How We Process the Data

If you are reading a temperature to compare it against something, you should know what was done to it first. This page is the whole path, including the parts that are discarded.

1. Where the readings come from

Every reading originates as a METAR published by the airport itself and distributed by the U.S. National Weather Service / NOAA. We do not measure anything, we do not model anything, and we do not blend sources. One station, one report, as issued.

We read NOAA's bulk cycle files — a single file per hour containing every station's latest report worldwide — rather than requesting each of our five stations separately. One request covers all of them, sent with an If-Modified-Since header so an unchanged file comes back as a 304 with no body at all. If a station is missing from the bulk file, that one station is fetched individually as a fallback.

The raw line for the station you are viewing is shown on the dashboard, unedited. Everything else on the page is derived from it, and you can check the derivation against the original.

2. What gets discarded

Airports publish routine observations on a fixed schedule, and SPECIs — unscheduled special reports — in between when conditions change sharply.

We keep the routine reports and discard the SPECIs. Each station has its own schedule:

These schedules were not assumed from a specification — they were confirmed against this site's own recorded history for each station.

The reason for discarding SPECIs: a special report can catch a momentary spike that never appears in the hourly record. Including it would produce a daily high that the station's own routine record disagrees with. Anyone comparing our figure against the airport's hourly observations would find a number that does not exist there.

3. Precision, and where it comes from

The temperature group in a METAR body is rounded to a whole degree Celsius. United States stations additionally publish a remarks group carrying tenths, and where that group exists we use it in preference to the rounded one.

So San Francisco is stored to a tenth of a degree, and the other four to the nearest degree. That is a property of what the stations publish, not a choice we made.

4. Storage

Readings go into SQLite keyed on station and observation time, so the same report arriving twice — from the bulk file and again from a fallback fetch — is stored once. Observation time means the timestamp inside the METAR, not the time we happened to read it.

Nothing is ever rewritten. If a station reissues a corrected report for a time we already hold, the stored reading stands.

5. What the charts actually show

• The large figure is the most recent routine reading, converted to your chosen unit.
• H and L are the highest and lowest readings recorded for that station on the current local day, taken from the readings themselves.
• Today · 24 h lists every individual routine reading of the local day, in order. Nothing is averaged or smoothed — each cell is one report.
• Last 7 days · hourly gives each hour one cell, and that cell carries the highest reading in that hour. For stations reporting twice an hour this matters: an hour that peaked and then cooled shows the peak, so the cell agrees with the strip above it rather than showing whichever report happened to arrive last.
• Empty cells mean no routine report was recorded for that hour. They are not zeroes and are not interpolated.

Local day and local hour are computed in the station's own timezone, not yours and not UTC. Lucknow's day boundary is midnight in Lucknow.

6. What we do not do

• We do not forecast.
• We do not adjust, correct, calibrate or bias-shift any reading.
• We do not fill gaps by interpolating between readings.
• We do not merge multiple stations into a "city" figure. An airport is not a city, and airports frequently read a degree or two away from the urban centre.