Table of Content

    15 June 2019, Volume 21 Issue 6 Previous Issue    Next Issue
    Spatiotemporal Estimation of High-Accuracy and High-Resolution Meteorological Parameters based on Machine Learning
    Ying FANG, Lianfa LI
    2019, 21 (6):  799-813.  doi: 10.12082/dqxxkx.2019.190014
    Abstract ( )   HTML ( )   PDF (19422KB) ( )   Save

    The meteorological stations are sparsely distributed across Mainland China. In terms of generating high-resolution surfaces of meteorological parameters, the estimation accuracy of existing models is limited for air temperature, and is poor for relative humidity and wind speed (few studies reported). With the measurement data of 824 monitoring stations covering the mainland of China in 2015, this study compared the typical Generalized Additive Model (GAM) and autoencoder-based residual neural network (here after, residual network for short) in terms of predicting three meteorological parameters, i.e. air temperature, relative humidity, and wind speed. The performances of the two models were evaluated through 10-fold cross-validation. Basic variables including latitude, longitude, elevation, and the day of the year are used in the air temperature models. In addition to the basic variables, the relative humidity models use air temperature and ozone concentration as covariates, and the wind speed models use wind speed coarse-resolution reanalysis data as covariates. In our spatiotemporal models, spatial coordinates capture the spatial variation and time index of the day captures the time variation. Compared with GAM, residual network significantly improved the prediction accuracy: on average, CV (Cross Validation) R2of the three meteorological factors increased by 0.21, CV RMSE decreased by 37%, and the relative humidity model improved the most (CV R2: 0.85 vs. 0.52, CV RMSE: 7.53% vs. 13.59%). With incorporation of the monthly index in the relative humidity models, the accuracy was greatly improved, indicating that the different levels of time factors are important for the relative humidity models. Furthermore, we also discussed the effectiveness and limitations of coarse resolution reanalysis data and nearest neighbor values as covariates. This study shows that the residual network model can greatly improve the accuracy of national high spatial (1 km) and temporal (daily) resolution meteorological data as opposed to traditional GAMs. Our findings provide implications for high-accuracy and high-resolution mapping of meteorological parameters in China.

    Figures and Tables | References | Related Articles | Metrics
    Urban Tourism Itinerary Planning from a Spatiotemporal Perspective
    Yang CAO, Junlian GE, Yi LONG, Ling ZHANG
    2019, 21 (6):  814-825.  doi: 10.12082/dqxxkx.2019.190062
    Abstract ( )   HTML ( )   PDF (11018KB) ( )   Save

    Traditionally, tourism itinerary planning is implemented as a spatial arrangement issue, which lacks the consideration of the spatiotemporal coupling and the flexibility for tourists to make choices. In this paper, the understanding of the itinerary planning problem was extended from the perspective of space to the perspective of tourist activities. From the time-space coupling relationship and reconstruction mode of tourism nodes, the multi-dimensional attributes such as time, space, and topic involved in the travel were organically organized, and then the travel's spatiotemporal chain was proposed. The conceptual model and the method of space-time convergence of the stroke elements. The proposed method was applied to the case study of Nanjing, Jiangsu Province of China. Results show that the match between the model and the traditional itinerary design method in terms of node name, number of nodes, and node order exceeds 80%, indicating good methodological reliability. Compared with existing itinerary planning studies, this research took the basic information of the itinerary (such as travel time, cost budget, departure place, destination, etc.) as the precondition, and considered the travel itinerary from the perspective of tourists. Specifically, the spatiotemporal characteristics of the tourist nodes were organically integrated. While satisfying the spatial order of tourist routes, the rationality of the time arrangement of each tourism element was also considered. The proposed algorithm is mainly used to serve independent tourists. Meanwhile, this algorithm has the advantage of arranging route and schedule flexibly. It should be noted that the specific application of the model is still constrained by the basic travel itinerary conditions. The flexibility of the proposed model for meeting tourists' individualized needs is currently not strong. Considering that tourists' demands are in reality often changing during traveling, it is necessary to further optimize the adaptability, flexibility, and stability of the proposed model.

    Figures and Tables | References | Related Articles | Metrics
    Detecting Urban Hotspot Region Association by Navigation Big Data Mining
    Rui CHEN, Mingjian CHEN, Xiang YAO, Jianguang WANG
    2019, 21 (6):  826-835.  doi: 10.12082/dqxxkx.2019.180406
    Abstract ( )   HTML ( )   PDF (26320KB) ( )   Save

