Based on the data of GF-1 and GF-2 in China, the freeze-thaw disaster distribution data of Qinghai Tibet project corridor is produced by using the deep learning classification method and manual visual interpretation and correction. The geographical range of the data is 40km along the Xidatan Anduo section of Qinghai Tibet highway. The data include the distribution data of thermokast lakes and the distribution data of thermal melting landslides. The dataset can provide data basis for the research of freeze-thaw disaster and engineering disaster prevention and reduction in Qinghai Tibet engineering corridor. The spatial distribution of freezing and thawing disasters within 40km along the Xidatan-Anduo section of Qinghai Tibet highway is self-made based on the domestic GF-2 image data. Firstly, the deep learning method is used to extract the mud flow terrace block from GF-2 data; Then, ArcGIS is used for manual editing.
NIU Fujun, LUO Jing LUO Jing
We propose an algorithm for ice fissure identification and detection using u-net network, which can realize the automatic detection of ice fissures of Typical Glaciers in Greenland ice sheet. Based on the data of sentinel-1 IW from July and August every year, in order to suppress the speckle noise of SAR image, the probabilistic patch based weights (ppb) algorithm is selected for filtering, and then the representative samples are selected and input into the u-net network for model training, and the ice cracks are predicted according to the trained model. Taking two typical glaciers in Greenland (Jakobshavn and Kangerdlussuaq) as examples, the average accuracy of classification results can reach 94.5%, of which the local accuracy of fissure area can reach 78.6%, and the recall rate is 89.4%.
LI Xinwu , LIANG Shuang , YANG Bojin , ZHAO Jingjing
We propose an algorithm for ice crack identification and detection using u-net network, which can realize the automatic detection of Antarctic ice cracks. Based on the data of sentinel-1 EW from January to February every year, in order to suppress the speckle noise of SAR image, the probabilistic patch based weights (ppb) algorithm is selected for filtering, and then representative samples are selected and input into the u-net network for model training, and the ice cracks are predicted according to the trained model. Taking five typical ice shelves（Amery、Fimbul、Nickerson、Shackleton、Thwaiters) in Antarctica as an example, the average accuracy of classification results can reach 94.5%, of which the local accuracy of fissure area can reach 78.6%, and the recall rate is 89.4%.
LI Xinwu , LIANG Shuang , YANG Bojin , ZHAO Jingjing
Fractional Vegetation Cover (FVC) refers to the percentage of the vertical projected area of vegetation to the total area of the study area. It is an important indicator to measure the effectiveness of ecological protection and ecological restoration. It is widely used in the fields of climate, ecology, soil erosion and so on. FVC is not only an ideal parameter to reflect the productivity of vegetation, but also can play a good role in evaluating topographic differences, climate change and regional ecological environment quality. This research work is mainly to post process two sets of glass FVC data, and give a more reliable vegetation coverage of the circumpolar Arctic Circle (north of 66 ° n) and the Qinghai Tibet Plateau (north of 26 ° n to 39.85 °, east longitude 73.45 ° to 104.65 °) in 2013 and 2018 through data fusion, elimination of outliers and clipping.
YE Aizhong
NDVI reflects the background effects of plant canopy, such as soil, wet ground, snow, dead leaves, roughness, etc., and is related to vegetation cover. It is one of the important parameters to reflect the crop growth and nutrient information. According to this parameter, the N demand of crops in different seasons can be known, which is an important guide to the reasonable application of N fertilizer. Correct NDVI (C-NDVI) is the value of NDVI after excluding the influence of climate elements (temperature, precipitation, etc.) on NDVI. Taking precipitation as an example, studies on the lag effect of precipitation on vegetation growth show that the lag time of precipitation effects varies in different regions due to differences in vegetation composition and soil types. In this study, we post-processed the MODIS NDVI data and firstly correlated the NDVI value of the current month with the precipitation of the current month, the average value of the precipitation of the current month with that of the previous month, and the average value of the precipitation of the current month with that of the previous two months to determine the optimal lag time. The NDVI was regressed on precipitation and air temperature to obtain the correlation coefficients, and then the corrected NDVI values were calculated by the difference between the MODIS NDVI and the NDVI regressed on climate factors. We corrected NDVI using climate data to give reliable vegetation correction indices for the circum-Arctic Circle (range north of 66°N) and the Tibetan Plateau (range 26°N to 39.85°N and 73.45°E to 104.65°E) for 2013 and 2018. The spatial resolution of the data is 0.5 degrees and the temporal resolution is monthly values.
