PMSL Anomaly Correlation Verification


This document describes the procedure employed to generate anomaly correlation values for the NCEP Medium Range Forecast Desk (MRFD) and the available model guidance. The first section discusses the data source and computation of climatological parameters. Section 2 deals with the comparison of individual forecasts to the NCEP surface analysis and generation of the verification database.

1. The Climatology

The climatological parameters required for the computation of standardized anomaly correlations are daily means and standard deviations of the pressure at mean sea level (PMSL). Monthly means of PMSL are available for more than 50 years from the NCEP-NCAR reanalysis (Kistler, et al., 2001) on a 2.5o by 2.5o latitude-longitude grid. A mean grid for each month of the year was obtained by averaging over 41-years (1958-1998). Similarly, a standard deviation was computed for each month of the year using the 41 values from the series of monthly mean grids. To get daily values, it is assumed that the parameters computed for each month apply at the middle of the month, and both the mean and standard deviation are interpolated linearly to a specific day. Leap days are treated as 01 March.

This method of computing the standard deviation may yield a smaller value than would be obtained by using daily data. This is not considered problematic because the standard deviation is used as a normalizing factor to compensate for the tendency of the anomalies to be greater at higher latitudes where the standard deviation is also larger (Hughes, 1992). Even if daily data had been used, significant running averaging would have been necessary to produce a smoothly varying series of daily standard deviation values.

2. The Verification Procedure

The verifying analysis is that produced by the WPC surface analysis desk. It is gridded on a polar stereographic grid with approximate resolution of 78 km. This grid covers the area within the outer box in Fig. 1. The inner box in Fig. 1 is the area covered by the MRFD grid, which has approximately 80-km resolution. All of the guidance models except for the UK Met Office model (UKMET), the European Center for Medium-Range Weather Forecasting (ECMWF) model, and the NCEP Global Forecast System (GFS) Medium Range Forecasting (MRF) model ensemble mean are on a one degree resolution (approximately 71-km mean longitudinal resolution) latitude-longitude grid. The UKMET, ECMWF, and GFS MRF ensemble mean output is on a 2.5o by 2.5o latitude-longitude grid having an approximate longitudinal resolution of 179 km. For the latter models, the 78-km resolution analysis is smoothed using a Gaussian weighted filter before interpolation to the verification grid. This slight smoothing results in a fairer comparison of features depicted at a resolution of 179 km against features depicted at 78-km resolution. The verification grid is the MRFD grid (~80-km resolution).

As necessary either or both the analysis and forecast are interpolated bilinearly to the verification grid. Then the climatological values are obtained and interpolated bilinearly to the verification grid. The standardized anomaly is computed at each point for both the forecast and the analysis by subtracting the climatological mean and then dividing by the climatological standard deviation (Hughes, 1992). The correlation between the forecast and analyzed standardized anomalies are computed over the entire verification grid (8829 grid points).


Hughes, F. D., 1992: Skill of Medium Range Forecasts. NCEP Office note 385, 32 pp. [Available from NCEP, 5200 Auth Road, Washington, DC 20233.]

Kistler, R., and Coauthors, 2001: The NCEP-NCAR 5-year reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82, 247--267.