Aboveground biomass of winter wheat is an important physiological indicator reflecting crop growth status and yield potential. Accurate and efficient monitoring of biomass is essential for implementing precision agricultural management， optimizing fertilization and irrigation strategies， and ensuring food security. In this study， a field experiment was conducted in the southern part of Gaocheng District， Shijiazhuang， Hebei Province， China. The aboveground biomass and canopy hyperspectral reflectance were measured at regreening stage， jointing stage， and grain filling stage of winter wheat. Sensitive wavelengths significantly correlated with aboveground biomass were preliminarily screened from the original spectra， multiplicative scatter correction spectra， First Derivative（FD） spectra， and continuum-removed spectra of winter wheat at different growth stages using Pearson correlation analysis. Based on this， feature band selection was performed using the Competitive Adaptive Reweighted Sampling （CARS） algorithm and the Successive Projections Algorithm （SPA）， and biomass estimation models were then constructed by integrating the selected features with three machine learning methods： Partial Least Squares Regression （PLSR）， Gaussian Process Regression （GPR）， and Support Vector Machine （SVM）. The results indicated that： （1） Spectral data processed using the first derivative method overall outperformed other spectral preprocessing approaches in both correlation analysis and machine learning modeling. （2） Both CARS algorithm and SPA effectively eliminated redundant information in hyperspectral data， reducing data dimensionality. Among them， SPA showed more prominent performance in simplifying bands， but the model estimation accuracy based on CARS algorithm was higher than that of SPA. The highly sensitive feature bands selected by both algorithms， which were mainly concentrated in red-edge and near-infrared spectral regions， were strongly related to aboveground biomass. （3） At different growth stages， the FD-CARS-GPR model demonstrated superior overall performance， exhibiting good adaptability and robustness. Particularly， the model achieved the highest estimation accuracy at jointing stage， with validation set R² of 0.802， RMSE of 50.97 g/m2， and RPD of 2.30， followed by regreening stage， and the lowest at filling stage. The aboveground biomass estimation model for winter wheat， developed based on first derivative spectra in combination with the CARS algorithm and constructed using GPR， demonstrated high accuracy and stability. It is suitable for quantitative estimation of aboveground biomass across different growth stages of winter wheat， providing reliable technical support for growth monitoring and precision agricultural management.
Timely and precise mapping of imitate wild Chinese medicinal materials is of importance for fostering local economic development and safeguarding traditional Chinese medicine resources. To harness the potential of integrating GaoFen series and Sentinel-2 remote sensing imagery for mapping these materials， an innovative approach was adopted. This involved fusing the Segment Anything Model （SAM） and the Simple Non-Iterative Clustering （SNIC） model to achieve high-precision extraction of field boundaries and leveraging long-term time series data of the Normalized Difference Vegetation Index （NDVI） to constructed identification features from both inter-annual and intra-annual perspectives. The findings reveal that fusing the segmentation results of the SAM and SNIC models enables high-precision identification of cultivated land boundaries in mountainous regions， with Dice Coefficients ranging between 0.94 and 0.97. Furthermore， the coefficient of variation and variance have been identified as the most effective inter-annual change indicators for differentiating imitate wildness Chinese medicinal materials from other categories. Notably， May， August， and October emerge as critical intra-annual time windows for their accurate identification. The constructed identification system boasts an impressive accuracy rate of 88%. In Huachi County， the cultivated area of imitate wild Chinese medicinal materials has witnessed a cumulative growth of 49.7% over the past five years， with a consistency coefficient of 0.79 when compared to statistical data. This underscores the high identification accuracy of the technology and suggests its potential applicability in regions with similar agricultural structures， serving as a valuable reference.
