From forecast to decision: how Nafas writes its advice

Nafas turns a 48-hour PM10/PM2.5 forecast into one line of plain English — for example, "move outdoor events indoors, hazardous dust 6–11am, clears by 2pm" — using a fixed set of rules, not a trained language model. Every forecast hour is scored against US EPA AQI breakpoints, the worst window in the coming day is located, and a pre-written sentence is filled in with the actual hours.

Step one: score every hour

For each of the 48 forecast hours, Nafas checks both PM10 and PM2.5 against the US EPA's AQI breakpoints (the 2024 PM2.5 update) and assigns a severity band from 0 to 5: Good, Moderate, Unhealthy for Sensitive Groups, Unhealthy, Very Unhealthy, Hazardous. Whichever pollutant is worse in that hour sets the band — the logic explicitly takes the max of the two, so a spike in either one is never averaged away.

Step two: find the worst window

Next, Nafas looks across the first 24 hours of the forecast and finds the worst contiguous run of hours: it identifies the peak severity band reached, then expands outward from the first hour that hits it to find where that same severity holds. That window, not a single worst instant, is what the advice speaks to — a five-hour hazardous stretch reads differently than a single bad reading.

Step three: write the sentence

Each of the six severity bands maps to one pre-written advice sentence, ranging from "Good air today — no restrictions on outdoor activity" up to "Hazardous dust conditions — cancel outdoor events and stay indoors during the hours below." For the three most severe bands — Unhealthy, Very Unhealthy, and Hazardous — the generic "during the hours below" is replaced with the actual time window, and if conditions are forecast to clear within the 48-hour window, the sentence appends exactly when: "Clears by Thu 14:00." Below that severity threshold, the sentence stays as a straightforward daily statement — sensitive groups get an explicit callout starting at "Unhealthy for Sensitive Groups," since that's the band where children, older adults, and people with respiratory conditions are the ones who need to act differently from everyone else.

Why rules, not a model, write the advice today

This is a deliberate choice, not a placeholder waiting to be replaced quietly. Nafas's product thesis is "decisions, not numbers" — the value is in naming specific hours and specific actions, which a fixed, auditable rule set does reliably and explainably. A generative model introduces a failure mode this product can't afford: a plausible-sounding sentence that doesn't match the actual forecast. The rule-based approach means every advice line traces directly back to the forecast numbers behind it, every time.

What is on the roadmap is the forecast itself, not this layer: a Gulf-tuned dust forecasting model, trained on Nafas's own daily validation archive rather than relying on global-default numbers. Seehow the underlying forecast is builttoday, or read the About page for where that roadmap stands. The advice line above is live right now — see it for any of the 10 Gulf cities on the live demo.