Some topographic factors such as slope, aspect, and land cover may cause errors on TanDEM-X 90 m Digital Elevation Model (DEM) product when collecting and processing of these data. In order to better understand the error distribution and serve the research in this field, the comparison between TanDEM-X 90 m DEM and ICESat/GLA14 DEM was conducted over the entire China. The findings are summarized: ① The average absolute error, Root Mean Square Error (RMSE) and Standard Deviation (STD) of TanDEM-X 90 m DEM over the entire China are about 3.89, 9.03, and 8.85 m, respectively. ② The error increases when the slope increases. The mean absolute error is about 1.29 m and the STD is about 2.84 m when the slope is below 3°. In comparison, the mean absolute error is above 20 m and the STD is about 30 m when the slope is above 25°. ③ For the aspect, the mean value of absolute error in the north-south direction is obviously smaller than that in the east-west direction, indicating the influence of aspect on TanDEM-X 90 m DEM product. ④ For the land cover, the uncultivated land shows the smallest error with the mean absolute error of 1.85 m, while the region covered with snow and glacier show the largest error with the mean absolute error of 12.68 m. Comparisons of the contour map and profile between TanDEM-X 90 m DEM and UAV-derived DEM suggest that the TanDEM-X 90 m DEM can reflect the real topography. However, due to the influence of resolution in some areas, it can not be expressed for some detailed terrains, especially for valley and ridge. The absolute error distribution of TanDEM-X 90 m DEM over the entire China is produced and evaluated based the weights of different influencing factors, which are considered to be reliable. Through the analysis of error distribution map, it is found that the accuracy of TanDEM-X 90 m DEM shows a trend of high in the north and low in the South over the entire mainland of China. In the North China region, the overall accuracy is higher, while the error in the northwest region is smaller, but the overall accuracy in the Central South region is poor. By referring to the relevant data, when using the data of TanDEM to generate DEM, its accuracy has a great relationship with the vegetation coverage rate of the area. High forest coverage rate will seriously affect the coherence of SAR data, and then affect the accuracy of generated DEM.

Global climate change has increased the impact and destructive power of marine environmental disasters. Real-time disaster information acquisition and analysis have become a key process in disaster emergency response. Compared with traditional earth observation network with delay effect, the crowdsourcing geographic information based on the Internet has been widely valued in disaster situation assessment and emergency response for its real-time nature. Marine environmental disasters include not only one single disaster but also a series of secondary and derived disasters caused by a single disaster. The latter are the extension and expansion of a disaster across the space and time dimensions, which continue to affect human activities. The analysis of a single disaster is difficult to fully explain the temporal and spatial evolution of a disaster to further explore its scope and degree of impact. In order to obtain information on marine environmental disasters implied in the Internet texts and explore the impact of disasters on human activities, this article builds a marine environmental disaster chain ontology based on collected marine environmental disaster-related knowledge and using five-tuples (concepts, relationships, properties, rules, and instances) to represent the logical structure of the ontology, which integrates the marine environmental disaster ontology describing the marine environmental disaster knowledge system, the geographical object ontology affected by the marine environmental disasters, and the corresponding human emergency ontology generated from the spatiotemporal processes of a disaster's occurrence, development, and end together. This article not only analyzes the concept of marine environmental disasters from the perspective of geographical events but also focuses on the characteristics of temporal and spatial processes of the development of marine environmental disasters. Many factors involved in the development of disasters from the perspective of the disaster chain are also analyzed. Thus, our paper breaks through the limitation of using a single disaster to represent the entire disaster process in traditional analysis and comprehensively analyzes the disaster process-related information. Finally, this article takes typhoon disaster as an example based on the typhoon disaster chain to illustrate the disaster information extraction from the Internet texts and analyze the spatiotemporal changes of typhoon disaster, so as to provide effective support for disaster prevention and emergency response.

