最新刊期
卷 27 ， 期 6 ， 2023
-
摘要：Sustainable ecosystem management requires considerable wetland spatial information given the evident climate change impacts and human disturbances on wetlands. Remote sensing of wetlands, as an important interdiscipline, has increasing publications. Here, we searched for published papers in the past 50 years from the “Web of Science Core Collection” database. We summarized the changes in the number of publications and citations and the development process and trend in remote sensing of wetlands. We divided the development history into three research periods including potential exploration phase, framework emerging phase, and rapid growth phase. Based on the development history over the past 50 years and facing the background of wetland ecosystem protection demand in the era of big data, studies on remote sensing of wetlands have developed in the direction of fast, multisource, and fine, such as wetland intelligent classification, remote sensing inversion of large-scale wetland vegetation ecological parameters, and wetland ecosystem health assessment. However, the spectral and backscattering characteristics of wetlands are complex due to the interaction of water, vegetation, and soil, and their annual/inter-annual variation characteristics are notable, aggravating the difficulty of remote sensing detection of wetlands. This condition is a key issue that requires resolution at present. Thus, the multimodal remote sensing experiments of wetlands should be strengthened. We also concluded the main research topics and data sources in different phases and analyzed the hotspots in remote sensing of wetlands by the extracted keywords from 500 latest and top-cited papers. Three outlook bullets were presented from the wetland classification and landscape dynamics in the era of big data, the fine remote sensing observations in wetland ecological variables, and the spatial decision support for sustainable wetland management. Based on cloud platforms (such as GEE), carrying out large-scale and long-term wetland mapping and landscape dynamic analysis by means of time series remotely sensed data (i.e., Landsat and Sentinel), investigating the application potential of diverse machine learning algorithms (i.e., random forest and deep learning) for wetland ecological parameter inversion at different geographic scales, establishing a scientific indicator system, and fully applying the multisource and multiplatform remote sensing observation technology to solve the actual ecological environment problems are important development trends and research hotspots of future wetland remote sensing studies. We hope that this review not only provides a glimpse, but also a framework understanding of wetland remote sensing research. With the improvement on the awareness of the importance of wetland ecosystem, the number of scholars engaged in research of remote sensing of wetlands increases. The review is expected to be beneficial for understanding the development history and international frontiers for studies in remote sensing of wetlands and to support their layout domestically and abroad.关键词：global scale;remote sensing of wetlands;review;long time series;big data;artificial intelligence;cloud platform;sustainable development- 1324
- |
- 784
- |
- 0
发布时间：2023-07-13 -
Research progress and prospects of hyperspectral remote sensing for global wetland from 2010 to 2022
摘要：Wetlands are transitional zones between terrestrial and aquatic ecosystems and play important roles in maintaining ecological balance, protecting ecological diversity, conserving water sources, and regulating climate. However, traditional field investigations and panchromatic and multispectral remote sensing technologies cannot meet the practical needs of current wetland monitoring. Hyperspectral remote sensing technology has become an important approach for wetland monitoring owing to its advantages of high spectral resolution and rich spectral information. This review summarizes the related literature on the hyperspectral application of wetlands from 2010 to the present.First, the literature was analyzed using CiteSpace software. Then, the country/institution of authors, international cooperation, keywords, research hotspots, and research trends were clarified. Finally, the feature extraction and dataset processing methods of hyperspectral datasets and their progress in wetland mapping and quantitative inversion were determined.China and the United States are the top two countries in terms of the number of hyperspectral wetland studies, but only a few international collaborations have been pursued. In addition, the classification of vegetation in wetlands is a hot research topic. Spartina alterniflora, reed, water quality, and soil properties have become the focus of hyperspectral wetland research. Machine learning methods represented by Random Forests (RFs) play an important role in wetland hyperspectral research. However, studies on classification and inversion based on deep learning methods are limited. Furthermore, under the background of global warming, coastal wetlands have received widespread attention from researchers worldwide. For hyperspectral remote sensing sensors, China’s spaceborne hyperspectral platforms have developed rapidly, but foreign countries have dominated ground and near-ground hyperspectral remote sensing platforms, with a spectral coverage range of 350—1000 nm. In terms of hyperspectral information extraction and image processing, studies have mainly focused on traditional feature extraction and classification methods, such as PCA, MNF, RF, decision trees, and spectral angle mappers. The processing and feature extraction of hyperspectral data based on deep learning feature extraction is expected to be an important research direction in the future. Hyperspectral wetland mapping mainly focuses on wetland vegetation, mangroves, and salt marsh vegetation. Nonetheless, the scale of existing research has been limited to small areas, such as nature reserves or national wetland parks, and the mapping algorithm continues to rely on traditional methods, such as RF and support vector machines. More refined tree species identification and mapping from the use of