Solar-Induced chlorophyll Fluorescence （SIF）， a proxy for vegetation photosynthetic activity， has gained widespread applications. This review synthesizes the principles， progress， and key frontiers in satellite SIF remote sensing. Firstly， we introduced the SIF retrieval principles and algorithms at both ground and spaceborne platforms. The SIF retrieval methods can be divided into two categories： physically-based inversion and data-driven algorithms. The accurate separation of SIF from reflected radiance in upwelling radiation is the key challenge. The current ground-based retrievals are still limited by spectrometer resolution and Signal-to-Noise Ratio （SNR）， developing instrument-agnostic， high-robustness algorithms remains a research priority. Data-driven approaches dominate satellite SIF retrievals. However， huge uncertainties persist in red-band SIF retrieval， demanding transformative algorithmic breakthroughs. Second， we analyze global SIF satellite developments over 30 years， highlighting China’s rapid progress （e.g.， successful experiments with TanSat and Goumang）. Nevertheless， gaps persist in satellite longevity， data sharing， and scientific utilization compared to international counterparts. The next-generation TanSat-2 （to be launched in 2026） is expected to revolutionize SIF remote sensing， offering 2-km resolution， global daily coverage， and dual-band （red/far-red） SIF data—resolving critical limitations of low resolution， SNR， and revisit frequency. Finally， we investigated the progresses of spatiotemporal fusion of SIF satellite data. Machine Learning （ML）-based simulation methods have achieved high-precision simulation of SIF data and have been widely used. However， the ML-based SIF datasets represent the modeled signals driven by reflectance and meteorology， not observations. Spatial downscaling of satellite SIF products preserves observational fidelity despite lower spatiotemporal continuity than ML counterparts. Emerging multi-sensor， long-term （1995—2024）， 0.05°-resolution downscaled products hold potential to accelerate SIF science applications. Therefore， despite the inherent challenge of high-precision retrieval of weak SIF signal， advances in payload technology and quantitative remote sensing are rapidly transforming SIF monitoring capabilities. China’s SIF remote sensing program （TanSat-2） is positioned to play an increasingly pivotal role in guiding global SIF science and applications.
Approximately 60% of global methane emissions originate from anthropogenic sources， making their effective control a critical aspect of greenhouse gas reduction efforts. The energy sector is a major contributor to anthropogenic methane emissions， with emission events traceable to specific facilities and characterized by a heavy-tailed distribution of emission rates. High-resolution hyperspectral satellite remote sensing data enable methane emission monitoring at the point-source scale and facilitate attribution to specific facilities. A high-resolution remote sensing monitoring system for methane point sources in the global energy sector has been developed based on hyperspectral retrieval methods and a WebGIS platform. The system comprises a point-source retrieval algorithm， a methane emission retrieval dataset， and a monitoring platform. By the end of 2024， effective monitoring of 573 methane emission events worldwide has been achieved. The system is designed to detect sudden methane emissions from coal mining and oil and gas extraction and storage， enabling facility-specific attribution. Support and validation data for emission source identification and estimation in the energy sector are provided， contributing to global methane reduction efforts.
Frequent extreme high-temperature weather under the background of urbanization and climate warming poses a threat to the health of urban residents. The Park Cooling Effect （PCE） is vital for alleviating these adverse effects. However， in land-scarce urban areas， it is unrealistic to expand the size of parks without limits， and maximizing the cooling effect per unit area becomes an urgent issue.Taking China’s hottest “furnace city”， Fuzhou， as the study area， this research quantifies PCE using three indicators： cooling amplitude （∆Tmax） ， cooling distance （L∆max） and cooling gradient （Gtemp）.It calculates the PCE indicators of 50 urban parks within the city and analyzes the influencing factors from the perspectives of external park morphology and internal landscape patch characteristics. Results reveal that： ①Among 50 parks， 42 exhibit a PCE， whereas 8 do not； ②Larger parks are not always better； it is crucial to consider both external morphological and internal patch attributes； ③Regarding external morphology， parks with simple and regular boundary shapes are more conducive to cooling effects； ④With regard to the internal patch characteristics of the parks， low impervious surface proportions， high proportions of water bodies and vegetation， and complex patch morphologies enhance PCE， while excessively high edge densities and landscape fragmentation weaken it. Therefore， efforts should be made to maintain the continuity and integrity of vegetation and water body coverage， design diverse and multi-tiered vegetation boundary structures， or leverage terrain changes to increase the vertical complexity of impervious surface boundaries to optimize cooling and alleviate urban heat island effects.
