Wet Snow Mapping in Southern Ontario with Sentinel-1A Observations

Abstract

Wet snow is defined as snow with liquid water present in an ice-water mix. It can be an indicator for the onset of the snowmelt period. Knowledge about the extent of wet snow area can be of great importance for the monitoring of seasonal snowmelt runoff with climate-induced changes in snowmelt duration. Moreover, effective monitoring wet snow cover has implications for operational hydrological and ecological applications. Spaceborne microwave remote sensing has been used to observe seasonal snow under all-weather conditions. Active microwave observations of snow at C-band are sensitive to wet snow due to the high dielectric contrast with non-wet snow surfaces. Synthetic aperture radar (SAR) is now openly available to identify and map the wet snow areas globally at relatively fine spatial resolutions (~100m). In this study, a semi-automated workflow was developed from the change detection thresholding method of Nagler et al. (2016) using multi-temporal Sentinel-1A (S1A) dual-polarization observations of Southern Ontario. Regions of Interest (ROIs) were created for agricultural lands to analyze the factors influencing backscatter responses from wet snow. To compare with the thresholding method, logistic regression and Support Vector Machine (SVM) classifications were applied on the datasets. Weather station data and visible-infrared satellite observations were used as ground reference to evaluate the wet snow area estimates. Even though the study merely focused on agricultural land, the results indicated the feasibility of the change detection method with a threshold of -2dB on non-mountainous areas and addressed the usefulness of Sentinel-1A data for wet snow mapping. However, with the capability of identifying non-linear characteristics of the datasets, classification methods tended to be a more accurate method for wet snow mapping. Moreover, this study has suggested using Sentinel-1A data with large incidence for wet snow mapping is feasible.