The near field high cloud characteristics of stratospheric airships are closely linked to stratospheric flight safety. Existing methods for cloud detection based on ground-based， airborne， and space-based systems are introduced， and their respective advantages and limitations in high cloud detection are analyzed and compared. To address the shortcomings of current detection techniques， two cloud detection methods based on stratospheric airships are proposed： near field detection and tethered payload-based detection. The development status of meteorological payloads and tethered equipment is also reviewed. To ensure the safety of stratospheric airship operations， a collaborative cloud detection system integrating ground， airborne， and space-based components is required， providing comprehensive cloud characteristic data to support the stability and reliability of the airship platforms. Moreover， in light of existing data gaps and technological limitations in high cloud detection in China， the development of meteorological payloads and related technologies should be accelerated to enhance the accuracy and reliability of meteorological sensing equipment， thereby improving the country’s independent technological capabilities and data collection capacity.
Accurately obtaining long-term evolution information of coastal wetlands is of great significance for ecological protection and resource management. Taking the buffer zone and core area of the Yancheng Rare Birds Nature Reserve as the study area， this research monitored the wetland evolution process and explored its spatiotemporal heterogeneity by employing Landsat remote sensing data from 1996 to 2024 on the Google Earth Engine cloud platform. An optimized method integrating the Continuous Change Detection and Classification （CCDC） model， vegetation phenological features， and Markov Random Field （MRF） post-processing （CCDC-MRF） was developed. The results show that： （1） The proposed method achieved high classification accuracy， with an average overall accuracy exceeding 86% and a Kappa coefficient over 0.85 for wetland identification， effectively suppressing salt-and-pepper noise. （2） Wetland evolution in the study area exhibited stage-specific characteristics， following a sequence of "reclamation - stabilization - returning ponds to wetlands - Spartina alterniflora control". Changes were more pronounced in the buffer zone， where artificial wetlands largely transitioned to reed marshes， while the core area experienced smaller changes dominated by natural vegetation succession， with recent changes primarily driven by S. alterniflora control actions. （3） The mechanisms governing water body and vegetation gain/loss showed spatial differences： water body dynamics in the buffer zone were mainly influenced by the conversion of aquaculture ponds back to wetlands， whereas large-scale vegetation loss in the core area was directly related to S. alterniflora eradication. The study demonstrates that the optimized continuous change detection algorithm enhances the accuracy of coastal wetland identification. The differential evolution characteristics of coastal wetlands in the Yancheng Reserve， shaped by policy regulation， natural processes， and S. alterniflora invasion， deepen the understanding of coastal wetland evolution mechanisms and can provide technical support for precise monitoring and zonal management of coastal wetlands.
Aiming at the challenges of landscape fragmentation， blurred boundaries， and limited discriminability caused by sample scarcity in coastal wetland semantic change detection， this paper proposes a language-guided spatial topology-aware network， termed GeoLingua. First， a patch-based Graph Convolutional Network （GCN） is utilized to aggregate non-local contextual information， which enhances the modeling capability of semantic consistency in fragmented regions. Second， a boundary-guided attention mechanism is introduced to improve the sensitivity and recognition accuracy for fuzzy boundaries. Furthermore， a contextualized semantic guidance module is designed to construct dynamic semantic constraints that guide visual feature learning， effectively compensating for the poor discriminability of ground objects under insufficient training data. Empirical analysis conducted on the Yellow River Delta coastal wetland dataset demonstrates that the proposed method achieves an Intersection over Union （IoU） exceeding 97% for the change detection task and a mean IoU （mIoU） exceeding 93% for the semantic segmentation task. The results indicate that GeoLingua exhibits superior semantic change perception in complex ecological environments， validating the effectiveness of integrating structural modeling with semantic constraints.
Wetlands are important habitats for biodiversity， with key ecological functions such as regulating hydrological cycles， purifying water quality， and mitigating climate change. However， the global wetland area has decreased by 35% since 1970， and the loss and degradation of wetlands are still ongoing. This article is based on the Google Earth Engine （GEE） platform， using Sentinel-2 high-resolution remote sensing images， combined with phenological assisted random forest algorithm and phenological feature extraction method， to conduct a detailed analysis of the dynamic changes in vegetation cover in Xixi Wetland， Hangzhou from 2019 to 2024. The research results indicate that the vegetation coverage of Xixi Wetland has undergone significant spatiotemporal changes， with the woody wetland area decreasing from 61.74% in 2019 to 57.48% in 2024， while the herbaceous wetland area increased from 15.31% to 22.66%， and the water body area remained relatively stable. The random forest wetland classification model based on GEE platform proposed in this article shows good accuracy （OA>0.8）， and the introduction of phenological features significantly improves the accuracy. It can provide technical support for long-term monitoring and management of wetland ecosystems， and provide reference for global wetland protection.
