海岸带遥感
Theme Keywords： wetland restorationvegetation typesvegetation dynamicvegetation distribution patterntime series analysisspecies invasionscientific issuessample generation methodsample automatic augmentationsample adaptive transfer
- The Paper
- Abstract：Generating samples from long-term remote sensing imagery is crucial for land cover classification, surface change monitoring, and land use pattern analysis. However, traditional supervised classification methods require extensive labeled samples, thereby increasing time and labor costs while limiting classification accuracy and reliability. Aiming to overcome these challenges, this study proposes a generation method for long-term remote sensing imagery samples based on the combination of automatic sample augmentation and adaptive transfer. The objective is to achieve “one-time sample annotation with multiple reuses” to enhance efficiency and applicability in large-scale remote sensing classification tasks.The proposed method comprises two key components sample automatic augmentation and adaptive sample transfer. First, a local clustering algorithm is used to identify potential sample pixels based on spectral similarity, thereby expanding the labeled dataset. Change analysis between bi-temporal remote sensing images is then conducted to identify transferable samples, and a domain similarity rule is adopted to ensure consistency across different temporal images. This approach ensures the effective reuse of sample information from one time period to another, reducing the need for repeated manual annotations. The above steps are integrated into an interactive algorithm, thereby allowing sequential image processing in long-term remote sensing datasets. The proposed method was validated using Landsat 8 OLI time-series imagery of the Hangzhou Bay area, spanning from 2013 to 2022.Experimental results demonstrate the following1) The proposed automatic sample augmentation strategy effectively increases the quantity and quality of training samples, leading to improved classification performance and accuracy. 2) The adaptive sample transfer strategy facilitates successful sample migration across different temporal images, thereby eliminating the need for annual manual labeling and markedly enhancing sample generation efficiency. 3) The proposed approach is also robust across multiple classifiers, including support vector machine and k-Nearest Neighbor, indicating its broad applicability in classification tasks involving remote sensing.The proposed method presents a considerable advancement in long-term remote sensing image classification by successfully supplementing and transferring samples, thereby reducing annotation costs and enhancing classification efficiency. Through the combination of automatic augmentation and adaptive transfer, this approach presents a scalable solution for large-scale, long-term remote sensing image classification, ensuring reliable and cost-effective land cover analysis. The experimental results highlight the effectiveness of the proposed method in enhancing classification accuracy and efficiency, making it a valuable tool for remote sensing applications.Keywords：sample generation method;land cover classification;sample automatic augmentation;sample adaptive transfer;Hangzhou Bay;long-term remote sensing imagery689|1059|1
<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30 - Abstract：Islands feature unique geographical characteristics and distinct ecosystems. Analyzing the temporal dynamics of carbon sequestration in island ecosystems is critical for supporting the development of low-carbon and ecological islands. This study uses GF-1 and Sentinel-2 multispectral satellite imagery to classify vegetation types across the Changdao Archipelago. Combined with topographic and geomorphological subdivision data, these datasets were analyzed using a carbon sequestration rate approach to quantify vegetation carbon storage capacity during 2015—2022. The methodology further examines spatiotemporal change patterns, identifies the drivers behind vegetation carbon sink enhancement, and evaluates the potential for afforestation-induced carbon sequestration. Results show the following: (1) From 2015 to 2022, vegetation coverage in Changdao exhibited an overall declining trend. Areas dominated by pure Pinusthunbergii stands and mixed forests decreased, while pure Robinia pseudoacacia stands and shrublands increased. In the southern area of the North Five Islands, vegetation coverage declined substantially, demonstrating the transition of extensive shrublands to non-forest grasslands, which represents a primary form of vegetation degradation. (2) Interannual vegetation carbon sequestration in Changdao showed a fluctuating but increasing trend, revealing a “multicore growth and banded weakening” spatial pattern. Substantial north-south differences were observed in the inflection points of vegetation carbon sequestration across islands. “Stable zones” of vegetation carbon sequestration exhibited continuous enhancement, with key sequestration inflection points occurring in 2016 and 2019. Additionally, vegetation carbon sequestration demonstrated notable variations with