Recurrent Snowmelt Pattern Synthesis Using Principal Component Analysis of Multiyear Remotely Sensed Snow Cover

Snow acts as a vital source of water especially in areas where streamflow relies on snowmelt. The spatiotemporal pattern of snow cover has tremendous value for snowmelt modeling. Instantaneous snow extent can be observed by remote sensing. Cloud cover often interferes. Many complex methods exist to resolve this but often have requirements which delay the availability of the data and prohibit its use for real-time modeling. In this research, we propose a new method for spatially modeling snow cover throughout the melting season. The method ingests multiple years of MODerate Resolution Imaging Spectroradiometer snow cover data and combines it using principal component analysis to produce a spatial melt pattern model. Development and application of this model relies on the interannual recurrence of the seasonal melting pattern. This recurrence has long been accepted as fact but to our knowledge has not been utilized in remote sensing of snow. We develop and test the model in a large watershed in Wyoming using 17 years of remotely sensed snow cover images. When applied to images from 2 years that were not used in its development, the model represents snow-covered area with accuracy of 84.9-97.5% at varied snow-covered areas. The model also effectively removes cloud cover if any portion of the interface between land and snow is visible in a cloudy image. This new principal component analysis method for modeling the interannually recurring spatial melt pattern exclusively from remotely sensed images possesses its own intrinsic merit, in addition to those associated with its applications. Plain Language Summary Mountain snow provides an important source of water. The ability to model snowmelt and the resulting streamflow helps predict the amount and timing of when water will be available for irrigation, drinking water, and other uses. Satellite remote sensing can produce maps of snow-covered area that can improve our ability to model snowmelt, but during the melt season, clouds often block the view of watersheds from space. However, snowmelt follows a repeatable spatial pattern as it melts, year after year. We used this observation to develop a model of the spatial pattern of melt using multiple years of satellite snow cover images. This model can remove the cloud interference from daily satellite snow cover images. Our model achieved 85-98% accuracy in representing the spatial pattern of snow observed in satellite images even when starting with 95% cloud cover. Other models achieve similar accuracy but require much more data to accomplish this, which prevents them from being used for cloud removal in real time. We expect the model to improve our ability to model streamflow from snowmelt runoff.