As the area of study was narrowed down, significant improvement was made in terms of capturing both the magnitude and the spatial distribution. The results were also significantly better than previous research that considered the contiguous United States, due most likely to a more direct consideration of regional differences in weather regimes. A contingency table analysis was done to analyze the accuracy of the results, and this included calculations of the hits, misses, false alarms, and correct negatives, as well as calculating skill scores such as the probability of detection, false alarm ratio, bias, and critical success index. Overall, the probability of detection was 0.477 and the false alarm ratio was 0.494. The probability of detection (POD) measures the ratio of correct forecasts to the total amount of times lightning was observed. The false alarm ratio (FAR) compares the forecasts that were incorrect (false alarms) to the total amount of lightning forecasts. Bias measures the ratio of lightning forecasts to lightning observations and was 0.942, which means that lightning was predicted about 6% less than it occurred. The critical success index (CSI) was 0.326, and is calculated as the ratio of the hits to the total number of times lightning was predicted or observed.
Comparisons of the lightning counts in grid boxes with hits, and total lightning counts in grid boxes with hits, false alarms, and misses were made to evaluate the results. Generally, while the power function was more successful than the exponential function for grid boxes with hits, it was prone to overestimating the magnitude of lightning counts in terms of false alarms. The contingency table skill scores show that the false alarm ratio for the two equations was the same, but this only takes into account the number of times there was a false alarm, not the amount of lightning estimated each time there was a false alarm. While both functions had the same number of false alarms, the power function estimated higher counts of lightning in the grid box when it was a false alarm.
In each grid box, the total counts of lightning as shown by NLDN for hits and misses as well as the total counts of lightning as derived from each function for hits and false alarms, was found. The total number of how many lightning counts per grid box there were for each time there was a hit for each function was compared with the total number of lightning counts for hits from the NLDN. For grid boxes with hits in the entire Southeast region, lightning derived from the exponential function had a correlation coefficient of 0.899 and a slope of 0.667, while the power function had a correlation coefficient of 0.899 and a slope of 0.798, both as compared with NLDN. The correlation coefficients of the two functions were similar and both showed a tendency to underestimate (i.e. the slope was less than 1) the amount of counts in a grid box for hits. For a comparison of the total amount of lightning counts, the magnitude of lightning counts per grid box with hits and false alarms was compared with the magnitude of lightning counts per grid box for NLDN hits and misses. For the total lightning counts, the exponential function had a correlation coefficient of 0.904 and a slope of 0.764 while the power function had a correlation coefficient of 0.900 and a slope of .999. The exponential function still had a slight tendency to underestimate, however, the power function went from underestimating to overestimating when all lightning counts were totaled for all contingencies. Although the slope of the trend line for the power function was slightly less than 1, plots of the the trend line alongside the one-to-one line, which illustrates the trend line if the relationship between derived and observed counts was a one-to one relationship, illustrate the trend line above the one-to-one line, indicating that the power equation may be overestimating counts. Mapping the strike counts for the different regions and for each equation confirmed this trend.
The high correlation and reasonable contingency table statistics in the comparison of estimated and observed counts for each region illustrates how microwave data can be extremely useful in creating a model to observe and understand CG lightning. However, the results also show that careful regional selection is an important component of the methodology.