In the agricultural field， dense and complete remote sensing time-series data are crucial for crop monitoring. Optical remote sensing， with its rich spectral information， has been widely used for crop observation. However， cloud and rain often lead to missing optical remote sensing time-series data. Synthetic Aperture Radar （SAR） remote sensing， with its all-weather imaging capability， serves as an effective complement but presents challenges in interpretation. Considering the characteristics of both data types and the limitations of traditional pixel-scale studies， this study proposes a parcel-constrained transformation method for optical and SAR remote sensing time-series features. Instead of using traditional pixels as the basic computational unit， farmland parcels are employed. A Transformer-based framework is constructed to establish the time-series feature transformation relationship between optical and SAR remote sensing data， which is then used for parcel-level crop classification. Experimental results show that the transformation achieves a MAE of 0.060， RMSE of 0.086， and R² of 0.889. In the crop classification scenario， compared to the original optical time-series features， the OA decreases by 2.8%， the Kappa coefficient decreases by 4.9%， while the Macro F1 score increases by 1.9%. These findings demonstrate the practical effectiveness of the transformation results. Additionally， the study explores the impacts of parcel constraints， crop types， and feature selection on transformation accuracy. This provides a potentially scalable solution for using SAR data to compensate for missing optical data and offers methodological insights for SAR-optical remote sensing synergy analysis in cloud-prone and rainy regions.
Cropping intensity of cropland is the number or acreage of crops planted on a unit area， which can effectively represent the way and efficiency of the utilization of cropland resources. Cropping intensity and its spatio-temporal change information are of great significance for remote sensing monitoring of cropland resources， adjusting agricultural planting structure， ensuring national food security and ecological security and achieving sustainable development of agricultural production. We have built up the cropping pattern spectrum including the major cropping patterns （Spring single-cropping， Summer single-cropping， Spring-and-Summer double-cropping， Summer-and-Autumn double-cropping， double-cropping paddy-rice） on the Jianghan Plain， and the Bayesian Network model for cropping intensity mapping by fusing remote sensing images from key phenological phases. The conditional probability table of each child node was defined based on the method of knowledge probability coding， and the vegetation index state of the key phenological periods was summarized. On this basis， the original Naive Bayesian Network used for cropping pattern mapping is extended in time dimension， and the cropping intensity change node is added as the child node of the cropping pattern node of the two time phases； Further the cropping intensity change detection model based on Dynamic Bayesian Network is constructed， through mining the semantic information of cropping pattern change， to achieve automatic and high-precision detection of cropping intensity change on the Jianghan Plain. The result shows that the Dynamic Bayesian Network model integrating remote sensing data and crop phenological knowledge can integrate the probabilistic coding and probabilistic inferencing of the prior knowledge， and can automatically extract the change information， improving the automation level of cropping intensity change detection； the extraction of semantic information of cropping intensity change based on cropping pattern spectrum can characterize the type of cropping intensity change more accurately， improving the accuracy of cropping intensity change detection； about 15% of the cropland has significant changes in cropping intensity， among which the decreasing trend of cropping intensity is the most significant， accounting for more than 60% of the total change on the Jianghan Plain. The Bayesian Network model of cropping intensity change detection has the advantage of fusing remote sensing data and crop phenological knowledge， achieving the integration of prior knowledge and probabilistic inference， and improving the automation level of cropping intensity detection. Based on the unique seasonal rhythm of crops， the proposed Dynamic Bayesian Network model can mine the multi-temporal cropping pattern change information with only a few key phenological phases， and reveal the spatio-temporal change patterns of cropping intensity on the Jianghan Plain from 2017 to 2021. The research demonstrates the validity and reliability of using the key phenological features to map the dynamics of cropping pattern on regional scale. The Dynamic Bayesian Network model for cropping intensity change detection has the advantage of integrating remote sensing data and crop phenological knowledge. The knowledge-based model does not require a large number of training samples， avoiding the problem of data dependency and local optimization of machine learning， effectively avoiding the accumulative error brought by the traditional change detection methods， which has great application value in monitoring the cropping intensity changes.