hyperspectral images is a relevant future research direction. Research on hyperspectral wetland quantitative inversion has mostly focused on chlorophyll and aboveground biomass. In the inversion process, the sensitive band is determined using the correlation coefficient between the ground measurement and the hyperspectral band or spectral index. Simple models, such as linear, quadratic polynomial, and logarithmic functions, are subsequently constructed to obtain the estimated biophysical parameters. Deep learning algorithms have good application prospects in hyperspectral band feature selection and inversion estimation models. Moreover, given the complexity of wetland vegetation, small-scale or point-scale parameter inversion is taken as the main research scale. Large-scale hyperspectral wetland quantitative inversion is difficult to implement due to the existence of high wetland heterogeneity. The resolution of hyperspectral images is not high enough, and mixed pixels exist.The fusion of multisource remote sensing data, such as multispectral-hyperspectral fusion, to improve the resolution or the development of corresponding spectral unmixing algorithms is the future direction of quantitative analysis for hyperspectral remote sensing applications in wetlands.关键词：hyperspectral remote sensing;wetlands;literature analysis;hyperspectral payload platform;information extraction;mangrove forest;salt marsh;quantitative inversion- 960
- |
- 864
- |
- 0
发布时间：2023-07-13
Review
-
Wetlands mapping in typical regions of South America with multi-source and multi-feature integration
摘要：Wetlands play an important role in maintaining ecological balance, conserving water resources, recharging groundwater, and controlling soil erosion. They are often called the “kidneys of the earth” because they help purify water by filtering out pollutants and sediments. South America has a vast area of wetlands, as well as a variety of wetlands types. While most of these wetlands were conserved in a relatively good condition until a few decades ago, pressures brought about by land use and climate change have threatened their integrity in recent years. However, no complete and uniform wetland map has provided adequate information on the location, distribution, size, and changing status of wetlands in South America. Remote sensing has been an effective tool for characterizing, mapping, and monitoring the complexity and dynamics of large areas of wetlands. Although fine wetland mapping may be done by combining data from many sources, the following two issues persist. On the one hand, given the complicated temporal dynamics and spectral heterogeneity of wetlands, large-scale wetland mapping remains a challenging task. On the other hand, supervised classification is a widely used technique for multi-category wetland mapping. However, selecting training samples is time consuming and labor intensive. Moreover, finer and more precise wetland information is currently unavailable for reference. In the study, we selected four study areas of typical wetlands in South America. First, an effective wetland sample collection process was proposed by using the existing land cover dataset to ensure the sample quality. Second, a multi-source feature set was constructed by combining Sentinel-1, Sentinel-2, and SRTM data. Then, feature selection is carried out on the basis of the random forest recursive feature elimination method (RF_RFE). We constructed a multi-feature combination scheme to compare the influence of multi-source features on wetlands classification. Finally, the random forest algorithm is used to classify wetlands in the study area. The research results show that the process facilitates the sample collection and improves the sample quality. The combination of Sentinel-1 and Sentinel-2 data can improve the accuracy of land cover mapping, and terrain features help greatly improve the overall classification accuracy and the accuracy of various types of objects. For wetland categories, the addition of multi-source data features can improve the separability of wetland categories. The feature selection based on RF_RFE can reduce the feature redundancy and improve the classification accuracy. The feature optimization results show that SAR polarization features and derived texture features can be used as an effective supplement to optical features. However, the dominant features account for less. The overall classification accuracy of the study area was 85.62%, and the Kappa coefficient was 0.8333. The study proposed an effective classification process and sample collection scheme to large-scale wetlands in South America. This study integrated Sentinel-1 synthetic aperture radar data, Sentinel-2 optical data, and terrain data to explore their importance to the extraction of different wetlands at a large scale. This work also verified the feasibility of feature selection based on random forest recursive feature elimination method. The research results reveal that the sample collection process facilitates sample collection and improves sample quality. The combination of Sentinel-1 and Sentinel-2 data can improve the accuracy of land cover mapping, and terrain features help greatly improve the overall classification accuracy and the accuracy of various types of objects. For wetland categories, the addition of multi-source data features can improve the separability of wetland categories. The feature selection based on RF_RFE can reduce the feature redundancy and improve the classification accuracy.关键词：remote sensing;Sentinel-1;Sentinel-2;Google Earth Engine;wetlands classification;feature selection;South America- 1009
- |
- 318
- |
- 0
发布时间：2023-07-13 -
