Assessing the Climactic and Spatial Sensitivity of the Snowmelt Pattern Derived Using Principal Component Analysis of Multi-Year Remotely Sensed Snow Cover From MODIS

Abstract:

Remote sensing has been used to quantify snow cover and verify snowmelt model output. Cloud cover poses a significant reduction in data availability, and has led to the production of a NASA cloud free dataset. The inter-annually recurrent snowmelt pattern can be extracted into a spatial model. A simple application of this model is cloud removal; however, it has proven to be a robust representation of snow cover against highly variable snow years over a test watershed. For this reason, applications of the model extend beyond cloud removal. The method for model synthesis uses Principal Component Analysis (PCA) applied to multiple years of the MODerate Resolution Imaging Spectroradiameter (MODIS) daily snow cover product. The output is a continuous spatial model. Each pixel value indicates the single pixel’s “relative” melt timing with regard to all other pixels within the watershed. Excellent results for the Upper Snake Basin watershed have been recorded, with accuracies ranging from 85-98% when compared with independent MODIS imagery. We apply the method to three new watersheds with varied climates. Study sites include watersheds within the Cascade Mountains, Bitterroot Mountains, and the Boise Mountains. Independent cloud free MODIS data is used to evaluate the accuracy of the PCA model. We then analyze the spatial sensitivity of the PCA method. The PCA model derived for the entire watershed is compared to a PCA model derived using a sub-basin. The two are directly compared, and is completed for all three watersheds. With defined boundaries, the model can be used with confidence to inform water managers of snow conditions over gauged and ungauged watersheds. Only a single MODIS image is required to identify “when” a watershed is in its melt cycle. As we plan to manage water in a changing climate, there is great value in recurrent spatial data.