25 December 2021, Volume 23 Issue 12 Previous Issue   
Theory, Method and Technological Application of Territorial Spatial Development Suitability Evaluation
LIU Xiaobo, WANG Yukuan, LI Ming
2021, 23 (12):  2097-2110.  doi: 10.12082/dqxxkx.2021.210037
Abstract ( )   HTML ( )   PDF (3245KB) ( )  

Suitability evaluation of territorial space is the premise and basis for scientific territory space planning, transition of territory space governance mode and creation of high-quality territory spaces. It is of important significance to optimize the development and protection pattern of territory space and perfect subjective functional orientation of regions. Based on literature review, summary, inductive and comparative analysis, this study reviewed conceptual connotation, development course, evaluation units, evaluation index system, evaluation method, modern technologies and application framework of suitability evaluation of territory space development. It pointed out shortages of existing studies and it suggested to deepening research fields and directions. At present, the theory and practice of suitability evaluation of territorial space have made great breakthrough, and the evaluation index system has formed a more comprehensive system, and the evaluation methods show a trend of diversification. Although many studies have discussed suitability evaluation of territory space development, there are inadequate attentions to multi-objective collaborative evaluation of territory space development. Specifically, it fails to achieve satisfying results in refining of evaluation index system, multi-scale comprehensive study and intelligence level of evaluation methods. The firstly, the construction of the index system is lack of uniformity and standardization. Different studies are based on different goals and perspectives, and the index systems constructed are obviously different. Different scholars have great differences in the evaluation of the same area, which affects the objectivity of the evaluation results. The basic information data can not meet the evaluation requirements of the whole area, the whole element and the whole temporal phase. The ability of Geo-information science and technology to support "smart evaluation" is insufficient. Secondly, in the existing studies, the micro scale evaluation is the majority, the macro scale evaluation is less, and the multi-scale comprehensive evaluation research is very rare. Finally, there is a lack of integration and application of existing evaluation methods for the suitability of territorial space development with intelligent frontier technologies such as spatiotemporal big data, cloud computing, unmanned aerial vehicles, Internet of Things, and 5G network, and there is a huge space for future mining. In future, we need to do well in the following aspects. The firstly, it is suggested to carry out more multi-scale suitability evaluations, promote the integration, transformation and transmission of multi-scale evaluation. Secondly, improve standardization and refining of evaluation index system. Attach importance to the positioning of regional development, pay attention to the special needs and industrial advantages of regional development. In addition, follow the pace of "smart society" construction closely, strengthen the integration and application of earth information technology and Internet, database, cloud computing and other emerging technologies, promote the formulation of land and space multi-source big data and the effective integration of suitability evaluation system; couple GIS, remote sensing and big data to carry out information mining, and provide strong technical support for "smart evaluation".

Figures and Tables | References | Related Articles | Metrics
Advance in Character Mining and Prediction of Ship Behavior based on AIS Data
ZHEN Rong, SHAO Zheping, PAN Jiacai
2021, 23 (12):  2111-2127.  doi: 10.12082/dqxxkx.2021.210495
Abstract ( )   HTML ( )   PDF (5604KB) ( )  

The mining and prediction of ship behavior characteristics is an important research content of maritime intelligent transportation system and a key scientific problem in the field of transportation engineering. In order to systematically study the research status and development trend of ship behavior characteristic mining and prediction, the Vosviewer is used to generate the clustering map and trend evolution map of high-frequency keywords of research content from the perspective of bibliometrics, based on literatures collected from WOS database and CNKI database. After comprehensive analysis, three topics of data mining of maritime traffic elements based on Automatic Identification System (AIS), ship behavior clustering research and ship behavior prediction research are summarized. The research contents, methods and existing problems of each topic are systematically analyzed. The research results show that: ① In the aspect of data mining of maritime traffic elements based on AIS, the research mainly focuses on the mining of spatial features of maritime traffic and temporal features of traffic flow,and the results are lack of sufficient association mining of time features AIS data and background environment features. Further exploration needs to be made on the mining of space-time characteristics and data fusion. ②In the aspect of ship behavior clustering, the research mainly uses the unsupervised clustering method to study the clustering of ship track points and ship track segments to obtain the spatial-temporal distribution of ship navigation behavior patterns and the maneuvering intention. The similarity calculation method of ship trajectory integrating multidimensional features, identification of ship the adaptive selection of clustering parameters and the semantic modeling of ship behavior need to be further studied. ③ In the aspect of ship behavior prediction, it mainly focuses on the prediction of ship behavior based on dynamic equation, traditional intelligent algorithm and deep neural network. Considering the characteristics of randomness, diversity and coupling of ship behavior, the use of hybrid neural network model and combining neural network with vector machine. In the end, the paper proposes the promising research area which include mining of ship behavior feature based on semantic model, the prediction of ship behavior based on deep convolutional neural network, the mining of ship behavior feature based on knowledge graph and the visualization of prediction results.