Ocean color remote sensing chlorophyll a concentration data are of great significance for a wide range of scientific issues， such as global change and biogeochemical cycle studies. However， the existing operational remote sensing data on chlorophyll-a concentrations suffer from spatial incompleteness and temporal discontinuity，significantly limiting their application potential. Research shows that data reconstruction is the primary approach to address this issue. Based on recent research findings on ocean color remote sensing data reconstruction， this research summarizes the theoretical foundations and development history of data reconstruction methods. It particularly discusses the hot topics in chlorophyll-a concentration data reconstruction， including common datasets， methods， and major research areas. The paper also identifies existing challenges in chlorophyll-a concentration data reconstruction， highlighting critical issues such as data reconstruction in extremely data-scarce regions， the development of short-time scale data reconstruction methods， multivariate combined reconstruction methods， improvements in Empirical Orthogonal Function data interpolation， and the integration of remote sensing with numerical simulation for data reconstruction. The paper concludes by outlining future research directions in this field.
In the context of global climate warming and rapid urbanization，the Urban Heat Island （UHI） effect exacerbates population heat exposure risk，accurately identifying high-risk areas of population heat exposure is crucial for adapting to and mitigating the threats posed by high temperatures．This study focuses on the area within the sixth ring road of Beijing，from the perspective of Local Climate Zones （LCZ），analyzing the temporal and spatial variations of Surface Urban Heat Island Intensity （SUHII） in different LCZs during summer，based on ECOSTRESS Land Surface Temperature（LST） data and Tencent mobility population data. The study also evaluates the daily dynamic changes in population heat exposure under different SUHII levels. The results show：①During the day，significant differences in SUHII are observed across different LCZ types. Except for LCZ 9 （Sparsely built），built-up LCZs，as well as LCZ E （Bare rock or paved） and LCZ F （Bare soil or sand），exhibit characteristics of heat sources，while other types are heat sinks．At night，the SUHII differences between LCZs decrease，and the heat sink effect of LCZ A （Dense trees） and LCZ B （Scattered trees） weakens，while LCZ G （Water） shifts from a heat sink to a heat source.②Population heat exposure is lowest between 6—7 AM and peaks between 10—11 AM.Built-up LCZs （except LCZ 9），along with LCZ E and F，show population heat exposure risk. Within the fourth ring road，where buildings are dense and population density is high，population heat exposure is particularly prominent.③When SUHII exceeds 2 ℃，the risk of population heat exposure significantly increases，particularly in areas with high building density and low green coverage，controlling SUHII to stay within 2 ℃ can effectively reduce the risk of population heat exposure. This study provides a theoretical basis and practical support for enhancing urban livability，improving living environments，and formulating urban heat risk control strategies．
Urban informal settlements are a global common phenomenon in the process of rapid urbanization. The research on their optical remote sensing identification is of great significance for achieving the 11th Sustainable Development Goal of sustainable cities and communities. Based on the statistics of relevant domestic and foreign literature from 2001 to 2024， this paper sorts out and analyzes the data sources and spatial resolutions of optical remote sensing images of urban informal settlements， single-class feature classification and multi-source feature fusion classification， as well as the identification methods and their advantages and disadvantages. The results show that data sources are classified into three categories according to spatial resolution： ultra-high/high， medium， and low； single-class features or data sources are classified into optical image features such as spectral， texture， geometry and context， and auxiliary data features such as GIS data； multi-source feature fusion is classified into three types： feature-level， data-level and decision-level. Compared with deep learning， traditional identification methods such as pixel-based and object-based methods have limitations such as weak adaptability and low recognition efficiency. Deep learning methods， with their strong feature extraction capabilities and good generalization ability， have broad prospects in the remote sensing identification of large-scale and long-term informal settlements. This paper aims to provide a reference for the future remote sensing identification of urban informal settlements and promote the construction and development of sustainable cities.
