Zone II • Versions EN2
Abstract: Snow cover over Tibetan Plateau plays an important role in regional water and energy circulation. Snow ablation also affects downstream rivers. Snow parameters and their long-term changes are sensitive factors affecting and responding to regional climate, influencing ecology and disasters. Moderate-resolution imaging spectrometer (MODIS) is widely used for remotely sensing snow due to its high spatio- temporal resolution. However, snow over Tibetan Plateau is distributed patchily and changes rapidly with unexpected atmospheric convection and precipitation. Also, because optical remote sensing is influenced severely by clouds, daily snow cover monitoring is a challenge requiring to remove cloud cover instances. Engaged in Tibetan Plateau’s terrain complexity and snow spatio-temporal characteristics, this paper presents a compound method by combining different cloud removal algorithms, giving a MODIS daily cloud-free snow cover algorithm for Tibetan Plateau, as well as MODIS daily cloud-free snow cover products (2002 – 2015). The accuracy of the snow cover products is then verified against experimental data observed from 145 ground stations during two winter periods from October 1, 2009 to April 30, 2011. Results show that, when snow depth exceeds 3 cm, the general classification accuracy is 96.6% and the snow classification accuracy is 89.0%. Accuracy was well controlled in each step, which provided a good algorithm for removing clouds from the MODIS snow cover imagery. A multi-language operational process was developed and the daily, cloud-free climatological snow cover products over Tibetan Plateau were released as a free utility online.
Keywords: daily snow cover products; cloud removal algorithm; cloud free; Tibetan Plateau; MODIS
|Chinese title||青藏高原 MODIS 逐日无云积雪面积数据集|
|English title||MODIS daily cloud-free snow cover products over Tibetan Plateau|
|Corresponding author||Qiu Yubao (email@example.com)|
|Data authors||Qiu Yubao, Guo Huadong, Chu Duo, Zhang Huan, Shi Jiancheng, Shi Lijuan, Zheng Zhaojun, Laba Zhuoma|
|Time range||July 2002 – June 2018|
|Geographical scope||The study spans an area of 25° – 45°N and 67° – 107°E, including the entire Tibet Autonomous Region and Qinghai Province, parts of Sichuan, Yunnan, Xinjiang and Gansu provinces, as well as parts of foreign territories in southern and western Tibetan Plateau|
|Spatial resolution||500-meter||Data volume||15.99 GB|
|Data service system||<h ttp://www.sciencedb.cn/dataSet/handle/55>|
|Sources of funding||Special Fund for Meteorological Scientific Research in the Public Interest “Constructing a Remote Sensing Product Dataset for Snow Pack over Tibetan Plateau” (No. GYHY201206040); State Key Program of National Natural Science Foundation of China (ABCC Grant No. 41120114001); the National Natural Foundation of China (No. 41371351); Thirteenth Five-Year Plan of the Chinese Academy of Sciences|
|Dataset composition||The dataset consists of three parts of data: the “MODIS daily cloud-free snow cover products over Tibetan Plateau from 2002 to 2015”, the updated “MODIS daily cloud-free snow cover products over Tibetan Plateau from 2002 to 2018” (produced based on the MODIS Snow Products Collection 6), and the vector data of the research area. They are, respectively:|
(1) MODIS_Dysno_Cloud-free_2002-2015.zip (6.9 GB), i.e., the daily cloud-free snow data;
(2) MODIS_Dysno_Cloudfree_C6_2002-2018.zip (9.09 GB), i.e., MODIS daily cloud-free snow cover products over Tibetan Plateau produced based on MODIS Snow Products Collection 6.
(3) Tibet_Range.zip (24 KB), i.e., the auxiliary vector data storing the boundary of the study area over Tibetan Plateau.
As an important part of the cryosphere, snow is one of the most active natural elements on the Earth’s surface, profoundly impacting regional and even global climate change, energy balance, and water cycle1. For Tibetan Plateau – the highest plateau in the world which is known as the “roof of the world” and “the third pole”, snow is an important feature of the land surface. With strong reflectivity and poor thermal conductivity, snow changes the radiation balance of the ground surface and the heat condition of the atmosphere by absorbing a lot of heat when melting, which causes changes in atmospheric circulation, thus impacting regional climate2. Hydrological effects of snowmelt also have profound impacts on surface runoff and atmospheric thermal conditions3. Snow’s impact on the Asian monsoon and the droughts and floods in South and East Asia has been a focus of domestic and international climatologists for more than a century4 – 5.