Terrace is a common agricultural landscape and effective land use mode in hilly and mountainous areas， which helps to improve the yield and quality of agricultural production. Transfer learning technology can realize the maximum reuse of existing samples， models and knowledge， but its performance in terrace extraction is less studied. Five typical high-performance deep learning frameworks， Unet， HRNet， SwinUNet， TransUNet and Segmenter， were selected to carry out transfer learning cross-regional terrace monitoring research based on remote sensing images with 2 meter high spatial resolution obtained by GF-6 satellite. The results show that Unet performs best in the feature extraction task of terrace， with the overall accuracy of 81.67% and the mean pixel accuracy of 81.97% in the target domain. In 2020， the total area of terraced fields in Sandu River Basin was 1 274 square kilometers， accounting for 61.8% of the cultivated land. Using transfer learning and pre-training model to obtain terrace distribution information can not only effectively reduce labor and material costs， but also provide a theoretical basis for cross-regional terrace monitoring and analysis， and lay a foundation for large-scale remote sensing monitoring， land use planning and other related applications.
The cropping intensity index is a key metric for assessing the degree of multiple cropping. To address the challenges posed by persistent cloud cover in agricultural regions of Southwest China， we developed a simple and efficient method for extracting this index to support precision agriculture monitoring. The approach integrates large-scale downscaled data from multi-source fusion and employs Whittaker smoothing to reconstruct the NDVI time series. Peak points are identified using a neighborhood comparison method， while the Normalized Burn Ratio 2 （NBR2） is used to detect bare soil occurrences between peaks， allowing for the removal of false peaks. A complete remote sensing workflow was established for the Sichuan Basin， enabling full-coverage mapping of cropping intensity for the year 2020.Validation based on visually interpreted samples showed an overall accuracy of 89.0%， with a Kappa coefficient of 0.793. The estimated sown area exhibited a strong correlation with county-level agricultural statistics （R² = 0.921 5，MRE=21.81%）. These results provide valuable data and decision-making support for local agricultural authorities in optimizing fallow and land development strategies， and in advancing precision agricultural monitoring.
The Guangdong-Hong Kong-Macao Greater Bay Area （GBA） stands as a paramount symbol of China's reform and opening-up， characterized by its exceptional economic vitality. Understanding its spatiotemporal urbanization patterns is crucial for deciphering the evolution of mega-city clusters and their interplay with economic development. To address the inconsistencies between DMSP/OLS （1992—2012） and NPP/VIIRS （2012—2021） nighttime light data， this study constructed a continuous， long-term dataset by evaluating two integration approaches： simulating DMSP/OLS from NPP/VIIRS and vice versa. The optimal method was selected based on the performance of linear regression models with Gross Domestic Product （GDP）， resulting in a consistent nighttime light dataset （1992—2021） for the GBA that shows a higher consistency with socioeconomic indicators. Our analysis revealed that： （1） From 1992 to 2021， the nighttime light intensity in the GBA showed a continuous increasing trend， with distinct spatial dynamics across three periods： initial rapid growth was concentrated in the core urban areas of Guangzhou， Hong Kong， and Macao （1992—2001）； subsequently， the growth hotspot diffused into the suburbs （2002—2011）； and finally， it expanded further into county-level towns and rural areas （2012—2021）. （2） The expansion rate of urban built-up areas was higher in the northern and western parts of the GBA than in the eastern and southern parts. Moreover， the primary direction of urban expansion underwent a notable shift， transitioning from a “southeast-northwest” orientation to a predominantly “east-west” orientation.
The uneven development within a country or region poses a long-term threat to social and economic progress and hinders poverty alleviation efforts. Accurately measuring the balanced development of regional economies is an urgent priority to mitigate and eliminate disparities， narrow regional gaps， and has been a focal point in regional economics research for an extended period. To address the issue of regional economic development， this study employs Nightlight-based Gini coefficients， derived from night light remote sensing data and population statistics， as an indicator to evaluate the economic equilibrium in Shaanxi Province. The paper provides a quantitative analysis of the spatio-temporal evolution characteristics of economic development balance across the entire province， its three natural regions （Northern， Central， and Southern Shaanxi）， 10 prefecture-level cities， and 107 county-level administrative regions from 2000 to 2020. Additionally， it utilizes coupled random forest models and SHAP methods to assess the importance of driving factors and explore their response characteristics. The results indicate that the Nightlight-based Gini coefficients in Shaanxi Province underwent a dynamic evolution process， characterized by generally high levels in 2000， followed by a gradual decline from 2000 to 2010， and a rapid trend towards equilibrium from 2010 to 2020. Significant multi-scale differences and linkage features in economic development balance are observed， with over 93% of county-level administrative districts showing a decreasing trend in Nightlight-based Gini coefficients. The Guanzhong urban agglomeration exhibits notably better internal economic balance compared to Northern and Southern Shaanxi， where cities like Xi'an and Xianyang， characterized by higher average light values and robust development vitality， demonstrate stronger internal economic equilibrium， with inter-city disparities gradually diminishing. At the regional scale， Guanzhong and Northern Shaanxi show strong economic balance， while Southern Shaanxi remains skewed from equilibrium. Furthermore， the balance of economic development in Shaanxi Province responds nonlinearly to driving factors， with road network density being the most critical factor influencing economic development. These findings offer a novel research perspective and provide a scientific foundation for achieving regional balanced development and promoting common prosperity.