Soil， as the most valuable natural resource， faces challenges of global， regional， and local degradation， with issues ranging from quality deterioration to salinization leading to significant losses of high-quality farmland. These problems impact agricultural productivity and ecological balance， disrupting food security and sustainable development. Therefore， timely monitoring and accurate mapping of salinization processes are crucial， especially in semi-arid and arid regions where the impact of climate change has reached alarming levels. With the rapid development of remote sensing technology， soil salinization mapping techniques are showing great potential. This paper systematically reviews the application prospects of remote sensing technology in soil salinization assessment. Analysis of relevant literature in the Web of Science database reveals that the United States and China have conducted in-depth research on remote sensing technology for soil salinization. Keyword searches indicate that recent research focuses on salt-alkaline remote sensing monitoring based on machine learning and artificial intelligence. Bare soil and vegetation information are widely used in current soil salinization detection， introducing the concept of spectral indices. By studying the reflectance of visible （VIS）， near-infrared （NIR）， and shortwave infrared （SWIR） bands， soil salinity changes are monitored. The aim is to integrate multi-source， multi-scale， multi-platform remote sensing data with ground information to establish a comprehensive "sky-ground" soil salinization investigation and monitoring technology system. However， large-scale soil salinity estimation based on remote sensing technology remains a major challenge due to the cost and coverage of acquiring high-resolution， hyperspectral， high signal-to-noise ratio， and multi-modal remote sensing data， as well as their revisit periods. In addition， efficient acquisition of ground survey data matching multiple-source multi-modal remote sensing data， extraction of effective salt response variables from multi-modal data， effective fusion of multi-dimensional （one-dimensional， two-dimensional， three-dimensional） multi-modal information to build inversion models for multiple characterization parameters of saline-affected land， and the application of developed remote sensing products in agricultural production and ecological environment protection practices are pressing issues that require further research.
Spartina alterniflora， as a typical invasive species in China's coastal wetlands， poses a severe threat to ecosystem structure and function due to its rapid expansion. There is an urgent need to establishan efficient， accurate， and operational large-scale monitoring system to support ecological risk assessment， invasive species control， and wetland conservation decision-making. Given the limitations of traditional monitoring methods—such as low efficiency and challenges in large-scale applications—and the constrained accuracy of remote sensing in complex environments， this study systematically reviews recent progress in remote sensing monitoring of S. alterniflora invasions， focusing on three key aspects： identification methods， spatiotemporal analysis， and ecological impact assessment.In terms of identification methods， remote sensing relies primarily on the unique spectral and phenological characteristics of S. alterniflora. The application of traditional machine learning and deep learning methods has significantly improved identification accuracy， while phenology-based strategies effectively mitigate interference from the "different objects with similar spectra" phenomenon. Numerous studies indicate that the spread of S. alterniflora along China's coast has undergone distinct stages—from scattered distribution to rapid expansion—with spatial patterns exhibiting clear regional differences. Its invasion systematically damages wetland ecosystems such as salt marshes， tidal flats， and mangroves， significantly compressing native vegetation habitats， degrading habitat quality， reducing waterbird habitat ranges， and altering tidal flat sedimentation processes. This leads to biodiversity loss， soil degradation， and ultimately undermines the stability and service functions of wetland ecosystems.Overall， while significant progress has been made in identification accuracy and understanding invasion mechanisms， challenges remain， including difficulties in identifying small patches， insufficient model generalizability， and inadequate synergistic application of multi-source data. Future research should integrate high spatiotemporal resolution and multi-source remote sensing data to enhance monitoring capabilities， develop transfer learning and few-shot learning methods to improve model generalizability， deepen the fusion of phenological， spectral， and texture features， and further elucidate the driving mechanisms and ecological effects of invasion. These efforts will provide scientific support for the control of invasive species in coastal wetlands and the maintenance of ecological security.
