Skip to content

How do you build a contingency table for a supercell forecast?

Build the four cells of a supercell forecast contingency table, then practice matching forecast windows, observations and event thresholds.

A forecast contingency table pairs a yes-or-no forecast with a yes-or-no observation for every eligible time and place. The four counts are hits, false alarms, misses and correct negatives. Define the event and match the samples before tallying.

Prerequisite. Use a 2 × 2 table and add row or column totals.

Example. One binary classification per zone-hour.
ForecastSupercell observedSupercell absent
Yes8 hits12 false alarms
No4 misses76 correct negatives

Define the sample and count the cells

Use the same zone-hour boundaries for the forecasts and observations. A hit means a forecast of at least one supercell in the zone-hour paired with a verified supercell there. A false alarm means yes was forecast and no was observed. A miss means the forecast was no and the observation yes. For the table above, both were no in 100 − 8 − 4 − 12 = 76 windows. Use a consistent classification procedure and mark windows with insufficient observing coverage.

Practice questions

  1. Reconstruct the correct-negative cell above.
    HintSubtract the three known outcomes from 100.
  2. Find observed-yes and forecast-yes marginal totals.
    HintAdd down one column and across one row.
  3. In 80 windows, 14 are forecast yes, 10 are observed yes and 6 are hits. Build all four counts.
    HintA hit contributes to both marginal totals.
  4. Combine the original sample and a disjoint sample with H = 4, FA = 1, M = 3 and CN = 12.
    HintAdd matching cells and check the grand total.
  5. How should the tally handle a zone-hour with a forecast of no and unavailable radar verification?
    HintCheck the observation requirement.
  6. One storm crosses two verification zones in the same hour. Does it count once?
    HintRead the unit of analysis.

Worked answers

Answer 1

100 − 8 − 4 − 12 = 76 correct negatives.

Answer 2

Observed yes: 8 + 4 = 12. Forecast yes: 8 + 12 = 20.

Answer 3

H = 6, false alarms = 14 − 6 = 8, misses = 10 − 6 = 4 and correct negatives = 80 − 6 − 8 − 4 = 62.

Answer 4

The counts become 12 hits, 13 false alarms, 7 misses and 88 correct negatives, for 120 windows.

Answer 5

Mark it unverified under the predefined protocol and report its exclusion. It has no confirmed event outcome.

Answer 6

If each zone-hour is a separate case, the storm can produce an observed yes in both units. A storm-object analysis would use a different design.

Sources