NOAA's station archive holds 34 Philippine thermometers going back to 1945. In 1965–69, 98.36% of the stations that reported filed a complete year. In 2020–24, 2.25% did — out of more station-years, not fewer.
The sibling page to this one reports a reanalysis: a model that reconstructs the weather everywhere, every day, whether or not anyone was watching. It therefore always has a number. This page reports thermometers, which frequently do not.
The archive holds 1,075,325 Philippine values that passed quality assurance, against 309,112 recorded as missing outright. A missing day is not a mild day. It is a day nobody wrote down, and averaging around it quietly assumes the days that were recorded stand in for the ones that were not.
So the first question is not what the temperature did. It is how much of the record exists.
Counting a year as measured when a station filed at least 330 days of minimum temperature: in 1965–69, 60 of 61 station-years cleared it. In 2020–24, 2 of 89.
The last bar is not a five-year block. 2025–29 holds only 2025 and a part of 2026 — 29 station-years against 89 in the block before it. Its 31.03% is a real uptick, and several stations did resume filing complete years, but it is computed on a third of the exposure and will move as the block fills.
station-years in 2020–24, against 61 in 1965–69. There is more reporting now. It is the completeness that went.
of the 34 stations have never filed a single complete year of minimum temperature in the whole archive.
complete years is the most any one Philippine station has managed, out of eight decades on record.
A national trend needs the same stations measured at both ends, or it measures which stations existed rather than what the weather did. Requiring fifteen complete years in 1961–90 and fifteen in 1996–2025 leaves 1 station out of 34.
That station is Surigao. Across 19 complete years in the baseline and 15 in the recent period, its mean night warmed by 0.6°C, and its nights that never fell below 25°C went from 30 a year to 74 — an increase of 44 such nights annually.
One station is not a country. That figure is what the Philippine station record can support, and this page will not average it into a national number, because there is no second station to average it with.
GHCN-Daily flags values that fail its own quality checks, and the flag sits in a single character next to each value. Honouring it removes 13,136 Philippine values across 6 kinds of failure.
One check dominates: 12,467 values — 94.91% of everything discarded — failed the duplicate check, meaning the value repeats another where a repeat should not occur. A pipeline that reads the value column and ignores the flag beside it inherits all of them, silently, and they are exactly the values most likely to flatten a trend.
Four limits, and the first is the whole shape of the page.
1 station clears a like-for-like comparison. There is no honest Philippine average here, and the absence is the finding rather than a gap to be filled.
The archive records that days are missing, not why. Reporting practice, digitisation backlog and what each station transmits to the global network would all look identical here.
330 days makes a year complete and 25°C makes a night hot. Both are stated rather than tuned, and both are published in the data so a different choice can be checked against this one.
34 points, heavily weighted to airports and coasts, with one mountain station at 1,296 m. They describe themselves, not a national average.
Every figure on this page traces to one of these, through a CSV in
data/ph-stations/. Each is checked against its source query on every
build.
Every daily observation from the 34 Philippine stations in the Global Historical Climatology Network, 1945 to 2026: maximum and minimum temperature, precipitation, and the quality flag on each value — NOAA work, United States public domain under 17 U.S.C. 105. No key, no account, no registration -- a plain HTTP GET. 18.2 MB of fixed-width .dly files, one per station, cached and never republished.
The record layout the parser depends on, the units of each element, and the fourteen quality-assurance flags — Downloaded and kept beside the data rather than recalled. Every column position in the parser -- ID 1-11, YEAR 12-15, MONTH 16-17, ELEMENT 18-21, then 31 repetitions of VALUE 5, MFLAG 1, QFLAG 1, SFLAG 1 -- is quoted from section III, as are the units (tenths of a degree C, tenths of a mm) and the meaning of every flag published on this page.
Name, latitude, longitude and elevation for every station, and the FIPS country code that identifies the Philippine ones — The RP prefix is what defines the station set here. A build that found fewer than 20 RP stations stops rather than publishing a thinner country.
Nothing on this page — The source of the sibling weather page, which reports modelled reanalysis at nine points rather than station observations. Licensed CC BY 4.0, an attribution obligation, so it is out of bounds under the public-domain-or-CC0 rule this page is built to. It is named here because the two pages answer the same question from opposite kinds of evidence: a model that always has a value, against thermometers that frequently do not.
I wrote a companion post — “We Have More Weather Stations Than Ever. Almost None Finish The Year.” — in simple, everyday words.