Figures and Tables | References | Related Articles | Metrics
Multi-scale Model of Point Group Elements in Network Space Combined with Blondel and k-shell Decomposition Hybrid Algorithm
WANG Xupan, ZHANG Heng, ZHOU Yang, HU Xiaofei, PENG Yangzhao, QI kai
2021, 23 (12):  2128-2138.  doi: 10.12082/dqxxkx.2021.210397
Abstract ( )   HTML ( )   PDF (9748KB) ( )  

In the 21st century, with the increasing dependence of people on cyberspace, Internet technology and network infrastructure develop rapidly. The elements that make up the cyberspace are complex and the data of nodes are large. It is difficult to directly use the form of numbers or tables for the overall planning and management of cyberspace, and it is not easy to find some key information hidden in cyberspace. So it is very important to construct a multi-scale model of point group elements in cyberspace for multi-scale analysis and visualization of data in cyberspace. If the nodes and topological relations in the cyberspace are directly visualized, a large number of points overlap and lines cross, resulting in the confusion of the information in the cyberspace. In this paper, based on the community and hierarchical characteristics of cyberspace, and referring to the characteristics of cyberspace stratification algorithm based on community division and cyberspace stratification algorithm based on node importance, a multi-scale model building algorithm of point group elements in cyberspace is proposed, which combines Blondel algorithm and K-shell decomposition hybrid algorithm. By automatic community division and combining nodes in the same community to build a new network, this algorithm effectively solves the problem of low automation degree of cyberspace stratification algorithm based on node importance. Core nodes in the cyberspace are extracted by k-shell decomposition hybrid algorithm. Core nodes are used to replace the whole community structure, which significantly retains the attributes of nodes in the cyberspace. For example, basic attributes such as the number, importance, and geographical location of nodes in the cyberspace. Experiments show that this algorithm can make the comprehensive proportion of each level of network space point group elements less than 30%, and achieve the clustering and stratification of network space point group elements. Compared with Blondel's algorithm, it is found that the proposed algorithm can preserve the hierarchy of network space with a high degree of importance. The multi-scale model of cyberspace point group elements is applied to geographic space which reduces the consumption of system resource, and the map of network space is realized. The multi-scale model of cyberspace can provide a data synthesis method for generating multi-scale cyberspace map. The different levels divided by this algorithm have different node numbers and edge numbers. The multi-scale model of point group elements can also provide node data with different levels of detail, which provides the basis for multi-scale analysis of network space.

Figures and Tables | References | Related Articles | Metrics
River Systems Geometric Similarity Measurement Method under Hierarchical Analysis
YANG Fei, WANG Zhonghui
2021, 23 (12):  2139-2150.  doi: 10.12082/dqxxkx.2021.210279
Abstract ( )   HTML ( )   PDF (11820KB) ( )  

The existing line groups geometric similarity measurement methods are mainly based on mathematical statistics. Geometric similarity is calculated through the statistics of overall change information. These methods lack the expression of local features and they are not suitable for river systems with highly fractal features. River systems are the basic elements of maps, and they have obvious fractal features and complex structural features. When we calculate the geometric similarity of river systems, the difference of the overall characteristics before and after cartographic generalization should be included, and the change of the local characteristics should also be considered. To this end, this paper divides geometric characteristics of river system into three levels of information characteristics: The shape characteristics of a single river, the structural characteristics of local area, and the distribution characteristics of global scope. Firstly, Included Angle Chain and Hausdorff distance are combined to calculate the shape similarity of a single river. Then, the local characteristic regions are determined according to the "80/20 Principle". This method is extended to calculate the local structure similarity of M: N river system through coordinate conversion. Finally, the overall feature descriptor is integrated to obtain the global distribution similarity. On this basis, a difference index is constructed to perform similarity calculation and quality evaluation of generalization. The results show that the calculation result of this method is better than the mean index method, and this method can be effectively applied to quality evaluation of generalization. The details are as follows. Firstly, the proposed method makes up for the shortcoming that the mean index method could only measure the overall change information. The changes of global distribution characteristics and local structure characteristics before and after cartographic generalization are expressed in the proposed method. Secondly, the method in this paper is more sensitive than the mean index method, and the calculation results of geometric similarity are more in line with human psychological cognition. Thirdly, the difference index P reflects the degree of generalization, the irrationality of deletion and simplification before and after generalization are detected.