Changes in vegetation phenology significantly impact the carbon cycle， water cycle， and energy exchange within ecosystems， making it a critical area of global ecological research. Due to its unique geographical location， the Tumen River Basin is sensitive to climate change and serves as one of China’s important ecological habitats and major rice-producing areas. Therefore， research on vegetation phenology in this basin can provide a scientific basis for the rational conservation of vegetation resources and the scientific management of agricultural production. Based on the MOD13Q1 NDVI dataset from 2001 to 2023， this study employed Savitzky-Golay （S-G） filtering， the dynamic threshold method， univariate linear regression trend analysis， and partial correlation analysis to examine the dynamic changes in vegetation phenology in the Tumen River Basin and the influence of various factors （temperature， precipitation， land surface temperature， and elevation） on these phenological shifts. The results indicated that the Start of Season （SOS）， End of Season （EOS）， and Length of Season （LOS） for vegetation in the Tumen River Basin are concentrated on days of 105~135， 280~300， and145~195， respectively. Additionally， from 2001 to 2023， the SOS for all vegetation types in the basin showed an advancing trend， with an overall rate of 2.93 days per decade. The EOS was generally delayed at a rate of 0.77 days per decade， with broadleaf forests， coniferous forests， and croplands exhibiting delayed EOS， while grasslands， shrublands， and wetlands showed an advanced EOS. The LOS extended at a rate of 3.87 days per decade， with only wetland vegetation experiencing a shortened LOS. The influence of different factors on vegetation phenology revealed that SOS was primarily affected by January temperature， March precipitation， and February land surface temperature， while EOS was mainly influenced by September temperature and October precipitation. Overall， SOS and LOS exhibited different trends with increasing elevation during 0~600 m， 600~1 500 m and above 1 500 m， whereas EOS was delayed as elevation increased. For different vegetation types， phenological patterns varied with changes in elevation. During 2001—2023， the vegetation growth period in the Tumen River Basin was prolonged. Temperature acted as the key driving factor for vegetation phenological changes in the basin， and the vegetation phenology showed significant elevational differentiation. Based on the above findings， climate-adaptive planting strategies should be formulated and differentiated resource management along elevation gradients should be implemented in the future.
Vegetation phenology exhibits regular variations along urban-rural gradients determined by population distribution. To investigate changes in vegetation phenology and its natural influencing mechanisms across these gradients， this study reconstructed time series using MOD13Q1 Enhanced Vegetation Index （EVI） data from 2001 to 2020. Threshold method was employed to extract three phenological parameters： Start of Season （SOS）， End of Season （EOS）， and Length of Season （LOS）. This revealed patterns of vegetation phenology across a ten-tier urban-rural gradient determined by population density. Subsequently， partial correlation analysis was used to investigate influencing factors. Results indicate： （1） Along the urban-rural gradient， SOS advances by an average of 3.08 days per tier， EOS delays by 3.31 days per tier， and LOS extends by 6.46 days per tier. （2） Rising surface temperatures advance SOS and delay EOS； increased precipitation and radiation delay SOS and advance EOS. （3） For each increment in the urban-rural gradient， the maximum absolute value of the partial correlation coefficients between SOS， EOS and influencing factors decreased by approximately 0.02 on average. These findings show that vegetation phenology undergoes systematic changes along the urban-rural gradient， primarily driven by associated influencing factors. This conclusion is extra evidence that human activities impact ecosystems by altering the natural environment.
