Snow cover， snow depth/snow water equivalent， surface soil frozen/thaw state and soil moisture are the key variables in the three cycles including energy， water and carbon cycles. In order to better understand the remote sensing techniques of above parameters， this paper presents a comprehensive review of the progress in remote sensing of snow， soil frozen/thaw state and soil moisture， including the methods and theories of snow cover， snow depth / snow water equivalent， surface soil frozen/thaw and soil moisture remote sensing monitoring from visible， microwave techniques and the integration of multi-sources of remote sensing. The research progress of these parameters is summarized， and the prospects of these parameters are also discussed. The capability of snow， surface soil frozen/thaw state and soil moisture with remote sensing has been demonstrated to be improved greatly due to the retrieval algorithms development based on from single-sensor to multi-sensor combination， single-band to multi-band integration， especially on the virtual satellites constellation. Long time series data set of these surface parameters about 40~50 years were generated， then these products provide our better understanding on surface response to global climate change， and accelerating the application into the research of hydrology， climate and carbon cycles. This review will be helpful for the application of key parameters retrieval in water cycle with remote sensing.
The temporal and spatial variation characteristics of snow depth over the Northern Hemisphere and nine typical areas were analyzed based on the GlobSnow snow water equivalent datasets of European Space Agency and the NHSD sow depth datasets of the National Qinghai-Tibet Plateau Scientific Data Center. The results showed that： the Average Annual Snow Depth （AASD） over the Northern Hemisphere generally decreased significantly （p<0.01） during 1988 to 2018， with a change slope of -0.55 cm·（10 a）-1. For high latitudes， the AASD in the northern Canada and Alaska decreased significantly （p<0.01）， with a rate of 3.48 cm·（10 a）-1 and 3 cm·（10 a）-1， respectively； and the Average Monthly Snow Depth（AMSD） decreased significantly in winner. The AASD decreased in the West Siberian Plain and Eastern European Plain with a significant change rate of -2.3 cm·（10 a）-1 in the latter （p<0.01）， and the AMSD decreased significantly in spring， especially in May. The AASD in the Eastern Siberia showed an increased trend， except in Kamchatka Peninsula， and the AMSD increased significantly in winner. For high mountainous areas， the AASD showed a slow increase rate in the Alps and Rockies， and slight decrease change in the Qinghai-Tibet Plateau （QTP）. The AMSD in Alps increased significantly in winner and decreased significantly in May. The variation of snow depth in the Rockies and the QTP presented spatial heterogeneity. During the whole study period， the AMSD decreased in the north of the Rockies and most areas of central region of QTP， while increased in the central and south of Rockies and the mountains on the northern edge of the QTP. The snow depth increased on the north slope of The Himalayas， while decreased on the south slope， with the absolute change rates of less than 0.5 cm·a-1. The AMSD of Nianqing Dangla Mountains which has deep snow showed a significant downward trend in winner. The seasonal variation analysis of snow depth （average snow depth from 2001 to 2010） in 9 typical areas showed that the peak values of snow depth in high mountainous areas are much smaller than those in high latitudes. The snow melting dates in high mountainous areas are obviously earlier than those in high latitudes except for the QTP.
The 8 scenes SAR imagery acquired by Sentinel-1A satellite in TOPS mode are employed in surface motion extraction and analysis of South Inylchek Glacier during period from January to December in 2018 with Small Baseline Subset Pixel Tracking Technique （SBAS-PT）， which could overcome the limit of the temporal incoherence. The results show that its overall movement rate is relatively small during period from January to March. The ice motion increased significantly since April and reached the maximum during period between July and August， then the movement rate began to slow down from September. And the ice motion became slow again during the period between October and December. The average surface velocity of the whole year is about 30 cm·d-1. In general， the ice motion rate in the upstream part of South Inylchek Glacier is significantly higher than that in the lower part of the glaciers. Both the reduction of the ice material supplement and the increase of moraine on glacier surface have made the ice gradually become to be stable in downstream part of the glacier.
