Fractional snow cover estimation in complex alpine-forested environments using an artificial neural network

There is an undisputed need to increase accuracy of Fractional Snow Cover (FSC) estimation in regions of complex terrain, especially in areas dependent on winter snow accumulation for a substantial portion of their water supply, such as the western United States. The main aim of this research is to develop FSC estimation in complex alpine-forested environments using an Artificial Neural Network (ANN) methodology as a fusion framework between multi-sensor remotely sensed data at medium temporal/spatial resolution (e.g.16-day revisit time; 30m; Landsat), and high spatial resolutions (e.g.1m; IKONOS). This research is the first known attempt to develop a multi-scale estimator of FSC from surface equivalent reference data derived from IKONOS multispectral data. It is also the first endeavor to estimate FSC values by combining terrain and snow/non-snow reflectance data. The plasticity of the developed ANN Landsat-FSC model accommodates alpine-forest heterogeneity, and renders unbiased, comprehensive, and precise FSC estimates. The accuracy of the ANN Landsat based FSC is characterized by: (1) very low error values (mean error~0.0002; RMSE~0.10; MAE~0.08 FSC), (2) high correlation with the ground equivalent reference datasets derived from 1m resolution IKONOS images (r²~0.9), and (3) robust FSC estimation that is independent of terrain/vegetation alpine heterogeneity. The latter is supported by a spatially uniform distribution of errors, and lack of correlation between terrain (slope, aspect, terrain shadow distribution), Normalized Difference Vegetation Index, and the error (r²=0).