Existing PhenoCam studies are limited by their focus on single-landscape monitoring and lack of rigorous comparative analyses. To resolve this， this study uses high-frequency PhenoCam Network imagery（2014-2023） to compare the phenological traits and environmental response mechanisms of three Vegetation Functional Types （VFTs）： deciduous broadleaf forests， evergreen needleleaf forests， and grasslands. We extracted the Green Chromatic Coordinate （GCC） by defining Regions of Interest （ROIs） and derived key phenological parameters using a double logistic function fitting method. Furthermore， a multivariate regression model incorporating meteorological data was constructed to analyze the influence of environmental factors on vegetation phenology. The results indicate that： （1） Deciduous broadleaf forests exhibited the largest seasonal amplitude in GCC with high interannual stability， while evergreen needleleaf forests showed subdued seasonal dynamics； conversely， grasslands displayed the most significant interannual fluctuations in phenological timing. （2） Multi-factor analysis revealed that the ecosystems in the study area are primarily “energy-limited”. After decoupling seasonal autocorrelation， air temperature remained the dominant driver of growth for all vegetation types （standardized coefficient β > 0.74）， whereas water availability factors showed no significant effect. （3） Grasslands exhibited a significant positive response to shortwave radiation （β = 0.24）， suggesting that herbaceous communities respond more directly to light resources compared to forest ecosystems. This study validates the feasibility of synchronously extracting multi-vegetation phenology from single-field-of-view imagery. It highlights the differentiated adaptive strategies of distinct vegetation types under identical climatic conditions， providing essential near-surface observational support for the refinement and validation of ecosystem model parameters.
Residual Trend Method （RESTREND） is one of the key approaches for attributing vegetation changes. However， RESTREND has long been associated with considerable uncertainty due to the collinearity between climatic factors and human activities （referred to as CH collinearity）. To address the CH collinearity issue， this study introduces the concept of a spatiotemporal benchmark and an iterative scheme for selecting spatiotemporal benchmarks， thereby establishing a Spatiotemporal Benchmark RESTREND framework （STR-RESTREND）. This framework was applied to attribute vegetation changes across China. The results indicate that：①From 1982 to 2022， vegetation in China generally showed an increasing trend， with the most rapid growth occurring on the Loess Plateau and its surrounding areas. In contrast， areas with faster vegetation degradation were mainly concentrated in economically developed major urban agglomerations.②Climate change had a promoting effect on vegetation in regions primarily located south of the Qinling–Huaihe Line， whereas areas to the north were largely characterized as having no significant climatic impact；Human activities promoted vegetation growth across 68.7% of China's land area， whereas they exerted an inhibitory effect in 18.2% of the area， with significant inhibition observed in only 0.7% of the total. ③ The impact of human activities on vegetation change was more profound than that of climate change across China. Notably， in approximately 90% of the pixels within both improved and degraded vegetation regions， the contribution rate of human activities exceeded that of climatic factors. ④ The proposed STR-RESTREND method effectively resolved the CH collinearity problem， which prevents the conventional RESTREND approach from misestimating the role of climate drivers. The study clarifies the relative roles of climate change and human activities in vegetation change， providing an important scientific basis for evaluating the effectiveness of ecological engineering and optimizing strategies for ecological conservation and restoration.
For a long time， satellite data from the red and near-infrared bands have been central to vegetation index calculations， providing critical support for global vegetation monitoring. However， directly integrating multi-source satellite data without prior intercalibration may introduce artificial errors， distort vegetation indices， generate false trends and ultimately reduce monitoring reliability. This study employs Simultaneous Nadir Overpass （SNO） cross-calibration technology to calibrate FY-3D/MERSI-II， using the radiation-validated AQUA/MODIS as the reference. Considering that traditional MERSI-II and MODIS SNO intersection points are predominantly distributed in high-latitude snow/ice regions， which have high reflectance but narrow dynamic range， the extended SNOx points are also derived by including MERSI-II and MODIS orbital intersections in low-latitude regions. Simultaneously， integrating SNO and SNOx crosspoint data significantly expands sample representativeness and diversity. High-precision calibration coefficients for the visible and near-infrared bands of MERSI-Ⅱ were generated from matched MERSI-II and MODIS data in 2019. These coefficients are then applied to recalibrate the reflective bands of MERSI-II in July 2020， and standard algorithms are then used to generate Normalized Difference Vegetation Index （NDVI） products， aiming for quantitative consistency with MODIS NDVI. Results demonstrate the 2019 intercalibration model exhibits excellent accuracy （R²>0.99）， sensor biases in red/near-infrared bands are effectively reduced， and corrected MERSI-II NDVI shows strong consistency with MODIS NDVI at 16-day and monthly scales and with correlation coefficients greater than 0.9. This study provides robust data foundations and technical support for collaborative and consistent processing of global-scale remote sensing data.
