Procedures that may be used to evaluate the operational performance of a wide spectrum of geophysical models are introduced. Primarily using a complementary set of difference measures, both model accuracy and precision can be meaningfully estimated, regardless of whether the model predictions are manifested as scalars, directions, or vectors. It is additionally suggested that the reliability of the accuracy and precision measures can be determined from bootstrap estimates of confidence and significance. Recommended procedures are illustrated with a comparative evaluation of two models that estimate wind velocity over the South Atlantic Bight.

We describe and validate a model that retrieves fractional snow-covered area and the grain size and albedo of that snow from surface reflectance data (product MOD09GA) acquired by NASA's Moderate Resolution Imaging Spectroradiometer (MODIS). The model analyzes the MODIS visible, near infrared, and shortwave infrared bands with multiple endmember spectral mixtures from a library of snow, vegetation, rock, and soil. We derive snow spectral endmembers of varying grain size from a radiative transfer model specific to a scene's illumination geometry; spectra for vegetation, rock, and soil were collected in the field and laboratory. We validate the model with fractional snow cover estimates from Landsat Thematic Mapper data, at 30 m resolution, for the Sierra Nevada, Rocky Mountains, high plains of Colorado, and Himalaya. Grain size measurements are validated with field measurements during the Cold Land Processes Experiment, and albedo retrievals are validated with in situ measurements in the San Juan Mountains of Colorado. The pixelweighted average RMS error for snow-covered area across 31 scenes is 5%, ranging from 1% to 13%. The mean absolute error for grain size was 51 µm and the mean absolute error for albedo was 4.2%. Fractional snow cover errors are relatively insensitive to solar zenith angle. Because MODSCAG is a physically based algorithm that accounts for the spatial and temporal variation in surface reflectances of snow and other surfaces, it is capable of global snow cover mapping in its more computationally efficient, operational mode.

No abstract

Abstract:Our understanding of snow distribution in the mountains is limited as a result of the complex controls leading to extreme spatial variability. More accurate representations of snow distribution are greatly needed for improvements to hydrological forecasts, climate models, and for the future testing and validation of remote-sensing retrieval algorithms. In this study, the relative performances of four spatial interpolation methods were evaluated to estimate snow water equivalent for three 1 km 2 study sites in the Colorado Rocky Mountains. Each study site is representative of different topographic and vegetative characteristics. From 1 to 11 April 2001, 550 snow depth measurements and approximately 16 snow density profiles were obtained within each study site. The analytical methods used to estimate snow depth over the 1 km 2 areas were (1) inverse distance weighting, (2) ordinary kriging, (3) modified residual kriging and cokriging, and (4) a combined method using binary regression trees and geostatistical methods. The independent variables used were elevation, slope, aspect, net solar radiation, and vegetation. Using cross-validation procedures, each method was assessed for accuracy. The tree-based models provided the most accurate estimates for all study sites, explaining 18-30% of the observed variability in snow depth. Kriging of the regression tree residuals did not substantially improve the models. Cokriging of the residuals resulted in a less accurate model when compared with the tree-based models alone. Binary regression trees may have generated the most accurate estimates out of all methods evaluated; however, substantial portions of the variability in observed snow depth were left unexplained by the models. Though the data may have simply lacked spatial structure, it is recommended that the characteristics of the study sites, sampling strategy, and independent variables be explored further to evaluate the causes for the relatively poor model results.