Abstract
A comprehensive and hemispheric-scale snow cover and snow depth analysis is a prerequisite for all related processes and interactions investigation on regional and global surface energy and water balance, weather and climate, hydrological processes, and water resources. However, such studies were limited by the lack of data products and/or valid snow retrieval algorithms. The overall objective of this study is to investigate the variation characteristics of snow depth across the Northern Hemisphere from 1992 to 2016. We developed long-term Northern Hemisphere daily snow depth (NHSnow) datasets from passive microwave remote sensing data using the support vector regression (SVR) snow depth retrieval algorithm. NHSnow is evaluated, along with GlobSnow and ERA-Interim/Land, for its accuracy across the Northern Hemisphere against meteorological station snow depth measurements. The results show that NHSnow performs comparably well with a relatively high accuracy for snow depth with a bias of −0.6 cm, mean absolute error of 16 cm, and root mean square error of 20 cm when benchmarked against the station snow depth measurements. The analysis results show that annual average snow depth decreased by 0.06 cm per year from 1992 to 2016. In the three seasons (autumn, winter, and spring), the areas with a significant decreasing trend of seasonal maximum snow depth are larger than those with a significant increasing trend. Additionally, snow cover days decreased at the rate of 0.99 day per year during 1992–2016. This study presents that the variation trends of snow cover days are, in part, not consistent with the variation trends of the annual average snow depth, of which approximately 20% of the snow cover areas show the completely opposite variation trends for these two indexes over the study period. This study provides a new perspective in snow depth variation analysis, and shows that rapid changes in snow depth have been occurring since the beginning of the 21st century, accompanied by dramatic climate warming.
Highlights
Seasonal snow cover is an important component of the climate system and global water cycle, and has significant impacts on the surface energy, hydrological processes and water resources, heat exchange between the ground surface and the atmosphere, and the ecosystem as a whole [1,2,3,4].On account of the high albedo and low thermal conductivity properties, snow cover may directly modulate the land surface energy balance [5], influence the soil thermal regime [2,6], and significantly impact on the atmospheric circulation [7,8]
This study attempts to address the following questions: (1) How consistent are Northern Hemisphere daily snow depth (NHSnow) and other sources snow cover datasets with in-situ Snow depth (SD) measurements? (2) What is the spatiotemporal variability of SD in the Northern Hemisphere from 1992–2016? another purpose of this paper is to provide a comprehensive description of two snow characteristics (SD and snow cover days) from 1992 to 2016
Step 5. we chose support vector regression (SVR) as the retrieval function (Equation (1)), with specific kernel functions and parameters [24]; the optimization of parameters for the selected method were performed in the training stage to improve the model performance in this study
Summary
Seasonal snow cover is an important component of the climate system and global water cycle, and has significant impacts on the surface energy, hydrological processes and water resources, heat exchange between the ground surface and the atmosphere, and the ecosystem as a whole [1,2,3,4]. Microwave remote sensing data can be used to estimate the SD and SWE by providing dual-polarization information at different frequency channels [15,16,17,18]. Some limitations for these retrieval algorithms that use only passive microwave brightness temperature remain, due to the diversity of the land cover types and the spatiotemporal heterogeneity of the snow properties. A previous study [24] developed an SD retrieval algorithm with a machine learning method using several source datasets over the Eurasia area, showing this algorithm to have great advantages in snow depth estimation when compared with the other four common SD retrieval methods (e.g., Chang algorithm, spectral polarization difference algorithm, artificial neural networks algorithm, and common linear regression algorithm).
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