Vegetation dynamics in the Lijiang River Basin are a key indicator of regional ecological security， reflecting responses to both natural environmental changes and human activities. Based on a continuous NDVI time series， this study integrates trend analysis， the Hurst exponent， and the Geodetector model to establish a multi-scale analytical framework of “trend identification–sustainability assessment–driving mechanism analysis”， aiming to systematically reveal the spatiotemporal evolution of vegetation and its driving mechanisms.The results show that： （1） NDVI exhibits a significant fluctuating upward trend， with the annual mean increasing from 0.586 to 0.675 at an annual rate of approximately 0.011； （2） spatially， vegetation in the central–northern karst hilly regions has significantly recovered， while localized degradation occurs in the southern region under human disturbance， leading to enhanced spatial heterogeneity； （3） the Hurst exponent shifts from anti-persistence to persistence， indicating a sustained improvement trend in the future； （4） topographic factors dominate vegetation dynamics， temperature influence has strengthened， precipitation plays a relatively weak role， and human activities show limited overall contribution but stage-wise intensification during key development periods； （5） multi-factor interactions exhibit significant nonlinear enhancement effects.These findings suggest that vegetation evolution is primarily governed by natural factors and modulated by human activities， with nonlinear coupling among multiple drivers jointly shaping its spatiotemporal differentiation， providing a scientific basis for ecological zoning and differentiated spatial management.
Under the combined influence of climate change and human activities， vegetation coverage in Shandong Province has undergone significant changes in recent years. Investigating the spatiotemporal characteristics of vegetation dynamics and their driving factors is crucial for regional planning and environmental protection. Based on MODIS NDVI data from 2003 to 2020， this study employed methods including Theil-Sen Median trend analysis， Mann-Kendall significance test， partial correlation analysis， and residual analysis to explore the spatiotemporal variation of NDVI and its responses to climate change and human activities in Shandong Province. The results showed that： （1） From 2003 to 2020， vegetation NDVI in Shandong Province was dominated by mild growth （0.02/10 a < slope < 0.05/10 a） and stable trends （-0.02/10 a < slope < 0.02/10 a）， with an average annual growth rate of approximately 0.03/10 a. The most significant increase occurred in northwestern Shandong， while declining NDVI areas were mainly concentrated in urban expansion zones. （2） Both mean temperature and precipitation exhibited positive correlations with NDVI， with temperature exerting a stronger influence than precipitation. （3） Climate change and human activities contributed 43.71% and 56.29% to vegetation NDVI changes， respectively. At the pixel scale， synergistic effects of climate change and human activities drove vegetation growth in the largest proportion of the study area， accounting for 81.57% of the total area.
To address the common limitations of traditional Interferometric Synthetic Aperture Radar （InSAR） sub-canopy topography inversion methods—such as insufficient characterization of scattering mechanisms， strong model nonlinearity， and poor parameter convergence—an improved approach based on the Interferometric Water Cloud Model （IWCM） is proposed. The method simultaneously accounts for both volume and surface scattering effects， incorporates the Normalized Difference Vegetation Index （NDVI） to impose prior constraints on model parameters， utilizes TanDEM-X dual-baseline InSAR data to increase observational dimensionality， and adopts a two-step nonlinear least-squares optimization strategy to enhance model convergence performance. The Genhe forest region in Inner Mongolia is selected as the study area， and the inversion results are validated using airborne Light Detection and Ranging （LiDAR） data. The RMSE of the original InSAR DEM is 6.43 m， while the RMSE of the sub-canopy topography obtained after removing the scattering phase center height based on the IWCM-based method is reduced to 3.81 m， representing an improvement of 40.75%. The experimental results demonstrate that the proposed method substantially improves both the accuracy and stability of sub-canopy topography inversion， offering new insights for dual- and multi-baseline InSAR-based sub-canopy topoyraphy inversion.
