Production background: The source area of the Yellow River is the main water producing area and water source conservation area of the Yellow River Basin. Snow melt water is one of the important water sources in the source area, and high-precision snow cover area datasets are the foundation for ecological and hydrological simulation, climate change research, and other research in the source area. However, a large amount of cloud coverage in MODIS snow products results in almost half of the information being missing. Due to the seasonal snow cover in the source area of the Yellow River showing shallow snow layers, patchy distribution, and rapid melting, traditional statistical methods are difficult to accurately capture the spatiotemporal characteristics of snow cover in the source area. Advanced deep learning techniques can better explore the spatiotemporal characteristics hidden behind the data
Production method: This dataset utilizes daily MODIS Normalized Snow Index (NDSI) products from 2000 to 2021, and uses the MODIS NDSI cloud pixel reconstruction model based on Partial Convolutional Neural Network (PCNN) developed by Xing et al. (2022) to first generate spatiotemporal continuous MODIS NDSI data products; Secondly, the standard algorithm of NASA's Snow Cover Ratio (FSC) product was used to prepare a seamless daily MODIS FSC remote sensing monitoring dataset for the Yellow River source area from 2000 to 2021
Data content: The element included in the data is FSC, with a spatial coverage of the entire Yellow River source area. The data starts from the 2000-2001 snow season (i.e. November 1, 2000) and ends on April 30, 2021, 2020. It includes 21 complete snow seasons. The spatial resolution is 0.005 degrees (approximately 500m), and the temporal resolution is daily. The naming convention is: YYYYDDD.tif, where YYYY represents the year and DDD represents the Julian day (001-365)
Data advantages, characteristics, and application scope: Based on the verification of snow depth observation data from six surface meteorological stations in the source area of the Yellow River and the "cloud hypothesis", the overall accuracy of the dataset can reach 94%, with overestimation and underestimation of 1%, an average absolute deviation of 10.43%, and an average determination coefficient of 0.86. This indicates that the data has high accuracy and is comparable to MODIS snow products under clear skies, This dataset can provide data support for research on the distribution of snow cover, estimation of snow water reserves, analysis of snow cover changes, and snow disaster risk assessment in the source area of the Yellow River
Using MODIS/Terra and MODIS/Aqua daily snow cover L3 level products (MOD10A1 and MYD10A1, V6). From NASA's Snow and Ice Data Center（ https://nsidc.org/ ）Free download
The production process of the MODIS FSC dataset in the Yellow River source area is mainly divided into five steps: obtaining MODIS snow products, data preprocessing, MODIS NDSI cloud pixel reconstruction (including MODIS/Terra and MODIS/Aqua data synthesis, near three day time filtering, and PCNN based cloud pixel spatiotemporal reconstruction), FSC estimation, and accuracy verification
(1) This is the most direct verification method based on site snow depth observation. Using the snow depth values observed at six meteorological observation stations in the source area of the Yellow River (Maduo, Dari, Henan, Jiuzhi, Ruoergai, and Hongyuan) from January 1, 2000 to March 31, 2020 as "true values", the calculated MODIS FSC was evaluated. Extract the MODIS FSC values of the corresponding pixels at six stations, compare them with the measured SD values at the stations, construct a confusion matrix, and define three evaluation indicators: overall accuracy (OA), overestimated snow cover event (MO), and underestimated snow cover event (MU). The results indicate that due to severe data loss caused by cloud coverage, the overall accuracy of MODIS original products is less than 50%. Both binary synthesis and near time window filtering can improve the overall accuracy of MODIS FSC products by nearly 10%. After reconstructing all cloud pixels using PCNN, the overall accuracy of spatiotemporal continuous MODIS FSC products is significantly improved, reaching 94%. Compared with the original product under clear sky pixel conditions, there is no significant increase in the probability of overestimating and underestimating snow events (both overestimating and underestimating increase by 1%).
(2) Based on cloud mask validation, this is an accurate quantitative evaluation of the MODIS FSC estimation model based on PCNN. Using 31089 independent test blocks, accurately quantitatively evaluate the FSC estimation results, and define the mean absolute deviation (MAE) and determination coefficient (R2) of two quantitative evaluation indicators. The results show that the average MAE of the reconstructed MODIS FSC value is 10.43%, and the average R2 is about 0.86
| # | number | name | type |
| 1 | 21JR7RA068 | Natural Science Foundation of Gansu Province |
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.ncdc.nieer.db2115.2022
10.12072/ncdc.NIEER.db2115.2022