A forecast that says "3 to 6 inches" is not a guess wrapped in a safety margin. It's a probability statement, and treating it like a single number with padding on both sides is the reason so many people feel lied to every winter.
I've spent a decade reading National Weather Service snow guidance for operational decisions where the outcome mattered — road crews, facility closures, staffing calls. The complaint I hear every single storm season is the same one: "They said 3 to 6, we got 3, and it felt like nothing happened." That reaction comes from doing math the forecast was never built to support. You can't average a range and call it a prediction. Once you see what's actually behind that range, you stop getting surprised by the low end or the high end, because you were never supposed to expect the middle in the first place.
The Range Isn't an Error Bar, It's a Distribution
Here's the definition dispute at the center of this whole issue: most people read "3 to 6 inches" the way they'd read a shipping estimate — "somewhere between these two numbers, probably close to the middle." Meteorologists don't write it that way. The range is drawn from a spread of plausible outcomes generated by dozens of model runs, and it's reported as a band that's likely to contain the result, not a band centered on the most likely result.
Those two ideas sound similar. They're not. A shipping estimate assumes uniform likelihood across the range. A snowfall range is almost never flat like that. The probability of getting exactly 4.5 inches might be lower than the probability of getting close to 3, or close to 6, depending on which side of the storm's track your location sits on. The range tells you the boundaries of what's reasonable. It does not tell you where inside those boundaries the outcome is most likely to land.
Where the Range Actually Comes From: The National Blend of Models
The 3-to-6 number you see in a forecast discussion or on a graphic almost always traces back to guidance from the National Blend of Models, a system that combines output from several individual weather models into one set of numbers. NBM doesn't spit out a single snowfall total. It generates a distribution of possible totals for each location, then reports that distribution using percentiles.
A percentile in this context answers a specific question: out of all the plausible outcomes the model sees, what total falls below this percentile threshold? The 50th percentile is the median outcome — half the plausible scenarios end up lower, half higher. The 10th percentile is a low-end outcome that only 10% of scenarios fall below. The 90th percentile is a high-end outcome that 90% of scenarios fall below.
When a forecaster writes "3 to 6 inches," they are usually rounding off two of these percentile values — often something close to the 25th and 75th, sometimes wider. The number in the middle of that written range is not the median. It's just the midpoint of two percentile values, which is a completely different thing mathematically.
Percentiles Aren't Guesses, They're Odds
This is the part that gets skipped in almost every explainer on this topic. A percentile isn't a softer version of a guess — it's an odds statement, and odds statements are useful precisely because they let you calculate something specific: the chance of clearing a number that matters to you.
Say the NBM percentile guidance for your location looks like this for a given storm:
| Percentile | Snowfall Total | Meaning |
|---|---|---|
| 10th | 2.0 inches | Only 10% of plausible scenarios end up lower than this |
| 25th | 2.8 inches | A quarter of scenarios fall below this |
| 50th (median) | 4.0 inches | Half of scenarios land above, half below |
| 75th | 5.5 inches | Three-quarters of scenarios fall below this |
| 90th | 7.2 inches | Only 10% of scenarios exceed this |
A forecaster looking at this same data might write "3 to 6 inches" as the public-facing range, rounding the 25th and 75th percentiles. Notice that the median here is 4.0 inches, not 4.5. The written range and the actual center of the distribution don't line up, because the range was chosen to communicate reasonable bounds, not the statistical middle.
Once you have the full percentile table instead of the rounded range, you can answer a much better question than "how much snow will we get." You can answer "what's the probability we clear a specific number I care about."
The Number Nobody Shows You: Probability of Exceedance
This is the actual product hiding behind every rounded range, and it's the piece that separates someone who reads forecasts casually from someone who uses them to make a call. Probability of exceedance is exactly what it sounds like: the odds that snowfall will exceed a specific threshold, expressed as a percentage.
You don't have to reverse-engineer this from percentiles by hand. The National Weather Service publishes winter weather probability graphics for exactly this purpose — maps and charts showing the chance of exceeding thresholds like 1 inch, 4 inches, 8 inches, and so on, for a given storm. These are public, free, and updated with each forecast cycle. Almost nobody outside of people making operational decisions ever opens them, because the rounded range gets all the attention in the graphic that goes out to the public.