Urban spatial structure constitutes a core topic in the fields of urban geography and urban planning. Deepening research in this area holds significant theoretical and practical importance for promoting urban sustainable development. To investigate the evolutionary characteristics and driving mechanisms of Xi'an's urban spatial structure over the past decade， this study employs a spatial coupling analysis method integrating Black Marble nighttime light data and POI data. Through grid processing， kernel density estimation， and coupling coordination degree modeling applied to data from 2012 and 2022， the research systematically analyzes the dynamic changes in Xi'an's urban spatial expansion patterns and functional layout. The results indicate that： （1） The coupling analysis method using Black Marble nighttime light data and POI data proves effective in characterizing urban spatial structure. （2） Xi'an maintains an overall monocentric urban spatial structure， exhibiting gradual outward concentric expansion， with particularly rapid development in its northern regions. （3） The coupling coordination degree between the two datasets shows a distinct gradient decline from the city center towards the periphery. （4） High-High and High-Medium coupling zones are primarily distributed within and diffusing outward from the central urban area， while High-Low and Medium-Low zones are typically located in urban-rural transition zones， reflecting the spatial matching relationship between urban functions and economic activity intensity. With the exception of the “Medium-High” type， the area of all other coupling zones increased over the ten-year period， corroborating the continuous expansion and internal restructuring of urban space. This study reveals that the evolution of Xi'an's urban spatial structure results from the combined effects of macro-policy guidance， socio-economic development， and industrial spatial adjustment. The research validates the effectiveness of the multi-source geospatial big data coupling analysis method， providing a scientific basis for territorial spatial planning and sustainable development strategies in Xi'an and similar cities.
Long-term monitoring of the average dry-air mixing ratio （XCO2） of atmospheric carbon dioxide column is of great significance for accurately estimating carbon emissions and studying climate change. At present， "carbon sniffing" satellite is the best source for large-scale coverage and obtain long-term time series data. However， satellite products are mostly represented by orbits， resulting in extremely uneven global distribution and large blank areas. It is urgent to adopt appropriate methods to grid them in order to give full exploit to the advantages of satellite detection.Therefore，based on the Level2 product data V11.1 version of OCO-2 （OCO2_ L2_Lite_FP）， and considering the spatiotemporal variation characteristics of XCO2， spatiotemporal interpolation mapping method is used to product the long-term gridded dataset. The product sum function variation model is established for the empirical semi-variance of residual data and the model parameters are fitted. The ordinary Kriging interpolation principle is applied to interpolate the 5° latitude zone separately to obtain the land 1°×1° longitude and latitude grid XCO2 dataset in the range of 55°S~60°N from 2015 to 2022. Compared with TCCON data， there is a significant linear relationship between them， with the correlation coefficient of up to 0.980 6， and the average absolute error within 1ppm. The results show that the dataset has high accuracy and can provide data support for evaluating carbon emissions and achieving the goal of “double carbon”.