YE Aizhong
This dataset consists of four files including (1) Lake ice thickness of 16 large lakes measured by satellite altimeters for 1992-2019 (Altimetric LIT for 16 large lakes.xlsx); (2) Daily lake ice thickness and lake surface snow depth of 1,313 lakes with an area > 50 km2 in the Northern Hemisphere modeled by a one-dimensional remote sensing lake ice model for 2003-2018 (in NetCDF format); (3) Future lake ice thickness and surface snow depth for 2071-2099 modeled by the lake ice model with a modified ice growth module (table S1.xlsx); (4) A lookup table containing lake IDs, names, locations, and areas. This daily lake ice and snow thickness dataset could provide a benchmark for the estimation of global lake ice and snow mass, thereby improving our understanding of the ecological and economical significance of freshwater ice as well as its response to climate change.
LI Xingdong, LONG Di, HUANG Qi, ZHAO Fanyu
Precipitation over the Tibetan Plateau (TP) known as Asia's water tower plays a critical role in regional water and energy cycles, largely affecting water availability for downstream countries. Rain gauges are indispensable in precipitation measurement, but are quite limited in the TP that features complex terrain and the harsh environment. Satellite and reanalysis precipitation products can provide complementary information for ground-based measurements, particularly over large poorly gauged areas. Here we optimally merged gauge, satellite, and reanalysis data by determining weights of various data sources using artificial neural networks (ANNs) and environmental variables including elevation, surface pressure, and wind speed. A Multi-Source Precipitation (MSP) data set was generated at a daily timescale and a spatial resolution of 0.1° across the TP for the 1998‒2017 period. The correlation coefficient (CC) of daily precipitation between the MSP and gauge observations was highest (0.74) and the root mean squared error was the second lowest compared with four other satellite products, indicating the quality of the MSP and the effectiveness of the data merging approach. We further evaluated the hydrological utility of different precipitation products using a distributed hydrological model for the poorly gauged headwaters of the Yangtze and Yellow rivers in the TP. The MSP achieved the best Nash-Sutcliffe efficiency coefficient (over 0.8) and CC (over 0.9) for daily streamflow simulations during 2004‒2014. In addition, the MSP performed best over the ungauged western TP based on multiple collocation evaluation. The merging method could be applicable to other data-scarce regions globally to provide high quality precipitation data for hydrological research. The latitude and longitude of the left bottom corner across the TP, the number of rows and columns, and grid cells information are all included in each ASCII file.
HONG Zhongkun , LONG Di
Aiming at the 179000 km2 area of the pan three rivers parallel flow area of the Qinghai Tibet Plateau, InSAR deformation observation is carried out through three kinds of SAR data: sentinel-1 lifting orbit and palsar-1 lifting orbit. According to the obtained InSAR deformation image, it is comprehensively interpreted in combination with geomorphic and optical image features. A total of 949 active landslides below 4000m above sea level were identified. It should be noted that due to the difference of observation angle, sensitivity and observation phase of different SAR data, there are some differences in the interpretation of the same landslide with different data. The scope and boundary of the landslide need to be corrected with the help of ground and optical images. The concept of landslide InSAR recognition scale is different from the traditional spatial resolution and mainly depends on the deformation intensity. Therefore, some landslides with small scale but prominent deformation characteristics and strong integrity compared with the background can also be interpreted (with SAR intensity map, topographic shadow map and optical remote sensing image as ground object reference). The minimum interpretation area can reach several pixels. For example, a highway slope landslide with only 4 pixels is interpreted with reference to the highway along the Nujiang River.
YAO Xin
This data set is a code file set of TCA (triple collision analysis) algorithm, which is used to generate the global daily-scale soil moisture fusion dataset from 2011 to 2018.
XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, XIE Qiuxia, JIA Li , HU Guangcheng
This data is the land cover data at 30m resolution of Southeast Asia in 2015. The data format of the data is NetCDF, and the variable name is "land cover type". The data was obtained by mosaicing and extracting the From-GLC data. Several land cover types, such as snow and ice that do not exist in Southeast Asia were eliminated.The legend were reintegrated to match the new data. The data provide information of 8 land cover types: cropland, forest, grassland, shrub, wetland, water, city and bare land. The overall accuracy of the data is 71% (Gong et al., 2019). The data can provide the land cover information of Southeast Asia for hydrological models and regional climate models.
LIU Junguo
Photosynthetically active radiation (PAR) is fundamental physiological variable driving the process of material and energy exchange, and is indispensable for researches in ecological and agricultural fields. In this study, we produced a 35-year (1984-2018) high-resolution (3 h, 10 km) global grided PAR dataset with an effective physical-based PAR model. The main inputs were cloud optical depth from the latest International Satellite Cloud Climatology Project (ISCCP) H-series cloud products, the routine variables (water vapor, surface pressure and ozone) from the ERA5 reanalysis data, aerosol from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) products and albedo from Moderate Resolution Imaging Spectroradiometer (MODIS) product after 2000 and CLARRA-2 product before 2000. The grided PAR products were evaluated against surface observations measured at seven experimental stations of the SURFace RADiation budget network (SURFRAD), 42 experimental stations of the National Ecological Observatory Network (NEON), and 38 experimental stations of the Chinese Ecosystem Research Network (CERN). The instantaneous PAR was validated at the SURFRAD and NEON, and the mean bias errors (MBEs) and root mean square errors (RMSEs) are 5.6 W m-2 and 44.3 W m-2, and 5.9 W m-2 and 45.5 W m-2, respectively, and correlation coefficients (R) are both 0.94 at 10 km scale. When averaged to 30 km, the errors were obviously reduced with RMSEs decreasing to 36.3 W m-2 and 36.3 W m-2 and R both increasing to 0.96. The daily PAR was validated at the SURFRAD, NEON and CERN, and the RMSEs were 13.2 W m-2, 13.1 W m-2 and 19.6 W m-2, respectively at 10 km scale. The RMSEs were slightly reduced to 11.2 W m-2, 11.6 W m-2, and 18.6 W m-2 when upscaled to 30 km. Comparison with the other well-known global satellite-based PAR product of the Earth's Radiant Energy System (CERES) reveals that our PAR product was a more accurate dataset with higher resolution than the CRERS. Our grided PAR dataset would contribute to the ecological simulation and food yield assessment in the future.
TANG Wenjun
This data includes the land cover data of Central Asia, South Asia and Indochina Peninsula in the from 1992 to 2020 with a spatial resolution of 300mLand cover data includes 10 primary categories, which are combined from the secondary categories of the original data. The data source is the surface coverage product CCI-LC of ESA, where the spatial distribution of cropland, built-up land, and water for the land cover data from 1992 to 2020. Combined with the Tsinghua university global land cover data (FROM GLC, 30 m grid), NASA MODIS global land cover data (MCD12Q1, 500 m grid), the United States Geological Survey (USGS global land data (GFSAD30, 30 m), Japanese global forest data (PALSAR/PALSAR - 2, 25 m), the training sample dataset of land cover interpretation were built from the consistent areas of multiple products. The Google Earth Engine and random forest algorithm were used to correct the cropland, built-up land, and water of temporal CCI-LC data. Using the high resolution images in Google Earth at 2019 and 2020, the accuracy of change areas of cropland, built-up land, and water was validated by the stratified random sampling. A total of 3,600 land parcels were selected from 1,200 land parcels of the three land cover types, indicating that the accuracy of our corrected product increased in the range of 11% to 26% for the change areas compared to the CCI-LC product.
XU Erqi
This dataset includes the Antarctica ice sheet mass balance estimated from satellite gravimetry data, April 2002 to December 2019. The satellite measured gravity data mainly come from the joint NASA/DLR mission, Gravity Recovery And Climate Exepriment (GRACE, April 2002 to June 2017), and its successor, GRACE-FO (June 2018 till present). Considering the ~1-year data gap between GRACE and GRACE-FO, we extra include gravity data estimated from GPS tracking data of ESA's Swarm 3-satellite constellation. The GRACE data used in this study are weighted mean of CSR, GFZ, JPL and OSU produced solutions. The post-processing includes: replacing GRACE degree-1, C20 and C30 spherical harmonic coefficients with SLR estimates, destriping filtering, 300-km Gaussian smoothing, GIA correction using ICE6-G_D (VM5a) model, leakage reduction using forward modeling method and ellipsoidal correction.
