In order to assess the accuracy of the Storm Prediction Forecast algorithm within the SCIT, 48 storms in New York State were chosen and the error between their forecasted tracks and their actual paths was determined. The storms were chosen based on their severity, all 48 being tornadic at some point during their lifespan. The forecasted positions and actual positions of the storm cells were obtained from archived radar data stored by the National Climactic Data Center. The Storm Position Forecast algorithm acts to predict positions for an identified cell in 15-minute intervals, out to a maximum of 60 minutes. Accordingly, the error was determined for each forecasted position available – the 15-minute, 30-minute, 45-minute and 60-minute forecasted position of the storm cell. To determine whether or not the algorithm improves with time with respect to a given storm cell (as it should), successive scans were also analyzed to verify an improvement in the forecast tracks.
In addition to the 48 New York State cells, 24 tornadic storm cells were chosen in Kansas to be analyzed in a similar fashion. The purpose of this was to look for any significant differences between radar performances in two disparate geographical regions.
Once the errors for each position forecast were calculated, comparisons were done between several different parameters to search for correlations with the errors. Some of these comparisons to error included distance to the radar station, dBZ value, lead-time, year and time of day. The errors were also broken down further based on direction (i.e., whether a storm's actual path went North, South, East or West of its predicted path).
The results of the analyses were as expected for some comparisons but quite surprising for others. The lead-time versus error was as expected – the error increased with the amount of time being forecasted for (the smallest error occurred with the 15 minute predicted position and the largest error occurred with the 60 minute prediction). A number of the other comparisons showed no distinct correlations, but some showed minor relationships, specifically the dBZ, year and time of day vs. error. The most distinct result, however, was discovered in the separation of error into individual directions. For both the New York and Kansas storms, there was a prominent trend for the storm's actual path to move to the right of the projected path. For example, 58.27% of the storms in New York tracked SW, S or SE of their projected paths, far greater than the 25% that would be expected for one quadrant of the potential directions for error.
The reason for this preferential rightward error is not known at this point, but would be an area for further research. One possibility for exploration would be the tendency for severe storms to move to the right of surrounding cells in a region of convective precipitation. The Storm Position Forecast must be looked at in further detail to determine whether or not this phenomenon could be inducing this directional error.