摘要：Remote sensing technology is widely used in mangrove forest mapping with the advantages of real-time mapping, accuracy, multiscalability, and repeatability. Until now, the mangrove forest dataset with the highest spatial resolution in China is produced by 10 m Sentinel data. In addition, most existing mangrove forest datasets in China ignore the influence of tide, leading to low spatial resolution and inaccurate mapping. On the basis of Chinese GF-2 images, this study aims to map Chinese mangrove forests in 2020 with a spatial resolution of 1 m under the tide. Specifically, 312 scenes of GF-2 image covering China's coastline (24 scenes of GF-1 image covers the areas without GF-2 images) were selected. First, the selected images were segmented using the object-based multiscale method, and the submerged mangrove recognition index was used as a tidal influence indicator. Finally, high-resolution mapping of mangrove forests in China was conducted using the random forest classifier. Results show that the mangrove area in China in 2020 was 29,576.48 ha, 95% of which was mainly distributed in the Guangxi, Guangdong, and Hainan Provinces. The overall classification accuracy was 92%, and the Kappa coefficient was 0.89. The mangrove area without tidal influence was 2,531.24 ha less than that with tidal influence. The high-resolution mangrove dataset generated in this study can provide high-precision data support for the monitoring, management, and evaluation of mangrove ecosystems in China, and it is valuable for practical application.关键词：remote sensing;GF-2;Mangrove forests;object-oriented;tide;random forest classifier- 1017
- |
- 447
- |
- 0
发布时间：2023-07-13 -
摘要：High-precision land cover data are an important research basis for ecosystem monitoring and assessment and regional sustainable development. However, studies on high-resolution land cover data (10 m) at the urban scale are few. With the establishment of wetland cities, high-quality and high-precision land cover data have become essential to provide information for relevant research.Taking the first wetland cities in China, as the study area, this study selected three sets of global 10 m land cover data (Dynamic World, ESA WorldCover, and Esri Land Cover) in 2020 and 2021, for spatial consistency evaluation and precision evaluation. First, the consistency and confusion between any two sets of data were calculated using the spatial superposition method, and a spatial consistency distribution map was drawn to analyze the spatial consistency of the three sets of data. Second, the confusion matrix was calculated by constructing verification sample points through visual interpretation to evaluate the accuracy of the three sets of data. Finally, based on the results of the spatial consistency analysis and accuracy evaluation, the land cover data sets were fused to produce a new set of data, and a fusion method based on spatial consistency analysis and accuracy evaluation was proposed.Results showed that (1) in any two data sets, the consistency of the water, forest, cropland, and urban areas was high, whereas the confusion of the wetland, grassland, and bare areas was high, and (2) the spatial consistency between ESRI and DW was the highest, with the highest proportion of consistent areas (more than 60%) and lowest proportion of inconsistent areas (less than 6%), which were distributed mainly in areas along the coast and rivers and with extensive wetlands. Such areas had strong heterogeneity and complex land cover types. (3) ESA had the highest overall accuracy, whereas DW and ESRI had a similar overall accuracy. The ESA wetland types had relatively high precision and classification details, which are suitable for urban wetland-related research. (4) Multisource land cover data can be effectively integrated, and the data accuracy of widely heterogeneous regions can be improved using the fusion method based on spatial consistency analysis and accuracy evaluation. The overall accuracy of the fusion results (more than 80%) and kappa coefficient was higher than that of the three original data.Therefore, the research results can provide data support and auxiliary decision-making support for wetland city certification and related research.关键词：remote sensing;land cover data set;wetland city;spatial consistency assessment;precision assessment;data fusion- 383
- |
- 429
- |
- 0
发布时间：2023-07-13
Wetland mapping at broad scale
-
摘要：The accurate identification of the wetland dynamic range is the basis for the protection and restoration of the wetland ecosystem. To maximize the advantages of high timeliness and large-scale repeated observation of remote sensing technology and in consideration of the characteristics of high spatial heterogeneity and high temporal dynamics of wetlands, all Landsat OLI time series image datasets available for Google Earth Engine were used to study the accurate identification of inland wetland dynamic range under water level fluctuation.Three typical inland wetlands were selected as research areas on the basis of the genetic factors and the hydrological factors of wetlands. In combination with the diagnostic characteristics of mature wetlands, i.e., hygrophyte and wet soil, the combination of water-wetness indices for defining wetland scope was selected. Image composition was used to determine the high and low water levels in one year. A modified fuzzy C-means algorithm was proposed to reduce the spatial heterogeneity of wetland background and improve the contrast and distinction between wetland and nonwetland boundaries. The maximum between-class variance (OTSU) was selected to determine the adaptive threshold of wetland disinflation boundary and then combined with the superposition rules of the water-wetness index dynamic combination scheme to identify wetlands within one year. Finally, a set of accurate identification technology of wetland dynamic range based on the “Elements-Index-Threshold” technology System (EITS) was constructed.Typical inland wetlands, such as the Guanting Reservoir, the Zoige wetland, and the Poyang Lake wetland, were selected as experimental areas to verify the applicability and accuracy of the set of technology system. Results showed that the extraction accuracy of wetland range was higher than 94%, and the Kappa coefficient was greater than 0.88.This research improves the accuracy and efficiency of wetland spatiotemporal dynamic range identification, hoping to provide effective support for long-term and large-scale wetland dynamic monitoring and mapping.关键词：remote sensing;defining the dynamic range of wetland;water-wetness index;dynamic process;water level fluctuation;time series data set;inland wetland- 881
- |
- 304
- |
- 0
发布时间：2023-07-13 -
