|Tornado||STP ≥ 1|
|Hail||MUCAPE ≥ 1000 J/kg, Eff. Shear ≥ 20 kt|
|Wind||MUCAPE ≥ 250 J/kg, Eff. Shear ≥ 20 kt|
The probability of at least one cloud-to-ground (CG) lightning strike within 12 miles (20 km) of a point over the forecast period is determined by:
[0.6 * P(4 km Refl > 40 dBZ) + 0.4 * P(QPF > 0.08 in.) + 0.1 * P(MU LI < -1)]
The forecast probabilities are calibrated to be statistically reliable based on CG lightning strikes recorded by the National Lightning Detection Network in 2017-2019.
Overlays of mean storm motion only use HRW members and do not include NAM or HRRR members.
Conceptually, PMM (probability-matched mean) is a variation of the ensemble mean with the original ensemble amplitude restored. At each grid point, the ensemble mean value is replaced with a value from the full distribution of individual member forecasts whose rank matches the point's rank within the ensemble mean distribution. See Ebert (2001, MWR) for more details.
LPMM (localized PMM) is a new technique from Clark (2017, WAF) wherein the PMM calculation is restricted to grid points inside some radius of influence, preventing precipitation in geographically distant areas from influencing the local value. On this site, we use r=110 km.
RF probs (random forest threshold exceedance probabilities) report the probability (PQPF) that the mean 24-h precipitation totals over the surrounding 20-km x 20-km area will exceed the given threshold (currently 0.1-, 0.5-, 1-, or 3-in.). The PQPFs are created using one random forest (RF) for each threshold. Inputs to the RFs include latitude and longitude as well as daily mean HREF ensemble mean forecast values of: temperature and dewpoint temperature at four atmospheric levels, 1-km above-ground-level reflectivity, surface-based CAPE and CIN, PWAT, hourly maximum wind components, 2-5 km updraft helicity, and forecast precipitation. These RF-based PQPFs perform best for the lower precipitation thresholds, which have the greatest climatological frequency.
Percentile values (e.g., P99.85) are derived from a spring 2019 climatology of 3-km CAMs where various experimental runs were binned by dynamical core. The P99.85 (intended to replace old 75 m2 s-2 products) values are: 75 m2 s-2 for ARW; 100 m2 s-2 for NMMB; and 160 m2 s-2 for FV3. The P99.95 (intended to replace old 150 m2 s-2 products) values are: 150 m2 s-2 for ARW; 175 m2 s-2 for NMMB; and 250 m2 s-2 for FV3..
The substantially different UH magnitude distributions for the three dynamical cores represented in HREFv3 necessitate a transition from fixed-value to climatological percentile thresholds in order to treat members equitably.
For now, the old fixed-value UH thresholds remain available as paintballs and as optional neighborhood probability contour overlays. When using these plots, note that FV3 members may exceed the fixed thresholds in more marginal cases than ARW and NMMB members.