Machine learning has been widely applied to analyze regional landslide susceptibility, such as the artificial neural network and decision tree model. Model selection depends on both the reliability and accuracy of model results, therefore comprehensively evaluating the performance of a model is necessary. Previous studies of landslide susceptibility focused more on the prediction accuracy of a model. However, model stability and model sensitivity to data volume also reflect important model performances in different aspects. In this study, we employed Back-Propagation (BP) artificial neural network and Classification and Regression Tree (CART) model for model performance comparison in landslide susceptibility prediction. We evaluated model performance from three aspects: Data sensitivity, prediction accuracy, and model stability. The Caiyuan basin in Fujian Province was taken as the study area and 11 landslide-related factors were selected. Additionally, Beichuan county in Sichuan Province was taken as the verification area and 12 landslide-related factors selected. Firstly, two models were both trained using different amounts of data as input. With increasing data volume, the prediction accuracy of BP artificial neural network increased faster than that of CART model. Specifically, in Caiyuan basin, the prediction accuracy of BP artificial neural network and CART decision tree model increased by 5.22% and 2.11%, respectively, for every additional 10 000 samples. In Beichuan county, the prediction accuracy of these two models increased by 4.88% and 3.40%, respectively. Secondly, 100 sets of training data and validation data generated by random sampling were fed into two models for training. The experimental results show that, for Caiyuan basin, the mean prediction accuracy was 81.60% and 72.97% for BP artificial neural network and CART model, respectively, and the standard deviation was 0.32% and 0.35% for BP and CART, respectively. For Beichuan county, the mean prediction accuracy of two models was 77.45% and 72.61%, respectively, and the standard deviation was 0.47% and 0.61%, respectively. Finally, landslide susceptibility maps were generated based on two models. Compared to real landslide spatial distribution map, the result of BP artificial neural network was more consistent with the actual landslide distributions. In general, our study demonstrates that BP artificial neural network is more sensitive to the increase of data volume and has better model stability and prediction accuracy than CART model. But it is worth noting that the performance of two models is close with small data volume. The study provides a new perspective of model selection for landslide susceptibility analysis.

Forest fire occurs frequently and suddenly. Therefore, it is essential to carry out the rapid warning of forest fire danger for the reduction of the loss caused by forest fire and the promotion of sustainable development of forest resources. This paper designs an early-warning model based on GIS spatial analysis and visualization technology and the construction of real-time meteorological monitoring network using ground meteorological stations, which can achieve timely and rapid warning of forest fire danger. To build the model, this paper first determines the forest fire danger early-warning factors, which are the input parameters of the model. Secondly, a hierarchy model of the importance of early warning indicators is constructed to determine the weight of the early warning factors via using the AHP method and combining the analysis of early warning factors. Then, the thresholds and grade division criteria of the early-warning factor are determined according to the national, industrial, and local regulations for determining forest fire danger levels, which is suitable for the model. Finally, the Voronoi Diagrams are used to establish a meteorological monitoring network based on weather stations and real-time weather data. The Overlay Analysis technology is used to calculate the early warning result. Based on the model and real-time acquisition and processing of data, a rapid warning system for forest fires was constructed. This paper took Qinghai Province as the experimental area where the feasibility and applicability of the system were verified, which indicates that early warning of forest fire danger can be realized by the model comprehensively, accurately, and rapidly. Results show that: (1) According to the early-warning model, the real-time early-warning indicators which were set before, and real-time meteorological monitoring data, the early-warning signal can be sent in time, which can quickly realize early warning and timely response of forest fire risks at the county and forest farm levels; (2) Via introducing GIS visualization methods, the thematic map of forest fire risk spatial distribution can be generated by the model quickly, which is conducive to observe changes in early-warning levels visually. The rapid warning of forest fire risks has important guiding functions for effective prevention, interruption management, and prevention measures of forest fire, and has great significance for forest fire prevention work, protection of forest resources, and safety of human life and property.

