A new spatiotemporal NDSI method to maximize snow cover mapping accuracy over Switzerland
Using Landsat-8 Normalized Difference Snow Index (NDSI) snow cover datasets from the Swiss Data Cube, researchers from the University of Geneva have investigated the relation snow-NDSI with different environmental variables (i.e., elevation, land cover type and seasons) and evaluate the accuracy of the common NDSI threshold of 0.4 against in-situ snow depth measurements over Switzerland for the period 2014-2020.
It was found that there is a consistent relationship between NDSI values and land cover type, elevation, seasons, and snow depth measurements. The global NDSI threshold of 0.4 may not be always optimal for the Swiss territory and tends to underestimate the snow cover extent. They therefore propose a new spatiotemporal NDSI method to maximize snow cover mapping accuracy by using a generalized linear mixed model (GLMM). This model uses three environmental variables (i.e., elevation, land cover type and seasons) and raw NDSI values and improves snow cover mapping accuracy by 24% compared to the fixed threshold of 0.4. By using this method omissions errors decrease considerably while keeping a very low value of commission errors. This method will then be integrated in the Snow Observation from Space (SOfS) algorithm used for snow detection in Switzerland.
To illustrate the model’s findings and the spatial comparison with the global reference NDSI threshold, the model has been applied in a small test site in Valais during the winter period (14.02.2019). In this study area, the model maps a snow cover of approximately 164 km2. The snow is mainly found at mid- to high-altitude with some snow-free areas in forested regions (Figure 1 bottom left panel). For the method considering a NDSI threshold of 0.4 (Figure 1 bottom right panel) the snow covers an area of about 134 km2. The disagreements (red area in the bottom right panel Figure 1) illustrating the difference between the snow cover area from the GLMM and a NDSI threshold of 0.4 mainly take place in the transition areas from snow to land because of the patchy snow conditions and in forest areas. Some inaccurate classification of water in snow were also held at low altitude. Indeed, 46% of misclassified pixels are located at altitudes between 1000 to 1500 m a.s.l. and 70% of misclassified pixels are in forested regions. The case study shows that the use of the model increases the snow cover area of 22% compared to the binary NDSI threshold of 0.4 methodology.
This new approach allows snow cover detection at low NDSI values while avoiding false snow detection. Moreover, the model permits to consider the field complexity in the snow-NDSI relationship and can be adapted to other regions possessing physiographic characteristics information. This work sheds light on the benefits and limitations of using Landsat snow cover datasets and will contribute to the understanding of the distribution pattern of snow at regional and local scale.
For more information, the whole article is available here: https://doi.org/10.1016/j.srs.2023.100078