Bathymetric information in coastal shallow-water regions is essential for coastal geomorphological mapping， port channel maintenance， and marine ecological monitoring. However， conventional vessel-based surveys are limited by shallow-water constraints and high operational costs， making wide-area measurements difficult to achieve. To address the limitations of traditional empirical models that are highly sensitive to optical variability， the lack of physical constraints in machine learning approaches， and the instability of deep learning models in complex turbid waters， proposes an enhanced deep learning bathymetry inversion framework， DRT-OptiNet， which integrates optical radiative transfer mechanisms. Built upon a U-Net encoder–decoder architecture， the model incorporates optical-parameter regression and a differentiable radiative transfer layer to establish a physically consistent mapping between spectral reflectance and water depth. In addition， residual connections， wide-bottleneck modules， and batch normalization are introduced to enhance feature representation and improve model stability and generalization. Experiments conducted in both clear-water environments （Ganquan Island） and highly turbid waters （North Channel of Shanghai Port） demonstrate that DRT-OptiNet significantly outperforms traditional empirical and machine learning models under varying optical conditions. In the Ganquan Island region， the model achieves R² = 0.98， RMSE = 0.54 m， and MRE = 6.51%， accurately reconstructing continuous depth gradients and fine-scale seabed features. Even in the optically complex North Channel of Shanghai Port， the model maintains high overall accuracy （R² = 0.84， RMSE = 1.72 m） and effectively captures the deep-trough and shoal structures of the main channel. Overall， DRT-OptiNet maintains strong accuracy and stability across diverse water types， offering an effective and physically interpretable solution for shallow-water bathymetry inversion.
Surface deformation such as subsidence， ground fissures and landslides induced by coal mining in northern Shaanxi have significantly impacted the ecological environment and safety of residents in mining areas. In order to comprehensively understand the characteristics of surface deformation and corresponding geological hazards in the mining area， Shennan mining area in Shenmu was selected as study area. The characteristics and formation processes of surface deformation were analyzed by employing the surface deformation monitoring results obtained from SBAS-InSAR from March 12， 2017 to September 18， 2022. Additionally， the development process and influencing factors of surface deformation were investigated based on the cumulative subsidence value along the profile and the curve of time-series cumulative subsidence. Furthermore， drone surveys confirmed that surface subsidence has caused other geological hazards， such as ground fissures and landslides. The development trends and related factors of ground fissures and landslides were discussed by utilizing the displacement monitoring data from ten monitoring points between December 2021 and December 2022. The results indicate that surface deformation is primarily resulted from mining activities and influenced by the thickness of overlying strata. Strong spatial extensibility and temporal continuity exists between surface deformation and mining activities， which also exhibit a spatiotemporal correlation. The study provides a reference for understanding the influence of mining on surface deformation and their spatiotemporal correlation.
The Yellow River Basin is abundant in water resources， and the water conservancy hubs within this region significantly influence the living conditions of residents and the economic development of the area. However， the geological environment in the Yellow River Basin is fragile， and many dams have been in operation for extended periods. Conducting health assessments of these dams is of great value and importance. This article employs spaceborne Synthetic Aperture Radar （InSAR） technology to investigate and analyze the spatiotemporal deformation characteristics of typical dams in the Yellow River Basin. To achieve high-precision measurements of dam deformation， this study utilizes InSAR phase unwrapping assisted by network optimization. The research focuses on the Gongboxia Dam in the upper reaches of the Yellow River and the Xiaolangdi Dam in the middle reaches. The experimental data is derived from Sentinel-1A， covering the period from 2017 to 2024. The findings indicate that the deformation rate in certain areas of the Gongboxia hydropower station dam exceeds 2 millimeters per year， while some areas of the Xiaolangdi water conservancy dam exhibit a deformation rate of 20 millimeters per year. The periodicity of the dam deformation time series correlates with fluctuations in the water level of the reservoir area. The results of this study provide essential support for the safe operation of water conservancy hubs in the Yellow River Basin and offer valuable insights for the health assessment of water conservancy dams in other regions of the Yellow River Basin.
A method for detecting collapsed buildings based on pre- and post-earthquake remote sensing imagery， which does not rely on single post-event data or elevation change information， is of great significance for timely emergency response. The polarization and scattering features contained in post-earthquake SAR imagery can effectively reflect the changes in microwave scattering characteristics caused by building collapses. However， collapsed buildings exhibit lower scattering intensity than intact ones， making them susceptible to interference from surrounding undamaged structures during detection. Furthermore， given the limitations of SAR imagery in representing spatial details， incorporating the rich spatial information from optical imagery is crucial for improving the detection accuracy of collapsed buildings.To this end， this paper proposes a collapsed building detection method， named PreOPT-SAR， which fuses visual features from pre-earthquake high-resolution optical imagery with polarimetric scattering features from post-earthquake SAR imagery. To address the registration challenge， an Optical-SAR feature point matching strategy based on Inscribed Centers and Boundary Points （OSICBP） is designed. This is combined with an improved region-growing algorithm incorporating a Boundary Compensation Factor and a Direction Compensation factor （BCFDC） to uniformly extract building objects from both types of imagery. Moreover， to mitigate the interference from adjacent intact buildings on the identification of collapsed areas， a Multi-factorial Polarimetric Decomposition method based On Yamaguchi decomposition （MPDOY） is introduced to enhance the characterization capabilities of the scattering components. Experimental results on multiple real-world datasets demonstrate the superior performance of the PreOPT-SAR method， achieving an Overall Accuracy （OA） consistently above 83.86% and significantly outperforming several state-of-the-art comparison methods.