Measuring the urban growth boundary is important to control the disorderly expansion of urban constructed land. How to define the urban growth boundary scientifically is a hot topic of current researches. This study attempts to introduce Baidu dynamic traffic data and POI data to improve the FLUS model to simulate land use changes. Taking the central of Changsha city as an example, the simulation accuracy of the improved FLUS model is first verified by comparing with the land use data of 2000, 2010, and 2018. Then, the land use of central Changsha in 2030 is simulated based on two scenarios using the improved FLUS model. The urban growth boundary is finally defined based on land suitability evaluation. The results show that: (1) Compared with the original FLUS model, the kappa coefficient of the improved FLUS model with dynamic data increases by 2.90% and 2.74% in 2010 and 2018, respectively, and the overall accuracy increases by 1.79% and 1.83%, respectively, which indicates a higher simulation accuracy of the improved model; (2) Based on the simulated land use of central Changsha in 2030, the area of constructed land is 930.06 km² and 881.36 km² respectively in benchmark scenario and ecological protection scenario. The largest proportion of land converted to construction land is cultivated land; (3) The area within the rigid growth boundary of central Changsha is 1479.59 km², accounting for 37.38% of the total area of the central city. These areas include Furong District, Tianxin District, Yuhua District, Yuelu District, and Kaifu District; (4) The area within the elastic growth boundary of central Changsha is 799.35 km² and 742.92 km² under the benchmark scenario and ecological protection scenario, respectively. The expanded construction mainly occurs in Changsha County and Wangcheng District, which is consistent with the development direction of 2010 Changsha urban master plan. The improved FLUS model with dynamic data can simulate the urban growth boundary in multiple scenarios, which provides a scientific basis for future planning decision.

Climate comfort level has great significance to human activities and regional suitability assessment, and temperature humidity index is important for climate comfort evaluation. Traditional temperature-humidity index is obtained based on the observed data from some sations, which cannot reflect the spatio-temporal characteristics of climate comfort in large-scale areas. In this paper, the modified temperature humidity index model is proposed using Land Surface Temperature (LST) and Precipitable Water Vapor (PWV) from 2005 to 2018 retrieved from MODIS. Using this new index, the spatio-temporal characteristics of climate comfort level in China are calculated and analyzed. The results are shown as follows: (1) The GWR method is used to fit the surface temperature and air temperature. The fitting accuracy (Adjusted R² = 0.90~0.98, RMSE = 0.14~1.89 ℃) is ideal, which indicates that LST, NDVI, and DEM are used as the independent variables for geographical weight Regression analysis can more accurately fit the air temperature; (2) The statistical results of the annual average temperature and humidity index from 2005 to 2018 show that the cumulative number of comfortable months in Yunnan Province is the most, up to 167 months, and the central provinces are relatively comfortable compared to the southeast coastal provinces, and the difference between the highest comfort months can reach 41 months. The spatial distribution of China's average annual comfort level is basically the same. Except for parts of Xinjiang, Tibet, and the northeast, the comfort level in China is from south to north, and the comfort level changes from comfortable to cold. Judging from the changes in the area of each comfort level, the national comfort level showed a trend from cold to comfortable from 2005 to 2018; (3) The month with largest comfortable area in 2018 is May, followed by October. Uncomfortable months are concentrated in January and July when the country is extremely cold or hot. The spatial distribution characteristics of spring and autumn are similar, showing a gradual decreasing trend from southeast to northwest; except for the Qinghai-Tibet Plateau, summer and winter show a decreasing trend from south to north. The comfort zone is mainly concentrated in low-latitude and middle-altitude areas.