elevation and slope gradients. (3) Key factors that influence carbon sequestration variation included the spread of pine wilt disease, extreme climate events, drought stress, and tree aging. The afforestation suitability index across Changdao ranged from 17.96 to 81.98, with the North Five Islands identified as the priority area for afforestation efforts. This study provides theoretical and empirical support for the development of Changdao as a “zero-carbon island.”Keywords：island city;vegetation types;Carbon sequestration;Carbon sink enhancement potential;Changdao381|460|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：Accurate monitoring of salt marsh vegetation phenology is crucial for understanding the carbon cycle in “blue carbon” ecosystems. High spatiotemporal resolution satellite remote sensing technology facilitates detailed monitoring of vegetation phenology; however, the presence of “salt-and-pepper” noise is an inevitable challenge. This study focuses on the Yellow River Estuary Wetland and adopts an object-oriented method combined with high-resolution remote sensing data to investigate coastal salt marsh phenology. First, multiscale segmentation is applied to Jilin-1 images to extract salt marsh vegetation objects, serving as basic units for phenological parameter extraction. Using time-series NDVI from PlanetScope images, phenological parameters, including the start date of the growing season (SOS), end date of the growing season (EOS), and length of the growing season (LOS), are extracted using S–G(Savitzky-Golay) filtering, a double-logistic model, and dynamic thresholding methods. Results are assessed from the following three aspects: (1) the fitting accuracy of the time-series NDVI, (2) the spatial heterogeneity of the extracted phenological parameters, and (3) consistency with observations from a phenological camera. Results indicate the following: (1) Compared to pixel-based approaches, the object-based time-series NDVI fitting achieves lower root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Specifically, the area with RMSE<0.05, MAPE<15%, and MAE<0.035 increased by 11.46%, 12.93%, and 10.72%, respectively, demonstrating improved fitting accuracy at the object scale. (2) The extracted phenological parameters are similar for object- and pixel-based approaches, which capture the spatial heterogeneity of salt marsh vegetation phenology. Conversely, object-based parameters are spatially smoother, mitigating the small-scale variability in pixel-level phenological parameters. Spatial heterogeneity analysis through semi-variogram functions reveals substantially lower nugget (C0) and partial sill (C) values for object-based parameters than those for pixel-based parameters. (3) Object-based phenological parameters exhibit a high degree of consistency with those obtained from the phenological camera (with SOS matching exactly, and EOS and LOS differing by only one day), whereas pixel-level parameters exhibit remarkable variations. This study reveals that object-oriented image analysis effectively reduces salt-and-pepper noise in high-resolution remote sensing images and holds strong potential for high-resolution phenology extraction in salt marsh wetlands.Keywords：Saltmarsh wetlands;phenology;Jilin-1;PlanetScope;high resolution969|2280|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：Ocean-Land Interfaces (OLIs) are instantaneous boundaries between fluctuating ocean surfaces and land and provide fundamental information for scientific research on ocean hydrology, ocean-land resource management, and sea level rise. However, detecting OLIs with high accuracy and resolution in automated manners is a challenging task. Airborne oceanic LiDARs (AOLs) are high-resolution, efficient, and flexible measurement systems that can be used for integrated ocean and land measurements. In areas where the ocean meets land, both ocean and land may exist in the laser spot of AOL, resulting in mixed ocean-land waveforms. If these mixed waveforms can be accurately identified, they can be used to detect the precise location of the ocean-land interface.Considering the existence of mixed ocean-land waveforms, a mixed waveform method based on AOL mixed ocean-land waveforms is proposed for ocean-land interface determination in this paper. First, the waveform features of the AOL infrared lasers are extracted, and principal component analysis is performed to reduce redundant features. Second, the waveform features are used to classify the AOL waveforms to obtain a membership matrix, and the Otsu method is used to determine the mixed ocean-land waveforms. Third, the DBSCAN algorithm is used to identify and eliminate misclassified mixed waveforms. Fourth, the PAEK algorithm is applied to smooth the laser points corresponding to the mixed ocean-land waveforms and output the ocean-land interface. Finally, the expression of the infrared laser-radar equation for mixed ocean and land is provided, and a method combining theoretical analysis and measured data verification is used to analyze the differences between ocean, land, and mixed ocean and land waveforms.The correctness and effectiveness of the methods proposed in this paper