-- EMC Tropical Cyclone Forecast Verification --

Spatial Variation of Track & Intensity Errors
for Atlantic Storms (1995 - 1999)

---

This page contains plots that show the spatial variation of track and intensity forecast errors for various models and for the official NHC forecast as well. The data used were the adecks and bdecks from 1995 - 1999 for the Atlantic basin only. The software used to do the verification is based on the code originally developed by Jim Gross and Mark DeMaria. A Cressman analysis was added to the code to provide the spatial analysis (this was suggested by Bob Tuleya, who had done a similar analysis for GFDL forecasts previously).

A few notes about these analyses and plots:

The analysis is NOT homogeneous; A separate analysis was done for each storm against CLIPER/SHIFOR. A balance had to be found between homogeneity of the sample and having enough data points for a valid Cressman analysis. If we had done a homogeneous comparison, we would have had to throw out all of 1995, since we have no UKMET tracks from 1995. With this in mind, these plots are most useful in identifying the strengths of individual models on their own merit, and somewhat less useful for doing inter-comparisons between different models for a specific region.

For the plots shown here, all errors throughout each forecast were assigned to the initial position of the storm for each forecast, as opposed to assigning the errors to the verification positions (plots for a verification- based analysis were also made, but are not shown here).

Spatial Variation of Track Errors

Spatial Variation of Track Errors
Relative to CLIPER (Skill)

Spatial Variation of Intensity Errors

Spatial Variation of Intensity Errors
Relative to SHIFOR (Skill)

Spatial Variation of Intensity Bias

Subjective Interpretation

Since it is difficult to objectively summarize the above plots into a nice, tidy figure, I looked over all of the plots and created a subjective summary of general trends in the spatial error analyses. Again, keep in mind that this spatial analysis was not a homogeneous one, and the purpose of these subjective summary plots is not to try to show that Model A is better than Model B in a particular area, but rather to highlight regions of interest (either strong or weak performance) for the various models.
General features in spatial distribution of track forecast error

General features in spatial distribution of intensity forecast bias