Land Surface Temperature （LST）， as one of the key physical quantities characterizing the energy exchange between the land surface and the atmosphere at regional and global scales， has been listed as one of the parameters prioritized for determination by the International Geosphere-Biosphere Programme （IGBP）， and high-resolution LST products are of great significance for the characterization of land features. Based on exploiting the advantages of SDGSAT-1 TIS high-resolution （30 m） data， this paper explores an LST retrieval algorithm suitable for SDGSAT-1 TIS high-resolution data by combining its spectral response function characteristics. To validate its reliability， LST retrieval was performed on 9 periods of SDGSAT-1 TIS data across 4 regions. The results were comprehensively evaluated through Temperature-based method with 3 SURFRAD stations data and Cross-validation method with different LST products. The results showed that：①The LST retrieved using this method exhibits high reliability， with an R² of 0.91 in three SURFRAD stations and a consistency of up to 0.98 with Landsat C2L2. ②The LST retrieved using this method shows some variability across different land cover types. The smallest retrieval differences are observed for water， with mean errors of -1.62℃ and -1.49℃ compared to Landsat C2L2 ST and MOD11A1， respectively. Followed by vegetation， cropland， and artificial surfaces. While bare land exhibits the largest differences， with mean errors of -1.51℃ and 3.23℃， which better than it in Landsat C2L2 ST and MOD11A1 （4.29℃）. ③The LST products retrieved using this method are able to more accurately reflect temperature differences among various land features， better distinguish internal details of land features， and more precisely delineate land feature boundaries. Comprehensive analysis indicates that the LST retrieval method proposed in this paper based on SDGSAT-1 thermal infrared data meets the application requirements for generating land surface temperature products from thermal infrared remote sensing data. It effectively enhances the capability to characterize LST with higher precision， which holds significant implications for advancing research in global climate change and global carbon balance.
Thermal infrared remote sensing of Land Surface Temperature （LST） is one of the important means for monitoring Soil Moisture （SM） changes. However， due to cloud cover， LST data often suffer from significant gaps. To address the limitations of missing LST data， scholars have proposed two major categories of filling methods： “spatiotemporal interpolation” and “multi-source data fusion.” Current evaluations of filling methods predominantly rely on ground-based temperature measurements， yet there is a lack of research assessing these methods from the perspective of soil moisture monitoring， which hinders the development of SM monitoring technologies based on LST. Therefore， this study compares and analyzes the performance of filling products based on spatiotemporal interpolation methods （LST_Zhang， LST_Shiff） and multi-source data fusion methods （LST_Gao， LST_Yu） in soil moisture monitoring， using aircraft and in-situ SM data as the basis for analysis. The study uses Soil Evaporation Efficiency（SEE），spatially calculated from the two-stage trapezoidal space of Land Surface Temperature and Fractional Vegetation Coverage， as an indicator for soil moisture monitoring. A scheme was designed to calculate SEE by combining different LST filling products and four meteorological reanalysis data （ERA5-Land， GLDAS， MERRA-2， NLDAS-2）. The SEE was calculated using aircraft SM and in-situ SM as the basis for analysis， and the consistency between SEE and aircraft SM was used as the analysis criterion. The results show that： （1） Compared with spatiotemporal interpolation methods， the LST_Zhang filling product， based on daily LST changes， performs better than the LST_Shiff filling product， which is based on annual LST changes. Among the multi-source data fusion methods， the LST_Gao filling product， which uses auxiliary data with a spatiotemporal resolution similar to the original LST， performs better than the LST_Yu filling product， which uses coarser auxiliary data. （2） When combined with different meteorological reanalysis data， the performance of LST_Zhang and LST_Gao is superior to that of LST_Shiff and LST_Yu， respectively. Moreover， the SEE effect of the combination calculation with MERRA-2 is better than other combination schemes. The average correlation coefficient R between the aircraft SM and these combinations reaches 0.39， with p<0.01， indicating a high consistency in the trend of temporal changes between these combinations and the in-situ SM. （3） For the purpose of soil moisture monitoring， it is recommended that future LST filling algorithms should consider the diurnal and annual spatiotemporal variations of LST， as well as select reliable multi-source auxiliary data with a spatial and temporal resolution close to the original LST. The matching scheme proposed in this study can obtain high-precision spatially continuous SEE data， and provide a reference for the production of LST filling products used for soil moisture monitoring.