In recent years, PM_2.5 has become one of the main pollutants in the haze outbreak. The risk of long-term exposure to PM_2.5 of high concentration may greatly increase the risk of disease and endanger the physical and mental health of residents. In this study, Beijing was taken as the research area where air pollution was serious and the population was highly concentrated. Based on the data of PM_2.5 concentration, the grid data of population spatial distribution, and the long-term respiratory volume of different populations in Beijing in 2019, an assessment model of PM_2.5 population exposure was established. Furthermore, the spatial distribution characteristics of PM_2.5 population exposure intensity and the differences of exposure-response among different populations in Beijing in 2019 were analyzed. The results show that: (1) In 2019, the PM_2.5 concentration in Beijing is the highest in winter, and the daily average concentration is 48.89 μg/m³, which shows an overall trend of low in the north and high in the south; (2) There are significant spatial differences in PM_2.5 population exposure, and the PM_2.5 exposure of different populations shows an overall trend of weakening from the center of the city to the surrounding areas, and the high exposure areas were mainly concentrated in urban areas; (3) There are obvious spatial differences in the exposure intensity of PM_2.5 population in different gender and age groups, and there were also significant differences in the response of PM_2.5 exposure among different populations in the city; (4) The exposure risk of PM_2.5 was not entirely determined by the concentration of pollutants, but by the concentration of pollution sources and exposure receptors, the high-risk area of population exposure to PM_2.5 in Beijing urban area was the high-risk area, and it was the core area for the government to effectively prevent and control pollution hazards in the future.

After the Chinese government put forward the Rise of Central China Plan, it rapidly facilitates the economy development of Henan province, Hubei province, Hunan province, Jiangxi province, Anhui province and Shanxi province which has gradually become the fourth growth pole driving national economic growth, has caused the built-up area to expand and arable land to decrease, which not only threaten food security, but also impose resource and environmental constraints. In the context, based on the panel data of 80 prefecture-level cities in six provinces of central China from 2007 to 2018, this paper analyzes the green-efficiency of land use and its evolution, the trajectory of gravity center change, influencing factors of green-efficiency of land use and its influence degree applying Malmquist-Luerberger index, gravity center model, spatial econometric regression model and geographical detector model. The results show that ① the green-efficiency of land use and technological progress in the six provinces of central China from 2007 to 2018 fluctuated frequently and their change pace was basically the same, while the technological efficiency was relatively stable, indicating the green-efficiency of land use was “single-track” driven by technological progress. ② The green-efficiency of urban land use showed obvious spatial differentiation characteristics, and the center of gravity generally moved to the northeast part of Central China. ③ It showed spatial dependence and spatial spillover effects on the green-efficiency of land use at the provincial level and prefecture level, the green-efficiency of land use among the prefecture-level cities is mainly in the high-high and low-low level. ④ In addition to the area of urban construction land, urbanization rate, the advanced level of industrial structure, the level of economic development, and the amount of foreign direct investment all positively affect the green-efficiency of land use in 80 prefecture-level cities of six central provinces. Among them, the influence degree of various factors on the green-efficiency of land use from strong to weak, in order, is the advanced level of industrial structure, the amount of foreign direct investment, the area of urban construction land, the urbanization rate and the level of economic development.

Tick Borne Encephalitis (TBE) is an acute infectious disease involving central neuropathy caused by the bite of a Tick borne Encephalitis Virus (TBEV) infected tick, which is a typical zoonotic disease. TBE occurs in areas with a wide distribution of tick and its distribution is related to the environment. Besides, it is also a typically natural focus disease in a special ecosystem composed of pathogens, vectors, and host animals. As a natural focus disease, TBE threatens human health and impedes socioeconomic development in northeastern China. Therefore, analyzing the spatial-temporal distribution of TBE and its influencing factors are of vital importance for TBE control. This paper selected Heilongjiang Province, Jilin Province, and the Inner Mongolia Autonomous Region in northeastern China, which are typical TBE endemic areas, to study the spatial-temporal distribution of TBE and its influencing factors. Firstly, we explored the spatial-temporal distribution of TBE in the study area from 2005 to 2015 through statistical analysis and spatial autocorrelation, Secondly, we used Geo-Detector to investigate the factors that influence the spatial distribution of TBE and its indicative role. Our results show that: (1) the incidence of TBE in the endemic areas of Northeastern China had an obvious growth trend and seasonal incidence characteristics from 2005 to 2015. The incidence of TBE in the study area had a strong spatial clustering pattern with two main hot spots, Hulunbuir city in the Inner Mongolia autonomous region and the Greater Khingan Range region in Heilongjiang province; (2) vegetation type, land use, average annual temperature, soil type, average temperature from May to August, slope, elevation, and annual rainfall were the main influencing factors of spatial prevalence of TBE. In general, the influence of natural environment was stronger than that of social environmental; (3) for the whole study area, the relationship between each risk factor and TBE was different, and the incidence of TBE was different with each factor. Besides, the interaction between various factors was significantly enhanced, that is, the impact of two factors was stronger than that of a single factor. The common interaction between some factors exceeded 0.5, and most factors exceeded 0.3. Particularly, the main interaction enhancement effect was manifested in the interaction of each factor with land use and elevation. Our results provide scientific basis and decision support for the effective control of TBE in the study area and the whole country.