Forest canopy height is a key parameter for evaluating forest structure and ecosystem functions， playing a crucial role in forest resource management and carbon stock monitoring. Satellite borne LiDAR has been widely applied in the retrieval of forest canopy information because of its high vertical penetration accuracy. However， studies based on China's domestically developed spaceborne full-waveform LiDAR data remain limited. China's forests resources are abundant and mostly located in areas with complex terrain， and different vegetation cover and complex topography can have varying degrees of impact on the feature extraction of full-waveform LiDAR signals. This study employs multi-beam full-waveform LiDAR data acquired by “Goumang”， China's first Terrestrial Ecosystem Carbon Inventory Satellite， from which the original footprint waveforms are filtered and key feature points are extracted， followed by class-specific thresholding to optimize the identification of waveform start and end positions. Based on the optimized signal endpoints and a simulated LiDAR waveform model of forest canopies， the effects of terrain slope are then corrected， followed by the inversion and validation of the forest canopy height in two typical demonstration areas， namely， the Northeast China Tiger and Leopard National Park and the Hainan Tropical Rainforest National Park. The results show that， under the terrain condition of slope ≤15°， the forest canopy height obtained from the optimized Goumang laser altimetry data exhibit strong consistency with the maximum canopy height from the CHM reference data of the sample plots， with correlation coefficients all exceeding 0.85， and the RMSE values decreasing significantly from 4.52 m to 2.56 m in the Northeast China Tiger and Leopard National Park and from 5.51 m to 3.20 m in the Hainan Tropical Rainforest National Park， which further proves that Goumang has a great potential for application in the estimation of the forest canopy height at the laser-footprint scale.
Forest canopy height is a critical parameter for estimating forest biomass and carbon sequestration. The Terrestrial Ecosystem Carbon Inventory Satellite， equipped with a full-waveform LiDAR system （referred to as the "GouMang full-waveform" data）， provides valuable information on forest vertical structure. Accurately extracting forest canopy height from GouMang full-waveform data is therefore of great significance. This study was conducted in the Genhe area， where preprocessed GouMang full-waveform data and UAV-LiDAR data were used to generate a Canopy Height Model （CHM） and extract footprint data. After data screening， footprints within the CHM coverage were selected， and the average CHM value within each footprint （hereafter referred to as footprint-average CHM） was calculated. The full-waveform data were then subjected to noise filtering， quality evaluation， and threshold processing to identify peak regions， from which waveform features were extracted. Subsequently， four machine learning models （XGBoost， AdaBoost， CatBoost， and GBRT） were constructed using the waveform features to predict the footprint-average CHM. A ten-fold cross-validation approach was employed， and model performance was evaluated using R²， RMSE， MAE， ME， Acc. The results showed that 108 high-quality footprints containing 201 full-waveform sequences were selected， with 109 waveform features extracted from each waveform. Among the models， XGBoost achieved the highest performance， followed by CatBoost， GBRT， and AdaBoost. The optimal model yielded R²=0.67， RMSE = 3.13 m， MAE = 2.31 m， ME = 0.12 m， and Acc = 68.54%. The findings demonstrate that through data screening， noise filtering， threshold processing， feature extraction， and model construction， forest canopy height can be accurately derived from GouMang full-waveform data， offering a new technical pathway for forest carbon sink estimation and biomass monitoring.
The ICESat-2 ATL08 data of spaceborne LiDAR are of great significance for monitoring global forest canopy structure. However， the performance variations of the data at different resolutions have not yet been comprehensively evaluated. This research employed ICESat-2 ATL08 data as the study material， and the canopy height retrieval accuracies of ATL08 data at two spatial scales （100 m × 12 m and 20 m × 12 m） were systematically evaluated based on high-precision airborne LiDAR data from 38 NEON （National Ecological Observatory Network） stations of the United States. Moreover， the study further investigated the influence of multiple factors on canopy height retrieval accuracy. The results show that： （1） At the 100 m × 12 m spatial scale， the canopy height retrieval accuracy is RMSE = 5.67 m， Bias = 0.03 m， and %RMSE = 32.7%； after filtering out low-quality segments lacking canopy-top photons， the retrieval accuracy at the 20 m × 12 m spatial scale is RMSE = 5.28 m， Bias = -0.43 m， and %RMSE = 34.2%. （2） The retrieval accuracy of nighttime data is better than that of daytime data， and strong-beam data exhibit superior in terms of penetration and signal-to-noise ratio （SNR）. The nighttime strong-beam data yield the highest retrieval accuracy （RMSE < 4.0 m， R² > 0.83） when using the data of these two spatial scales for canopy height retrieval. （3） The canopy height retrieval accuracy is significantly affected by terrain slope， vegetation characteristics， canopy photon counts， and spatial resolution. This study also validated the effectiveness of the canopy height uncertainty parameter （h_canopy_uncertainty） as an indicator for quality screening of high-resolution data.