Satellite remote sensing technology has gradually become the main approach for snow monitoring because of its short revisit period and large coverage area. Frequently used snow remote sensing products include Landsat, SPOT6,7, AVHRR8, VEGETATION9, MODIS snow products10, SMMR SSM/I11,12, AMSR-E13 – 15 and other microwave snow products, wherein MODIS daily snow data (MOD10A1 and MYD10A1) are most widely used due to its high spatio-temporal resolution.
However, since optical remote sensing products are highly influenced by weather conditions, clouds restrict the acquisition of daily cloud-free snow products. This study combines the cloud removal algorithms for MODIS standard snow products from scholars in China and the international community16 – 25. Based on the snow- cover and topography characteristics of Tibetan Plateau, this paper presents a snow identification methodology according to the geographical and climatological findings in the region. Cloud pixels were reclassified as snow or land through eight steps: daily combination of MODIS morning and afternoon observations, adjacent three-day temporal synthesis, “permanent” snow and land recognition, neighboring four-pixel method, elevation filtering, correction of misclassified shaded areas, maximum snow and land masking, and finally, snow lines fitting. By the seventh step, the cloud area has fallen to an average of about 5.1% with a minimum loss in the snow classification precision. By the final step – expected snow lines fitting, all clouds have been removed and the daily cloud-free MODIS snow products can be obtained. The products provide a reference for dynamically monitoring the Plateau’s snow at daily scale.
2.1 Original data products
(1) Original MODIS snow cover products MOD10A1 and MYD10A1, downloaded from the official website of the US National Snow and Ice Data Center (http://nsidc.org) which span from July 2002 to April 2015. The track numbers of the raw data are shown in Figure 1.
(2) Digital Elevation Model (DEM) data, downloaded from the website of the Geospatial Data Cloud (http://www.gscloud.cn). There are 54 scenes in total.
2.2 Data processing
The characteristics of Tibetan Plateau – its large area, complex and diverse terrain, and heterogeneity of snow distribution – determine the regional applicability of cloud-removal methods. In the last step, Tibetan Plateau is zoned into seven explicit sub-regions, according to the slope complexities and differences, namely, Indo-Gangetic Plain, North of Tibetan Plateau, Hinterland of Tibetan Plateau, Mountainous area in eastern Tibetan Plateau, Pamirs, Tianshan Mountains, and Himalayas, as illustrated in Figure 2.
There are eight steps for reclassifying clouds from MODIS snow products of Tibetan Plateau; the cloud is gradually reduced to obtain daily cloud-free products. Figure 3 shows the flow chart of the algorithm.
By combining the MODIS Reprojection Tool (MRT) open source software (downloadable at: <http://nsidc.org/data/modis/tools.html>) with customized algorithms, the MOD10A1 and MYD10A1 images were mosaicked and clipped through reprojection. ENVI 5.1 software was used for the mosaicking and clipping of DEM data, yielding MOD10A1, MYD10A1, and DEM data in the WGS84 geographic coordinate system.
Steps for cloud removal
Steps for cloud removal are described in Table 1.
|1) Daily combination of MODIS morning and afternoon observations17||We assume that the snow-cover condition at the same place does not change on the same day. The EOS-AM Terra snow data (MOD10A1) and EOS-PM Aqua Data (MYD10A1) are maximally synthesized in a sequence of priority: snow > ice lake > lake > land.|
|2) Adjacent three-day temporal synthesis24||On cloudy days, snow receives less solar radiation and therefore stands longer after landing on the ground. Considering temporal continuity, there is a higher possibility that the day will be snowy if it snows the day before and the day after, which also applies to land24.|
① If a pixel of the same place is presented as snow the day before and the day after, then the cloud pixel of the day is reclassified as snow.
② If a pixel of the same place is presented as land the day before and the day after, then the cloud pixel of the day is reclassified as land.