Offshore oil and gas platforms are important equipment for the exploration， exploitation and transportation of marine oil and gas resources. Monitoring offshore oil and gas platforms is of great significance for oil spill monitoring， greenhouse gas emission estimation， and marine ecological protection. Satellite remote sensing imagery have been widely used to monitor offshore oil and gas platforms. However， the “dynamic and static separation” strategy adopted in existing studies ignores the situation of ships passing by the vicinity of offshore oil and gas platforms， and additionally it is difficult to reflect the short-term changes of mobile offshore oil and gas platforms. To this end， this study proposed a novel dynamic and static separation strategy that considers the spatiotemporal overlap of targets， and monitored dynamics of offshore oil and gas platforms using time-series Sentinel-1 synthetic Aperture Radar （SAR） imagery. Firstly， the time-series Sentinel-1 remote sensing imagery and the Faster R-CNN deep learning model were used to extract bright targets at sea， and the spatial overlap of the target bounding box in the time-series images was calculated. The moving targets were removed by combining the residence time， and the target area threshold was set to remove small targets and to capture the extraction results of offshore oil and gas platforms. The dynamics of oil and gas platforms in typical areas were further monitored according to the appearance time and disappearance time of the platforms. This method is applied to monitor the dynamics of offshore oil and gas platforms in the central area of Weizhou oilfield during from 2015 to 2025. The extraction results were verified using Gaofen series satellite imagery and Sentinel-2 satellite imagery， yielding an extraction accuracy of 98.6%. A total of 80 offshore oil and gas platforms were identified in the study area， including 13 fixed platforms that existed throughout the monitoring period， 4 newly built platforms， and 63 mobile platforms that changed position during the monitoring period. The residence time of the 54 changed platforms was ≤30 days， indicating that offshore oil and gas platforms in this area are active， there are many short-term drilling and exploration operations. In summary， our method can accurately monitor the dynamics of offshore oil and gas platforms， and especially can capture the short-term dynamics of mobile offshore oil and gas platforms.
Yak is a key animal husbandry resource in the alpine ecosystem of the Qinghai-Tibet Plateau. The acquisition of its quantity and distribution information is of great significance for grassland resource management， grazing optimization and ecological monitoring. The traditional yak identification mainly relies on manual investigation or field remote sensing point interpretation， which has the limitations of heavy workload， lagging update and limited space coverage， and is difficult to meet the needs of large-scale and automated yak positioning. In order to realize the rapid and accurate identification of yaks， this paper constructs the multi temporal yak remote sensing data set of Sibu ranch based on the multi spectral remote sensing images of GF-2 and GF-7， uses the improved YOLOv8 neural network model to carry out target detection， and introduces four channel input to enhance the spectral discrimination ability. Aiming at the similarity between yak and static objects in morphology and spectrum in remote sensing images， a static target elimination method based on spatial constraints， radiation dark spot detection， confidence filtering and historical time comparison is designed to effectively suppress false detection and missed detection. The results showed that the average accuracy of the model was 0.92 and the average recall rate was 0.74； The average accuracy of the images in the dry season was the same as 0.92， and the average recall rate was 0.75. This study verified the effectiveness of the deep learning method combined with multi-source remote sensing and static target screening mechanism in the task of yak recognition， and provided ideas and methods support for species remote sensing recognition under similar weak targets and high background interference.