    Navigation big data is a set of massive data produced by navigation and positioning technology with both the characteristics of PNT (position, navigation, and time) and 5V (volume, velocity, variety, value, and veracity). Urban environment affects the travel behavior of residents. Consequently, navigation big data generated in daily travel activities contain abundant spatiotemporal information about both residents’ behavior and environment. The spatial distribution of hotspots and their latent association are key aspects of urban dynamics. This kind of characteristics is influenced by objective urban environment condition and subjective residents' activities. The knowledge about urban hotspot region association, which can be detected via mining navigation big data, is useful for urban environment management and intelligent transportation, etc. However, most studies focus on only determining hotspots. To better understand urban dynamics, a method based on spectral clustering and ant colony algorithm was proposed in the present study to mine the knowledge of association among urban hotspots from navigation big data. Implementation of this method included three main steps. Firstly, data preprocessing including data cleaning and OD (Origin and Destination) points extraction was performed on raw data. Then, the hotspot regions were extracted by two-step density-based clustering of the OD points. These hotspot regions were discretized by k-means clustering and Voronoi polygon. The proposed discretization strategy efficiently retained the intrinsic spatial distribution characteristics of the OD points. Lastly, we defined the degree of association among hotspot regions based on travel frequency. This measurement intuitively described the relationship among different regions embedded in travel activities. The association among hotspot regions was investigated using spectral clustering and ant colony algorithm. The two algorithms converted the association mining into the issues of graph partition and optimal path solution. The proposed method takes advantage of using navigation big data as a proxy of urban dynamics. We applied it to real-world taxi trajectory data in Shanghai. Results show the spatial distribution characteristics of hotspot regions with tight association and the pattern of residents' travel behavior. Discretized hotspot regions frequently visited by residents were clustered into 15 groups. Regions of each group formed a strongly associated structure. With land-use and road network information, the intrinsic mechanism of this characteristics was also analyzed. Findings of this paper could provide decision-making support and useful knowledge for layout design of urban function region and the construction of smart cities.

    Figures and Tables | References | Related Articles | Metrics
    Simulating the Optimal Migration Paths between Prehistoric Settlement Sites
    Ning LI, Lin YANG, Jiangwei SHEN, Fangzihao ZHENG, Qing YANG
    2019, 21 (6):  836-843.  doi: 10.12082/dqxxkx.2019.180540
    Abstract ( )   HTML ( )   PDF (9434KB) ( )   Save

    As the relics of the ancients for production and other activities, prehistoric settlement sites reflected objectively the strong geographical features and differentiation patterns of long-term human-environment interactions. Due to the low productivity of the ancients, more natural factors were likely taken into account in between-settlement migration and resource exploration around their residences, which was consistent with the idea of "optimal path." This paper aimed to simulate the optimal path of the ancients in migrating between settlement sites, and to explore their behavioral pattern based on archaeological and geographic data. We emphasized the determination of migration cost, the integration of cost raster layers, and the optimal path analysis. The basic geographical and archaeological data of more than 1200 sites of the sequential cultural period, involving Daxi, Qujialing, and Shijiahe Cultures in the late Neolithic period in the middle reaches of the Yangtze River, were included for this study. Elevation, slope/slope direction, topographical relief, and distance from water were selected as the independent variables to analyze and construct the distribution model of the settlement sites in the study area, based on binary classification logistic regression model. To analyze the spatiotemporal evolution of the settlement groups, firstly, the clustering and centers of the settlement groups were identified based on the Expectation Maximization method and Voronoi algorithm; then, the natural factor coefficients in the distribution model were used as weights to generate the cost raster layers; lastly, the layers were input for simulating the optimal path of between-settlement migration constructed based on the cultural sequence. The results were consistent with the basic rule that the ancients were inclined to walk through valleys and hills in the process of migration at the lowest cost. The simulated path was proved reasonable when compared to the other available site information. The method proposed in this paper can address the limitation of relying on only subjective weighting in generating and integrating the cost raster layers, and can enhance the science of archaeological research. Our method and conclusion help explore the spatiotemporal evolution patterns of prehistoric settlement sites and human-environment relationship.

    Figures and Tables | References | Related Articles | Metrics
    Spatiotemporal Analysis of the Trajectories of Guqin Celebrities based on Crowdsourcing Data
    Ju LIU, Can CHEN, Jun XU
    2019, 21 (6):  844-853.  doi: 10.12082/dqxxkx.2019.180575
    Abstract ( )   HTML ( )   PDF (15298KB) ( )   Save

    Guqin is the most classical Chinese musical instrument. In its more than 3000-year history, guqin has developed many genres with specific characteristics in different regions of China, with each genre having its representative players. If the lifeline trajectories of guqin celebrities in history can be collected, the spatiotemporal distribution of each genre can be analyzed which will help to know the development and evolution of this ancient art. Due to the lack of specialized historic literature of guqin and that the information of guqin are scattered in other literature, it is difficult to collect all the information efficiently. With the rapid development of network technology and the increasing number of internet users, more and more volunteers on the Internet are willing to participate in crowdsourcing. A spatiotemporal trajectory collection and retrieval system of guqin celebrities was built by combining Chinese guqin art, crowdsourcing, and GIS. Representations and the databases of the trajectory data and knowledge data were presented in this study. There are three modules of the system, a data collection module, a knowledge base module, and a spatial retrieval and visualization module. The data collection module collects crowdsourcing input data. The knowledge base module is used to store and retrieve knowledge of guqin. There are complex relations between genres, places, and people, so the graph database Neo4j is used to represent guqin knowledge and the rich relationships among guqin players. The spatial retrieval and visualization module displays trajectories in 2 or 3 dimensions. With the collected trajectories, the spatiotemporal distribution of locations on the trajectories was analyzed. Results show that the trajectories of guqin celebrities were consistent with the trend of population migration in China's history, and that guqin celebrities tended to stay in historically famous cities and landscapes, which were conducive to spreading the guqin culture and creating guqin music.