were verified via raw AOL datasets collected by the Optech CZMIL system. Compared with the traditional AOL elevation threshold method, the proposed AOL mixed waveform method reduced the mean and standard deviation of the ocean-land interface bias by 24.07% and 9.76%, respectively, and improved the SSIM index by 0.031, providing a new approach for detecting the ocean-land interface on tidal flats via AOL.The coexistence of water and land within AOL laser spots generates mixed ocean-land waveforms, and identifying these mixed waveforms has important theoretical and practical value. This study proposes an identification method for infrared laser mixed waveforms based on waveform fuzzy classification and Otsu threshold determination and an ocean-land interface extraction method using those identified mixed waveforms. Furthermore, this study extends the laser-radar equation by proposing an infrared laser-radar equation for infrared laser interactions with mixed ocean and land, providing a theoretical basis for studying mixed infrared laser waveforms. On the basis of this equation, a differential analysis was conducted on the ocean, land, and mixed ocean-land waveforms. The correctness and practicality of the infrared laser-radar equation for mixed ocean and land were verified via visualization analysis results of raw waveform data.Keywords：Airborne oceanic LiDAR;infrared laser;ocean-land waveform classification;mixed ocean-land waveform;ocean-land interface635|1416|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：As a critical link between terrestrial and aquatic ecosystems, wetlands provide essential ecological services and are vital for biodiversity conservation. However, the complex vertical layering of vegetation and unique hydrological conditions in karst wetlands pose remarkable challenges for effective vegetation distribution monitoring, thereby limiting deeper insights into wetland ecology. Aiming to address this issue, this paper proposes and implements a 3D vegetation mapping and analysis method based on LiDAR point cloud semantic segmentation, using the Huixian Karst Wetland of International Importance in Guilin, China, as the study area. This method adopts a deep learning point cloud semantic segmentation algorithm, DWS-KP-FCNN, to perform detailed 3D mapping of wetland vegetation. Several post-processing methods are applied to refine the 3D mapping results and improve their quantification potential. Based on the 3D vegetation map, the proposed method quantifies the volume distribution, proximity to water, and inundation frequency of each vegetation type, revealing the relationship between vegetation distribution and wetland hydrology. Results of the study reveal the following: (1) The deep learning algorithm, DWS-KP-FCNN, accurately identifies and classifies various vegetation types from LiDAR point clouds, effectively addressing challenges such as vegetation overlap and water body detection through post-processing techniques. Thus, this algorithm ultimately produces fine-grained high-precision 3D vegetation distribution maps. (2) Using the 3D vegetation distribution map, a quantitative analysis of vegetation volume, proximity to water, and inundation frequency reveals clear distribution patterns along the hydrological gradient. These patterns include vegetation clustering, mutual shading, optimal proximity to water, inundation frequency ranges, and the sensitivity of vegetation to hydrological changes. (3) Using a hierarchical clustering algorithm, the study area is divided into zones featuring distinct vegetation patterns based on proximity to water and inundation frequency. In shallow and near-water areas, karst wetland endemics, such as some bamboo and shrub species, are dominant, contributing to high species diversity and ecological value. Conversely, deeper zones reveal dense populations of invasive species such as water hyacinth, thereby posing potential management challenges that require effective control measures. These findings emphasize variations in vegetation communities across hydrological environments and provide valuable data for targeted wetland ecosystem management.Overall, the 3D vegetation mapping method, based on LiDAR point cloud semantic segmentation, offers an efficient, accurate, and comprehensive remote sensing approach for monitoring vegetation in karst wetlands, offering substantial benefits for ecological protection and management.Keywords：karst wetland;UAV LiDAR;3D vegetation mapping;point cloud semantic segmentation;deep learning;vegetation distribution pattern599|1280|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：Controlling the invasive plant Spartina alterniflora is a crucial aspect of ecological protection and restoration in coastal wetlands in China. Over the past decades, S. alterniflora has rapidly colonized tidal flats and estuarine areas, and led to a remarkable decline in biodiversity, degradation of wetland ecosystem services, and caused increased challenges for coastal management. Multiple coastal provinces in China are currently implementing S. alterniflora control projects, adopting methods such as physical removal and chemical control. The former is effective