As a powerful meteorological disaster, tropical cyclones have a great impact on the maritime shipping of the Maritime Silk Road. Base on the best track data of tropical cyclones over the North Indian Ocean and the Northwest Pacific Ocean from 1990 to 2017, this paper analyzes the spatio-temporal characteristics of the tropicalcycloneshazard. First, the tropical cyclones are classified into 4 grades. Next, the Modified Rankine Vortex (MRV) model is used to simulate the wind field of each tropical cyclone. Then, the frequency of each grade of tropical cyclone that occurred in the research area can be obtained. Finally, a tropical cyclone hazard assessment model is used to assess the tropical cyclone hazard level in the main sea areas of the Maritime Silk Road.The main conclusions are as follows: (1) The sea areas of the Maritime Silk Road are seriously affected by tropical cycloneswith high occurrence frequency of tropical cyclones. The Northwest Pacific Ocean is more severely affected by tropical cyclones than the North Indian Ocean. (2)The sea areas between 15~30° North and 120~145° East are at the highest hazard level. (3) The seasonal change of tropical cyclone hazard is obvious. The tropical cyclone hazard levels of the sea areas of the North Indian Ocean and the Northwest Pacific Oceanare higher in autumn and summer than winter and spring. Moreover,in the summer and autumn, the sea areas in July, August, September and October have the relatively higher hazard levels. (4) Among the main sea areas, Eastern China Seais at the highest hazard level, followed by the South China Sea, the Sea of Japan, the Bay of Bengal, and the Arabian Sea, while the Red Sea and the Persian Gulf are not affected by tropical cyclones. Among the main channels, the Luzon Strait is at the highest hazard level, followed by the Taiwan Strait, the Tsushima Strait,the Soya Strait, the Tartar Strait, the Paco Strait, and the Strait of Hormuz, however the Strait of Malacca and the Strait of Mande are not affected by tropical cyclone.

Planning and construction of China-Pakistan Economic Corridor (CPEC) is inseparable from the scientific cognition of the spatial patterns and changing processes of land resources and eco-environment in this area. Land Use and Land Cover Change (LUCC) simulation can provide reliable prediction data for regional land resources management, eco-environment sustainability, and eco-environment risk assessment. In this paper, based on the coupled System Dynamics Model (SD) and future Land Use Simulation Model (FLUS), combined with the China-Pakistan Economic Corridor construction and regional eco-environment policies, various scenarios were set up to simulate the land use change of the China-Pakistan Economic Corridor, taking full advantages of the two models in macro land demand simulation and micro land allocation. Firstly, the SD-FLUS model was constructed and validated using the historical data of CPEC in 2009-2015. Then the land use changes from 2016 to 2030 under three different scenarios, namely Baseline Development (BD)scenario, Investment Priority Oriented (IPO) scenario, and Harmonious Development (HD) scenario, were simulated. Results show that: (1) The relative error of demand simulation was less than 9%, and the overall accuracy and Kappa coefficient of the simulation were over 90% against the actual land use data in 2015, which indicates the SD-FLUS coupling model effectively reflected the land use change pattern of the China-Pakistan Economic Corridor. The model could be used for further simulation of land use changes in CPEC under different scenarios; (2) There are significant differences in simulated land use under different scenarios until 2030. Construction land expanded under all three scenarios but at different speeds. The expansion speed of HD scenario was in the middle. Under this scenario, the construction land in Kashgar and Pakistan increased by 235.17 km² and 4942.80 km², respectively. The expansion speed under the IPO scenario was the fastest, with the construction land in Kashgar increased by 265.23 km² and construction land in Pakistan increased by 5918.91 km². Under the BD scenario, the construction land in Kashgar and Pakistan increased by 163.71 km² and 2861.84 km², respectively. Under the HD scenario, increment of Pakistan's cultivated land area was less than half of that under BD scenario. Kashgar's cultivated land area increased the most in IPO scenario (about 882.54 km²), which was about three quarters of that in the HD scenario. The forest land was effectively restored only under the HD scenario. Generally, the HD scenario taking both social-economic development and eco-environment protection into account is the most ideal scenario among the three scenarios. Our simulation results can provide useful data support for the construction of China-Pakistan Economic Corridor and the assessment of eco-environment in the future.