Qinghai Lake is the largest brackish lake in China and an important water resource for maintaining ecological security in the northeastern part of the Qinghai-Tibetan Plateau. In recent years， under the influence of climate warming and humidification， Cladophora blooms have appeared in some regions near the shore in the western and northern of the lake. Timely and accurate acquisition of the spatial distribution range of Cladophora blooms is crucial for the effective protection of the water environment of Qinghai Lake and the scientific management of the ecological problems caused by Cladophora blooms. Based on Sentinel-2 MSI remote sensing images and UAV orthophotos， the distribution information of Cladophora bloom was extracted by combining nine algal spectral indices with the typical region on the north side of the Buha River estuary as the sample area and comparing the extraction effect and applicability. On this basis， the outstanding spectral indices are used to extract the Cladophora blooms in the whole Qinghai Lake， and the optimal indices are selected to invert the distribution characteristics of the Cladophora blooms in Qinghai Lake in 2019 and analyze the change rule. The results show that： the six spectral indices， including SB， BD， NDVI， MCI， SABI and VB-FAH， have better extraction effects in typical regions （Kappa>0.9）， with lower misclassification and omission rates； the VB-FAH index has the optimal extraction effect in the whole lake （Kappa=0.88）， followed by the NDVI index （Kappa=0.86）， and the SABI and BD index had comparable extraction effects （Kappa=0.79）， and MCI index had the worst extraction effect （Kappa=0.67）. The maximum area of the Cladophora blooms in Qinghai Lake in 2019 was 4.10 km2， and the minimum area was 1.79 km2； the area with the largest distribution of Cladophora blooms was the north side of the mouth of the Buha River， followed by the south side of the mouth of the Buha River， and the area of Cladophora blooms in the other regions was relatively small. Spatially， the Cladophora blooms migrated outward with the expansion of the Qinghai Lake area by wind force， and was distributed near the lake shore in the form of strips or patches. The results identified the VB-FAH index as the optimal method for extracting Cladophora blooms， providing a reference for long-term monitoring and extraction of Cladophora bloom distribution information.
In recent years， with the accelerating pace of urbanization， the Impervious Surface Area （ISA） in Ji′nan has expanded rapidly， intensifying the urban heat island effect. Using multi-source data from Sentinel-2 MSI， Landsat-8 TIRS， and MODIS， this study extracted ISA spatial patterns from 2015 to 2021 and derived Impervious Surface Percentage （ISP） through linear spectral mixture analysis. From the perspective of landscape pattern metrics， the relationship between ISA and Land Surface Temperature （LST） was analyzed， and functional relationships between graded ISP and LST were explored. The results reveal that LST exhibited distinct spatiotemporal characteristics， with high values concentrated in the eastern region in June and in the city center in September， and showed an overall declining trend over time. ISP was unevenly distributed， with values generally above 0.6 in the urban core and below 0.2 in peripheral areas. The ISA exhibited a U-shaped density pattern， with both low- and high-density ISA areas exceeding 200 km²， significantly larger than the intermediate classes. ISP was significantly correlated with LST （r = 0.657~0.734）， and polynomial regression provided the best fit， with the highest coefficient of determination （R²） reaching 0.904 in 2018. Landscape pattern analysis showed that patch density and aggregation index were positively correlated with LST， while the landscape shape index was negatively correlated. These findings provide scientific support for urban thermal environment regulation and spatial planning.
Guizhou Province is one of China’s major tobacco-producing regions； however， in its karst mountainous areas， persistent cloud cover and fragmented land surfaces significantly limit the accuracy of tobacco cultivation mapping using remote sensing. To address this issue， Jinsha County of Bijie City， Guizhou Province， was selected as the study area. Based on the Google Earth Engine platform， Sentinel-1 synthetic aperture radar imagery and Sentinel-2 multispectral data were fused using the HSV-PCA method to construct a multi-source remote sensing feature dataset. Tobacco cultivation areas were extracted using Random Forest， Maximum Likelihood Classification， Support Vector Machine， Neural Network， and Object-Oriented methods， and their classification accuracies were comparatively evaluated. The results indicate that Sentinel-1/2 data fusion effectively mitigates the influence of cloud interference and surface fragmentation， leading to a substantial improvement in classification accuracy. Under the fused data condition， the Neural Network method achieved the best performance， with a producer’s accuracy of 95.10%， an overall accuracy of 98.96%， a Kappa coefficient of 0.972， and a user’s accuracy of 97.22%. Compared with official annual tobacco planting statistics for Jinsha County， the area difference of the fused classification result was 4.92%. These results demonstrate that HSV-PCA based fusion of Sentinel-1 and Sentinel-2 imagery enables high-accuracy extraction of tobacco cultivation areas in karst mountainous regions.
