A Global Analysis of Snow Depth for Numerical Weather Prediction

Abstract

The operational analysis of snow depth at the Canadian Meteorological Centre is described. The analysis makes use of forecasts of precipitation and analyses of screen-level temperature to estimate snowfall and snowmelt for a global domain, and assumes persistence of the mass of the snowpack between melting and/or snowfall events. In addition, wherever snow depth observations are available, these are incorporated using the method of statistical interpolation, performed every 6 h on a ⅓° grid. Correlations between two observations or between observations and grid points are taken as functions of spatial separation in both the horizontal and vertical with an e-folding distance of 120 km in the horizontal and 800 m in the vertical. Observations undergo three separate tests designed to eliminate 1) false reports of snow when none is present, 2) systematically understated snow depths, and 3) reports that violate temporal continuity. Verification of the analysis is presented using three sources of independent data. The analysis obtained when observations are withheld is compared with quality-controlled snow depth reports, with the result that the analysis shows more skill than climatology in all regions and periods examined. The analysis is also compared with Special Sensor Microwave/Imager snow cover, where it again shows more skill than climatology. Finally, a verification over a 134-week period using the NOAA weekly hemispheric snow cover product based on visible imagery is presented. Major differences between analysis and independent data are explained.