Linear mixed modelling of snow distribution in the central Yukon
Abstract Remote sensing estimates of snow water equivalent (SWE) in mountainous areas are subject to large uncertainties. As a prerequisite for testing passive microwave algorithm estimations of SWE, this study aims to collect snow depth (SD) data and provide an understanding of its complex spatial...
| Published in: | Hydrological Processes |
|---|---|
| Main Authors: | , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
Wiley
2011
|
| Subjects: | |
| Online Access: | http://dx.doi.org/10.1002/hyp.8168 https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fhyp.8168 https://onlinelibrary.wiley.com/doi/pdf/10.1002/hyp.8168 |
| Summary: | Abstract Remote sensing estimates of snow water equivalent (SWE) in mountainous areas are subject to large uncertainties. As a prerequisite for testing passive microwave algorithm estimations of SWE, this study aims to collect snow depth (SD) data and provide an understanding of its complex spatial structure as part of the Canadian International Polar Year observations theme. Snow accumulation, redistribution and ablation are controlled by processes that depend on a variety of topographic factors as well as land surface characteristics, which leads us to modelling SD as a function of proxy variables derived from digital elevation model and Landsat data. Field measurements were performed at 3924 locations compromising 184 sites in 50 transects over 2 years. These measurements were used to predict SD over the study area using a spatial linear mixed‐effects model, a model type capable of handling the hierarchical structure of the field data. The model, built using stepwise variable selection, uses as predictor variables transformed elevation, slope, the logarithm of slope, potential incoming solar radiation and its transform; the normalized difference vegetation index, and a transformed tasseled cap brightness from Landsat imagery. A second, simpler model links SD with density giving SWE. The cross‐validated root mean squared error of the SD distribution model was 14 cm around an overall mean of 80 cm over a domain of 250 × 250 km. This instantaneous end‐of‐season peak‐accumulation snow map will enable the validation of satellite remote sensing over a generally inaccessible area. Copyright © 2011 John Wiley & Sons, Ltd. |
|---|