Xinjiang is rich in glacier resources， and the development potential of glacier tourism is huge， but its current tourism development is seriously lagging behind. This study established a evaluation system of glacier tourism service potential from five aspects： glacier resource endowment， tourism resource diversity， climate comfort， transportation accessibility， and population economic conditions. By using the analytic hierarchy process （AHP） and ArcGIS spatial analysis tools， the study established an evaluation system of glacier tourism service potential， and comprehensively evaluated the level of glacier tourism service potential in the 54 counties or cities， finally put forward corresponding spatial development strategies. The results show that： （1）the glaciers in northern Xinjiang are small in scale and have poor endowments， but have better accessibility； （2）the glaciers in southern Xinjiang have large scale and high endowments， and the conditions for combining with surrounding landscapes are superior， but glaciers have poor accessibility； （3）Currently， the overall level of glacier tourism service potential in the central and western regions of the Tianshan Mountains is relatively high， while other regions are relatively low； （4）In the 1930s and 1950s， with climate change and social development， the areas with high glacier tourism service potential will be shifted from Tianshan mountains to the Pamirs and Karakoram Mountains. Generally， the current level of glacier tourism service potential in the Tianshan Mountains， Xinjiang is relatively high， but the service potential of glacier tourism will be significantly reduced with the rapid retreat of small-scale glaciers in the future. Based on this， the current glacier tourism service potential dividend should be actively utilized， and the ice and snow tourism industry should be vigorously developed to promote the development of Xinjiang's all-season and all-for-one tourism.
This study proposed a recognition method for snow cover based on feature selection using GF-3 fully polarimetric data. The study area was selected from the typical area of the Kelan River Basin in the southern piedmont of the Altai Mountains， Xinjiang Province. First， we obtained 22 polarization features of GF-3 data by polarization decomposition. The importance of each feature was calculated by using Random Forest （RF） method. Then， we designed the rules of feature selection to generate the optimal feature sets， which were used to recognize snow cover with RF method. Analyzing the importance of the features， we can find that， for snow cover recognition， the contribution of the same polarization backscattering coefficient is greater than that of the cross polarization backscattering coefficient， and the contribution of the surface scattering or volume scattering is greater than that of the dihedral angle scattering. Finally， a comparison with the Maximum Likelihood， Support Vector Machine， and BP neural network was made for testing the performance of the proposed method. It is found that the optimal feature sets using RF method to recognize snow cover have the highest accuracy （F-score is 0.86， overall accuracy is 0.79）. From the selection of classifiers and the results of features selection， the proposed method is very effective in recognition of snow cover.
Fractional Snow Cover （FSC） is the ratio of the Snow Cover Area （SCA） to the spatial area in a unit pixel， which can provide quantitative information of snow cover distribution for regional climate simulation and hydrological model. MODIS FSC products are calculated according to the empirical model， without considering the impact of environmental factors such as topography， vegetation and surface temperature. The accuracy in the Tibetan plateau is low. Therefore ， the effects of environmental factors （topography， vegetation， and surface temperature） on FSC preparation were taken into account in Tibetan plateau， based on Multivariate Adaptive Regression Splines （MARS） and linear regression model， and established a non-parametric regression model and an empirical regression model respectively based on Multivariate Adaptive Regression Splines （MARS） and linear regression model. The reference dataset of FSC was prepared with Landsat 8 surface reflectance data and SNOMAP algorithm. A part of reference dataset is selected as the training samples of the model， and the other part as the validation dataset of the model. The results show that the accuracy of the MARS method is significantly higher than that of the linear regression model and the original MODIS FSC preparation method. The total R， RMSE and MAE of MARS were 0.791， 0.103 and 0.058， respectively. In the linear regression model， the overall R， RMSE and MAE with the highest accuracy are 0.647， 0.128 and 0.072， respectively. The overall R， RMSE and MAE of the original MODIS FSC mapping method are 0.595， 0.221 and 0.170 respectively. MARS method with environmental information is more suitable for FSC preparation in Tibetan plateau. This study provides a new idea for preparing FSC data with higher accuracy in Tibetan plateau.
