Probability of detection vs. false alarm ratio: storm exercises
Probability of detection is the share of observed events you caught. The false alarm ratio is the share of forecast events that did not occur. The denominators refer to different groups.
Start with How do you build a contingency table for a supercell forecast?.
| Outcome | Count | Use |
|---|---|---|
| Hit | 12 | Numerator of POD and included in both denominators |
| Miss | 3 | Observed yes |
| False alarm | 8 | Forecast yes |
Worked example
With H hits, M misses and FA false alarms, POD = H/(H+M) and FAR = FA/(H+FA). In this synthetic supercell zone-hour sample, H = 12, M = 3 and FA = 8. POD is 12/15 = 0.80 and FAR is 8/20 = 0.40. If either denominator is zero, flag that metric as undefined. False alarm ratio also differs from the false positive rate, whose denominator is observed non-events.
Hint for the exercises
For POD condition on observed events. For FAR condition on forecast events.
Practice questions
- Calculate POD for the worked sample.
- Calculate FAR for the worked sample.
- Add 5 misses to the worked sample. How do both statistics change?
- Add 4 false alarms to the original sample and leave both other counts fixed.
- A classmate adds 77 correct negatives and reports FAR as 8/(8+77). Diagnose the denominator.
- Why report both statistics when comparing two forecast thresholds?
Worked answers
Answer 1
12/(12+3) = 0.80.
Answer 2
8/(12+8) = 0.40.
Answer 3
POD falls to 12/20 = 0.60. The forecast-yes denominator is unchanged, so FAR stays 0.40.
Answer 4
POD stays at 0.80. FAR becomes 12/(12+12) = 0.50.
Answer 5
8/85 is the false positive rate. FAR uses the 20 forecast-yes cases and remains 8/20.
Answer 6
A lower forecast threshold can catch more events and also generate additional false alarms. The two statistics show how the threshold performs for observed events and forecast events.
- 12/(12+3) = 0.80.
- 8/(12+8) = 0.40.
- POD falls to 12/20 = 0.60. The forecast-yes denominator is unchanged, so FAR stays 0.40.
- POD stays at 0.80. FAR becomes 12/(12+12) = 0.50.
- 8/85 is the false positive rate. FAR uses the 20 forecast-yes cases and remains 8/20.
- A lower forecast threshold can catch more events and also generate additional false alarms. The two statistics show how the threshold performs for observed events and forecast events.