③ If a pixel of the same place is presented as lake at least one day either before or after, then the cloud pixel of the day is reclassified as lake.
④ If a pixel of the same place is presented as ice-covered lake at least one day either before or after, then the cloud pixel of the day is reclassified as ice-covered lake.
|3) “Permanent” snow and land recognition||Given a season of snow as the hydrological period (the snow season of Tibetan Plateau is from October 1 to April 30 the next year), the period between July 1 of the first year to June 30 of the next year is divided into three parts (hereinafter taking 2003 – 2004 as an example): summer during July 1, 2003 – September 30, 2003; winter during October 1, 2003 – April 30, 2004; summer during May 1, 2004 – June 30, 2004.|
By comparing clearer images of snow cover on Google Earth and a statistical analysis of the distribution based on the DEM, it was found that, for elevation above 5,800 m, the area was completely covered with snow (the area above 5,800 m was included in the permanent snow-covered area); the cloud above the snow was more than that above the land, and because of the possible misjudging of MODIS, it was stipulated that as long as the cloud-covered time plus snow-covered time of a certain pixel exceed 95% of the total synthesis time in the snow season, the cloud can be reclassified as snow. Similarly, if during this period only land and cloud (no snow) appear on a certain pixel, and there are just a few cloud-covered days (e.g., when the proportion is less than a certain threshold value), then the cloud is reclassified as land. A comparison with Google Earth shows that when the number of cloud days is less than 20% of the total days, clouds can be removed to the maximum and the accuracy can be guaranteed with relatively small losses.
For statistical analysis of the “permanent” snow and “permanent” land in each time period, determinant conditions are as follows:
① Pixels with an elevation > 5,800m are “permanent” snow-covered areas.
② For cloud pixels with an elevation between 3,000 m and 5,800 m, if cloud days + snow days > total days * 0.95, the cloud pixels are reclassified as snow.
③ If a pixel satisfies: a. cloud days + land days = total days, or b. cloud days < total days * 0.2, the cloud pixel is reclassified as land.
|4) Neighboring four-pixel method24||① If at least three of the four pixels adjacent to the cloud pixel are snow, then the value of the central cloud pixel is assigned as snow.|
② If at least three of the four pixels adjacent to the cloud pixel are land, then the value of the central cloud pixel is assigned as snow; lake and ice-covered lake are assigned in the same way.
|5) Elevation filtering||When the central pixel is snow, if there is a cloud pixel in the eight pixels most adjacent to it, and the elevation of the cloud pixel is higher than that of the snow, then the value of the cloud pixel will be assigned as snow.|
|6) Correction of misclassified|
|In the Pamirs, part of the shaded area was misclassified by MODIS as lake or ice-covered lake while it was actually land or snow. For this situation, the existing lake boundary was masked and the lake or ice-covered lake within the boundary were retained; the misclassified lake or ice-covered lake beyond the boundary were re-assigned as cloud-covered, and would be determined next step.|
|7) Maximum snow and land masking||The products of eight-day maximum snow cover (similar to the MODIS eight-day snow products) and eight-day maximum land cover, based on the above elevation filtering products, was synthesized. The sequences of priority are: snow > ice lake > lake > land, land > ice lake > lake > snow. The synthetic results obtained are referred to as Image a (maximum snow synthesis, i.e., the minimum land synthesis), and Image b (maximum land synthesis, i.e., the minimum snow synthesis). Snow gradually melts or suddenly descends within eight days, displaying certain laws of gradual spatial change – the daily snow area is greater than the minimum area of eight-day snow, and the land area is greater than the minimum area of eight-day land (Figure 4).|
Algorithm for removing clouds: Snow products from the previous steps are to be reclassified in accordance with the following rules:
① If the pixel is cloud, and the pixel in the maximum snow synthetic
Image a is land, then the value of the cloud pixel is assigned as land.
② If the pixel is cloud, and the pixel in Image a or Image b is ice-covered lake, then the value of the cloud pixel is assigned as ice-covered lake.
③ If the pixel is cloud, and the pixel in Image a or Image b is lake, then the value of the cloud pixel is assigned as lake.