The use of GPS-Interferometric Reflectometry （GPS-IR） can realize the monitoring of surface environmental parameters. Based on the relationship between multi-path reflected signals of GNSS and snow depth， this paper proposes a multi-star fusion product based on multiple linear regression based on the consideration of the influence of multi-star fusion on the inversion effect. Snow depth inversion model. In order to verify the reliability of the algorithm， the snow depth inversion research was performed using the continuous monitoring data from the P101 station in the PBO observation network in the United States. Theoretical and experimental results show that the inversion results have a significant correlation with the snow depth reference value； multi-star fusion can effectively synthesize the inversion performance of each single satellite， and the correlation coefficients are all greater than 0.940， which is at least 13.6% higher than the single star； Both RMSE and MAE are less than 0.08 and 0.165.
Qilian Mountains is a typical area of snow cover research in China because of its rich snow types and complex identification. Therefore， it is of great significance for the regional ecological environment and socio-economic development to accurately monitor the change of snow area and its spatiotemporal evolution in the Qilian Mountains. FY-3C MULSS uses the multi-threshold snow index model to provide global daily snow cover products， and the FY-4A AGRI sensor provides a global multispectral image every 15~60 min on average. By using the characteristics of FY-4A AGRI with the high temporal resolution， a dynamic multi-threshold and multi-temporal snow detection method suitable for FY-4A data was constructed， which greatly reduced clouds impact on snow detection in optical images. Then combining with the advantages of the higher spatial resolution of snow-covered daily products FY-3C MULSS， the FY3C4 snow-covered data after cloud removal is obtained. Using Landsat 8 OLI high-resolution satellite data to verify the accuracy of the fused snow cover， the results show that the fused FY-3C and FY-4A can better identify the snow cover in the Qilian Mountains. Taking MODIS MOD10A2 snow cover product as the real value， the recognition accuracy of the fused daily snow cover product from March 2018 to March 2019 is tested randomly， and the overall accuracy is as high as 85.25% without the clouds. Further research shows that the snow area in the Qilian Mountains is extremely uneven in time and space under the influence of altitude， climate and slope direction. In general， the snow cover area in winter and spring is larger than that in summer and autumn， and the snow cover area in the east is larger than that in the west.
Rock glacier is a geomorphological landform with ligule accumulation texture that formed from mixture of ice and debris. Investigating the distribution of rock glaciers can provide effective information for studying the environment and climate change in cold regions. The development of remote sensing technology provides an effective way for identifying rock glacier. However， its execution is difficult due to the large area of rock glacier distribution as well as the similarity between rock glacier and its surroundings in spectral surface reflectance. Comparing to the traditional visual interpretation approach， this paper presented a more effective method for automatically identifying rock glaciers in high-resolution images. The method was implemented by integrating the Deep Learning development framework to build the model through interactive training from the ResNet network. The model was then applied to identify rock glaciers in GaoFen-1 images that collected in West Nyainqentanglha Mountains， where are rich of rock glaciers. Gaofen-1 images were used as the satellite data source， and 96 rock glaciers were identified in the West Nyainqentanglha Mountains. Accuracy of the results were assessed by comparing to human interpreted data， and it reported 98.72% Overall Accuracy， 89.48% Producer's Accuracy， and 81.77% User's Accuracy， suggesting that the presented method is very effective for identifying rock glaciers， and it provides a potential capability for mapping the distribution of rock glacier in large areas.
Snow cover recognition with high temporal resolution plays an important role in the development of agriculture and animal husbandry and snow disaster warning in Xinjiang pastoral areas. To solve the problem that existing snow cover products are susceptible to complex topography， landform， underlying surface type and cloud cover， which leads to the reduced accuracy of snow cover recognition， a deep learning method is proposed to use the data of Fengyun-4A Star Multichannel Radiation Scanner （AGRI） and the number of geographic information.Based on the method of multi-feature time series fusion， a new snow cover recognition model based on convolution neural network is constructed and trained， which takes the multitemporal FY-4A/AGRI multispectral remote sensing data， terrain topographic information such as elevation， aspect， slope， and surface cover type as the input of the model， and the high-resolution snow cover map extracted by Landsat 8-OLI as the "true value" label.Clouds， snow and snow-free surfaces in Xinjiang's complex terrain and underlying areas ultimately lead to hourly snow cover products. It is verified by the data set and the snow coverof meteorological station in 2019 the accuracy of this method is higher than that of MOD10A1 and MYD10A1， the main international MODIS snow products， which significantly reduces the misclassification rate of cloud and snow.
