Outlooks and forecast models

What the SPC categories mean, and the most common way people misread a convection-allowing model.

Radar tells you what is happening. This is about the few hours before that — and about the single most common way people misread a model.

The SPC categorical outlook

The Storm Prediction Center issues convective outlooks on a five-step categorical scale. The categories are frequently misunderstood as intensity ratings. They are a probability of severe weather occurring near a point, not a promise about how bad it will be.

Category	Short	Roughly means
1 — Marginal	MRGL	Isolated severe possible, limited in duration and coverage
2 — Slight	SLGT	Scattered severe possible
3 — Enhanced	ENH	Numerous severe expected
4 — Moderate	MDT	Widespread severe, often with a significant threat
5 — High	HIGH	Widespread severe with a high-end, long-lived threat

A Marginal day can produce a tornado. The category describes coverage and confidence, not a ceiling. Low-category days with a single dangerous storm are a recurring source of surprise.

High risk is genuinely rare — issued a handful of times a year — and its issuance is itself information. The SPC is saying it has unusual confidence in a widespread, significant event.

Alongside the categorical, the SPC issues probabilistic outlooks for tornado, wind and hail separately. Those are more useful once you are comfortable with them: a 10% hatched tornado probability is a materially different day from a 2%, and both can appear under the same categorical label.

An outlook is also time-stamped and superseded, updated through the day as the picture sharpens. It is only meaningful with its issuance time attached — which is exactly why a tool showing an outlook without its date is showing you something misleading.

HRRR: what it is

The High-Resolution Rapid Refresh is a NOAA model at 3 km grid spacing, updated hourly. At that resolution it is convection-allowing: it does not parameterize thunderstorms as a statistical average over a grid box, it attempts to simulate individual storms.

That is what makes its output look like radar. It is also the source of the most common misreading.

Simulated reflectivity is not a forecast of where storms will be

This is the most important idea on this page. When HRRR shows a line of 55 dBZ echoes crossing a county at 4 p.m., it is not predicting that a storm will be in that county at that hour. Convection-allowing models produce realistic-looking storms with realistic structure, timing and intensity — but the placement of an individual cell is not skillful at that precision.

What the output is skillful at:

Mode — discrete supercells, a line, or scattered pulse storms. Genuinely useful and often right.
Timing, broadly — initiation around midday or after dark.
Coverage — isolated or widespread.
Environment — the thermodynamic and shear fields, which are more reliable than the individual storms drawn on top of them.

The professional read is “discrete cells developing along the dryline in the late afternoon, becoming linear after dark” — a statement about character. Not “there will be a storm here at 5:15.” Run-to-run consistency matters more than any single run: a solution persisting across several hourly cycles deserves more weight than one that appears once and vanishes.

Why the seam matters

Observation and prediction should never be presented as the same kind of thing. Where a loop continues past the current time, there has to be an unmistakable marker of where measurement ends and simulation begins.

And forecast frames should not be interpolated down to the observed cadence — smoothing an hourly model into two-minute steps invents radar data that was never computed. The honest presentation keeps the model at the cadence it was actually run.

The Storm 3D window showing reflectivity surfaces with a legend reading 18 dBZ, 35 dBZ and 50 dBZ. — Every threshold on screen is a number you chose. The same discipline applies to a forecast tail: what is measured and what is modelled must never look alike.

CAPE, shear, and why both are needed

Two quantities do most of the work in severe weather forecasting.

CAPE (convective available potential energy, J/kg) — how much buoyant energy a parcel has. Loosely: fuel. Higher CAPE permits stronger updrafts.
Shear — how wind changes with height. Deep-layer shear organizes storms and lets them persist; without it an updraft is undercut by its own precipitation and collapses. Low-level shear, and specifically storm-relative helicity, relates to rotation near the ground.

Neither alone is sufficient. High CAPE with no shear gives brief, intense pulse storms that fall apart. Strong shear with no CAPE gives nothing at all, because there is no convection to organize. Significant severe weather generally requires both — and the classic dangerous setup is moderate CAPE with very strong shear, which is why days that look unimpressive on a CAPE map alone can still be serious.

CIN (convective inhibition) is the counterweight: a layer of warm air aloft acting as a lid. A strong cap can suppress convection entirely, or hold it back until the atmosphere is extremely unstable and then break, producing explosive development. “Will the cap break?” is one of the genuinely hard forecast questions.

What none of this settles

Models have systematic biases and they change with each upgrade. Two convection-allowing models can disagree completely on the same afternoon. Ensembles exist because single deterministic runs are not trustworthy on their own.