Figures and Tables | References | Related Articles | Metrics
Green View Index Estimation Method based on Three-dimensional Simulation of Urban Tree Landscape
JIANG Feng, TANG Liyu, LIN Ding, CHEN Xiaoling, FENG Xianchao, CHEN Chongcheng
2021, 23 (12):  2151-2162.  doi: 10.12082/dqxxkx.2021.210063
Abstract ( )   HTML ( )   PDF (20096KB) ( )  

Urban ecological landscape has sensory functions, which describe the visual effect of green plants for human beings. Green view index is considered as a relatively good indicator for measuring the visibility of urban green space, which can reflect different levels of urban vegetation space directly. Green view index is usually calculated using static images or street view data. However, green view index is a variable quantity since the variant of view point location results with very different visual effects and the phenological change of plants. What is more, plants, as organism with life characteristics, are the most important elements in the urban green space landscape, which can change their morphology with time factors constantly and affect the amount of visible green space. In the paper, a green view index calculation method was proposed based on the three-dimensional simulation landscape of garden trees driven by spatial information data of geographic entities and tree architecture and growth model. This method comprises three steps. Firstly, using virtual geographical environment, virtual plants, and other technologies, a three-dimensional urban vegetation landscape was generated according to hard landscape data (e.g. roads, building) and tree models. Secondly, based on visual mechanisms of seeing, virtual cameras were constructed to set observation points and generate the landscape visual images. Thirdly, the visibility analysis was conducted to identify vegetation information visible from each observation point at the pixel level, which can compute the value of green view index. A three-dimensional tree landscape simulation and green view index estimation prototype was developed. Taking urban road greenery scenes (e.g. Jinshan avenue in Fuzhou) as an example, the green view index was estimated and analyzed. The results are closed to those derived from street view images. Therefore, it can effectively reflect the visual feeling of vehicle passengers. It is useful for quantitatively evaluating the visual effects of urban forest states of past, present, or future at different growth stages. It is also suitable to simulate and calculate green view index dynamically by setting view points everywhere and in arbitrary directions. The method can be used as a potential tool to assess the simulation results of urban green space design schemes before they were carried out. It is also helpful for the rational planning of urban green space. It can provide references for the science and rationality of the future landscape of different engineering design schemes, thereby promoting the sustainable development of the city.

Figures and Tables | References | Related Articles | Metrics
Mapping Method of Fishing Grounds based on Marine AIS Data
CHEN Renli, WU Xiaoqing, LIU Baijing, WANG Yueqi, GAO Meng
2021, 23 (12):  2163-2173.  doi: 10.12082/dqxxkx.2021.210106
Abstract ( )   HTML ( )   PDF (10335KB) ( )  

Automatic Identification System (AIS) is a ship-to-ship and ship-to-shore system used for ship monitoring. It can automatically and continuously broadcast static and dynamic information about the vessel's position and movement, as well as voyage-related information. More and more research suggests that AIS is not only an effective tool for maritime traffic supervision, but also a good data source for the study of maritime traffic and its related industrial activities. In China, the application of AIS equipment is relatively late. The research based on AIS data mainly focuses on the maritime traffic safety management, such as vessel collision avoidance and traffic flow statistics, while there are few reports in the fishery field. In order to provide technical support for the protection and recovery of inshore fishery resources in China, it is urgent to excavate the information value of AIS data and carry out the research on the spatiotemporal pattern of inshore fishing activities based on AIS data. Therefore, based on AIS data of fishing vessels at sea, this study used the Gaussian Mixed Model (GMM) to identify fishing behavior of fishing vessels and determine the speed threshold of fishing vessels in fishing activities. This study proposed a method combining Kernel Density Estimation (KDE) and Hot Spot Analysis (HSA) to map fishing grounds. The results show that, firstly, compared with single KDE or HSA method, the combination method combined the distance attenuation effect of KDE with quantitative statistical index of HSA, which can not only avoid the problem of range definition of fishing grounds in the KDE process, but also improve the efficiency of data analysis in the HSA process and the scattered distribution effect of fishing grounds in the extraction results of the HSA. The combination method had better application effects and higher efficiency in the mapping of fishing grounds. It provided a quick and simple evaluation method for the rapid acquisition of marine fishing information and the effectiveness evaluation of fishery resources protection and management measures. Secondly, based on the AIS data from September to December 2018, the combination method was used to map the spatiotemporal distribution of fishing grounds around the Bohai Strait. The study found that the fishing grounds around the Bohai Strait were mainly distributed in the inshore areas of Yanwei and the Bohai Strait. The distribution range of concentrated fishing activities in traditional fishing grounds of the Bohai Sea is relatively small, and the distribution range and spatial morphological characteristics of fishing grounds had some variabilities in different months. The results can provide technical methods and decision support for fishing management and marine ecological protection around the Bohai Strait.