    Figures and Tables | References | Related Articles | Metrics
    Determining the Distribution of Unmanned Aerial Vehicles Airports for the Emergency Monitoring of Floods in China
    Ming LU, Xiaohan LIAO, Huanyin YUE, Shifeng HUANG, Chenchen XU, Haiying LU, Yiqin BAI
    2019, 21 (6):  854-864.  doi: 10.12082/dqxxkx.2019.180177
    Abstract ( )   HTML ( )   PDF (7507KB) ( )   Save

    Frequent flood hazards affect large areas and cause great losses, which have posed a serious threat to economic development and people's lives and properties. Unmanned Aerial Vehicles (UAVs) have proven to be useful in monitoring disaster status and providing decision-making support for emergency rescue, because they are able to arrive at floods area timely, obtain flood images and videos quickly, low-risk to operate and flexible to carry out different sensors. The important roles of UAV in emergency rescue have been widely recognized. However, the lack of available UAV resources nearby at the sudden onset of floods seriously limits the capability of UAVs' rapid response to flood disasters. For addressing this challenge, a multi-UAVs remote sensing observation network is highly required, and now has been planned in China to enhance our ability of emergency response. Key problems include where and how to allocate UAVs resources such that the UAVs can reach destination timely when floods occur. To help close this gap, we proposed to build a number of UAV airports in China to establish a remote sensing observation network of UAVs. Field stations of Chinese Academy of Science (CAS) were considered as the potential locations of UAV airports because of their extensive geographical distribution and good cooperation with the UAV application and regulation research center of CAS. With the available flood risk prevention data, administrative division data, CAS field stations data and the UAV database as data sources, we created a fishnet of 0.5° multiply 0.5° to discretize administrative divisions and regarded the central points of these grids as potential demand points, and then calculated the importance of UAV to those demand points based on their flood risk prevention level. Based on this analysis, a Maximum Covering Location Problem (MCLP) model was adopted to determine the optimum stations for UAV airports and a cost-effectiveness curve was used to determine the optimum number of UAV airports. In the end, 81 field stations were selected from 268 field stations, thereby ensuring that UAV airports would be allocated near flood-prone areas and most floods in China could be monitored with UAVs within two hours, which is critical for saving lives and reducing losses. The construction of UAV airport networks will surely contribute to an integrated disaster emergency observation system combining satellite, airplane, UAV and ground observations in China. Additionally, the methods and results in this study can serve as a basis for building a more comprehensive national UAV remote sensing observation network.

    Figures and Tables | References | Related Articles | Metrics
    Assessment of China's High-Temperature Hazards: Accounting for Spatial Agglomeration
    Ting ZHANG, Changxiu CHENG
    2019, 21 (6):  865-874.  doi: 10.12082/dqxxkx.2019.180580
    Abstract ( )   HTML ( )   PDF (8584KB) ( )   Save

    Hazardness assessment is the basis of high temperature hazard (HTH) research. The spatial aggregation of multiple extreme high temperature events will magnify the impacts of extreme events, which not only strengthen high temperature risk, but also increase the vulnerability of disaster-bearing bodies. Previous studies mainly use high temperature days and high temperature intensity to depict HTH, yet seldom consider the the spatial agglomeration of high temperature events. To address this gap, the HTH assessment in this paper integrates the traditional HTH assessment and the spatial agglomeration of high temperature events. In so doing, the evaluation of HTH is more comprehensive by accunting for the synergy effect of high temperature hazard agglomeration. Based on high temperature days, high temperature intensity, and high temperature spatial agglomeration, HTH in China from 1979 to 2017 was evaluated integratively. The spatial distribution characteristics of these indices were illustrated and their inter-annual variations were analyzed. Finally, areas with the highest integrative high temperature intensity and regions with simultaneous enhancement of the three indices were idenified separately. The results show that the strength of HTH in western and northeastern Inner Mongolia and northern Shanxi is underestimated due to the lack of consideration of high temperature spatial agglomeration, where the Grade 4 HTH is underestimated to be Grade 2 or 3. On the other hand, if HTH assessment considered only the number of high temperature days, the high temperature hazardness in most areas of Inner Mongolia, Heilongjiang, Shandong, and the northern part of western Shaxi would have been underestimated. This indicates that HTH assessment needs to account for the three indices simultaneouly. From the results of integrative assessment, for now, the most dangerous areas due to high temperature are located in southern Tianshan Mountains and eastern Hunan, where high temperature days are 20-36 days, high temperature intensity ranges from 1.190 ℃ to 2.180 ℃ and high temperature spatial agglomeration ranges from 13.390 to 18.710. From the results of inter-annual variations, all of the three indices significantly increased in the junction of Inner Mongolia and Gansu, Jiangsu, and the junction of Sichuan and Chongqing. These areas may become the most dangerous areas of high temperature hazards in the future, where from 1979 to 2017 the variability of high temperature days is 0.419-0.740 days/year, the variability of high temperature intensity is 0.30-0.42 ℃/10 years and the variability of high temperature spatial agglomeration is 0.250-0.390 per year. The proposed HTH assessment method is helpful to improve the accuracy of risk assessment of high temperature hazards, and the results of our trend analysis of these indices can help predict the high-risk areas. In addition, the method can identify the high-risk areas of each hazard factor and also the dominant factor of each high-risk area. It provides a scientific basis for targeted high temperature prevention and mitigation, as well as resource allocation over the whole China.

