Error Analysis of MODIS Snow Data

Snowmelt runoff forecasting in mountainous areas, like the Western US, use forecasting models for water resource management. However, current seasonal forecasting models are unequipped to manage water in a changing climate and during extreme events. The need for better assessment of the snow covered resources in mountain areas has been made available through new Earth System Data Records using MODIS data. This MODIS data still requires validation and investigation of uncertainties and possible errors, and this project aims to undertake the necessary validation. The spatially and temporally dense MODIS data represents fractional snow cover, fractional snow albedo and snow water equivalent (SWE). This data’s use in snowmelt models and reservoir operations will be significantly advanced by this investigation, which would validate the products, analyze the structure of errors, and advise users of caveats and likely accuracy.

Investigators: Noah Molotch, James Frew of UCSB and Jeff Dozier of UCSB

Fractional Snow-Covered Area

Snow-covered area in mountainous terrain usually varies at a spatial scale finer than that of the ground instantaneous field-of-view of the remote sensing instrument. This spatial heterogeneity poses a “mixed-pixel” problem, because the sensor may measure radiance reflected from snow, rock, soil, and vegetation. To use the snow characteristics in distributed hydrologic models, we must therefore map snow-covered area at subpixel resolution in order to accurately represent its spatial distribution; otherwise, systematic errors may result.

Daily maps of fractional snow cover and albedo have missing values because of cloud cover or sensor noise, and some less reliable values because of the highly off-nadir viewing geometry. We can view the snow data as a sparse time-space cube that needs filtering, smoothing and interpolation. When clouds do not obscure the pixels, viewing angles can perturb the signal. Finally, the algorithms for fractional snow-covered area probably have imperfections, especially at large view angles where the effect of subpixel topographic variability is amplified. The method is to fill in the space-time cube by smoothing the values along the time axis.

Strategy for Validation and Error Analyses

Our validation approach includes covering a time series of wet and dry years, assembling a set of products and their validation datasets, and making the validation data freely available. Therefore, as new versions of algorithms emerge during the period of the investigation, we can immediately run comparisons with the previous version. The post-validation data set will provide comprehensive test cases to future researchers as they explore new ways to analyze MODIS data and integrate them with other sensors and with field measurements.