④ If the pixel is cloud, and the pixel in the maximum land synthetic Image b is snow, then the value of the cloud pixel is assigned as snow.
|8) Expected snow lines fitting||Since snow cover is related to longitude, latitude, slope inclination, slope direction and elevation, and high altitudes are more likely to have snow cover than low altitudes, a method of fitting expected snow lines was adopted for the whole region. The algorithm is described as follows:|
① Extracting snow line samples. Snow line refers to the boundary between
snow and land; data concerning the elevation, slope inclination, slope direction, longitude and latitude of the snow cover above the snow line were extracted.
② Fitting the elevation of expected snow lines. The elevation of expected
snow lines was fitted by applying the multiple linear regression method with elevation set as the dependent variable and the other four factors as independent variables.
③ Reclassifying the cloud pixels. Expected snow lines may be calculated
by the slope inclination, slope direction, longitude and latitude. Comparing expected snow lines with the actual cloud elevation, if the actual elevation is higher, the cloud will be assigned as snow, or otherwise, as land.
3.1 Naming format
The file names of the “MODIS daily cloud-free snow products over Tibetan Plateau” dataset conform to the format “MODIS_Dysno_Cloudfree_ YYYYMMDD.tif”. The following information of the dataset can be acquired from the name.
(1) “MODIS_Dysno_Cloudfree” represents MODIS daily cloud-free snow products over Tibetan Plateau;
(2) “YYYY” indicates the year;
(3) “MM” indicates the month;
(4) “DD” indicates the day.
3.2 Classification description
Daily snow depth data – 61,480 pairs of observed values provided by 145 ground stations – were selected as samples for data quality evaluation. The data span 424 time phases in two snow seasons over Tibetan Plateau from October 1, 2009 to April 30, 2011. The confusion matrix and formulas (1) and (2) were adopted to evaluate the snow classification images and analyze the MODIS data (MOD10AI, MYD10A1) under clear weather. The accuracy of all types of synthetic products during cloud processing, including general classification accuracy and snow classification accuracy, was analyzed. For this purpose, the following samples were considered during accuracy evaluation: ① sample (a) with snow images and records from ground stations (snow depth > 0); ② sample (b), recorded snow from ground stations classified as snow-free, namely omission; ③ sample (c), recorded snow-free from ground stations classified as snow; and ④ sample (d) with snow-free images and records from ground stations (Table 3).
General classification accuracy, namely accuracy, reflects the capacity of classification algorithms in identifying snow as snow-covered and land as snow- free across the whole research area. Snow classification accuracy, namely precision, reflects the proportion of real snow pixels among all the snow pixels identified by the classification algorithm. They may be expressed by the following formulas:
Raw data of MOD10A1, MYD10A1 and all processed products in the confusion matrix are indicated in Figure 4. The general classification accuracy of MODIS data was above 98%, while the snow classification accuracy approximated to 82%. Since there were few pixel elements modifying the erroneous judgment of lake and lake ice, statistical results were omitted. The following factors might result in errors during precision validation: thin snow cover, dispersed snow distribution, differentiated spatial scales employed for generating snow depth data at ground stations and MODIS snow image pixels, optics’ penetrating effect in thin snow-covered areas, and spectral mixing effect caused by mottled thin snow. When snow cover is thin (for example, less than 3 cm) during snow-melt periods, snow underestimation might occur, in which circumstance snow is observed within the small area of ground stations but the MODIS image (500 m * 500 m) displays to be snowless. It might also be that the station is located in a city and the snow quickly melts due to the temperature which is higher than its surrounding areas, in which circumstance the MODIS image overestimates the situation and erroneously suggests snow. In the case where snow is thicker than 3 cm, the samples recorded as snow by both images and ground observations are denoted by a3, and the samples recorded as snow by images but as land by ground observations are denoted by b3. A calculation by formulas (1) and (2) demonstrated an improved general accuracy of all synthetic products, with Oa and Sa of MODIS data respectively rising to about 99% and 94%.