but prone to recurrence, while the latter is cost-effective but may have negative environmental impacts. Timely monitoring of S. alterniflora control dynamics and identification of the control methods are crucial for evaluating the effectiveness of control projects and assessing their environmental impacts. This study focuses on the S. alterniflora area in the coastal wetlands of Zhejiang and Shanghai and aims to introduce a method based on time-series optical satellite imagery to identify the removal status and control methods (unremoved/physical removal/chemical control) of S. alterniflora and estimate the dates of physical removal.A novel framework based on dense time-series optical satellite imagery (Sentinel-2 MSI and Landsat 8 OLI) is proposed. Specifically, a time-series spectral index dataset was constructed through the fusion of Sentinel-2 and Landsat 8 observations, followed by cloud masking and spectral harmonization to ensure consistency. First, periods that are affected by control measures were identified using a sliding window combined with a rule-based decision approach, and unremoved pixels were separated from those experiencing removal. Subsequently, a random forest classifier was trained using field survey data and high-resolution validation imagery to distinguish physical removal from chemical control. The exact removal dates for physically removed areas were further estimated by analyzing rapid decline in vegetation indexes, including NDVI, EVI, and LSWI, as well as abrupt increments in DFI.The results indicate that this method achieved high classification accuracy for control status and control method classification, revealing an overall accuracy of 98.8% and a Kappa coefficient of 0.979. The Mean Absolute Error of estimated physical removal dates was only 3.91 days, and date recognition accuracy was 93.67%. Spatial analysis revealed substantial differences between regions: in 2023, Shanghai achieved a removal rate of 4.2%, with nearly equal proportions of physical and chemical control, while Zhejiang achieved a substantially higher removal rate of 62.7%, dominated by physical removal operations.The proposed framework not only provides a reliable means of tracking S. alterniflora control projects but also facilitates differentiation of control strategies at regional scales. Such information is crucial for evaluating the effectiveness of invasive species control, supporting ecological restoration planning, and minimizing unintended environmental impacts. Furthermore, the methodology is scalable and can be extended to other coastal provinces or adapted for monitoring different invasive species across various environmental conditions.Keywords：optical remote sensing;species invasion;time series analysis;wetland restoration;vegetation dynamic988|1426|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：Over the past four decades, remote sensing technology has made remarkable progress, leading to unprecedented resolution and coverage in coastal zone observations, and ushering in the era of big data. However, when addressing practical challenges, a major issue lies in the effective processing and accurate analysis of large-scale coastal remote sensing data. Artificial Intelligence (AI) has rapidly developed in recent years, leading to the emergence of numerous Deep Learning (DL) models and their extensive application in big data analytics and real-world problem solving. The integration of AI with coastal remote sensing has driven progress in various application fields, introducing considerable value and benefits to society.This paper reviews major AI-driven advancements in coastal zone remote sensing, placing emphasis on model efficiency for coastal flood monitoring, waterline extraction, raft aquaculture zone management, green tide monitoring, and coastal wetland monitoring. For instance, AI algorithms can process Synthetic Aperture Radar (SAR) data in real time to assess flood extent. Thus, these models can accurately detect inundated areas and track their progression, offering timely information to support emergency response efforts.The article highlights waterline extraction as another crucial application of AI in coastal zone remote sensing. Aiming to achieve accurate and automated waterline delineation, AI-based algorithms can efficiently analyze large volumes of remote sensing data. They are especially effective in complex coastal environments, facilitating the detection of subtle changes in dynamic shorelines. Through the integration of multi-source data, AI also enables real-time monitoring and forecasting of coastal erosion, supporting effective coastal management and protection.AI has also demonstrated remarkable potential in monitoring the distribution and temporal dynamics of raft aquaculture zones, contributing to highly efficient resource management. Deep learning models can accurately outline aquaculture boundaries and monitor water quality and facility conditions. Thus, AI enhances aquaculture oversight, optimizes resource allocation, and helps mitigate environmental pressures by leveraging remote sensing data and real-time analytics.Moreover, AI plays a crucial role in green tide monitoring. Through