Abstract
Snow is a vital component of available freshwater world-wide. Remote sensing of snow offers daily world-wide imagery at moderate resolutions; however, the data availability is reduced due to cloud coverage. The spatial and temporal availability of remote sensing allows us to extract the recurrent pattern of snowmelt. Patterns are exceptionally useful for snowmelt runoff modeling. In this work, we expand, apply, and determine sensitivities of these patterns to better understand and utilize them. The recurrent spatial pattern of snowmelt can be extracted with multiple years of remotely sensed data with a Principal Component Analysis (PCA), which we term the PCA Model. The qualities of the PCA model allow us to represent snowmelt linearly in time over 16 years (r-squared 0.946 - 0.992). With a linear and repeatable relationship, we can normalize snowmelt and compare of multiple years and adjust snow depletion for secondary snowfall events. We also use another multivariate technique to extract the recurrent pattern of snowmelt. The K-means Cluster method produces a model of the recurrent pattern of melt which can be used to remove cloud cover with spatial accuracies of 84.6 – 96.0% when compared against independent data. Sensitivities of the PCA model to user decisions were investigated. By varying the user decision of what is considered snow or no snow we increased average spatial accuracies of the PCA model versus cloud free (<10% cloud cover) images from the previously published 92.61% to 95.46%. Spatial and temporal scaling sensitivities of the model were also presented. The model performs with excellent average spatial accuracy (>90%) when compared against four years of independent data for large Hydrologic Unit Code (HUC) 6 boundary sizes. An analysis of the number of years required to develop a robust pattern demonstrated with only three years the pattern captured by the model is exceptionally similar with minimum correlations greater than 0.97. The intrinsic attributes of PCA model are a significant step forward in understanding and spatially representing snowmelt.