NPP-VIIRS night lighting data is a stable data source for the study of urban development and change on medium and large scale. Based on the night lighting data of NPP-VIIRS from 2012 to 2018， taking Shandong Peninsula urban agglomeration as the research object， extracting urban built-up area patches by reference comparison method and selecting nine landscape pattern indices to quantitatively analyze the urbanization development characteristics of Shandong Peninsula urban agglomeration. The results showed that： ①As a whole， the total area of patches in Shandong Peninsula increased at a rate of 4.5%， the total length and density of the edges increased by 3.15% annually， and the number and density of patches increased rapidly （1.95% and 1.98%， respectively）， indicating that the overall urban area of Shandong Peninsula urban agglomeration increased rapidly and the urban area continued to expand. ②According to the changing trend of different indicators， qingdao and dongying cities （9.66% and 6.01% respectively） had the fastest growth in the total area of patches； qingdao had the fastest increase in the number and density of patches （9.54% and 8.55% respectively）， and rizhao had a significant decrease in the number and density of patches at the rate of 3.65%； the overall growth rate of landscape shape index was slow； the average radius of gyration had a high annual growth rate in rizhao city （5.99%）. ③From the differences in the development characteristics of various cities， the average patch area and gyration radius of qingdao decreased by 0.56% and 1.53% respectively， while other indicators increased significantly， indicating that there were more emerging towns in qingdao and the urban area continued to expand. Urban areas in jinan， rizhao and dongying cities growed rapidly， and the number of patches， landscape shape index and other indicators grow slowly. The urban development of jinan， rizhao and dongying cities is dominated by the expansion of old urban areas. Around 2015 and 2016， weifang， zibo and yantai experienced the emergence of emerging towns and urban integration， with rapid urban development. Generally speaking， the urbanization of Shandong Peninsula urban agglomeration develops rapidly， but the spatial difference is obvious.
The problems of low resolution and poor timeliness of nightlight remote sensing images are the resistance to the development of light pollution research. In this regard， a novel method to capture nightlight remote sensing image called Volunteered Passenger Aircraft Remote Sensing （VPARS） is used to obtain high-resolution nighttime light imagery from Changsha City， combined with the commercial POI data in 2018， the source and the patterns of light pollution was analyzed， and the preliminary application of VPARS in light pollution research was explored. The results showed that the Volunteered Passenger Aircraft Remote Sensing can effectively obtain high-precision night-time remote sensing data of cities， and has great potential in light pollution monitoring applications. According to the preliminary analysis of the VPARS night-light remote sensing image of Changsha City， the POI of shopping service， living service and catering service are responsible for more than half of the detected light output， and the brightness coefficient of them is much higher than the average value， which makes them the main source of light pollution in Changsha. Although the POI of travel service and public service facilities has a low luminous ratio， the brightness coefficient reaches a maximum of 1.82， and the degree of light pollution is high. Light pollution in Changsha is mainly concentrated in commercial districts and urban cores.