Total Suspended Matter （TSM） is one of the critical indicators to evaluate inland water quality. In this study， the TSM concentrations and in situ water reflectance spectra from 92 water samples （covering multiple seasons） were collected in four representative study areas of Poyang Lake. Retrieval algorithms to estimate TSM in Poyang Lake were developed based on simulated Landsat-8 image bands with measured spectral data. The results indicated that the inversion models based on remote sensing reflectance of Landsat-8 B4， B5 and B4/B3 had better accuracy， and the optimal fitting coefficients （R2） were 0.87， 0.90 and 0.86， respectively. The three optimal inversion algorithms based on B4， B5 and B4 /B3 were applied to Landsat-8 image which was quasi-synchronous with in-situ sampling date. These algorithms were used to extract TSM concentrations in Poyang Lake， followed by a verification and evaluation process. The application results demonstrated that the inversion algorithm with the highest validation accuracy was the exponential model based on B4. The Root Mean Square Error （RMSE） of TSM was 9.92 mg/L， and the percentage root mean square error （%RMSE） reached 36.6%. In this study， the index model based on the B4 band was employed to analyze the spatiotemporal distribution of TSM in Poyang Lake during 2022. Spatially， the areas with high TSM concentrations were primarily located in the northern channel connecting with the Yangtze River and the region extending from Songmen Mountain to the central lake. Temporally， TSM concentrations were significantly higher during the dry season compared to the wet season. These spatial and temporal variations of TSM in Poyang Lake indirectly reflect the influence of human activities on the water quality of the lake. This study evaluated and demonstrated the potential capability of Landsat-8 satellite imagery for effectively monitoring TSM in Poyang Lake， which can provide data and model support for protection of water environment in Poyang Lake.
The land-based discharge of suspended sediment in the highly human-activity-intensive Pearl River Delta region is a critical factor for the coastal ecological environment of the South China Sea. Its accurate remote sensing monitoring holds significant implications for regional environmental management and pollution control. In this study， based on the principles of radiative transfer and considering the influence of organic pollution， developed a remote sensing retrieval model for SSC with clear retrieval mechanisms and well-defined parameter significance. Through the retrieval and spatiotemporal variation analysis of SSC in the Pearl River Delta from 2021 to 2023， the results indicate that： （1）When considering the influence of organic pollution， the remote sensing retrieval values of SSC showed a high correlation with measured values， with a coefficient of determination （R²） of 0.741， a Root Mean Square Error （RMSE） of 5.969 mg/L， and a Mean Absolute Error （MAE） of 4.951 mg/L， indicating high model accuracy and reliable results. （2）Through visual interpretation， it was observed that areas such as the Zhongtang Waterway in the Dongjiang Delta， Hanxi River， and the middle-upper reaches of the Xijiang Delta exhibited high SSC， possibly related to terrestrial pollution from industrial areas and sediment deposition in downstream river channels. （3）The SSC in the Pearl River Delta showed seasonal variations， with higher concentrations in spring and lower concentrations in autumn. This study can provide a scientific basis for the remote sensing monitoring of water environments and the integrated land-sea management of the Pearl River Delta and similar estuarine and coastal regions.
Aiming at the issues of image degradation effects that do not align with real-world scenes and the limited ability of models to perceive multi-scale feature information in remote sensing image super-resolution tasks， this paper proposes a remote sensing image super-resolution reconstruction algorithm combining a degradation quality network and multi-scale feature fusion. Firstly， to account for the influence of multiple degradation factors on image quality in real-world scenarios， a new degradation network is proposed. Secondly， to address the insufficient feature information capturing ability under multi-scale conditions， the paper designs an effective generative network and discriminative network. In the generative network， the Dense Multi-scale Feature Extraction Structure （DFEMS）， which is the core of the network， adopts the form of densely connected Adaptive Feature Enhancement Blocks （AFEB）. This design aims to reduce the model's computational burden while enhancing its ability to perceive multi-scale feature information. In the discriminative network， the U-Net network structure is optimized to improve the model's ability to distinguish between global structure and local details. The paper conducts ablation and comparison experiments based on the UCMerced and WHU-RS19 datasets. The experimental results show that the proposed method outperforms existing methods and demonstrates strong generalization capability in remote sensing image super-resolution tasks.