Structural parameters and biomass constitute core indicators for assessing forest carbon storage and carbon cycling. Light Detection and Ranging （LiDAR） serves as a pivotal technology for the precise quantification of multi-type individual tree parameters. This study targeted a Metasequoia forest on Chongming Island， Shanghai. LiDAR point cloud data were acquired from three platforms—Handheld Laser Scanning （HLS）， Terrestrial Laser Scanning （TLS）， and Airborne Laser Scanning （ALS）. A stem-leaf separation technique combining point cloud intensity， density， and 3D connectivity clustering was proposed. Methods for individual tree structural parameter extraction and biomass estimation were systematically compared by combining the Tree Quantitative Structure Model （TreeQSM） with the random forest algorithm， and performance variations， applicability， and complementary advantages across different data sources were rigorously analyzed. Results demonstrated that： （1） By integrating coarse-to-fine individual tree segmentation， stem–leaf separation， TreeQSM reconstruction， and random forest machine learning， the proposed workflow enabled accurate extraction of structural parameters and reliable biomass inversion from different LiDAR platforms. （2） Distinct platform advantages were observed： TLS yielded the highest point cloud density with superior branch structure delineation， albeit at elevated costs； ALS efficiently covered extensive areas， excelling in tree height and canopy characterization（with relative completeness of tree height reaching 100%）； Exceptional efficacy for rapid Diameter at Breast Height （DBH） measurement was demonstrated by HLS（with DBH closure rates all falling within the 75%~100% range）. （3） Multi-source data fusion unlocked significant synergistic benefits， markedly enhancing key parameter extraction accuracy. For instance， integrating ground-platform data elevated ALS-derived DBH estimation accuracy from R²=0.19 to>0.82. （4） For biomass prediction， random forest models leveraging TreeQSM-derived multidimensional parameters exhibited strong performance （all R²>0.7）. Fused datasets consistently outperformed single-source counterparts， with the HLS+TLS fusion achieving peak model accuracy （R²= 0.94）. Overall， high-precision stem and branch parameters derived from ground-based TreeQSM are key to biomass estimation. When precise ground-based data are available， the marginal gain in biomass accuracy from supplementary ALS canopy information is limited. These findings provide theoretical guidance and technical references for the optimal combination and application of multi-platform LiDAR data in forest inventories.
Individual trees are the fundamental units of a forest and are critical factors in forest resource surveys and ecosystem monitoring. With the development of UAV technology and deep learning methods， the efficient extraction of individual tree parameters has become possible. Based on this， the present study focused on a eucalyptus plantation in Jiajiang County， Leshan City， Sichuan Province. Using UAV-based visible light imagery， two deep learning algorithms—SSD （Single Shot MultiBox Detector） and Faster R-CNN （Region-based Convolutional Neural Networks）—were employed for individual tree detection and crown width classification. The two models were compared across multiple dimensions， including fitting accuracy， validation accuracy， and training time. The results showed that both models achieved fitting accuracies above 83%， indicating good performance. The Faster R-CNN model achieved an average precision of 95.13%， an average recall of 88.62%， and an average F1-score of 91.76%， while the SSD model achieved an average precision of 91.03%， an average recall of 82.66%， and an average F1-score of 86.64%. Faster R-CNN outperformed SSD in terms of accuracy and stability， whereas the SSD model showed clear advantages in training time and detection efficiency. Using the Faster R-CNN model， a total of 129 045 eucalyptus trees were identified. Applying the interquartile range method， the growth conditions of all trees were classified into three levels. The results indicated that normally growing trees （Level II） accounted for 96.07%， vigorously growing trees （Level I） for approximately 3.84%， and slow-growing trees （Level III） for only 0.09%. These findings confirm the strong application potential of deep learning algorithms for individual tree detection using UAV visible light imagery， and highlight the practical significance of crown width classification in guiding forestry production and formulating tending measures.