Figures and Tables | References | Related Articles | Metrics
Analysis and Comparative Study of the Evolution of Public Opinion on Social Media during Typhoon for Different User Groups
JIN Cheng, WU Wenyuan, CHEN Bairu, YANG Xuchao
2021, 23 (12):  2174-2186.  doi: 10.12082/dqxxkx.2021.210065
Abstract ( )   HTML ( )   PDF (14934KB) ( )  

Social media has been successfully applied to typhoon monitoring, on-site rescue, and disaster loss assessment. Preview studies mostly utilized topic modeling and sentiment analysis technique to analyze the focus of public opinion and sentiment evolution in the social media platform during the typhoon period. However, the existing studies were usually conducted at large spatial scales and long time spans. Moreover, the difference in behavior pattern among user groups was ignored. Firstly, a case study of Typhoon Lekima was implemented for verifying the effectiveness of microblog's response to typhoon disaster in Zhejiang province from three perspectives: word frequency, spatiotemporal change of public attention to typhoon, and public response to specific events. Secondly, the Latent Dirichlet Allocation (LDA) topic model was adopted to mine the text topics, whose community structure were divided by Louvain algorithm. Thirdly, a custom emotion dictionary was developed to calculate the sentiment index, and subsequently compared with SnowNLP in sentiment polarity prediction. Finally, we investigated the difference between official microblogs and public microblogs in topic concern and sentiment evolution. The results indicated that microblogs were capable of tracking typhoon dynamics and reflecting the spatiotemporal distribution of hazards within the provincial region. The LDA model result showed that the percentage of microblogs on public dynamics topic was large in days and small in nights; the percentage of microblogs on warning topic was on a downward trend; the disaster event rose significantly after typhoon landed; and the peak of that on rescue activities appeared in the late period of typhoon. The topic of official microblog had a clearer community structure than the public microblog, but this characteristic may be blurry when mixing the microblogs from two groups. The negative emotion on Sina Weibo significantly deepened in the typhoon landing period, and the public had a more timely emotional response to typhoon disasters, while the sentiment index of official microblog was always higher.

Figures and Tables | References | Related Articles | Metrics
The Temporal Influence Difference of Drug-related Personnels' Routine Activity on the Spatial Pattern of Theft
LIU Lin, SUN Qiuyuan, XIAO Luzi, SONG Guangwen, CHEN Jianguo
2021, 23 (12):  2187-2200.  doi: 10.12082/dqxxkx.2021.210069
Abstract ( )   HTML ( )   PDF (7779KB) ( )  

According to the routine activity theory, the spatiotemporal pattern of crime is strongly related to routine activity of victims and offenders. However, due to the difficulty of data acquisition, there is a lack of research on offenders' routine activity and the spatiotemporal pattern of crime events. The existing literature shows that there is a great correlation between drug-related persons and property crimes such as theft. Based on this, this study verifies the role of the routine activity of offenders in shaping the spatial-temporal pattern of theft through analyzing the impact of the routine activity of drug-related persons on theft. In this paper, taking XT police district with 150 m×150 m grids in ZG city in southern China as an example, the theft data, routine activity data of drug-related persons, POI data, and patrol and interrogation data were used. Poisson regression models were established respectively in different periods. The results show that, firstly, compared with traditional static arrest or policing events data, active routine activity data of potential offenders and victims could promote goodness of fit in models effectively. Secondly, compared with total amount of people in whole day, active real-time activity data of drug-related personnel and residents could explain the spatial pattern of theft better. Thirdly, static land use density has a different influence on theft events in different periods. The above results verify the relationship between the routine activity of drug-related persons and the spatiotemporal pattern of theft. The research conclusions verify and enrich the routine activity theory, which can provide a certain reference for the actual crime prediction and police deployment.