    Figures and Tables | References | Related Articles | Metrics
    Assessing the Spatial Layout Efficiency of basic Educational Resources by an Improved Coverage Model: A Case Study of Nanchang City, China
    Mengfang GUO, Bisong HU
    2019, 21 (6):  875-886.  doi: 10.12082/dqxxkx.2019.190034.
    Abstract ( )   HTML ( )   PDF (12423KB) ( )   Save

    :With the continuous improvement of people's material well-being, people's demands for educational qualities and equities also increase. It is a trend of China's educational reform to promote the integration of the compulsory educations in urban areas and rural areas. The study of educational resources has attracted growing interests. This paper firstly improved a coverage model; Then, based on the improved model and the proportional model-based fairness index, this paper quantitatively analyzed the spatial distribution and layout efficiency of the basic educational resources (i.e., the primary and middle schools) in 2018 of Nanchang City on the town/subdistrict scale. Results show that: (1) There are differences between the existing scale of primary and secondary schools in Nanchang City and the basic conditions of running schools in Jiangxi Province. The number of primary and secondary school class size in each jurisdiction area exceeds the standard. (2) The township with the most convenient access to primary and secondary education resources are located in Qingshanhu District and Xihu District, while the streets with the most convenient access to primary and secondary education resources are located in Honggutan District and Donghu District. (3) As to the coverage and overlap of the service areas of the primary and middle schools, they all show the core-periphery pattern. There exists a significant positive correlation between the coverage and overlap of the service areas. (4) For the spatial aggregation of the schools in Nanchang City, it is obviously stronger for the middle schools than the primary schools. Over half of the schools in towns/subdistricts, mostly located at the central urban regions, are of high-coverage and high-overlap. (5) With respect to the overall spatial layout efficiencies of the primary and middle schools, there also exists a significant difference among villages, towns, and subdistricts. The major layout type of the schools is the deviated type. The distribution of the basic educational resources is imbalanced. The situation of service areas of the primary and middle schools is bad in that they are of high-coverage, high-overlappe, low-coverage, or low-overlap. Further, the integrative service efficiency is low. To conclude, the current resource configuration and spatial layout of the primary and middle schools in Nanchang need to be substantially improved.

    Figures and Tables | References | Related Articles | Metrics
    Effect of Grid Size on Habitat Quality Assessment: A Case Study of Huangshan City
    Jian PENG, Feixiong XU
    2019, 21 (6):  887-897.  doi: 10.12082/dqxxkx.2019.180632
    Abstract ( )   HTML ( )   PDF (21886KB) ( )   Save