摘要：Plant communities play an important role in wetland elements and are vulnerable to human activities and climate change. Wetland plant community classification and mapping provide scientific important data support for wetland ecological monitoring and evaluation. This study aims to develop a classification scheme suitable for the wetland plant communities in the Poyang Lake wetland.Taking Poyang Lake National Nature Reserve as the research area and on the basis of the monthly Sentinel-1 and Sentinel-2 time-series data in 2019, this study extracts five types of image feature parameters, including water and vegetation index group, red edge index group, texture feature group, spectral feature group, and polarization radar backscatter group, with a total of 240 feature indexes, and uses Random Forest (RF), Support Vector Machine (SVM), and Deep Neural Network (DNN) algorithms for classification to explore a set of optimal feature combinations and a suitable classification scheme for wetland vegetation mapping in Poyang Lake.In conclusion, A classification scheme for wetland plant communities in the Poyang Lake wetland was proposed in this study using multi time-series Sentinel-2 and Sentinel-1 data. The optimal acquisition time periods of satellite data are in January, April, August, September, October, and December. The optimal image feature group can be red edge index group or water and vegetation index group for feature selection. The classification algorithm can select deep learning or RF algorithm to classify wetland plant communities according to the requirements. This classification scheme can effectively improve the accuracy of wetland vegetation mapping in the Poyang Lake and provide scientific and technical solutions for decision-making departments.Results show the following(1) Compared with radar data, the extraction accuracy of optical data is remarkably better than that of radar data in wetland plant community classification and mapping. Radar data can be used as a supplement to optical data when optical data are insufficient. (2) Screening the importance of each image feature of Sentinel-2 helps improve the classification accuracy. The preferred time periods are mainly distributed in January, May, August, September, October, and December. (3) Five groups of unitary image features are selected to classify separately, and the classification accuracy is as follows: red edge index group > water and vegetation index group > spectral feature group > radar polarization data group > texture feature group. (4) Comparing the combined image feature groups with the unitary image feature groups reveals that the combined image feature group is not necessarily helpful to improve the classification accuracy. The classification accuracy is as follows: red edge index group > water vegetation index group > combined image feature group. Among them, the overall accuracy of the classification scheme using the red edge index group and random forest method is 0.81, and the Kappa coefficient is 0.76. (4) By comparing the three classification algorithms, the classification accuracy is ranked as follows: DNN > RF > SVM. The overall accuracy of the deep learning method does not greatly improve, that is, only 2% higher than the RF algorithm. Thus, the DNN and machine learning method (RF) can be used as optimization algorithms.关键词：remote sensing;Poyang Lake;wetland vegetation Mapping;Image features selection;random forest algorithm;Deep Neural Networks (DNN);Multi-temporal optical and SAR data- 1372
- |
- 603
- |
- 1
发布时间：2023-07-13 -
摘要：The biomass and growth of coastal wetland vegetation vary greatly due to different water and salt conditions in the growing area, and the spectral features of certain vegetation at the peak biomass are highly similar, making it easy for coastal wetland vegetation to be misclassified. In response to this problem, this study proposes a new semantic segmentation network called MFVNet to be combined with vegetation index for the fine mapping of coastal wetlands.In the proposed MFVNet, an Enhanced Multiscale Feature Extraction (E-MFE) module was first constructed on the basis of atrous convolution and attention mechanism to capture features of different scales adaptively. Then, the E-MFE module was used to replace the double convolution operations in traditional encoder-decoder network architecture, such as UNet. It was also used to merge the semantic features and detailed features of different resolutions to enhance feature representation. Finally, some typical vegetation indices were selected and input into the proposed MFVNet to improve the ability of coastal wetland fine mapping.The experiments of this study were conducted using GF-2 remote sensing images to study the coastal wetlands of the Yellow River Estuary. Experimental results indicated that the proposed MFVNet achieved good performance with an overall accuracy of 93.89% and a Kappa coefficient of 0.9072. On typical vegetation, such as reeds, spartina alterniflora, tamarix mixed area, and seagrass beds in the Yellow River Estuary, the F1 scores of MFVNet were 0.91, 0.87, 0.82, and 0.76, respectively, which were better than that of other methods. Moreover, ablation experiments showed that the combination of the E-MFE module and the vegetation index can increase the overall accuracy from 91.46% to 93.89%.(1) Compared with deep semantic segmentation networks, such as HRNet, MFVNet can more effectively extract vegetation information of coastal wetlands. (2) The proposed EMFE module can adaptively capture features of different scales and improve the overall accuracy, which justified its effectiveness in coastal wetland mapping. (3) The inclusion of vegetation index can enhance the spectral features of coastal wetland vegetation and improve the accuracy of vegetation information extraction, indicating the importance of vegetation index in coastal wetland mapping. (4) Simultaneously splicing modified soil adjusted vegetation index, difference vegetation index, and ratio vegetation index in remote sensing images contributed the most to the extraction of coastal wetland information.关键词：remote sensing;Coastal Wetland Information Extraction;GF-2;Deep Convolutional Neural Network;MFVNet Model;vegetation index- 1189
- |
- 2400
- |