Figures and Tables | References | Related Articles | Metrics
Spatiotemporal Pattern Analysis of Fire Service Coverage Rate in Nanjing from Real-time Road Conditions
ZHU Kaixin, ZHANG Fengyan, LI Yuyu, CHEN Yuehong
2021, 23 (12):  2201-2214.  doi: 10.12082/dqxxkx.2021.210129
Abstract ( )   HTML ( )   PDF (16631KB) ( )  

The rapid urbanization in China results in the problem that urban fire facilities fail to keep pace with urban development. Assessing urban fire service coverage rate plays an important role in optimizing urban fire resource allocation. This paper aimed to propose an assessment for the fire service coverage rate using real-time road conditions to explore the spatiotemporal pattern of urban fire service coverage rate. By consideration of the coverage area of fire stations, the real-time rescue time of fire stations arriving at historical fire incidents was obtained by using the AutoNavi Maps API for three consecutive weeks in September 2020. The real-time travel time was then used to calculate the fire service coverage rate for investigating the spatiotemporal pattern of fire service coverage rate in Nanjing, China. Results show that: (1) The average travel time of fire stations was 10 minutes in urban fire-intensive area and 16 minutes in non-fire-intensive area. The average travel time for both areas was significantly longer than the national standard arrival time of five minutes. Thus, the fire service coverage rate of fire stations that met the national standard in Nanjing was only 8.2%; (2) As the average travel distance for fire stations in fire-intensive area of Nanjing was 37% of that in non-fire-intensive area, the waiting time for fire incidents rescue in fire-intensive area was significantly shorter than that in non-fire-intensive area, especially in the southwest and northeast of fire-intensive area and around some fire stations. The proportion of fire incidents with waiting time for rescue within five minutes in Nanjing was less than 7%, and fire incidents with waiting time for rescue between five to ten minutes were mostly affected by traffic congestion during morning and evening rush hours; (3) The fire service coverage rate in Nanjing was affected by the morning and evening rush hours which presented a "W" shaped pattern, resulting in lower fire service coverage rate in morning and evening rush hours than that in other time. The fire service coverage rate in fire-intensive area complying with the five minute standard arrival time decreased from 11.5% during the non-rush hours to 8.4% during the rush hours, and decreased from 6.1% to 5% in non-fire-intensive area. In the intersection area of Shimenkan and Dongshan fire stations and the surrounding area of Hanzhongmen and Maigaoqiao fire stations, the number of fire incidents with waiting time for rescue over 15 minutes in morning and evening rush hours was larger than that in other time; (4) The fire service coverage rate with the "W" shaped pattern had the smallest fluctuation with the use of 5 minutes as the standard arrival time, and the largest fluctuation occurred with the use of 10 minutes as the standard arrival time. The fire service coverage rate reached 43.5% using 10 minutes as the standard and 75% using 15 minutes as the standard arrival time. Finally, based on the analysis, suggestions were made for the future construction and planning of Nanjing fire facilities.

Figures and Tables | References | Related Articles | Metrics
Classification of County Leading Function Types and Pattern Recognition of Its Spatial Structure based on Multi-source Data
CHEN Zilong, WANG Fang, LI Shaoying, FENG Yanfen, Chen Jianguo
2021, 23 (12):  2215-2231.  doi: 10.12082/dqxxkx.2021.210050
Abstract ( )   HTML ( )   PDF (23369KB) ( )  

It is of great significance to study the types of dominant function at county level and analyze the spatial structure pattern for the guidance and planning regulation of regional comprehensive, coordinated, and sustainable development. Taking Guangdong Province as an example, this paper uses K-means clustering analysis to classify 124 dominant functional areas of county units. Based on the Tencent location big data, the spatial structure index was constructed using the rank-size rule. The spatial structure patterns of the counties with different dominant functions were identified and analyzed by combining remote sensing data. Research results show that, firstly, the use of statistics, remote sensing, social media, night lighting multi-source data, and K means clustering method in county dominant function classification can get nice results, which can objectively reflect the features of area. The dominant function of counties in Guangdong can be divided into five classes, including ecological leading, agricultural leading, industry leadership, center service, and balanced developmental. Secondly, the rank-size law method based on Tencent location big data can break through the city scale. It can be used together with the quantitative analysis of remote sensing data in the study of spatial structure pattern recognition at the meso-county scale. Thirdly, the average value of spatial structure index of agriculture-oriented and ecology-oriented counties is greater than 1, showing a trend of single center spatial structure. The mean value of spatial structure index of industrial dominated county is less than 1, showing a polycentric spatial structure trend. The mean value of spatial structure index of balanced development county is close to 1. There are differences in spatial structure characteristics of balanced development county in different regions. Fourthly, the spatial structure patterns of the counties with different dominant functions were significantly different. The ecology-dominated counties showed a single central pole core pattern, while the agriculture-dominated counties showed a single central pole core and scattered point pattern. The balanced development counties had obvious coexistence characteristics of multiple patterns, while the industrial-dominated counties showed a multi-center network distribution pattern.