Lake Water Surface Temperature （LWST） is a crucial parameter reflecting lake ecological environment and the impact of climate change， and it is of great significance for studying regional water-heat cycle and ecosystem changes. This study inverts LWST in three prominent lakes on the Qinghai-Tibet Plateau—Qinghai Lake， Siling Co， and Namco—using FY-3D satellite MERSI data and the Split-Window algorithm. After radiation calibration and brightness temperature conversion of MERSI data， Multiple Channel Sea Surface Temperature （MCSST） model parameters were fitted using measured data from hydrological stations. The inversion results were cross-validated using 122 sets of measured data from three hydrological stations between 2020 and 2021 and MODIS temperature products. The results indicate that the lake temperature inversion derived from FY-3D MERSI data shows strong consistency with both the measured values and MODIS products. The average water temperature from the hydrological stations is 3.23 ℃， while the average water temperature from the FY-3D inversion results is 4.51 ℃， with a correlation coefficient exceeding 0.85. The inversion error remains stable， demonstrating high reliability and continuity. This study provides a new technical solution for future monitoring of lake ecological environment changes at both regional and global scales.
In the context of global warming， understanding the response mechanisms of grassland Gross Primary Productivity（GPP） to drought in the Three-River Headwaters Region of the Qinghai-Tibetan Plateau is crucial for comprehending vegetation dynamics and regional ecosystem protection. This study utilizes the glass GPP product along with meteorological， soil moisture， and soil temperature data to analyze the response mechanisms of grassland GPP to drought in the Three-River Headwaters Region of the Qinghai-Tibetan Plateau from 2002 to 2020. The results indicate： （1） Over the past 19 years， the region has predominantly experienced warming and humidification trends， with grassland GPP increasing in most areas； （2） The response of grassland GPP to the Standardized Precipitation Evapotranspiration Index （SPEI） during the growing season shows significant spatial differences， with SPEI-1 and SPEI-12 having the greatest impact； （3） Along the elevation gradient， the response of grassland GPP to drought first increases and then decreases. In low-altitude areas， precipitation and soil moisture are the main limiting factors， while in high-altitude areas， temperature and soil temperature are the dominant constraints during drought.
Aerosols are key factors influencing climate change and regional air quality， making their accurate monitoring crucial. The high aerosol loading and complex types over Asia pose greater challenges for satellite retrieval techniques.To evaluate the applicability of the FY-4B Land Aerosol （LDA） product in the Asian region， this study conducts an accuracy validation using ground-based AERONET data from 58 stations， covering the period from August 2022 to December 2023. The FY-4B LDA product was compared with the Himawari-8 Aerosol Optical Depth （AOD） product to assess spatial distribution consistency. The findings indicate that： （1） The overall accuracy of the FY-4B LDA product in the Asian region is moderate， with R² = 0.52， MAE = 0.20， and RMSE = 0.29. Only 37% of the data fall within the expected range， with a significant overestimation observed. （2） On a monthly scale， the LDA product slightly underestimates aerosol levels in spring， while overestimation is more prominent in other months， especially in summer. The best accuracy is observed in October and November. （3） On an hourly scale， FY-4B and AERONET AOD values peak at UTC 08∶00 and 09∶00， showing a rise followed by a decline， with the highest MAE occurring at UTC 03∶00 and the lowest at UTC 21∶00. （4） The FY-4B LDA product performs best in forested areas in terms of surface type. （5） The spatial distribution of AOD from FY-4B and Himawari-8 are generally consistent. High AOD values are concentrated in regions such as the Ganges Delta， Northeast China， eastern Myanmar， and northern Thailand and Laos， while missing data primarily occur in central and southern China. The differences between the two products follow a normal distribution.By assessing the Asian accuracy of the FY-4B LDA product from multiple perspectives， this study offers a theoretical basis and data support for enhancing the capabilities of the Fengyun satellite series.