C.K. Shum
This dataset is blended by two other sets of data, snow cover dataset based on optical instrument remote sensing with 1km spatial resolution on the Qinghai-Tibet Plateau (1989-2018) produced by National Satellite Meteorological Center, and near-real-time SSM/I-SSMIS 25km EASE-grid daily global ice concentration and snow extent (NISE, 1995-2018) provided by National Snow and Ice Data Center (NSIDC, U.S.A). It covers the time from 1995 to 2018 (two periods, from January to April and from October to December) and the region of Qinghai-Tibet Plateau (17°N-41°N, 65°E-106°E) with daily product, which takes equal latitude and longitude projection with 0.01°×0.01° spatial resolution, and characterizes whether the ground is covered by snow. The input data sources include daily snow cover products generated by NOAA/AVHRR, MetOp/AVHRR, and alternative to AVHRR taken from TERRA/MODIS corresponding observation, and snow extent information of NISE derived from observation by SSM/I or SSMIS of DMSP satellites. The processing method of data collection is as following: first, taking 1km snow cover product from optical instruments as initial value, and fully trusting its snow and clear sky without snow information; then, under the aid of sea-land template with relatively high resolution, replacing the pixels or grids where is cloud coverage, no decision, or lack of satellite observation, by NISE's effective terrestrial identification results. For some water and land boundaries, there still may be a small amount of cloud coverage or no observation data area that can’t be replaced due to the low spatial resolution of NISE product. Blended daily snow cover product achieves about 91% average coincidence rate of snow and non-snow identification compared to ground-based snow depth observation in years. The dataset is stored in the standard HDF4 files each having two SDSs of snow cover and quality code with the dimensions of 4100-column and 2400-line. Complete attribute descriptions is written in them.
ZHENG Zhaojun, CAO Guangzhen
Snow cover dataset is produced by snow and cloud identification method based on optical instrument observation data, covering the time from 1989 to 2018 (two periods, from January to April and from October to December) and the region of Qinghai-Tibet Plateau (17°N-41°N, 65°E-106°E) with daily product, which takes equal latitude and longitude projection with 0.01°×0.01° spatial resolution, and characterizes whether the ground under clear sky or transparent thin cloud is covered by snow. The input data sources include AVHRR L1 data of NOAA and MetOp serials of satellites, and L1 data corresponding to AVHRR channels taken from TERRA/MODIS. Decision Tree algorithm (DT) with dynamic thresholds is employed independent of cloud mask and its cloud detection emphasizes on reserving snow, particularly under transparency cirrus. It considers a variety of methods for different situations, such as ice-cloud over the water-cloud, snow in forest and sand, thin snow or melting snow, etc. Besides those, setting dynamic threshold based on land-surface type, DEM and season variation, deleting false snow in low latitude forest covered by heavy aerosol or soot, referring to maximum monthly snowlines and minimum snow surface brightness temperature, and optimizing discrimination program, these techniques all contribute to DT. DT discriminates most snow and cloud under normal circumstances, but underestimates snow on the Qinghai-Tibet Plateau in October. Daily product achieves about 95% average coincidence rate of snow and non-snow identification compared to ground-based snow depth observation in years. The dataset is stored in the standard HDF4 files each having two SDSs of snow cover and quality code with the dimensions of 4100-column and 2400-line. Complete attribute descriptions is written in them.
ZHENG Zhaojun, CHU Duo
Contact Support
Northwest Institute of Eco-Environment and Resources, CAS
0931-4967287
poles@itpcas.ac.cn
Links
National Tibetan Plateau Data CenterFollow Us
A Big Earth Data Platform for Three Poles © 2018-2020 No.05000491 | All Rights Reserved
｜ 
No.11010502040845
Tech Support: westdc.cn
