anthropogenic variables within a MaxEnt modeling framework. The model showed moderate predictive performance and identified land cover as a key predictor of habitat suitability, with deciduous forests showing the highest probability of occurrence. This study highlights how multi-modal sensor data can be integrated for biodiversity monitoring and habitat assessment. Full article

The construction of the Hainan Free Trade Port (FTP) is guided by the core philosophy of “ecological priority and green development.” To meet the practical requirements of building a “Green and Beautiful FTP,” this study constructed a kernel Normalized Difference Vegetation Index (kNDVI) dataset using Landsat series satellite imagery. By integrating methods including the Mann–Kendall (MK) trend test, Hurst exponent, and coefficient of variation (CV), an in-depth analysis was conducted on the spatiotemporal evolution characteristics and trends of growing-season vegetation in Hainan Island from 1994 to 2023. Additionally, the Extreme Gradient Boosting (XGBoost) model and the SHapley Additive exPlanations (SHAP) interpretation method were employed to quantitatively unravel the driving mechanisms of climatic factors and human activities on kNDVI variations. The results indicate that: (1) Over the past three decades, the overall kNDVI of Hainan Island has exhibited a significant upward trend, characterized spatially by an evolution pattern of “stable recovery in the central region and localized degradation along the coast.” (2) The vegetation evolution demonstrates strong persistence and is highly consistent with community stability. The high stability in the central mountainous areas stems from a superior natural background and strict ecological protection; the transition zone is jointly influenced by vegetation types and anthropogenic management; meanwhile the coastal areas exhibit significant degradation characteristics driven by high-intensity human disturbances. (3) The analysis of driving mechanisms reveals a complex control logic of “topographical foundation—human reshaping—extreme climate triggering.” Static topographical factors, such as elevation and slope, occupy an absolute dominant position; human activities, represented by rubber plantation expansion and urbanization, exert a bidirectional reshaping effect characterized by “inland greening and coastal suppression”; furthermore, extreme drought and high temperatures in the later stages of the study demonstrated a significant pulse-like impact, exacerbating the risks of short-term climatic stress. This study clarifies the core patterns of vegetation evolution in Hainan Island, validates the effectiveness of ecological policies, and provides a quantitative scientific basis for ecological conservation and sustainable development in tropical island regions. Full article

In regions where severe topographic constraints coincide with intensive human activity, the mechanisms underlying landscape ecological risk (LER) and its spatial differentiation remain poorly understood. In particular, the nonlinear responses and threshold effects arising from the combined influence of complex natural gradients, urban expansion, and policy interventions have not been adequately characterized, limiting effective regional ecological management and policy formulation. Taking the Taiyuan section of the Fen River Basin as the study area, this study constructed an LER index using land-use data for 2014, 2019, and 2024. Landscape metrics and spatial autocorrelation analyses were used to characterize the spatiotemporal evolution of LER, and a LightGBM-SHAP model with spatial block cross-validation was employed to identify the nonlinear effects of natural and socioeconomic drivers. A four-quadrant zoning framework integrating current risk state and driver sensitivity was then developed. The results showed that: (1) LER followed a fluctuating trajectory, rising from 2014 to 2019 and declining from 2019 to 2024, with evident spatial differentiation. Low- and relatively low-risk zones dominated about 71% of the area, while medium- to high-risk zones clustered mainly in the northeast, south, and parts of the northwest; high-risk agglomerations gradually contracted. (2) LER was driven by both natural and socioeconomic factors, with natural factors playing the stronger role. Slope, NDVI, elevation, and GDP were the key drivers. (3) The effects of these drivers were strongly nonlinear: slope increased risk at 5–13° but reduced it above 13°; NDVI displayed an inverted U-shaped relationship, with the strongest positive contribution at 0.60–0.75.; and elevation shifted from a positive to negative contribution near 1200 m. Based on these results, the framework coupling risk state and driver sensitivity delineated differentiated management units, providing fine-scale spatial guidance for ecological protection, restoration, and development control in topographically constrained river-valley basins. Full article