Figures and Tables | References | Related Articles | Metrics
Building Extraction from High-Resolution Remote Sensing Image based on Res_AttentionUnet
LI Chuanlin, HUANG Fenghua, HU Wei, ZENG Jiangchao
2021, 23 (12):  2232-2243.  doi: 10.12082/dqxxkx.2021.210008
Abstract ( )   HTML ( )   PDF (25582KB) ( )  

To contribute to the current research of building extraction based on deep learning and high-resolution remote sensing images, we propose an improved Unet network (Res_AttentionUnet), which combines the Residual module of ResNet and Attention mechanism. We apply the Unet network to the extraction of buildings from high-resolution remote sensing images, which effectively improves the extraction accuracy of buildings. The specific optimization method can be divided into three parts. Firstly, in the traditional Unet semantic segmentation network convolution layer, the ResBlock module is added to enhance the extraction of low-level and high-level features. Meanwhile, the Attention mechanism module is added to the network step connection part. Secondly, in the whole net, the ResBlock module enables the convoluted feature map to obtain more bottom information and enhance the robustness of the convolution structure, so as to prevent underfitting. Thirdly, the Attention mechanism can enhance the feature learning of building area pixels, making feature extraction more complete, so as to improve the accuracy of building extraction. In this study, we use the open data set (WHU Building Dataset), provided by Ji Shunping team of Wuhan University, as the experimental data and select three experimental areas with different building characteristics and representativeness. Then, we preprocess the different experimental areas (including sliding, cropping, and image enhancement, etc.). Finally, we use four different network models of Unet, ResUnet, AttentionUnet, and Res_AttentionUnet to extract buildings from three different experimental areas. The experimental results are cross-compared and analyzed. The experimental results show that, compared with the other three networks, the Res_AttentionUnet proposed in this paper has higher accuracy in the building extraction from high-resolution remote sensing images. The average extraction accuracy of Res_AttentionUnet is 95.81%, which is 17.94% higher than the original Unet network, and 2.19% higher than ResUnet (the Unet with only residual module). The results demonstrate that Res_AttentionUnet can significantly improve the effectiveness of building extraction in high-resolution remote sensing images.

Figures and Tables | References | Related Articles | Metrics
Landslide Susceptibility Analysis based on Deep Learning
WANG Yi, FANG Zhice, NIU Ruiqing, PENG Ling
2021, 23 (12):  2244-2260.  doi: 10.12082/dqxxkx.2021.210057
Abstract ( )   HTML ( )   PDF (43266KB) ( )  

The formation mechanism of landslide disasters is complicated and there are many influencing factors. It is imperative to explore a low-cost and highly applicable method to manage and prevent landslide disasters. As a hot spot in the current artificial intelligence field, deep learning can better simulate the formation of landslide disasters and accurately predict potential slopes. Thus, to explore the application potential of deep learning, this paper constructs one-dimensional, two-dimensional, and three-dimensional forms of landslide data, and then introduces three Convolutional Neural Networks (CNN)-based landslide susceptibility analysis frameworks, including CNN-based classifiers, integrated models, and ensemble models. The proposed deep learning methods were applied to Yanshan County, Jiangxi Province for experiments. 16 landslide influencing factors were first selected for modelling based on the geomorphological, hydrological, and geological environment conditions of the study area. These factors include altitude, aspect, distance to faults, land use, lithology, normalized difference vegetation index, plan curvature, profile curvature, rainfall, distance to rivers, distance to roads, slope, soil, stream power index, sediment transport index, and topographic wetness index. Then, the multi-collinearity analysis and relief-F algorithm were used to analyze and screen the influencing factors. All CNN-based methods were constructed and validated based on several statistical measures of accuracy, root mean square error, mean absolute error, sensitivity, specificity, and the receiver operation characteristic curve. Finally, the susceptibility value of each pixel in the study area was predicted based on the CNN-based methods, and the entire study areas were reclassified into five susceptibility categories: very low, low, moderate, high, and very high. The factor analysis results show that the plan curvature, profile curvature, stream power index, and sediment transport index are redundant factors and should be removed from further modelling process. The model evaluation results demonstrate that all CNN-based models can obtain accurate and reliable landslide susceptibility mapping results. The two-dimensional CNN model achieved the highest prediction accuracy of 78.95% among single CNN models. Moreover, the performance of logistic regression was effectively improved by combining the two-dimensional CNN for feature extraction, with an accuracy improvement of 7.9%. Besides, the heterogeneous ensemble strategy can greatly improve landslide prediction accuracy when using CNN classifiers, with an accuracy improvement between 4.35% and 8.78%. Generally, the CNN has been proven to have huge application potential in landslide susceptibility analysis and can be implemented in other landslide-prone areas with similar geo-environmental conditions.