    Grid cell is the basic spatial unit to analyze the habitat quality based on remote sensing imagery. The choice of suitable grid size is very important for different regions, as the resolution must be fine enough to ensure the accuracy of habitat quality assessment. The purpose of this study is to identify differences in the habitat quality assessed at different grid scales, and to provide a basis for selecting appropriate spatial scales to analyze habitat quality in different regions, so as to improve the assessment accuracy. Taking Huangshan City as the example, this paper evaluated its comprehensive habitat quality in 2017 based on the model of ecosystem service value, habitat quality index consisting of NPP and NDVI, and InVEST habitat quality assessment model. With 30 m as the base scale and employing the elastic coefficient and spatial autocorrelation method, this paper assessed how changing grid sizes affect the habitat quality assessment results. The findings are as follows. (1) in 2017, the original assessment result of Huangshan's total habitat quality was 2.02×1010 yuan, with an average of 21126.1 yuan/hm2. And the last adjusted total habitat quality was 1.84×1010 yuan, with an average of 18627 yuan/hm2. (2) With the increase of pixel's side length, the total value of the comprehensive habitat quality assessment result decreased in the zigzag form. (3) The grid size effects on different land uses showed that the results of the habitat quality assessment were much more accurate when the scale of change was an odd multiple of the base scale, i.e., 30 m. And it was much clearer for the elevation range of 0~200 m and 200~400 m. (4) In terms of the elasticity coefficient of habitat quality calculated for scale change from one grid size to another, the coefficient was larger for the scale changes from 150m to 180 m, 270 m to 300 m and above 900 m. With the base scale as a reference, the elasticity coefficient of the habitat quality would decrease in the power-function form with increasing grid size. And the change of the habitat quality would be most sensitive when grid size changed from 30m to 60m. While the value of habitat quality change was relatively insensitive as the grid size changed from 30m to 210m. (5) The spatial distribution of habitat quality in Huangshan City showed a significant positive autocorrelation. With increasing grid size, the index of Moran's I decreased in a waving manner, and the Z value of the normal distribution also decreased in the power-function form. Our fidings can provide theoretical support for the selection of suitable grid sizes for habitat quality analysis in different places.

    Figures and Tables | References | Related Articles | Metrics
    Photon-Counting LiDAR Point Cloud Data Filtering based on the Random Forest Algorithm
    Bowei CHEN, Yong PANG, Zengyuan LI, Hao LU, Xiaojun LIANG
    2019, 21 (6):  898-906.  doi: 10.12082/dqxxkx.2019.190013
    Abstract ( )   HTML ( )   PDF (17508KB) ( )   Save

    The new generation of spaceborne laser satellite ICESat-2 (the Ice, Cloud, and land Elevation Satellite-2) of NASA (National Aeronautics and Space Administration) has adopted a newly designed micropulse photon counting system, which is the very first time that this technology gets applied in the space environment. Thanks to the high sensitivity of single photon detection technology, it can be seen from the currently released data product (both from the airborne simulators and the simulation data) that there is huge noise in the atmosphere and even below the ground. Therefore, preliminary research on these relevant experimental data to investigate the methods for separating signal photons from noise photons are important for the future applications. MATLAS data, which simulate the expected performance of the ICESat-2 ATLAS (Advanced Topographic Laser Altimeter System) instrument, was chosen to test our machine learning-based approach from two test sites in Oregon and Virginia in the United States. We first derived 12 features, such as the kNN (k-Nearest Neighbour) distance, based on the characteristics of photon point clouds data. Then we applied feature selection techniques by ranking variable importance using Random Forest. Three most representative features were chosen according to the variable importance ranking and we built a Random Forest classifier trained by the sample points we had selected. The established models were further applied to the whole study area. The final classification results indicate that the classifier we constructed had good performance to distinguish signal photons from noise photons. In terms of the mean values of the statistical indicators in the test sites, the overall classification accuracy was 96.79%, and the Kappa coefficient was 0.94. The producer and user accuracies were 97.1% and 96.8%, respectively. Additionally, the results show that our method not only worked well on data of relatively lower noise rate on flat terrain surfaces but also achieved good results for those with higher noise rate on complex terrain surfaces. To conclude, our method showes good potential to be applied to larger areas, for especially the classification of the photon counting LiDAR data in the future.

    Figures and Tables | References | Related Articles | Metrics
    Waterbody Extraction from SAR Imagery based on Improved Speckle Reducing Anisotropic Diffusion and Maximum Between-Cluster Variance
    Yu LI, Yun YANG, Quanhua ZHAO
    2019, 21 (6):  907-917.  doi: 10.12082/dqxxkx.2019.180525
    Abstract ( )   HTML ( )   PDF (29476KB) ( )   Save

    The use of remote sensing technology to obtain surface waterbody information is of great significance for water resource investigation, natural disaster assessment, watershed planning, and ecological environment monitoring. As a reliable data source for large-scale ground monitoring, SAR imaging has unique advantages that optical remote sensing systems of all-weather, all-weather, and wide coverage do not have, and has been widely used in waterbody mapping. However, due to the influence of speckle noise of SAR imagery, existing methods for waterbody mapping are difficult to extract the complex and fine natural water structures from SAR imagery quickly and accurately. In this paper, a new waterbody extraction method for SAR imagery based on improved speckle reducing anisotropic diffusion and maximum between-cluster variance was proposed. First, the SAR imagery were filtered by improved speckle reducing anisotropic diffusion. The iterative process was adaptively controlled by calculating the average structural similarity between imagery in the iterative filtering process, so that the fine edges and texture structure could be preserved simultaneously. Then, based on the criterion of maximum variance between classes, the threshold value was determined adaptively, and the binary segmentation of the filtered image was conducted. In the binarized segmentation result, connected foreground regions composed of the pixel points that have same intensity values and adjoin to each other spatially, are searched. In so doing, each connected region formed an identified block. By obtaining geometric parameters of these blocks, the false segmentation of the imagery was eliminated, and real waterbody areas were precisely identified based on the SAR imagery. To verify the accuracy of the proposed method, water boundaries extracted by this method were on manually drawn waterbody boundaries. Results of comparing the two methods show that they are pretty consistent with each other. Meanwhile, the results of our proposed method were compared with the results of three other kinds of water extraction algorithms commonly used for SAR imagery, in terms of the visual level, extraction accuracy, and running time. The running time of the proposed method meets the requirement of real-time application. The overlap degree of the extracted boundary in two pixels rating areas has reached 80%, which is obviously superior to other methods and the extraction results of the proposed method are also more significant in visual aspects such as boundary and detail information. The qualitative and quantitative evaluation shows the superiority of our proposed method.

