Classic thunder indices: K, Total Totals, Boyden · Experiments

The question. Do the K index, Total Totals or Boyden index — computed from each member's 850, 700 and 500 hPa temperatures, humidities and heights — add skill to the live day-0 thunder model (best lifted index + rain) or the day-3 model (member share)?

Why

Forecasters have used these indices for decades, and they need only mandatory-level data, which GEFS gives at every lead. They were not in the first comparison because the archives then lacked the 700 hPa fields.

Data

Summers 2023 and 2024 (184 runs, all 31 members, leads every 6 hours to 96 h), whose archives hold T, RH and geopotential height at 850, 700 and 500 hPa (and 1000 hPa height). Summer 2025's archives lack them, so it cannot be used. Event as live: thunder within 50 km of the airport. Indices per member and 6-hour step, from dewpoints computed from T and RH:

K index = (T850 − T500) + Td850 − (T700 − Td700)
Total Totals = T850 + Td850 − 2 × T500
Boyden index = (Z700 − Z1000)/10 − T700 − 200 (heights in m, T in °C)

How it will be judged (fixed before running)

1. Each index alone: ensemble mean of the day's most favourable value, ROC area at day 0 and day 3, against the best lifted index on the same rows. 2. Each index added to the live model: leave-one-week-out cross-validated BSS and AUC, fitted and scored separately on 2023 and on 2024. 3. An index is adopted only if it raises BSS by at least 0.01 in both summers and lowers AUC in neither. Otherwise it is reported here as not helping, with its numbers.