Once you know the threshold that actually matters to you — a road treatment trigger, a closure trigger, a point where conditions change from manageable to genuinely hazardous — the probability of exceedance for that specific number is more useful than the entire written range combined.
Worked Example: A Trigger Set at 4 Inches
Say your organization has a policy that kicks in once snowfall is expected to exceed 4 inches. The public forecast says "3 to 6 inches." Using the percentile table above, here's how you'd actually work the problem instead of guessing from the rounded range.
The median (50th percentile) in that table sits right at 4.0 inches. That alone tells you this storm is close to a coin flip relative to your trigger — roughly half of plausible scenarios clear it, half don't. That is a meaningfully different piece of information than "3 to 6 inches," which gives you no sense of where your 4-inch line sits inside the range.
Now compare that to a second storm where the percentile table instead reads 10th at 3.5, 50th at 5.0, 75th at 6.5, 90th at 8.0 — but the rounded public range still comes out to "3 to 6 inches" after rounding. Same written range, completely different odds against a 4-inch trigger. In the second case, well over half the plausible scenarios clear 4 inches, because the whole distribution sits higher even though the two-number range looks identical to the first storm.
This is the exact failure mode of averaging a range: two storms with wildly different odds of exceeding your trigger can produce the identical "3 to 6" headline number. The rounded range hides the thing you actually need. The percentile data and the exceedance probability behind it doesn't.
Why Snow-to-Liquid Ratio Wrecks the Simple Math
Even the percentile table isn't the end of the uncertainty, because snowfall totals themselves depend on a conversion that can shift the entire range after the fact: the snow-to-liquid ratio. Models are often more confident about how much liquid precipitation will fall than about how much snow that liquid becomes, because the ratio depends on temperature through the whole depth of the cloud, not just at the surface.
A storm producing precipitation that would fall as a 10-to-1 ratio (drier, colder, fluffier snow) produces a very different total than the same liquid amount falling at 15-to-1. That's the difference between a storm landing at the low end of the percentile range or the high end, even when the model's liquid-precipitation forecast barely changes. This is one reason totals can end up outside the original 10th-to-90th percentile band entirely — the ratio assumption built into the snowfall conversion turned out to be wrong, not the storm track or timing.
If you want a rough gut check on how much ratio uncertainty is baked into a given range, check how cold the forecast is through the column, not just at the surface. Marginal temperature situations near freezing carry a wider plausible ratio spread than a storm arriving in a deep cold airmass, and that translates directly into a wider gap between the low and high end of the percentile table.
Reasonable Worst Case Versus Most Likely Case
One more term shows up around these forecasts that gets confused with the range itself: reasonable worst case. This isn't the same thing as the 90th percentile, and it isn't the top of the written range either. It's a separate judgment call, usually made by a forecaster looking at the full spread of model output and asking what a plausible bad outcome looks like if several uncertain factors break the same direction at once.
Because it's a judgment call layered on top of the statistical distribution, the reasonable worst case can sit above the 90th percentile value in the raw guidance, particularly when there's a known risk of a mesoscale band or a track shift that the model ensemble doesn't fully capture. Treat it as a separate data point, not as another way of writing the top of the "3 to 6" range. When a forecast discussion mentions a reasonable worst-case scenario separately from the numeric range, that's worth reading on its own, not folding into your mental average.
What This Changes About How You Read the Next Forecast
Stop turning "3 to 6 inches" into 4.5 in your head. That number doesn't exist in the data behind the forecast — it's an artifact of your own arithmetic, not a statistical output. The two numbers in the range are rounded percentile boundaries, not error bars around a central guess.
If a specific number matters to your decision — a trigger, a threshold, a point where plans change — look for the percentile guidance or the exceedance probability tied to that exact number instead of the two-number range built for a general audience. A storm forecast as "3 to 6 inches" can carry a 30% chance of clearing your 4-inch line or a 70% chance, and the headline range will look identical either way. The percentile table and the exceedance probability are the only things that tell those two situations apart.
What's the widest gap you've seen between a published range and what actually landed at your location? I'd like to hear which storms made you distrust the forecast, and whether the percentile data behind it would have told a different story.
0 Comments