Long-term series of daily snow depth dataset over the Northern Hemisphere based on machine learning （1980-2019）- A Big Earth Data Platform for Three Poles

Supported by the Strategic Priority Research Program of the Chinese Academy of Science (XDA19070100). Tao Che, the director of this program, who comes from Key Laboratory of Remote Sensing of Gansu Province, Northwest Institute of Eco-Environment and Resources, CAS. They used machine learning methods combined with multi-source gridded snow depth product data to derive a long-time series over the Northern Hemisphere.

Firstly, the applicability of artificial neural network (ANN), support vector machine (SVM) and random forest (RF) method in snow depth fusion are compared. It is found that random forest method shows strong advantages in snow depth data fusion. Secondly, using the random forest method, combined with remote sensing snow depth products such as AMSR-E, AMSR-2, NHSD and GlobSnow and reanalysis data such as ERA-Interim and MERRA-2. These gridded snow depth products and environmental factor variables are used as the input independent variables of the model. In situ observations of China Meteorological Station (945), Russia Meteorological Station (620), Russian snow survey data (514), and global historical meteorological network (41261) are used as reference truth to train and verify the model. The daily gridded snow depth dataset of the snow hydrological year from 1980 to 2019 (September 1 of the previous year to May 31 of the current year) is prepared on the cloud platform provided by the CASEarth. Since the passive microwave brightness temperature data from 1980 to 1987 is the data of every other day, there will be a small number of missing trips in the data during this period. Using the ESM-SnowMIP and independent ground observation data for verification, the quality of the fusion data set has been improved. According to the comparison between the ground observation data and the snow depth products before fusion, the determination coefficient (R2) of the fusion data is increased from 0.23 (GlobSnow snow depth product) to 0.81, and the corresponding root mean square error (RMSE) and mean absolute error (MAE) are also reduced to 7.7 cm and 2.7 cm.

File naming and required software

The daily snow depth data is a matrix with 1440 columns and 360 rows, and the spatial resolution of the data is 0.25 °, NoData_Value is represented by - 9999. The snow depth datasets are stored according to the natural year folder by folder, the folder of each year contains the daily fused gridded snow depth dataset of that year. The data format of fused snow depth dataset is stored in GeoTiff format, the unit of snow depth data is centimeter, and the projection is WGS84. Among these years, the daily fused snow depth dataset from September 1, 1980 to December 31 of that year were contained in folder of 1980, and the folder of 2019 contains the daily fused snow depth dataset from 1 January 1, 2019 to May 31 of that year. Except for these two years, al folders contain the fused snow depth dataset from January 1 to May 31 and September 1 to December 31. Because of the quality of the original gridded snow depth datasets, daily snow depth datasets from June 1 to August 31 were not considered in our study. The file name is "ML_NHSD_YYYY-MM-DD.tif", where ML stands for machine learning, NHSD stands for Northern Hemisphere Snow Depth, YYYY stands for the natural years from 1980 to 2019, MM stands for January to December (except June to August), and DD stands for the date. The file can be viewed with software such as ArcMAP or QGIS and processed accordingly.
The original gridded snow depth datasets of NHSD and GlobSnow were produced with the passive microwave brightness temperature from SMMR sensor in 1980 to 1987. The SMMR dataset was the other day dataset and with large missing bands. Snow depth dataset of NHSD and GlobSnow were also inevitably have a large number of missing data, and the fused snow depth dataset have a similar situation. The fused snow depth is as complete as possible for the land area of the Northern Hemisphere, there are minor missing data on some days, and the coastal areas such as Greenland and Iceland are not within the scope of the study.

Data Citations Data citation guideline What's data citation?

Cite as:

Che, T., Hu, Y., Dai, L., Xiao, L. (2021). Long-term series of daily snow depth dataset over the Northern Hemisphere based on machine learning （1980-2019）. A Big Earth Data Platform for Three Poles, DOI: 10.11888/Snow.tpdc.271701. CSTR: 18406.11.Snow.tpdc.271701. (Download the reference： RIS | Bibtex )

Related Literatures:

1. Hu,Y.X., Che, T., Dai, L.Y., & Xiao, L. (2021). Snow depth fusion based on machine learning methods for the Northern Hemisphere. Remote Sensing, 13,1250.( View Details | Bibtex)

Using this data, the data citation is required to be referenced and the related literatures are suggested to be cited.

Support Program

CASEarth:Big Earth Data for Three Poles（grant No. XDA19070000） (No:XDA19000000)

Strategic Priority Research Program of the Chinese Academy of Sciences (No:XDA19070100) [XDA19070100]

Copyright & License

To respect the intellectual property rights, protect the rights of data authors, expand services of the data center, and evaluate the application potential of data, data users should clearly indicate the source of the data and the author of the data in the research results generated by using the data (including published papers, articles, data products, and unpublished research reports, data products and other results). For re-posting (second or multiple releases) data, the author must also indicate the source of the original data.

License： This work is licensed under an Attribution 4.0 International (CC BY 4.0)