Ground meteorological stations are essential for obtaining near-surface temperature data. In mountainous areas， there are significant differences in the spatial extent represented by the temperature data observed from different stations. Therefore， determining the optimal layout of monitoring stations in mountainous regions to improve the spatial representativeness of observational data is of great importance for the development of high-quality meteorological services in these areas. This study focuses on the Chongli District， Zhangjiakou City， Hebei Province. Using 30-meter resolution annual average Land Surface Temperature （LST）， Normalized Difference Vegetation Index （NDVI）， and topographic elevation—three factors affecting near-ground temperature—this research employs the Fuzzy C-Means （FCM） clustering algorithm and utilizes the information entropy index to determine the optimal positions for temperature monitoring stations in the region. Additionally， based on the principle of maximizing information and minimizing redundancy， an optimization scheme for adding the original observation station positions is developed. The spatial representativeness of the original and optimized meteorological station locations was compared and analyzed using GIS neighborhood statistical methods， with temperature distribution data from January and July. The results indicate that， compared to the original sites， the average spatial representativeness range of the stations determined by the Fuzzy C-Means clustering algorithm （FCM） combined with the information entropy index has increased by 21.90 km² in winter and 28.79 km² in summer. Moreover， after the increase of several stations， the monitoring coverage rate of the station network in winter and summer has increased by 14.87% and 16.79%， respectively. This study employs a site selection method that integrates factor analysis and information theory， providing a reference for the siting or expansion of temperature monitoring stations in mountainous areas.
To obtain rapid and effective extraction method of potato Seedling period， Squaring period and Tuber growth period， using UAV images with a ground resolution of 5 cm， computing the vegetation index of Visible-band Difference Vegetation Index（VDVI）， Excess Green Index（EXG）， Normalized Green-Red Difference Index（NGRDI）， Normalized Difference Vegetation Index（NDVI） 4 species， select Timing intersection， Otsu threshold and Support Vector Machines（SVM） combining with Vegetation Index（VI） histogram to determines the threshold segmentation method， and The results of the UAV with ground resolution is 1 cm.Then the vdvi index was selected. Using the threshold of single images of potato seedling， current bud and tuber expansion periods， the large range of experimental field images was extracted and the extraction effect was verified. The results show that SVM and VI histogram to determine the threshold segmentation classification map effect is good， the smallest error of vegetation coverage improvement， the average error is 3.05%， the timing intersection method classification map effect is second， the vegetation coverage extraction accuracy is high but not stable enough， the average error is 7.39%； the otsu threshold segmentation classification map effect is poor， the average error is 15.39%，And the threshold determined by SVM and VDVI histogram.
5G signal towers have become an indispensable infrastructure in people's lives， and their digital reconstruction is critical for safety monitoring and management. In this study， a novel method for multi-view point cloud registration based on improved Rotational Projection Statistical feature descriptors （RoPS） and maximum spanning tree is developed by taking advantage of the multi-site laser scan data. Prior to the coarse registration conducted by the improved RoPS descriptor and random sample consensus method， the noise was removed by applying the filter based on the distance-weighted radius. Iterative Closest Point （ICP） was employed for accuracy improvement and then calculating the overlap of point clouds and having the maximum spanning tree constructed based on the breadth-first search， a finer global registration combining points from multi-site was implemented at the proper pose. The Stanford dataset and ground 3D laser scanning data of signal towers were used for validation. The result shows that our method achieved 0.07 m in accuracy， indicating this reliable technique can support the 3D model reconstruction and safety monitoring of signal towers.
Global warming has led to frequent extreme weather events. As a vital ecological barrier and economic zone， Southwest China has experienced recurring droughts in recent years， posing severe threats to water security， agricultural production， and the ecological environment. Therefore， systematically investigating the spatiotemporal evolution patterns and future trends of drought events in Southwest China holds crucial theoretical and practical significance for deeply understanding regional drought dynamics and establishing effective disaster prevention and mitigation strategies. This study utilizes GRACE data to explore the spatial and temporal evolution characteristics of land water storage and groundwater storage in Southwest China between 2002 and 2022. It diagnoses the drought characteristics and future development trends in Southwest China based on the GWSA Drought Severity Index （DSI）. The results show that： （1） the reconstructed quantitative results of water storage and drought severity index are reliable， which can provide a scientific basis for drought evaluation； （2） both land water storage and groundwater storage have obvious seasonal fluctuations and continue to increase， with Guizhou and Chongqing being the most significant ones； （3） a total of 15 droughts have occurred during the 20-year period， with the highest frequency in spring and winter， and the most severe drought is the winter-spring drought in 2010； （4） The droughts in southwest Yunnan， west Sichuan， and south Guangxi have increased significantly during the past 20 years， and the regions of east Sichuan， Chongqing， Guizhou， and north Guangxi will face more severe droughts in the future. The study indicates that despite the overall increase in water storage in the Southwest， frequent and spatially unevenly distributed droughts necessitate enhanced water resource monitoring and drought prevention and control in high-risk areas to provide scientific support for regional disaster prevention and mitigation.