    Figures and Tables | References | Related Articles | Metrics
    Identifying Soybean Cropped Area with Sentinel-2 Data and Multi-Layer Neural Network
    Fuyou TIAN, Bingfang WU, Hongwei ZENG, Zhaoxin HE, Miao ZHANG, Bofana José
    2019, 21 (6):  918-927.  doi: 10.12082/dqxxkx.2019.180424
    Abstract ( )   HTML ( )   PDF (23843KB) ( )   Save

    As the most important oil crop in the world, soybean is a large-scale agricultural product that China imports. The accurate identification of its planting area is the basis for decision-making and planting structure adjustment, and is of great significance to national food security. Sentinel-2 was used as a data source and multi-layer neural network was employed to map soybean cropped area. Besides, visible and infrared bands, three red-edge bands were also selected after radiation and atmospheric correction using the Sentinel-2 Toolbox. According to our test, 8-hidden-layer conducted using Scikit-learn package in Python2.7 was the optimal structure for identifying soybean and other crops. Simple linear iterative clustering (SLIC), the state-of-art segmentation algorithm, was performed to segment the remote sensed image. This method combined five-dimensional color and the image plane space to efficiently generate compact and nearly uniform super pixels. To remove the “salt and pepper effect”, the pixel-based result was integrated with the object output from the SLIC. If the pixel as soybean in an object accounted for less than 50%, this object was eliminated in the fusion map. The results showed that the overall accuracy of multi-layer neural network was 93.95%, which was highest and followed by the support vector machine, decision tree, and random forest. Then, the neural network classification was selected as the best result to integrate with SLIC object-oriented segmentation, and the results ignored the small differences of the same land and distinguish the crop differences of different blocks compared with the segmentation in eCognition software. Sentinel-2 data is an appropriate data source for large-scale soybean planting mapping. According to feature importance derived from the random forest classifier, near-infrared band is the most critical feature for classification, followed by third red edge band (Band 7), fourth red edge 4 band (Band 8), red band, and second red edge band (Band 6). The reflectance values of soybeans and other crops in the second red edge band were different, indicating a huge potential in crop type identifying. In the future, the red edge band can be introduced more into crop type even landscape classification. The multi-layer neural network method performs well in the image classification task and had similar or better overall accuracy value compared with other outstanding machine learning classifier including SVM, decision tree, and random forest. Combined with the image segmentation algorithm, such as SLIC, multi-layer neural network can map soybean cropped area with an accuracy high up to 95.51%, which can serve for soybean planting area monitoring in a large area.

    Figures and Tables | References | Related Articles | Metrics
    Prediction Method of Tungsten-molybdenum Prospecting Target Area based on Deep Learning
    Huihui CAI, Wei ZHU, Zixuan LI, Yuanyuan LIU, Longbin LI, Chang LIU
    2019, 21 (6):  928-936.  doi: 10.12082/dqxxkx.2019.190032
    Abstract ( )   HTML ( )   PDF (11533KB) ( )   Save

    With the exploration of minerals from shallow mines to deep concealed mines, from easy-to-identify mines to difficult-to-identify mines, the difficulty of prospecting is increasing, and geological experts are paying more and more attention to the application of new theories, new methods, and new technologies. As a frontier field and technology of artificial intelligence, deep learning has a unique advantage in realizing the intelligent forecasting and evaluation of mineral resources. The method uses normalized geochemical data as the training data to extract outliers by a neural network called Autoencoder and identify the favorable mineralization areas, and then realizes the qualitative prediction of mineral resources prospecting prospect. The research results show that after classifying the original data of 957 single elements geochemical anomalies and labeling of the model, the whole process automatically completes the learning and prediction in the "black box" of the computer, compared with the traditional prediction research method, this method of research is highly automated and objective. In addition, this paper uses the known mine sites to construct the training dataset, and uses the random forest method to predict the mineral resources prospecting target area in the prediction area, which provides a scientific basis for further narrowing the scope of the prospecting target area.