spectral analysis, deep learning algorithms enable the rapid detection of algal bloom regions and mapping of their spatial spread with high precision, even under large-scale, data-intensive conditions. These capabilities support timely environmental assessments and the development of early warning systems to reduce ecological risks.In coastal wetland monitoring, AI enables rapid classification and change detection using multi-sensor and multi-temporal observations. Deep learning models help effectively track wetland dynamics, evaluate ecosystem health, and identify degradation patterns, thereby supporting biodiversity conservation and ecological restoration planning.Finally, this review outlines future directions for AI in terms of coastal remote sensing. With continuous advancements in computational capacity and algorithm design, AI applications are also expected to become highly accurate, scalable, and indispensable. They can offer critical support for addressing climate change, coastal erosion, and sustainable coastal management challenges.Keywords：Coastal Zone Remote Sensing;artificial intelligence;image processing;Flood Monitoring;Waterline Delineation;Raft Aquaculture Zone Monitoring;Green Tide Detection;Coastal Wetland Monitoring1313|5582|0<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
- Abstract：The coastal zone is an ecologically crucial zone where land and ocean interact, collecting a considerable amount of matter and energy. However, this area is currently facing unprecedented challenges due to intensified human activities, global climate change, and species invasion. Remote sensing science and technology provide effective means for the comprehensive and systematic monitoring of coastal zone resources, ecosystems, and environmental conditions. However, our understanding of key scientific issues relative to remote sensing applications in coastal zones remains unclear. Several issues and challenges are encountered in the design of remote sensing sensors, the interaction mechanisms between electromagnetic waves and surface objects, remote sensing data processing and information extraction, quantitative retrieval of ecological parameters, and the cross-application of remote sensing across various fields of coastal zones. Considering the uniqueness of coastal zones and the advantages of remote sensing technology, this paper analyzes the current status of domestic and foreign research on coastal remote sensing, using data from the Web of Science and CNKI databases. It explores the scientific and technological issues in coastal remote sensing, summarizes research progress, identifies opportunities and challenges, and discusses potential future development directions. Results show the following: (1) Considering the characteristics of high spatial and temporal heterogeneity, changing climate, and complex surface factors in the coastal zone, developing a new generation of remote sensing payloads is necessary. These payloads should feature specific spectral characteristics, observation modes, orbital modes, and orbital inclinations to provide support for large-scale, high-frequency monitoring and fine-scale detection of natural resources in the coastal zone. (2) Aiming to accurately analyze the physical and optical characteristics of the coastal zone environment, geoscience big data is integrated with numerical simulation technology to develop precise models of the scattering and absorption characteristics of atmospheric and marine materials and clarify radiation transmission process across the atmosphere-land-water interface. (3) Comprehensive methods should be developed to improve remote sensing image quality under complex coastal zone imaging conditions. through remote sensing AI model and cloud computing technology, to achieve high-precision and intelligent extraction of remote sensing information across the entire coastal zone. (4) Existing radiation transfer, light energy utilization, and process models should be optimized to analyze the radiation transfer mechanisms of coastal surface elements. A “mechanism-data” dual-driven quantitative inversion model of surface parameters, powered by an opportunistic AI model, is introduced to address inversion accuracy errors caused by differences in the radiation characteristics of multi-source sensors, as well inconsistencies in observation angles and observation times. (5) The cross-application of land-sea integration, ecological restoration, and disaster prevention and reduction in coastal zones has seen marked advancements, promoting deeper interdisciplinary integration and contributing to the development of a more comprehensive scientific and technological system. This paper can serve as a reference for understanding the key scientific challenges in coastal remote sensing and for identifying future directions in its development.Keywords：remote sensing of coastal zone;scientific issues;design of sensors;interaction between ground objects and electromagnetic waves;information extraction;quantitative inversion of remote sensing parameters;practice of interdisciplinary;challenge and opportunity1694|8186|1<HTML><L-PDF><Enhanced-PDF><Meta-XML>Updated：2025-12-30