Under the background of “Double carbon”， the Yellow River basin is an important energy base and the national ecological corridor along the Yellow River. The study takes the cities in the upper reaches of the Yellow River as an example， and builds a carbon emission estimation model based on luminous remote sensing data， a 21-year analysis of the spatiotemporal pattern of carbon emissions using gravity models， standard deviation elliptic models， and hot spot analysis， the carbon emission intensity of cities in the upper reaches of the Yellow River in 2030 was predicted by CA-Markov model. The results show that the carbon emission in the upper reaches of the Yellow River has a rising trend from 0.52×109 t to 1.78×109 t from 2000 to 2020， in general， there is “Strong gravity in the northeast， weak gravity in other regions” and the carbon emission center of gravity shifts to industrial cities such as Ordos， while the MOLAIN index is relatively stable at the city and county levels in the upper reaches of the Yellow River， cities such as Baotou， Hohhot and Shizuishan have high concentrations， while cities such as Longnan and Xining have low concentrations. The spatial distribution of carbon emission cold and hot spots shows significant changes in the degree of agglomeration， with the river source area gradually increasing and to the east， the canyon area relatively stable， and the alluvial plain area showing fluctuating changes.The carbon emission intensity of cities in the upper reaches of the Yellow River shows little change by 2030.Carbon emissions in the upper reaches of the Yellow River exhibit significant spatial heterogeneity， with industrial cities showing high emissions. Future mitigation efforts should focus on the northeastern and industrial areas， implementing clean energy substitution， regional coordinated reduction， and enhanced regulatory mechanisms to achieve basin-wide carbon reduction and promote green and sustainable urban development.
To address the challenge of food demand brought about by global population growth， detecting crop row centerlines automatically in precision agriculture is particularly important. However， the traditional method of individually detecting crops is inefficient. This study proposes a row centerline extraction method based on deep learning for the early stage of corn growth， aiming to improve crop production management efficiency. Based on the Faster R-CNN model for processing UAV remote sensing images， target detection boxes are obtained by selecting the optimal weights for validation. Perform binary segmentation on the image within the detection box， combine Susan corner detection method to obtain feature points， and optimize using dynamic circle optimization method. Finally， the least squares method and RANSAC algorithm were used to fit the row centerlines. After experimental verification， the average accuracy of detecting corn crop row centerlines is relatively high， about 0.004 8 °， which is better than the position clustering method. This not only proves the effectiveness of the method proposed in this study， but also provides an efficient and accurate technical means for crop monitoring in precision agriculture， which is expected to promote the intelligent and refined development of agricultural production.
The Datong River Basin is located at the border of Gansu and Qinghai provinces. It is the largest sub-basin of the Huangshui River Basin and an important component of the Qilian Mountains. The middle and lower reaches of the basin feature narrow， deep valleys and steep mountains， with abundant water and forest resources. Forest resources play a crucial role in water conservation， climate regulation， and soil and water preservation for this basin and its surrounding areas. Based on the Google Earth Engine （GEE） cloud platform， long-term time-series Landsat satellite imagery from 1987 to 2023 for the study area was obtained， and cloud-free images during the growing season （June to September） were composited annually. Using an improved VCT （Vegetation Change Tracker） algorithm combined with an automated forest sample extraction method， this study monitored the spatiotemporal changes and disturbances of forests in the basin over the past 37 years. The research indicates that forests are primarily distributed along the river valleys and mountain slopes in the middle and lower reaches of the basin. The improved VCT detection algorithm achieved an overall accuracy of 90.63% and a Kappa coefficient of 0.85 in identifying three categories： forest， non-forest， and forest disturbance， demonstrating the method's effectiveness and reliability in monitoring long-term forest change. Forest disturbances were more concentrated between 1989 and 2000， while they were relatively fewer from 2000 to 2023. The findings provide data support for forest resource management and ecological protection in the basin. The improved VCT detection algorithm combined with the GEE platform proves to be an effective tool for monitoring forest disturbances and dynamic changes.