    Figures and Tables | References | Related Articles | Metrics
    Mapping Paddy Rice in the Hainan Province Using both Google Earth Engine and Remote Sensing Images
    Shen TAN, Bingfang WU, Xin ZHANG
    2019, 21 (6):  937-947.  doi: 10.12082/dqxxkx.2019.180423.
    Abstract ( )   HTML ( )   PDF (10741KB) ( )   Save

    Rice is one of the main grain crops in China and East Asia, including China. The annual yield of rice has a significant influence on domestic livelihood. Therefore, timely and accurate assessment of rice distribution information is crucial for forecasting rice yields and optimize the allocation of agricultural resources. Remote sensing (RS) images can provide time series surface spectral, and other electro-magnetic, dynamics over a large-scale land surface, which are commonly used for large-scale crop monitoring. However, routine rice classifying strategies provided by the RS images during key growth stages, require spectral patterns at high frequency. This method appears to be impractical within South China, as the number of high quality RS images are difficult to obtain due to cloud contamination caused by the hot and wet weather. A combination of various RS images of rice classification from multi-platforms provide an indirect way of reducing the revisit period in routine rice classification, thus enabling successful crop mapping in cloudy regions. However, this causes difficulty with data manipulation and storage, especially when conducting classification work at province or large area levels. To address these issues, this research utilizes Google Earth Engine (a cloud-based geospatial analysis platform running on the Google server)to collect online optic RS data and micro-wave RS data at diverse resolutions for rice mapping. A distribution map of paddy rice at 10-m spatial resolution in the Hainan Province in 2016 was made by using the combined methods of random forest (RF) classification and a pattern-matching strategy based on conjunct features extracted at monthly level and histogram value distribution. Results showed this method was suitable for rice mapping in Hainan and could show clear feature divergence between the different land surface cover types. Spatial distribution results corresponded well with the actual edges of the field, along with texture information. The rice classification result of the Hainan Province was validated using sample points captured on the ground and achieved overall accuracy of 93.2%, indicating reliability for practical application. Overall, the automatic rice classifying strategy was able to map paddy rice with high efficiency and sufficient accuracy in the Hainan Province, and could be applied to other vast areas.

    Figures and Tables | References | Related Articles | Metrics
    Hyperspectral Estimation Models of Chlorophyll Content for Dicranopteris Dichotoma Leaves at Different Ecological Restoration Stages in the Eroded Red Soil Areas of Southern China
    Chao DENG, Zhibiao CHEN, Haibin CHEN, Zhiqiang CHEN
    2019, 21 (6):  948-957.  doi: 10.12082/dqxxkx.2019.180513
    Abstract ( )   HTML ( )   PDF (2650KB) ( )   Save

    Dicranopteris dichotoma is an important zonal herbaceous plant for ecological restoration in the eroded red soil areas of southern China, and is an effective control of soil erosion. Using remote sensing techniques to monitor the chlorophyll content can help diagnose the vegetation growth and healthy condition of dicranopteris dichotoma. Based on hyperspectral reflectance data and the corresponding chlorophyll contents of dicranopteris dichotoma leaf from six different ecological restoration stages in Zhuxi small watershed of Changting County, Fujian Province, this study analyzed the hyperspectral curve properties of leaf, transformed the original spectral into the first derivative, and selected the sensitive wavebands to create ratio (RVI) and normalized (NDVI, FDNDVI) hyperspectral indices. Then correlation analysis was conducted for the chlorophyll contents and hyperspectral indices which were selected from reported indices and newly constructed indcies with sensitive wavelengths. Based on the correlation coefficients, we can chose the best indices to create estimation models. The linear, exponential, multiplicative, quadratic polynomial, logarithmic, and multivariate regression models were constructed for comparison. Furthermore, the optimal estimation model was determined by the accuracy of each estimation model. Results showed that the sensitive wavelengths of the original spectral for dicranopteris dichotoma leaf at different ecological restoration stages were 407 nm, 603 nm, and that the optimal wavebands of the first derivative were 463 nm, 554 nm, 674 nm, and 739 nm. The relationship between the chlorophyll content of dicranopteris dichotoma leaf and the hyperspectral indices of red edge position (λr), NDVI[603, 407], Modified Red Edge Normalized Difference Vegetation Index (mNDVI705), Vogelmann Index (Vog) were very significant, and the correlation coefficients were over 0.85. The estimation models of chlorophyll content established by hyperspectral indices of mNDVI705, Leaf Chlorophyll Index (LCI), Vog, RVI603/407, NDVI[603, 407] showed better test results, and the R2 were over 0.8. The model established by FDNDVI[739, 463] index had the highest verification accuracy, and the R2 reached 0.741. The multivariate regression model based on hyperspectral indices got highest test results with the highest R2. Therefore, the LCI index and the multivariate regression model based on hyperspectral indices have the strongest ability for predicting chlorophyll concentration, which provides scientific basis for dynamic monitoring of dicranopteris dichotoma in the eroded red soil regions of southern China. It is significant for monitoring soil and water conservation plants. Meanwhile, the objective of this research was to provide effective technical support for ecological restoration by building hyperspectral estimation models of chlorophyll content, with a rapid and non-destructive method for monitoring vegetation growth.