The relative radiometric calibration of optical satellite sensors minimizes the inconsistency of linear or non-linear responses of the sensor detectors with different incident radiances. It is a prerequisite of various types of high-quality remote sensing products. The response of on-orbit sensor changes with time due to factors such as launch induced vibration, space environment changes, and the sensor degradation. However, both the one-time relative radiometric calibration and single relative radiometric calibration methods cannot guarantee the consistency in responses of satellite sensors at a specific time. Therefore, the normalized on-orbit radiometric calibration method with high frequency and high precision for satellite sensors is necessary for better applications of remote sensing products. In this study, we focused on the image radiometric calibration of the whole life cycle of optical satellite sensors, and the high-frequency, high-precision, wide or full dynamic range on-orbit relative radiometric calibration methods. Also, we summarized the existing accuracy evaluation methods of relative radiometric calibration methods as well as their application scenarios. In our study, the LJ1-01 nighttime satellite images were used to verify the dark current and uniform field calibration methods. The Zhuhai1 group 02 hyperspectral satellite images were used to verify the statistical calibration and yawing radiometric calibration methods as well as the normalized radiometric calibration method integrated by a variety of calibration methods. Our results show that the stripe coefficient was less than 0.25%, with a higher relative standard deviation of the images (3.00%) than images processed by each individual evaluation method. The normalized radiometric calibration method integrated by various calibration methods maximizes the advantages of individual calibration methods and is capable of high precision on-orbit calibration for common sensors, which further guarantees the quality of remote sensing products and meets the requirements of quantitative application.

High-intensity underground mining leads to huge surface deformation, and excessive deformation phase gradients could lead to interference unwrapping errors. Currently, the characteristics of surface subsidence cannot be obtained well by conventional DInSAR and its derivative technologies. In this paper, we propose a new method, which combines DInSAR and Offset-tracking technology to extract large-scale deformation accurately, and further restore settlement field shape using the GAUSS function model. We take Bu'ertai Mine, and Cuncaota No.1 and No.2 Mine in Shendong Mining Area as research areas and use the high-resolution RADARSAT-2 data with 5 m fine beam model (MF5) on February 13, 2012 and November 27, 2012 to obtain the mine subsidence boundaries at sub-centimeter level using DInSAR technology and the mine subsidence center at meter level using Offset-tracking method. The boundary subsidence value is -0.01~ -0.02 m and the central subsidence value is -1.0~ -4.0 m. The whole subsidence field is then retrieved by integrating above two results. Finally, the GAUSS function model is used to fit the mining subsidence boundary and central, and to reconstruct the characteristic curve of mining subsidence. Our results demonstrate that the Offset-tracking method could compensate the deficiency of DInSAR in large deformation extraction, and the combination of these two techniques could effectively and accurately extract large-scale subsidence field in mining areas.