Table 4 shows that the cloud removal procedure prior to elevation filtering had less influence on the classification accuracy; daily observation combination and elevation filtering slightly improved the general snow classification precision, whereas maximum snow and land masking brought down the general classification accuracy of synthesis products; in the case of the data with snow- depth being above 3 cm, the accuracy decreased to 98.24% from 98.97% (average accuracy value of MOD10A1 and MYD10A1) while the snow classification accuracy increased from 94.48% to 94.73%, with a larger amount of cloud removed; fitting expected snow lines had the lowest accuracy and was suitable for the last step of cloud removal; nevertheless, when the snow depth was above 3 cm, the general classification accuracy and snow classification accuracy of cloud-free snow products still reached 96.6% and 89% respectively, higher than previously published documentary records24 and basically met the accuracy requirements of MODIS standard snow products under clear weather. All of this indicates that the algorithm for cloud removal may be applied in the dynamic, daily, cloud-free snow monitoring over Tibetan Plateau.
|Accuracy of general|
|Accuracy of snow|
All sta- tions
Snow dep- th ≤ 3 cm removed
All sta- tions
Snowdep- th ≤ 3 cm removed
|Adjacent three- day temporal synthesis|
|“Permanent” snow and land recognition|
|Neighboring four -pixel method|
|7||Elevation fil- tering||4,059||722||495||42,761||3,656||151||97.47||98.63||84.90||96.03|
|Maximum snow and land masking|
|Expected snow lines fitting (cloud- free s now products)|
The release of the MODIS snow cover product C626 enabled us to recalculate the MODIS daily cloud-free snow cover from 2002 to 2018. The new cloud-free snow cover product is stored as a standalone supplement of the previously released data set titled “MODIS daily cloud-free snow cover products over Tibetan Plateau”27, with a data volume of 9.09 GB. Based on “Daily fractional snow cover dataset over High Asia”28, the new product is obtained by using the Normalized Difference Snow Index (NDSI), where the feature properties of 56<NDSI<101 are set as snow. The NDSI threshold is determined based on the High Asia dataset28 and the User Guide26.
The pixel values of this new dataset are shown in Table 5:
There are three major differences between the new updated products and the previous one27: first, the new dataset is produced by using MODIS snow product collection 6 (C6)26 and NDSI, while the previous product is based on MODIS C527. The quantitative image restoration (QIR) algorithm is used to recover the lost MODIS/Aqua band 6 data, so the new dataset is of higher quality and accuracy. Second, the new product is generated by using the method as per “Daily fractional snow cover dataset over High Asia”28, which adopts a certain threshold of NDSI to obtain the new cloud-free snow cover product. Third, the geographical coverage of the new product has been extended from 25°N – 45°N / 67°E – 107°E to 26°N – 46°N / 62°E – 105°E.
The dataset, based on moderate-resolution snow products MOD10A1 and MYD10A1, applies to studies on water and energy circulation, ecology, and disasters on Tibetan Plateau. It is especially of great significance for research on glacier snow model, climate change, run-off change, long-term laws for snow spatio-temporal distribution, ecological benefits, snow disasters, future tendencies and other aspects of the Tibetan Plateau environment.
Bin C, Qiu Y, Shi L et al. Comparative validation of snow depth algorithms using AMSR-E passive microwave data in China. Journal of Glaciology and Geocryology 35 (2013): 801 – 813.
Liu Y. Sea-Land Thermal Differences on the Impact of Climate Change in China. Beijing: Meteorological Press, 2005.
Barnett TP, Dümenil L, Schlese U et al. The effect of Eurasian snow cover on regional and global climate variations. Journal of the Atmospheric Sciences 46 (2010): 661 – 686.
Qiu Y, Guo H, Bin C et al. Comparison on snow depth algorithms over China using AMSR-E passive microwave remote sensing. IEEE International Geoscience and Remote Sensing Symposium (2014): 851 – 854.
Blanford HF. On the connexion of the Himalaya snowfall with dry winds and seasons of drought in India. Royal Society of London Proceedings 37 (1996): 3 – 22.
Rutger D & Steven MDJ. Monitoring snow-cover dynamics in Northern Fennoscandia with SPOT VEGETATION images. International Journal of Remote Sensing 25 (2004): 2933 – 2949．
Hendrix SD. Variation in seed weight and its effects on germination in Pastinaca sativa L. (Umbelliferae). American Journal of Botany 71 (1984): 795 – 802.