Figures and Tables | References | Related Articles | Metrics
Evaluation and Comparison of Downward Solar Radiation from New Generation Atmospheric Reanalysis ERA5 across Mainland China
ZHANG Junbing, SHEN Runping, SHI Chunxiang, BAI Lei, LIU Junjian, SUN Shuai
2021, 23 (12):  2261-2274.  doi: 10.12082/dqxxkx.2021.180357
Abstract ( )   HTML ( )   PDF (20821KB) ( )  

The European Centre for Medium-Range Weather Forecasts (ECMWF) has developed ERA5, a global atmospheric reanalysis product with high spatiotemporal resolution. The Shortwave Downward Radiation (SWDN) of ERA5 is an important atmospheric forcing dataset which has important applications in regional climate assessment, agriculture, and solar energy resource utilization. In this study, the observed SWDN dataset after quality control was collected from 91 official radiation monitoring stations across mainland China in 2011-2018 and was applied to evaluate the SWDN in ERA5 on different spatial and temporal scales, together with other three reference SWDN datasets from global atmospheric reanalysis products (i.e., ERA-Interim, CFSR, and MERRA2) and the CERES satellite inversion product (SYN1deg). Results show that: ① On the monthly mean scale, the ERA5 product had the highest correlation coefficient (Corr) with the station observation data (0.939) and the lowest Root Mean Square Error (RMSE) (28.309 W/m2), compared with other reanalysis products. The average bias of ERA5 (15.4 W/m2) was slightly higher than that of the ERA-Interim product (13.2W/m2). The Corr between CERES satellite inversion product and observation data was 0.955, the RMSE was 20.042W/m2, and the Bias was 5.3W/m2; ② The radiation values of all these five SWDN products were overestimated against the observation data. In general, the overall accuracy of the ERA5 product in mainland China was higher than the other reanalysis products, but was lower than the CERES satellite inversion product. The comparison of daily mean values between products also showed similar results; ③ Regional evaluation results show that the SWDN in ERA5 had a good consistency with observation data in four regions across mainland China. All five SWDN products performed poorly in the southern region. Compared to the northeastern and northern regions, the RSME and the bias of the ERA5 product and the CERES satellite inversion product relative to observations were larger in the western region.

Figures and Tables | References | Related Articles | Metrics
Extraction Method of Rapeseed Planting Distribution Area based on GF-6 WFV Image
JIANG Nan, ZHANG Xuehong, WEN Jianlong, GE Zhouhui
2021, 23 (12):  2275-2291.  doi: 10.12082/dqxxkx.2021.210233
Abstract ( )   HTML ( )   PDF (17096KB) ( )  

