Fractional Snow Cover Mapping through Polytopic Vector Analysis of MODIS Spectral Reflectance

Abstract

Snow cover plays an important role in the Earth’s climate systems. Accurately estimating snow cover is beneficial for predicting the runoff from snowmelt. Fractional Snow Cover (FSC) mapping computes the fraction of snow within a pixel of a remote sensing imager and provides a more precise snow cover extent estimate compared to binary comparing to binary identification of a pixel as snow or not. Linear mixture analysis has been commonly adopted to map FSC and multiple algorithms have been developed using this method. Polytopic Vector Analysis (PVA) is performed as an alternative to linear mixture analysis. PVA has some inherent advantageous over the standard linear unmixing method, which include that the generic PVA approach guarantees each endmember fraction falls within a physically realistic range (0 to 1 or 0 to 100%) and PVA automatically selects endmembers in an objective manner.

This study investigates the feasibility that applying PVA in mapping FSC. The PVA algorithm was developed in python, and was tested by using MODIS atmospherically-corrected spectral reflectance to determine snow-cover fraction. Reference fractional snow cover maps created from 30m-resolution Landsat images were used to assess the proposed method’s performance. The PVA method showed a R^2 of 0.63 and RMSE of 0.12. This result is comparable with the MOD10 binary FSC product, but not as good as the MOD10 FSC product and Artificial Neural Network. However, as a parsimonious approach, PVA showed its potential for FSC mapping.