The change of landscape pattern has an important influence on the material and energy flows of ecological environment. Quantifying the landscape pattern of rocky desertification in a karst area is very important for understanding the development of rocky desertification. Rocky desertification is a dynamic land degradation process, which is a comprehensive reflection of vegetation, bedrock, soil cover, and other surface factors. Particularly, exposed bedrock acts as a key indicator of karst rocky desertification. In this study, spatial distribution of rock patches with varied bare-rock ratio (11%, 20%, 29%, and 48%) is characterized using an Unmanned Aerial Vehicle (UAV) image in a rocky desertification area in Guizhou Province. The classification methods for this small-scale UAV image include unsupervised classification, supervised classification, and object-oriented classification. The existing supervised and unsupervised classification methods based on pixels cannot meet the requirements of accurate extraction of rocky desertification information in karst rocky desertification area with complex geological environment. So an optimal classification method is selected to classify the UAV image of rocky desertification in our study. Our results show that the object-oriented classification method has higher accuracy than the others, which reduces the “salt-and-pepper phenomenon” caused by complicated topography. Based on object-oriented classification, the UAV image is interpreted first, and the distribution characteristics of rock patches with different bare-rock rates (i.e., 11%, 20%, 29% and 48%) are quantified by combining various indices in landscape ecology including landscape patch index, landscape shape index, and landscape fragmentation index. Generally, the average patch area of rock is negatively correlated with bare-rock rate. With the increase of bare-rock rate in different rocky desertification areas, the number of rock patches gradually increase with increasing rock shape index and rock fragmentation index, which indicates the increase of rocky desertification. The exposed bare rocks in this karst area show different distribution patterns and characteristics under different rocky desertification rates. The higher the rate of bare rock is, the higher the degree of rock fragmentation is, with a relatively scattering distribution of rock patches. Analyzing the rock distribution for a rocky desertification area can provide support for the evaluation and management of rocky desertification areas. Since the changes of small-scale, small-patch landscape pattern in rocky desertification areas can affect the ecological processes, our small-scale study also provides better understanding of future processes of rocky desertification and the development of rocky desertification at regional scale.

Vegetation classification is the basis and premise of forest resource investigation and dynamic monitoring. Remote sensing techniques have long been important means of forest monitoring with their ability to quickly and efficiently collect the spatial-temporal variability of vegetation. Vegetation classification is a key issue for forest monitoring and is critical to many remote sensing applications in the domain of precision forestry such as vegetation area estimation. Remote sensing applications in vegetation classification have traditionally focused on the use of optical data such as MODIS. However, due to cloud and haze interference, optical images are not always available at phenological stages that are essential to vegetation identification, making it difficult to construct complete time-series vegetation growth and limiting the vegetation classification accuracy. Unlike passive visible and infrared wavelengths which are sensitive to cloud and light, active SAR (Synthetic Aperture Radar) is particularly attractive for vegetation classification due to its all-weather, all-day imaging capabilities. In addition, SAR provides information on the stem and leaf structures of vegetation and is sensitive to soil roughness and moisture content, making it effective in forest applications. In this study, a deep-learning-based time-series analysis method employing multi-temporal SAR data is presented for forest vegetation classification in the Taibai Mountain (the main peak of Qinling Mountains). Firstly, Sentinel-2 optical images and digital elevation data in the study area were used for multi-scale segmentation to produce a precise farmland map. Then pre-processed SAR intensity images were overlaid with the farmland map to construct time-series vegetation growth for each parcel. Finally, a deep-learning-based classifier using the Long Short-Term Memory (LSTM) network was employed to learn time-series features of vegetation and to classify parcels to produce a final classification map. The experimental results show that: (1) Compared with traditional machine learning methods (such as Random Forest and Support Vector Machine), the classification accuracy of the deep-learning-based method proposed in this paper was improved by more than 10%; (2) Among different combinations of Sentinel-1A polarizations, VV+VH performed best, having a similar accuracy with the VV+VH+VV/VH; (3) Time-series classification using all images in the whole year achieved the best performance, with an overall accuracy of 82% using VV+VH. The study shows that the combination LSTM network and time-series Sentinel-1A data can effectively improve the accuracy of vegetation classification and provide new ideas for forest vegetation classification from the perspectives of data source and classification method.