Rapeseed is the main agricultural cash crop and the main source of cooking oil in China. Timely and accurate acquisition of the spatial distribution of rape plants is important for understanding the status of rape planting, strengthening production management, and optimizing the spatial pattern of crop planting. The Wide Field View (WFV) sensor of Gaofen-6 (GF-6) adds a purple band, a yellow band, and two red edge bands to the visible-near-infrared bands, which provides more spectral information for rape identification from remote sensing, thus improving the identification accuracy compared with the "traditional bands" of blue, green, red and near-infrared. In this paper, the Gushi county, Henan province, a dominant area of rape, was selected as the research area. Two GF-6 WFV images within the flowering period of rape were mosaiced as the data source. Due to the phenomenon of "same spectrum with different species" between rape and other land objects and the phenomenon of "same species with different spectrum" for rape at different growth phases, we put forward four spectral indices including NDSI28、S34、NDSI23, and NDSI46, based on the unique spectral reflectance characteristics of rape at flowering phase and the algorithm of standardized close range between means. A decision tree model for rape identification was then constructed based on these indices. The results show that the decision tree model based on the combination of four indices achieved a high accuracy in extracting rape, with an overall accuracy of 96.17%, which was 0.31%, 0.88%, and 1.24% higher than that of random forest, Support Vector Machine(SVM), and maximum likelihood method, respectively. The cartography accuracy of decision tree model was 98.15%, which was 4.72%, 4.21%, and 5.59% higher than that of random forest, SVM, and maximum likelihood method, respectively. The user accuracy of the decision tree model was 86.89%, which was 2.2% lower than that of random forest, 1.63% lower than that of the maximum likelihood method, and 0.11% higher than that of the SVM. The cartographic accuracy of different classification methods was greater than 90%, and in particular, the decision tree model showed the highest cartographic accuracy. In terms of the user accuracy, the random forest showed the highest value (89.09%), the Support Vector Machine (SVM) showed the lowest value (86.78%), and the decision tree method showed an user accuracy of 86.89%. As a result, the new bands in the GF-6 WFV data can greatly enrich the spectral information of rape. Our results demonstrated the unique advantages and great potential of the new bands in the GF-6 WFV data in the extraction of crop planting region and distribution information, including rape.

Figures and Tables | References | Related Articles | Metrics
Fine Classification and Mapping of Mangroves in Guangxi Coastal Zone based on Spectral Characteristics of GF Images
MA Yunmei, WU Peiqiang, REN Guangbo
2021, 23 (12):  2292-2304.  doi: 10.12082/dqxxkx.2021.210494
Abstract ( )   HTML ( )   PDF (21760KB) ( )  

Accurate understanding of mangrove species composition in coastal zone of China is helpful for mangrove resource investigation, protection, and utilization. In this paper, based on GF-2 multi-spectral images of Guangxi Coastal zone from 2018 to 2020, the vegetation index method and first-order differential method were used to reconstruct spectral characteristic data. Based on the reconstructed data, the Support Vector Machine (SVM) classification method was used to study the interspecific classification of mangroves in Guangxi coastal zone. Taking Maoweihai as an example, the validity of the reconstructed data for the identification of mangrove species was verified by comparing with the classification results using original data and the first-order differential method. The results show that the classification accuracy of the reconstructed data based on spectral features was the highest (91.55%) and the Kappa coefficient was 0.8695, which was 6.92% higher than the classification accuracy using original data and 11.17% higher than the classification accuracy using first-order differential method. Based on this, mangrove species identification in Guangxi coastal zone was further carried out using the spectral feature reconstruction data. Mangroves in Guangxi can be divided into eight types, namely, Aegiceras corniculatum, Avicennia marina, Rhizophora stylosa, Sonneratia apetala, Kandelia candel, Bruguiera gymnorrhiza, Acanthus ilicifolius, and a salt marsh herbaceous plant Cyperus malaccensis. The total area of typical vegetation for all types of wetlands was 7402.98 hm2. The area of mangrove in Fangchenggang city, Qinzhou City, and Beihai City was 1826.16 hm2, 2496.18 hm2, and 3080.47 hm2, respectively. The dominant species of mangrove in Guangxi were Aegiceras corniculatum and Avicennia marina, with the largest distribution area of 3372.09 hm2 and 3445.17 hm2, respectively, accounting for 92.09% of the total area. Next came the Cyperus malaccensis with an area of 287.50 hm2, accounting for 3.88% of the total area of the mangroves, followed by Rhizophora stylosa and Sonneratia apetala, with an area of 135.97 hm2 and 126.52 hm2, respectively, accounting for 3.55% of the total area of mangroves. The area of Kandelia candel, Bruguiera gymnorrhiza, and Acanthus ilicifolius were all less than 20 hm2, which accounted for less than 1% of the total mangrove area. The total area of mangrove in Beilun Estuary, Shankou, and Maweihai sea mangrove nature reserves was 1009.21 hm2, 715.56 hm2 and 1546.62 hm2, respectively. In this paper, based on the spectral characteristic data reconstruction method using GF images, the fine classification of mangroves was investigated, providing technical and data support for the management, protection, and reconstruction of mangroves in Guangxi.

Figures and Tables | References | Related Articles | Metrics
Share: