The long-term record of anomaly correlations for WPC Medium Range forecasts of mean sea level pressure valid at 12 UTC begins with 1976 for forecast days 3 and 5 at projection times of 72 and 120 hours, respectively. This record ends with 1996. The record resumes for 2000 with day 7 (projection time 168 hours) added. The break in the record between 1996 and 2000 was due to unremediated loss of computational resources. The record resumes in 2000 using a climatology based on the NCEP-NCAR reanalysis as described elsewhere. The climatology used in the earlier part of the record was a 30-year climatology updated once per decade.Graphical Display of the Historical Record
The WPC issues forecasts of pressure at mean sea level (PMSL) valid at 1200 UTC (12Z) at projection times of 72, 96, 120, 144, and 168 hours (days 3--7). These forecasts are verified against the WPC surface analysis valid at 12Z. The verification is expressed as the correlation between the standardized anomaly from climatology computed for the forecast and that computed for the analysis. The contributions to the correlation are values on an array of grid points covering the continental United States along with some surrounding area. More details on the method of calculating the correlations are found here.
The time axis labels on the time series of anomaly correlations are given in the form YYMMDD/HH, where YY is the two digit year, MM is the month, DD is the day, and HH is the hour. Only the first label is complete, the following labels include only what is necessary, usually DD/, except when the month changes, then MMDD/ is given. Nothing follows the /, indicating that the hour of the day shown in the first label (12) is repeated. Gaps in the time series may occur when data is missing.
These cumulative anomaly correlations by season are displayed along with the number of forecasts contributing to the result. The lower the number of contributions, the less the significance of the correlation. All forecast systems show the same number of contributions for each season because the verification system makes sure that verification at the same set of valid times is done for all systems to assure a fair comparison. If a forecast system does not extend to a given forecast day, it is omitted from the graph. The box-whisker elements in these graphs show the statistical significance of the differences at the .05 level (.05 probability of false rejection of the null hypothesis of no difference). A random resampling method is used for the statistical significance assessment.
This graph shows the cumulative performance of models and the medium range desk forecasts over the past year as a function of forecast day. The number of contributions at each forecast day is the same for all forecast systems because the verification system does the comparison using the same set of valid times for all forecast systems. This makes the comparison at each forecast day fair. If a model forecast does not extend to a given forecast day, then the number of contributions is zero for that forecast day.Annual anomaly correlations by forecast day.
This graph also shows performance of models as a function of forecast day, but it includes box-whisker elements showing the ranges of anomaly correlation difference between each model and the WPC forecast. The boxes show the 95% confidence interval for significance corresponding to .05 probability of false rejection of the null hypothesis of no difference between the model and the WPC forecast. Any histogram bar ending above or below the box indicates a statistically significant difference. A random resampling method is used for this assessment of statistical significance. Only the models available through Day 7 are included in this graph.
Annual anomaly correlations by forecast day with statistical significance assessment.
This graph shows the cumulative performance of the three medium range desk forecast products: the overnight preliminary, the morning pre-final, and the final products. The corresponding 00 and 12 UTC GFS forecast standardized anomaly correlations are also shown.MEDR comparison anomaly correlations by forecast day.