Efficient extraction of spectral-spatial information is the core problem of hyperspectral remote sensing image classification. Transformers framework can effectively characterize the advantages of high-level semantic features， while CNNs can efficiently extract local features， and both of them have great potential in the field of hyperspectral image classification. In order to fully extract the spectral-spatial features， this paper proposes a Transformers-based high-frequency spectral-spatial information-enhanced hyperspectral image classification method （HFSST） by fusing the advantages of Transformers and CNNs networks. First， the low-level spatial-spectral features are extracted collaboratively by 3D-CNN and 2D-CNN. Second， the CAB （Converted Attention Block） attention mechanism is embedded in the Transformers Encoder （TE） to balance the channel and spatial information. Finally， a local convolution branch is added after the output features of the TE module to achieve better recovery of high-frequency information. The experiments are validated on Indian Pines， Pavia University， and Houston 2013 datasets， and the classification performance outperforms several current state-of-the-art methods， indicating that the method in this paper can strengthen the differentiability of inter-class features by enhancing the high-frequency spectral-spatial information. In addition， the correctness of the generalization of the hyperspectral image classification problem to the locally advanced semantic classification problem can be demonstrated.
Pansharpening technology for remote sensing images integrates the intricate texture details of high-resolution panchromatic images with the abundant spectral information of low-resolution multispectral images， culminating in high-resolution multispectral images with superior spectral fidelity. Existing deep learning-based pansharpening approaches rely on supervised learning， requiring downsampling to construct reference samples， inevitably leading to the loss of original feature information， whereas unsupervised methods， though trainable at full resolution， still suffer from limitations in spatial texture detail preservation. To address these issues， an unsupervised generative adversarial network incorporating spatial–spectral sliding features and dense residual connections for pansharpening is proposed. First， a sliding-window mechanism is introduced in both spatial and spectral domains to capture spatial–spectral features. Then， dense residual structures are employed to perform feature fusion and reconstruction. Furthermore， a joint loss function is designed to impose overall constraints on the model training process. Subjective and objective evaluations conducted on two different datasets demonstrate the performance in terms of spatial texture details preservation and spectral consistency， while ablation experiments analyze the contributions of network components and joint loss constraints. The results indicate that stable performance can be achieved in unsupervised pansharpening tasks， providing valuable insights for network architecture design and training constraint strategies in related models.
As a key atmospheric component， aerosols have profound effects on vegetation photosynthesis and the carbon cycle of ecosystems by altering surface radiation characteristics and microclimatic conditions. However， the interaction between aerosols and vegetation is highly heterogeneous across different regions and climates， and the mechanisms underlying this relationship remain inadequately understood. Leveraging high-resolution Aerosol Optical Depth （AOD） and sun-induced chlorophyll fluorescence data from 2000 to 2023 in Northeast China， this study employs nonlinear Granger causality analysis to systematically explore the impact of AOD on vegetation photosynthesis and its climate dependence. The findings reveal that AOD exerts a significant positive influence on forest vegetation photosynthesis by enhancing scattered radiation， particularly in vegetation types with high leaf area index and complex canopy structures. Low temperatures and moderate precipitation further amplify the positive effect of AOD on vegetation photosynthesis， while extreme climatic conditions may weaken or obscure this influence. This study not only underscores the complexity of the aerosol-ecosystem interaction but also provides a critical scientific foundation for the formulation of aerosol management and regional ecological protection strategies.