Improving the flood adaptability of urban waterfront spaces is an essential entry point for enhancing regional stormwater regulation capacity, scientifically preventing flood disasters, and stabilizing urban water security. Existing flood adaptability assessments mostly rely on single weighting methods and individual evaluation models, inevitably causing systematic bias and low result robustness. Against this limitation, this study integrates remote sensing intelligent interpretation, spatiotemporal landscape pattern analysis, and multi-criteria decision theory to construct a comprehensive flood adaptability evaluation system under the pressure–state–response (PSR) framework. Innovatively, a game-theoretic combined weighting scheme integrating the entropy weight method, CRITIC method, and standard deviation method is proposed, and three complementary models including TOPSIS, VIKOR, and EDAS are coupled for cross-verification evaluation, which effectively improves the objectivity and robustness of spatial flood adaptability quantification. Taking Anqing City as a typical case, this study adopts Sentinel-2 time-series remote sensing images from 2016 to 2023 and applies an optimized random forest algorithm to automatically classify land cover. Five underlying surface types, including water bodies, vegetation, farmland, built-up areas, and bare land, are accurately extracted with an overall classification accuracy of around 90% for most years. Core landscape metrics such as Shannon’s diversity index and patch density are selected to systematically analyze the spatiotemporal differentiation characteristics of waterfront landscape patterns during the study period. The results indicate the obvious spatial heterogeneity of flood adaptability in Anqing City. Yingjiang District and Yuexi County present high comprehensive flood adaptability, while Wangjiang County and Huaining County show relatively low performance. Urban areas gain strong flood resistance from complete disaster prevention infrastructures and economic resilience; mountainous areas possess natural advantages in flood retention and drainage due to high vegetation coverage and topographic relief; by contrast, plain districts are severely restricted by low-lying terrain and insufficient drainage systems, resulting in prominent flood vulnerability. The proposed method is helpful for providing reliable scientific support for waterfront landscape optimization, zoned flood disaster management, and resilient water space planning in riverine cities. Full article

Accurate prediction of plough layer thickness (PLT) in cropland is essential for soil quality assessment and sustainable land management, yet regional-scale PLT mapping remains challenging because PLT is a subsurface structural attribute that cannot be directly retrieved from surface spectral signals. This study developed an interpretable framework for province-scale mapping of cropland plough layer thickness (PLT) in Hubei Province, China, by integrating multi-source remote sensing observations, including Landsat 8 optical spectral bands, Sentinel-1 SAR backscatter, and crop dynamic spectral response metrics derived from multi-year Enhanced Vegetation Index (EVI) time series, together with topographic, climatic, soil physicochemical, and land-use variables. A total of 1926 cropland soil samples were used to train and validate random forest (RF), extreme gradient boosting (XGBoost), and Cubist models, while prediction uncertainty was quantified using 90% prediction intervals. The relative contributions of different environmental variable groups were assessed, and Shapley Additive Explanations (SHAP) were used to interpret key predictors. The all-variable scenario achieved the best overall performance, with RF showing the highest accuracy (R² = 0.46; RMSE = 3.12 cm) and the narrowest prediction interval. Climatic and topographic factors dominated PLT spatial variability, whereas other variable groups provided complementary predictive information. These findings demonstrate the potential of integrating multi-source environmental data and interpretable machine learning for regional PLT mapping, and the mapped distribution of cropland PLT provides a spatial basis for cropland quality assessment and targeted soil management, although further improvements will require spatially explicit agricultural management information and more direct PLT-related predictors. Full article

Accurate estimation of forest tree species diversity at regional scales is essential for biodiversity monitoring, forest resource management, and ecological conservation. Because tree species differ in canopy spectral responses and phenological dynamics, multi-season remote sensing observations can provide critical information for characterizing species composition and diversity patterns. However, the potential contribution of seasonal image features to improving remote-sensing-based tree species diversity estimation has often been insufficiently considered. In this study, the Yichun forest region in Heilongjiang Province, northeastern China, was selected as the study area. Sentinel-1, Sentinel-2, and topographic data were integrated to extract multi-seasonal spectral, vegetation index, texture, radar, and topographic features. The Boruta algorithm was used for feature selection, and random forest (RF), extreme gradient boosting (XGBoost), k-nearest neighbor (KNN), support vector regression (SVR), Bayesian regularized neural network (BRNN), and Stacking ensemble learning were developed to estimate and map Richness, Shannon, and Gini–Simpson indices. The results showed that: (1) Sentinel-2 optical features were the primary information source for tree species diversity estimation, topographic factors further improved model performance, and Sentinel-1 radar features mainly provided complementary structural information; (2) seasonal remote sensing features differed in their predictive ability, with Richness performing better in spring, while Shannon and Gini–Simpson achieved higher accuracy in winter. The four-season fusion scenario produced the highest accuracy for all three indices, with optimal R² values of 0.51, 0.63, and 0.57, respectively; (3) the Stacking ensemble generally improved estimation accuracy and model stability, although the optimal model differed among diversity indices, with Stacking, SVR, and RF performing best for Richness, Shannon, and Gini–Simpson, respectively; and (4) summer Sentinel-2 NDVI, GNDVI, and NDWI contributed strongly to all three indices, elevation was particularly important for Richness, and winter vegetation indices and autumn red-edge bands and texture features were also informative for Shannon and Gini–Simpson. These findings indicate that integrating multi-seasonal remote sensing features and multi-source data using machine learning models can effectively improve forest tree species diversity estimation, providing technical support for regional forest biodiversity monitoring and precision forest management. Full article