- 0
发布时间：2023-07-13 -
摘要：The ZY-1 02D satellite was successfully launched in 2019 and officially used in October 2020. It is the first civil hyperspectral satellite independently developed and operated by China and has a wide application prospect. Taking the Yellow River Delta wetland as study area, this work investigated the performance of ZY-1 02D Advanced Hyperspectral Instrument (AHSI) images in wetland landscape classification. First, the ZY-1 02D AHSI spectral curves of typical land cover types in the study area were evaluated. Second, two levels of wetland landscape classification systems, including seven basic classes and nine refined classes considering the difference of vegetation coverage, were developed with the assistance of field surveys and unmanned aerial vehicle images. Then, the random forest algorithm was used for classification, and the tree SHAP method was introduced to sort and select important bands. The most important ZY-1 02D AHSI bands overlapping with Landsat 8 OLI bands were selected for classification, and the classification results were compared. Our results showed the following: (1) ZY-1 02D AHSI data can represent the differences of spectral features of different landscape types. (2) For the two classification systems, the first 40 important bands achieved the highest overall classification accuracy. The classification accuracy of seven basic classes and nine refined classes were 92.18% and 90.76%, respectively. Most of the important bands were visible and near-infrared bands. (3) For the two classification systems, a total of 24 and 29 bands overlapping with Landsat 8 OLI multispectral bands were selected for classification. The classification accuracy reached 90.01% and 89.76%, which were remarkably higher than that of Landsat 8 OLI. (4) Blue and green bands are important for the identification of high- and low-density Spartina alterniflora and Phragmites australis; blue, green, and short-wave infrared bands are important for Phragmites australis identification, and the red band is important for the identification of high- and low-density Suaeda salsa. Results showed that ZY-1 02D AHSI data have advantages in distinguishing different landscape features and differences in vegetation coverage. Our study is expected to provide scientific basis for the application of ZY-1 02D hyperspectral data in wetland ecological resources.关键词：ZY-1 02D;AHSI satellite image;hyperspectral data;vegetation coverage;Random Forest;Tree SHAP- 1680
- |
- 506
- |
- 0
发布时间：2023-07-13 -
摘要：Salt marshes are the world’s most valuable and vulnerable ecosystem. Thus, accurate and timely monitoring of the distribution of salt marsh vegetation is essential. With the accumulation of multi-source remote sensing imagery, the time-series method has increasingly become important for monitoring coastal areas. However, effectively constructing time series is still challenging as the number of available observations is relatively low owing to the frequent cloudy weather in the coastal areas. In this study, we coupled multi-sourced Landsat images and constructed a pixel-level time series with XGBoost. Based on this, the feasibility and stability of classifying salt marsh vegetation were tested using the three typical sites in the Yangtze River Delta. Results showed that (1) Inter-calibration for multi-sourced images was necessary for not only improving the availability of images but also reducing the spectral differences among sensors. (2) The performance of salt marsh vegetation classification based on the pixel-level time series was favorable, reflected by 81.50% as the mean overall accuracy and 0.755 as the Kappa coefficient. The classification results were excellent, particularly for the widely distributed Suaeda salsa and Spartina alterniflora in the Yangtze River Delta. (3) Compared with the single-phrase classifications, the pixel-level time series–based classifications were stable, evidenced by an inter-annual absolute mean error lower than 3.27%. Therefore, our proposed method is expected for dynamic monitoring of salt marsh vegetation, which facilitates managing coastal resources and implementing ecological conservation effectively of China’s coasts.关键词：remote sensing classification;salt marsh vegetation;Landsat;pixel-level time-series;XGBoost;Yangtze River Delta;Red-crowned Crane Nature Reserve;Jiuduansha Wetland;Southern Coastal Wetland in Hangzhou Bay- 402
- |
- 388
- |
- 0
发布时间：2023-07-13 -
摘要：Wetland vegetation plays an important role in the process of carbon sequestration. As a typical alpine wetland ecosystem, the Ruoergai wetland has been attracting increasing attention due to its carbon sink function, which makes the classification and change detection of its vegetation coverage crucial.This study aims to present a method for mapping the vegetation of the Ruoergai wetland and monitor its change by integrating Sentinel-2 optical data and Sentinel-1 Synthetic Aperture Radar (SAR) data, taking advantage of their respective advantages.We utilize the spectral characteristics of Sentinel-2 MSI data and adopt the dynamic time warping algorithm to extract the time-series phenological characteristics of Sentinel-1 SAR data; in this manner, different wetland vegetations can be differentiated easily. The random forest algorithm was used to combine both data for classifying wetland vegetation types. In-site samples obtained by UAV in 2020 are used and migrated to 2017 to train and validate the classification results.The classification results have an overall accuracy of 97.43% and a Kappa coefficient of 0.96. Vegetation in the Ruoerge wetlands remains overall unchanged from 2017 to 2020, with the changed area exhibiting a recovery trend (~7% of the total area).On the basis of the principle of sample migration, this study solved the problem in which the field samples were not available in the historical period. By combining Sentinel-1 and Sentinel-2 data, their respective multitemporal and multispectral advantages were fully exploited to obtain reliable classification results.关键词：remote sensing;Change Detection;training sample migration;SAR;wetland vegetation classification;Zoige wetland- 1831
- |
- 458
- |
- 1
发布时间：2023-07-13 -