    Figures and Tables | References | Related Articles | Metrics
    Forest Vegetation Dynamics and Responses to Climate Change in a Southern Subtropical Monsoon Region in Jangle County
    Bingjie ZUO, Yujun SUN
    2019, 21 (6):  958-968.  doi: 10.12082/dqxxkx.2019.180686
    Abstract ( )   HTML ( )   PDF (9134KB) ( )   Save

    Vegetation dynamics and responses to climate change is a hot research topic in the fields of ecology and geography. This paper analyzed and compared the difference in the response of different forest vegetation to different time scales in a Southern subtropical monsoon region. We assessed the dynamics characteristics of forest vegetation and relevant meteorological factors, analyzed the differences in the response of different forest vegetation to climate change. We used the 2000-2017 MODIS-EVI data and meteorological site data, and conducted the maximum value composite (MVC), liner trends, and correlation analyses. We have five major findings. Firstly, in the 18 years, the forest vegetation coverage, EVI, precipitation, and humidity increased significantly, indicating that the forest vegetation has been growing better. Secondly, at the beginning and end of the growing season, there was a significant positive correlation (p<0.1) between EVI and precipitation, so was temperature. At the beginning of the growing season, EVI was more affected by precipitation; while at the end of the growing season, it was more affected by temperature. Thirdly, climate change in the January to March and over the whole year is critically important to forest growth. With increasing time scale, the correlation with EVI and SPEI also increased. The increase of humidity on long-term scales has a positive effect on forest growth. Fourthly, broadleaf-conifer mixed forests have larger correlation coefficients with EVI and meteorological factors than other forest types do. Their correlation with SPEI at different time scales is relatively larger than with meteorological factors. Broadleaf-conifer mixed forests are climate-sensitive; more attention should be paid to their production and management to prevent damages induced by climate change. Fifthly, the correlations of forest cover change with precipitation and with SPEI_24 are extremely significant, indicating that the long-term precipitation change is an important factor affecting the change of forest vegetation coverage.

    Figures and Tables | References | Related Articles | Metrics
    Extraction of Arctic Urban Land Use Information based on Multi-source Remote Sensing and Topograph
    Li LIANG, Xinyang LI, Qingsheng LIU, Gaohuan LIU, Chong HUANG, He LI
    2019, 21 (6):  969-982.  doi: 10.12082/dqxxkx.2019.180536
    Abstract ( )   HTML ( )   PDF (31142KB) ( )   Save

    As climate warms up and ice melts, the Arctic is drawing much more attention. It is undeniable that Arctic urban spatial information is critical for studying, understanding, and exploring the Arctic. Due to the special geographical situation, Arctic urban extraction has unique difficulties such as urban fragmentation and confusion with bare mountains. To overcome the problems of extracting Arctic urban, multi-source data including Landsat, DMSP/OLS, and ASTER-GDEM2 were used. Spectral features, texture features, nighttime light features, and topographic features were obtained after feature extraction. Apart from that, the AdaBoost algorithm was used to extract the urban areas at 1990, 2004 and 2016. To clearly and more completely understand the function of each feature, we divided features into four different groups, and compared their differences. The result shows that, adding terrain features or nighttime lighting features can improve the extraction accuracy, and that the combination of spectrum, texture, terrain, and nighttime lighting is the optimal combination of features. The overall accuracy (OA) and kappa values based on spectral and texture features are 86.20% and 0.68, respectively. After adding terrain feature, the accuracy increased by 2.7% (OA) and 6.21% (kappa) respectively. When only adding nighttime lighting feature, OA increased by 2.1% and kappa 0.50. The best result was reached when we added terrain feature and nighttime lighting simultaneously. In this case, the overall accuracy and kappa increased by 3.7% and 8.55%, respectively. So, it is the optimal combination of features. After identifying the optimal feature combination, the maximum likelihood method was used to extract urban areas to prove the effectiveness of the AdaBoost algorithm. Experiment results show that, with the optimal feature combination, extraction based on AdaBoost has its OA and kappa value 10% and 20% higher respectively than those by the maximum likelihood method. Finally, the urban expansion was analyzed. The intensity of the urban expansion in the study area is around 4.4×10-3 from 1990 to 2004 and this number is 4.5×10-3 from 2004 to 2016, which can be interpreted as slow expansion. The average level of expansion is 0.018, 1/2 of the global average. The urban expansion level between 1990 and 2004 is higher than that between 2004 and 2016. The difference in the dynamics during 1990-2004 and during 2004-2016 indicates that the study area is currently transitioning from a high-speed development period to a stable development period. Given the warming of the Arctic and the growing of population, Arctic urban is expected to continue expanding slowly.

    Figures and Tables | References | Related Articles | Metrics