The poverty area recognition is the key to formulate national poverty alleviation policies. Based on satellite-based nighttime light data （NPP-VIIRS data） of 119 counties in Shanxi Province from 2013 to 2017， the statistical significance of differences between poverty counties and other counties was tested by variance analysis in terms of total light intensity， average light intensity， maximum patch light intensity， total patch area， and patch agglomeration. The recognition model of poverty areas was then developed using the NPP-VIIRS data of 2013 and applied to recognize poverty counties in 2014~2017. The results showed that the recognition accuracy of the model for poverty counties is relatively high， ranging from 79.31% to 86.21%. For non-poverty counties， the recognition accuracy is relatively lower， ranging from 59.02% to 73.77%. The comprehensive recognition accuracy is between 71.43% and 77.31%. Besides parameters of light intensity， including parameters related to landscape characteristics of lighted patches helps to improve model accuracy. In addition， we analyzed the relationship between poverty probability and GDP， the reasons of the counties with incorrect cognition， and annual variation of the poverty probability for 58 poverty counties and 15 counties out of poverty. The results not only confirmed the applicability of nighttime light data in the poverty counties recognition and assessment of the counties out of poverty， but also highlighted the important role of landscape characteristics of lighted patches， which were not included in the existing studies.
To solve the issues of inaccurate bounding box in large-scale remote sensing image object detection， an accurate object detection and localization appoarch of remote sensing image based on Bayesian Optimization is proposed. The method consists of two stages： In the first stage， the EdgeBoxes which is based on edges information is adopted to generate object proposals. The classifier is applied to get initial object detection result. To obtain more accurate bounding box， a bayesian optimization based on gaussian process is applied to fine-tune the bounding box around each object in the second stage. Firstly， a set of boxes that intersect the initial bounding box around each initial box is selected to form a gaussian process. Secondly， a new bounding box is estimated through bayesian optimization and added to the set of boxes. Thirdly， the score of each box is calculated by the classifier， and the box with the highest score is set as the base box in the next iteration. At last， the bayesian optimization process is repeatedand and final bounding boxes is obtained. Experiments demonstrate the EdgeBoxes method can achive a better recall evaluation with less number of propsals. The bayesian optimization based on gaussian process can significantly improve the localization accuracy of the detection bounding box.
The Frequency Modulation Continuous Wave Inverse Synthetic Aperture Radar（FMCW ISAR） images by transmitting the Frequency Modulation Continuous Waves， due to the limited bandwidth of the transmitted waves， the imaging is affected by the sidelobes of the sinc function inevitably. The CLEAN algorithm is widely used in the astronomical images. In order to solve the problems of the sidelobes of the sinc function， the algorithm which combines of the windowing in time domain and the CLEAN algorithm is proposed， and improves the quality of the images obviously. Finally， the “dirty beam” and “clean beam” in the CLEAN algorithm are modified to improve the capacity of the radar detection. The results show that the proposed method has a certain reference for the applications.
Time series remote sensing data with moderate resolution is playing as an important role for surface process and surface disturbance observation. For Landsat data in tropical areas， there are invalid land observation data and data loss caused by cloud， fog or sensor defects. Based on the existing GNSPI algorithm， a triple repairing method is proposed. This method automatically identifies the effective reference pixels within 48 days that before and after the current invalid pixels， and then obtain effective reference pixels for the current data to be repaired， and then fill the current missing data using the three triple repair which was based on the GNSPI method. The overall average filling accuracy is up to 0.88 in our study area. This method makes up for the GNSPI filling algorithm's harsh requirement that the reference image needs the whole image without invalid pixels， and improves the utilization rate of observed high quality pixels. The method proposed in this paper has a great significance for the establishment of long time series data and the related research.
In order to solve the problems of traditional multi-scale infrared visible fusion image with blurred edges， low contrast and inconspicuous targets， an infrared visible image fusion algorithm based on mutually guided filtering and saliency map is proposed. Firstly， the original images are decomposed into structure layers with redundant information and texture layers with complementary information at different scales by means of mutually guided filter， because the filter can separate the consistent structure from the inconsistent structure， and has the awareness of scale and edge preservation. Secondly， the visual saliency map function is constructed to map saliency of structure layers and texture layers of different scales according to the visual characteristics of the over-light or over-dark regions that are more likely to attract attention. Finally， according to the structure and texture characteristics of different scales， the final fusion image is reconstructed. The experimental results on two data sets show that compared with the traditional multi-scale fusion methods， the proposed method has a better subjective and objective evaluation effect in maintaining the image edge， enhancing the image contrast and highlighting the target.