摘要：Obtaining refined wetland resource information is important for the restoration, protection, management, and utilization of wetlands in international wetland cities and for regional sustainable development. At present, refined wetland classification research for international wetland cities is lacking, especially for the detailed classification of water bodies in wetlands. Refined wetland classification results could provide vital information support for the nomination and assessment of potential or existing international wetland cities.This study takes Changde City, a typical international wetland city, as the case study area. On the basis of the Google Earth Engine (GEE) cloud computing platform and Sentinel 1/2 time series remote sensing data and terrain data in 2020, the minimum redundancy-maximum correlation algorithm and the recursive gradient boosting tree algorithm are first used to optimize the wetland classification feature set. Then, an intelligent model for refined urban wetland classification integrating pixel-based random forest and object-oriented knowledge rule decision model is constructed to realize the refined classification of wetland resources in Changde City.Using multisource remote sensing data, the GEE cloud computing platform, a machine learning algorithm, and knowledge-driven rule-based model, this study can accurately and efficiently extract refine wetland information of international wetland cities. The methodology developed in this study could be operationally transferred to other urban wetland mapping and has great application potential in the nomination and management of international wetland cities as well as the restoration, protection, and sustainable development and utilization of wetland resources.The results are as follows(1) The number of features before and after the optimization of wetland classification features is reduced from 63 to 16, and the overall accuracy is reduced by 0.9%. The characteristics of water index, vegetation frequency, antd built-up area index in the dry period are of great importance, and feature optimization can reduce the redundancy of feature data and improve the classification efficiency. (2) The results of the fine classification of wetlands in Changde City include eight wetland types: rivers, lakes, reservoirs, aquaculture ponds/pits, canals, mudflats, sedge and reed with an overall accuracy of 91.53% and a kappa coefficient of 0.89. These results meet the requirements for the fine wetland classification of international wetland cities, indicating that the precision of the urban wetland refine classification method framework is high. (3) Wetlands in Changde City are mainly distributed in the eastern and western parts of the Dongting Lake Plain, showing a spatial pattern of more in the east and less in the west.关键词：wetland classification;Random Forest;knowledge-based rules;Sentinel-1/2;international wetland cities;Changde city- 432
- |
- 523
- |
- 0
发布时间：2023-07-13
Wetland fine classification method
-
摘要：Salt marsh vegetation is an important part of the blue carbon ecosystem, which has strong carbon sequestration and carbon storage capacity. The Leaf Area Index (LAI) determines its growth, photosynthetic active radiation absorption ratio, and biomass. LAI is often used to simulate the photosynthesis, respiration, and transpiration of vegetation, thus playing a crucial role in improving the yield of Suaeda salsa. An accurate estimation of Suaeda salsa can provide an important basis for judging the growth status of alkali ponies and thus provide an effective aid for monitoring salt marsh wetlands.To improve the accuracy of LAI estimation accurately and rapidly, the Yellow River Delta Suaeda salsa shoal wetland was selected, and the indigenous plant Suaeda salsa was used as the research object. A UAV hyperspectral remote sensing image was obtained, and the ground spectrum was measured in combination with regional soil factors, vegetation spectral characteristics, hyperspectral image texture characteristics, and vegetation coverage. Multimodal data, through Random Forest (RF) feature selection for multimodal data, and the RF-PSO-DELM algorithm with a dual optimization strategy was developed to construct an inversion model of the LAI of Suaeda salsa in coastal wetlands.The coefficient of determination (R2) and the Root Mean Square Error (RMSE) were 0.9546 and 0.1341, respectively. Compared with the inversion model accuracy of Suaeda salsa LAI constructed on the basis of the five algorithms of SVM, BP, ELM, DELM, and PSO-DELM, R2 was increased by 0.2654 at most, and the RMSE was reduced by 0.0828 at most.Compared with the traditional inversion model (SVM), the RF-PSO-DELM model had better generalization; moreover, the fusion of multimodal data could effectively improve the accuracy of the inversion model. This study further enriched the theory and technology for the accurate monitoring of salt marsh vegetation based on UAV hyperspectral remote sensing technology. Multisource modal data, such as soil factors, texture features, spectral features, and vegetation cover affecting the growth of alkali ponies in coastal wetlands, were comprehensively considered, and the important influencing factors sensitive to the LAI of alkali ponies’ LAI were extracted by the random forest feature preference algorithm, which effectively reduced the complexity of model inversion and greatly improved the accuracy of model prediction.关键词：UAV;Multimodal Data;Suaeda salsa;leaf area index;Random Forest;particle swarm optimization;Deep Extreme Learning Machine;Yellow River delta- 943
- |
- 386
- |
- 0
发布时间：2023-07-13 -
