The Effect of Grain Size on Spectral Mixture Analysis of Snow-Covered Area from AVIRIS Data

We developed a technique to improve spectral mixture Results were verified with a high spatial resolution aerial analysis of snow-covered area in alpine regions through photograph demonstrating equivalent accuracy. Analysis the use of multiple snow endmembers. Snow reflectance of fraction under/overflow and residuals confirmed mixin near-infrared wavelengths is sensitive to snow grain ture analysis sensitivity to grain size gradients. Elsesize while in visible wavelengths it is relatively insensivier Science Inc., 1998 tive. Snow-covered alpine regions often exhibit large surface grain size gradients due to changes in aspect and elevation. The sensitivity of snow spectral reflectance to INTRODUCTION grain size translates these grain size gradients into specSnow covers over 30% of the Earth’s land surface seatral gradients. To spectrally characterize a snow-covered sonally and through its melt is the dominant source of image domain with mixture analysis, the variable specfresh water to many regions. An example is western tral nature of snow must be accounted for by use of North America, which receives as much as 75% of its multiple snow endmembers of varying grain size. We perannual fresh water from melt of high mountain snow formed numerical simulations to demonstrate the sensistorage (Steppuhn, 1981). Because snow has the highest tivity of mixture analysis to grain size for a range of sizes albedo of any natural and spatially extensive surface, it and snow fractions. From Airborne Visible/Infrared Implays an important role in the Earth’s radiation balance. aging Spectrometer (AVIRIS) data collected over MamThe areal extent of snow cover is likely to be a sensitive moth Mountain, CA on 5 April 1994, a suite of snow imindicator of climate change (Gleick, 1987). A warmer cliage endmembers spanning the imaged region’s grain size mate may force either a higher snow line and earlier onrange were extracted. Mixture models with fixed vegetaset of melt or increased convected moisture leading to tion, rock, and shade were applied with each snow endmore extensive snow cover in alpine regions (Williams et member. For each pixel, the snow fraction estimated by al., 1996). Thus, it is necessary to monitor snow-covered the model with least mixing error (RMS) was chosen to area and other snow properties at ranges of temporal and produce an optimal map of subpixel snow-covered area. spatial scales. Spatial distributions of snow-covered area (SCA) are * Department of Geography, University of California, Santa crucial inputs to models of hydrology and climate in alBarbara pine and other seasonally snow-covered areas (Dozier, † Institute for Computational Earth System Science, University of California, Santa Barbara 1989). SCA is necessary to parameterize energy balance ‡ Jet Propulsion Laboratory, California Institute of Technology, calculations in mesoscale and general circulation models Pasadena (Marshall and Oglesby, 1994), to initialize and validate § Donald Bren School of Environmental Science and Management, University of California, Santa Barbara distributed snowmelt modeling efforts (Harrington et al., Address correspondence to Thomas H. Painter, Inst. for Compu1995), and to estimate snow water equivalent from obtational Earth System Science, 6th Floor Ellison, Univ. of California, servations of snow-cover depletion (Martinec and Rango, Santa Barbara, CA 93106. E-mail: [email protected] Received 1 July 1996; revised 10 December 1997. 1981). Remote sensing is an effective and necessary