PS-InSAR technology is widely used for monitoring surface deformation； however， its results are often affected by spatial noise. To improve monitoring accuracy， this study introduces a regional stacking filtering method， constructs a residual grid based on a 0.1° × 0.1° mesh， and applies ISODATA spatial clustering to effectively eliminate Common-Mode Errors （CME）. Based on PS-InSAR data from southeastern Tibet collected between October 2014 and January 2024， the results show that removing CME significantly improves the model fitting accuracy， with the RMSE reduced by approximately 45% and R² increased by around 25% on average. The deformation velocity results reveal significant uplift （>5 mm/a） in the northwestern and central-eastern parts of southeastern Tibet， mainly controlled by fault activity. In contrast， subsidence zones are concentrated around Biru County， closely associated with the Anduo South Fault Zone and the 2021 Ms6.1 earthquake， with deformation patterns highly consistent with tectonic activity. In addition， this study extracts the Line-Of-Sight （LOS） coseismic deformation field of the earthquake and analyzes changes in surface deformation rates before and after the event. The results indicate that the epicentral region experienced up to 23 mm of subsidence， with abrupt coseismic displacement observed in some areas， reflecting fault rupture characteristics. The pre-seismic deformation rate was relatively stable （<5 mm/a）， whereas a post-seismic acceleration of approximately -10 mm/a was observed. Combined with focal mechanism solutions and InSAR inversion， the earthquake is confirmed to have been triggered by a strike-slip–oblique-slip normal faulting mechanism. This study significantly enhances the accuracy of PS-InSAR data and provides insights into the spatiotemporal distribution and geological causes of surface deformation in southeastern Tibet. The findings offer scientific support for geological hazard monitoring and hydropower development in the Yarlung Tsangpo River basin.
The small mafic-ultramafic rocks are the carrier of magmatic copper-nickel-cobalt deposit， but the mafic-ultramafic rocks are easy to be missed by field work because of their small scale. Therefore， rapid extraction of the small mafic-ultramafic rocks by remote sensing is of great significance for relevant ore exploration. In this study， ASTER data， SDGSAT data， and combination of convolutional neural network and Vison Transformer deep learning model were used to extract mafic-ultramafic rocks in the East Tianshan area. The extraction results of two different data were compared and field verified. The results show that： （1） Using the Huangshan area as the training area and the known mafic-ultramafic rocks in the Huangshan area as the training sample， the main mafic-ultramafic rocks in the study area have been successfully identified， which proves that the research route of using a small area as the training area to learn and then test and identify the surrounding large area is feasible. （2） The spatial resolution of ASTER data for infrared bands is only 90 m， and the extraction results based on ASTER data are unsatisfying for the identification of rocks with small exposed area； moreover， due to limited the swath width of ASTER data， the mosaic of muti-scenes is needed for large-scale application， which reduces the recognition accuracy to a certain extent. （3） SDGSAT-1 image has a spatial resolution of 30m in the thermal infrared band， which has a good effect on the identification of small mafic rocks in the study area. At the same time， the image has a large swath width， which is more suitable for large-scale research application. In the future， the identification accuracy of the model can be further increased by introducing negative samples of non-mafic rock， and it can be put into large-scale lithologic identification.
The economic value of various services and functions rendered by ecosystems to human society is quantified by ecosystem service value assessment； decision-making support for ecological protection， environmental management， and sustainable development policy formulation is provided； and an important foundation for promoting sustainable development is established. As China's crucial ecological barrier and economic belt， the Yellow River Basin confronts severe challenges such as soil erosion and habitat quality degradation， impacting the ecological well-being of hundreds of millions of people. The study focused on the densely populated middle and lower reaches of the Yellow River Basin as the research area. Based on land cover data from 2000， 2010， and 2020， as well as annual precipitation and elevation data， the InVEST （Integrated Valuation of Ecosystem Services and Tradeoffs） model was utilized to analyze the spatiotemporal variations of four ecosystem services—water yield， carbon storage， soil retention， and habitat quality—from the perspectives of provincial scales and land cover types， thereby revealing the interactions among these ecosystem services. The results indicated that， over the two study periods， water yield and soil retention services exhibited a gradual upward trend； carbon storage progressively declined； habitat quality demonstrated a dynamic change of first rising and then falling； significant differences in ecosystem service performance were observed at the provincial scale； the ecological service functions of forestland and grassland were notably prominent； ecosystem services generally exhibit synergistic relationships， with significant spatial heterogeneity. The study provided high-resolution assessments of ecosystem service dynamics and proposed region-specific ecological conservation strategies based on provincial differences， offering references for ecological management and sustainable development in the middle and lower reaches of the Yellow River Basin.