摘要：Wetlands, which are an important carbon pool on Earth, is crucial for human beings and the environment. An accurate estimation of wetland carbon storage and its temporal and spatial changes are conducive to understanding the sustainable development of wetland ecosystems. Net Primary Productivity (NPP) is the net accumulation of organic matter fixed by photosynthesis per unit time and per unit area of green vegetation and is an important indicator to characterize the status of carbon flux. Therefore, accurate estimation of the spatial patterns and temporal dynamics of wetland NPP at a regional scale is crucial to improving our understanding of the carbon dynamics and sustainable development of terrestrial ecosystems. In China, similar studies have mapped wetlands or estimated wetland NPP using optical data. However, only a few studies have used dense high-spatiotemporal-resolution multispectral images for wetland mapping and considered the accuracy of the light-use efficiency (ε) of wetland vegetation types for NPP estimation.In this study, we proposed an improved Carnegie-Ames-Stanford Approach (CASA) model to generate wetland NPP with high spatiotemporal resolution. First, spatiotemporal fusion algorithm process under remote sensing cloud computing was utilized to produce dense Landsat 8 reflectance images based on Landsat 8 and MOD09A1 images. Then, we explored the potential of the Landsat 8 dataset for vegetation type mapping in a subtropical wetland ecosystem using the adaptive stacking algorithm. Subsequently, the vegetation classification map was used to determine the final prior specification of a maximum ε (εmax) of each vegetation pixel. Finally, wetland NPP with CASA was estimated using the normalized difference vegetation index, LSWI and wetland vegetation map.Visually, the SpatioTemporal Fusion Algorithm (STFA) process based on Google Earth Engine (GEE) showed good performance in downscaling MODIS at low to high spatial resolutions, except for some minor flaws that did not affect the overall product. For the fused image, the STFA based on GEE produced an R2 value larger than 0.88, RMSE less than 0.05, and SAM less than 3, which indicated that the fused image was nearly consistent with original Landsat spectrally and spatially. Therefore, STFA based on GEE is suitable for image fusion in areas experiencing rapid change, such as wetlands and city suburbs. The overall accuracy of the wetland map was above 88%, which indicates the potential of the improved stacking algorithm for delineating different land cover types. Additionally, the user and producer accuracies of vegetation types varied within 85%—92% and 83%—91%, respectively. The classification accuracy associated with the proposed method are notably higher than those of the classical methods (e.g., SVM, RF, and kNN), indicating the superiority of the adaptive stacking algorithm for discriminating land cover in a wetland with complex conditions. The measured NPP values derived from field aboveground biomass data were used to validate the accuracy of simulated NPP. The high correlation coefficient (R2=0.85) and low RMSE (20.16 g C/m2) between the estimated and measured NPP demonstrated a significant linear relationship, and thus the estimated NPP based on Landsat data using the CASA model with the input parameters described above is creditable. The average NPP of sedges and reed wetland were 357.50 and 424.26 g C/m2, respectively. The mean NPP values of wetlands (reed and tussock) estimated by the modified CASA model in this study were also closer to those estimated by other models.In this study, time-series Landsat data were obtained on the basis of the STFA based on GEE, and the modified CASA model estimated the NPP of the Dongting Lake wetlands with high spatiotemporal resolution. The NPP estimation method in this study is expected to provide scientific data support for quantitative research on regional wetland carbon reserves and sustainable development.关键词：remote sensing;wetland;net primary productivity;CASA;spatio-temporal fusion;classification;Dongting Lake wetland- 1062
- |
- 408
- |
- 0
发布时间：2023-07-13 -
摘要：Invasion of Spartina alterniflora poses a serious threat to the biodiversity and ecosystem health of coastal wetlands in China. Many coastal provinces in China have initiated projects for clearance and treatment of S. alterniflora in recent years. The timely and accurate understanding of S. alterniflora clearance dynamics is crucial in coastal wetland management and decision making. The objective of this study was to propose a new method for monitoring S. alterniflora clearance dynamics on the basis of dense time series remote sensing images.The Yellow River Estuary wetland was taken as the study area in this work. First, Sentinel-2 MSI, GF-1 PMS, and GF-1 WFV images were fused to construct time-series Normalized Difference Vegetation Index (NDVI) dataset. Second, temporal variations of NDVI were analyzed, and the potential clearance periods were detected. Finally, tidal inundation was examined and S. alterniflora clearance date was identified by eliminating the influence of tidal inundation on NDVI time series. The map of S. alterniflora clearance dates with a spatial resolution of 10 m was obtained for the Yellow River Estuary.The overall accuracy of clearance dates was 88.24%, and the Kappa coefficient was 0.87. Results showed that the fusion of Sentinel-2 and GF-1 data can effectively improve the identification accuracy of clearance dates compared with the single Sentinel-2 date source. The cleared area of S. alterniflora from September to December 2021 was 4816.35 ha, which accounts for 92.81% of the total S. alterniflora region in the study area. Uncleared areas are mainly distributed in the coastal areas of the north shore with complex hydrology and interlaced tidal creeks. The project was completed in two stages because of the early October flood peak in the lower reaches of the Yellow River. The first stage was finished between early September and early October, and the second stage was concluded between mid-October and mid-December, with the majority of S. alterniflora being cleared in early December.The rapid and accurate observation of the dynamics of S. alterniflora clearance through the proposed method is crucial in the monitoring and evaluation of S. alterniflora treatment and wetland restoration projects in coastal wetlands across the country. This method is expected to be applied to the dynamic monitoring and comprehensive evaluation of the effectiveness of large-scale treatment projects.关键词：Sentinel-2;GF-1;Invasive species;Spartina alterniflora management;coastal wetland;time series- 822
- |
- 312
- |
- 1
发布时间：2023-07-13 -