Xiao X, Zhang Q, Boles S et al. Mapping snow cover in the pan-Arctic zone, using multi-year (1998 – 2001) images from optical VEGETATION sensor. International Journal of Remote Sensing 25 (2004): 5731 – 5744.
Grody NC & Basist AN. Global identification of snow cover using SSM/ I measurements. IEEE Transactions on Geoscience & Remote Sensing 34 (1996): 237 – 249.
Feng Q, Zhang X & Liang T. Dynamic monitoring of snow cover based on MOD10A1 and AMSR-E in the north of Xinjiang Province, China. Acta Prataculturae Sinica 18 (2009): 125 – 133.
Lu X, Wang X, Cui C et al. Snow depth retrieval based on AMSR-E data in northern Xinjiang Region, China. Journal of Glaciology & Geocryology 35 (2013): 40 – 47.
Dai L, Tao C, Jian W et al. Snow depth and snow water equivalent estimation from AMSR-E data based on a priori snow characteristics in Xinjiang, China. Remote Sensing of Environment 127 (2012): 14 – 29.
Gafurov A & Bárdossy A. Cloud removal methodology from MODIS snow cover product. Hydrology and Earth System Sciences 13 (2009): 1361 – 1373.
Gao Y, Xie H, Yao T et al. Integrated assessment on multi-temporal and multi-sensor combinations for reducing cloud obscuration of MODIS snow cover products of the Pacific Northwest USA. Remote Sensing of Environment 114 (2010): 1662 – 1675.
Paudel KP & Andersen P. Monitoring snow cover variability in an agropastoral area in the Trans Himalayan region of Nepal using MODIS data with improved cloud removal methodology. Remote Sensing of Environment 115 (2001): 1234 – 1246.
Dietz AJ, Kuenzer C & Conrad C. Snow cover variability in Central Asia between 2000 and 2011 derived from improved MODIS daily snow cover products. International Journal of Remote Sensing 34 (2013): 3879 – 3902.
Dietz AJ, Wohner C & Kuenzer C. European snow cover characteristics between 2000 and 2011 derived from improved MODIS daily snow cover products. Remote Sensing 4 (2012): 2432 – 2454.
López-Burgos V, Gupta HV & Clark M. Reducing cloud obscuration of MODIS snow cover area products by combining spatio-temporal techniques with a probability of snow approach. Hydrology and Earth System Sciences 7 (2013): 1809 – 1823.
Gao Y, Xie H, Lu N et al. Toward advanced daily cloud-free snow cover and snow water equivalent products from Terra-Aqua MODIS and Aqua AMSR-E measurements. Journal of Hydrology 385 (2010): 23 – 35.
Thirel G, Salamon P, Burek P et al. Assimilation of MODIS snow cover area data in a distributed hydrological model. Hydrology & Earth System Sciences Discussions 8 (2011): 1329 – 1364.
Huang X, Hao X, Wang W et al. Algorithms for cloud removal in MODIS daily snow products. Journal of Glaciology & Geocryology 34 (2012):1118 – 1126.
Parajka J & Bloschl G. Spatio-temporal combination of MODIS images - potential for snow cover mapping. Water Resources Research 440 (2008): 79 – 88.
George A. Riggs, Dorothy K. Hall．MODIS Snow Products Collection 6 User Guide, availble at: <http://nsidc.org/sites/nsidc.org/files/files/MODIS-snow-user-guide-C6.pdf>.
Qiu Y, Guo H, Chu D et al. MODIS daily cloud-free snow cover products over Tibetan Plateau (2002 – 2015). China Scientific Data 1 (2016), DOI: 10.11922/csdata.170.2016.0003
1. Qiu Y, Guo H, Chu D et al. MODIS daily cloud-free snow cover products over Tibetan Plateau. Science Data Bank. DOI: 10.11922/sciencedb.170.55
How to cite this article
Qiu Y, Guo H, Chu D et al. MODIS daily cloud-free snow cover products over Tibetan Plateau. China Scientific Data 1 (2016), DOI: 10.11922/csdata.170.2016.0003