摘要：Land Use/Cover Change (LUCC) is generally defined as the use of land by humans, which is the direct result of the interaction between humans and nature, and reflects the basic process of the interaction between the Earth’s environmental system and human production systems. A bay is an area where land and water interact, with a relatively fragile ecosystem and easily damaged resources and environment. Large-scale long-time series and high-precision LUCC mapping is the basis for territorial spatial planning and environmental protection in bay regions. Random forest algorithms received considerable attention in recent years owing to their high interpretability and reliability in handling complex data. However, room for improvement exists in optimizing the performance of random forest algorithms in processing long-time series datasets. Most existing mapping methods are aimed at original remote sensing images, and fully tapping and jointly utilizing the information potential of the feature space and transformation space are difficult, resulting in the poor application effect of traditional methods in bay areas with high surface heterogeneity. The combined application of the remote sensing spectral index can effectively increase the separability of object categories in bay areas, and principal component transformation can effectively eliminate correlations between features and achieve data compression and image enhancement. Based on Landsat long-term satellite and Google Earth Engine images of the Hangzhou Bay area, this study proposes a random forest remote sensing image classification method that integrates the remote sensing spectral index and principal component transformation and analyzes LUCC mapping spatiotemporal patterns from 1985 to 2020 (at 5-year intervals). Results show that (1) the random forest algorithm, integrating the remote sensing spectral index and principal component transformation, can accurately extract Hangzhou Bay LUCC information, and the average overall accuracy and kappa coefficient of the eight time phases are 92.83% and 0.9108, respectively. (2) During the study period, the construction land area (278.26 km2 to 2984.76 km2, with an average annual increase of 77.33 km2), water area (509.32 km2 to 680.21 km2, with an average annual increase of 4.88 km2), and bare land area (768.99 km2 to 1078.13 km2, with an average annual increase of 8.83 km2) showed an increasing trend, whereas the wood land area (2159.49 km2 to 1881.52 km2, with an annual average decrease of 7.94 km2), cultivated field area (6998.45 km2 to 4800.59 km2, with an annual average decrease of 62.80 km2), and tidal-flat area (181.65 km2 to 161.50 km2, with an annual average decrease of 0.58 km2) showed a decreasing trend. (3) During the study period, the cultivated field area was the main transfer source, whose total area proportion decreased from 64.23% to 41.43%. The transfer out of the cultivated field area was mainly to construction land (2268.05 km2) and bare land (630.20 km2), and the transfer in of the cultivated field area was mainly to water body (376.22 km2) and forest land (352.22 km2). This study provides data support for the scientific management of land resources in the Hangzhou Bay, and the obtained LUCC dataset is of considerable significance to the sustainable development of the region.关键词：Hangzhou Bay;land use/cover change;Google Earth Engine;Landsat;spatio-temporal characteristics- 1119
- |
- 671
- |
- 0
发布时间：2023-07-13 -
摘要：The Guangdong-Hong Kong-Macao Greater Bay Area is a developing competitive international bay area and world-class city cluster. One of the most important aspects of such a comprehensive development goal is efficient and eco-friendly resource usage. Coastal mangrove wetlands are vast in this area, where mangroves play an important ecological role in reducing waves and wind, thereby protecting biodiversity, purifying the sea, and sequestering carbon. The mangrove forests in the bay area were damaged by human activities but have been well-restored under the guidance of the wetland protection policy. However, a consistent and standard dataset for scientifically and objectively clarifying the historical changes in and latest status quo of mangrove wetlands at the regional scale is lacking owing to inconsistent investigation methods and limitations in timely monitoring. By utilizing status quo monitoring, combined with the dynamic updating method, this study constructs a standard and consistent database that is scientifically comparable across different years. Specifically, this study proposes a dynamic updating method based on a high-performance cloud computing platform, namely, Google Earth Engine (GEE), for last-time-period updating, which largely improves the updating efficiency. Using satellite remote sensing images, this study constructs a long-term mangrove distribution series for 1990, 2000, 2010, 2018, and 2020. In addition, this study quantifies the mangrove changes during the four time periods. Results show that (1) an efficient mangrove dynamic updating method can be designed utilizing the GEE platform, making timely and constant mangrove database construction and yearly updating at the regional scale feasible. The timely database can contribute to the efficient management of mangroves by corresponding departments. (2) Over the past three decades, the mangrove forests in the Guangdong-Hong Kong-Macao Greater Bay Area were well-restored and protected, with the total area increasing by 10.21 km2 from 1990 to 2020. However, during the period of 1990—2000, the area decreased by 4.60 km2 owing to the occupation of newly built fish/shrimp ponds and artificial construction. Since 2000, the mangrove area has increased steadily owing to the construction of mangrove parks and nature reserves. (3) Although natural-growing mangrove forests are mostly found in intertidal zones, the newly planted mangroves, restored as mangrove parks, demonstrated a tendency to extend inland slightly after 2010.关键词：mangrove;Change Detection;dynamic updating method;Google Earth Engine (GEE);random forest method;remote sensing- 2945
- |
- 439
- |
- 0
发布时间：2023-07-13
Ecological remote sensing of wetlands
- Postal code：100190
- Tel：010-58887052 Email：nrsb@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备20021838号-9
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0