NOAA's quarter-degree grid holds 540 monthly temperatures for every cell since 1981-09 — 1,641,600 values over the Philippine box. The sea has warmed 1.02°C across the record, at +0.23°C a decade, and not one of the 3,040 cells is cooling.
A quarter-degree box over an archipelago is 3,456 squares, of which 416 — 12.0% — are land or part land and carry a fill value rather than a temperature. The other 3,040 each have 540 monthly readings, and fitting a line through each one gives this.
of 3,040 cells have a positive trend. There is no cool patch anywhere in the box, which is the part that makes a single average worth quoting at all.
per decade, against +0.14°C for the slowest. A threefold spread across one archipelago, which an average over the whole box hides.
cells are excluded because they are land in at least one month of the record. They are not absent from the archive — they hold −9.96921×1036, and averaging without masking them returns nonsense.
A trend in degrees is hard to feel. Here is the same thing counted differently: a month is "warm" when its mean is above the 90th percentile for that calendar month across 1982–2011. By construction that happened a tenth of the time in the baseline.
warm months a year through the baseline — 36 months in thirty years, which is exactly the tenth a 90th percentile is supposed to mean.
a year over the last ten full years: 83 of the 144 warm months in the whole record — 57.6% of them — fall in the last decade.
warm months out of twelve, in 2024. What the baseline treats as a one-in-ten month was nearly every month of that year.
Every anomaly on this page is a difference against a seasonal cycle computed from 1982–2011, so the cycle itself is worth showing rather than assuming.
The recent line sits above the baseline's 90th percentile for most of the year. That is the whole finding in one picture: the threshold that used to mark an unusually warm month is now roughly the normal month.
It also gives the pipeline a way to check its own arithmetic. A 90th percentile is exceeded by a tenth of the sample it was computed from, so thirty baseline years must show about thirty-six warm months. They show 36. An off-by-one in the order statistic, or a climatology built from the wrong window, would move that number and almost nothing else — so a check asserts it, and it is the one test here of the percentile arithmetic rather than of the bookkeeping around it.
6 rectangles, each named for the sea it mostly covers. They are boxes rather than coastlines, and their bounds ship in the data so nobody has to trust the name.
per decade in the Philippine Sea, the open Pacific side. The slowest is the Sulu Sea at +0.2°C — a difference, but not a reprieve.
is the mean for the Celebes Sea over the whole record, against 28.26°C for the box as a whole.
is the mean of the last ten full years, against 27.91°C for 1982–1991. The same water, three decades apart.
Ranked by the box mean, against the baseline for that calendar month.
The warmest month in 45 years of record is 2020-07, at 30.4°C — +1.28°C above its baseline. Of the 15 warmest months, 12 fall in 2010 or later.
These are monthly means over an area of roughly two hundred thousand square kilometres of sea. A month that averages more than a degree above its own baseline across that much water is not weather.
The file behind this covers the whole planet at a quarter degree, every month for forty-five years. This page needs 0.16% of it.
So none of it is downloaded. OPeNDAP takes the index ranges in the URL — time, latitude, longitude — and the server returns just those cells as plain text. 1,641,600 values arrive in 5 requests of about 4.6 MB, and the pipeline needs no NetCDF library at all, because the response is numbers separated by commas.
It is the same idea as pushing a predicate into remote Parquet, applied to a raster: describe the slice you want in the request and let the server do the cutting. The alternative — fetching the archive and discarding ninety-nine point eight per cent of it — is the version that needs a machine with a lot of disk and a lot of patience.
Four limits, and the second is the one most likely to be over-read.
The standard marine-heatwave definition is daily: five or more consecutive days above the 90th percentile for that day of the year. This archive's daily files cap at about ninety days per request, so the full record would be some 180 requests against a free public server. What is counted here is a monthly analogue, defined in one sentence, and is never called a marine heatwave.
Warm water is what bleaches coral and what typhoons draw energy from, but this page does not join to either. Reef extent and bleaching observations carry attribution licences this site does not accept, so no reef is named and no storm is attributed.
OISST is an interpolated product built from satellite and in-situ measurements, not a direct reading. It is the standard record for exactly this question, and it is still a model output with its own uncertainty, which this page does not carry.
The 6 named areas are rectangles this
page defines, not hydrographic boundaries. Their corners ship in
ph_sst_area.csv so the name is a label on a box the reader can
check, rather than a claim about where a sea ends.
Every figure on this page traces to one of these, through a CSV in
data/ph-sst/. Each is checked against its source query on every
build.
Sea surface temperature for every quarter-degree cell in a box from 4°N to 22°N and 116°E to 128°E, every month from September 1981 to August 2026 — 1,641,600 values over 3,040 sea cells — Keyless: no account, no token, no form. The file behind this covers the whole planet and this page needs 0.16% of it, so nothing is downloaded whole: OPeNDAP takes the index ranges in the URL and returns only those cells as plain text, which is the same pushdown idea as reading remote Parquet and means the pipeline needs no NetCDF library. Five requests of about 4.6 MB stand in for a planetary archive
The terms this page relies on, quoted: “The data may be used and redistributed for free but is not intended for legal use, since it may contain inaccuracies” — That sentence and the warranty disclaimer after it are the whole of it: no attribution condition, no share-alike, no non-commercial term, no registration. Combined with NOAA being a federal agency whose works are not subject to copyright, the dataset passes this site's rule of public domain or CC0 only
The statutory position behind the statement above: a “government work is something created by a U.S. government officer or employee as part of their official duties” and such works are “not subject to copyright protection” — The third-party caveat that shaped several other pages in this series does not bite here. A gridded temperature analysis is computed by NOAA from instrument observations; there is no contributed prose anywhere in it, and this page publishes only numbers and coordinates
Would have provided ten-minute satellite imagery of the Philippines, and by far the best pictures in this series — Rejected on licence rather than access. NOAA states it “has rights to distribute this data freely and openly to the public”, but the data is produced and managed by the Japan Meteorological Agency and that is a distributor's permission rather than a dedication by the producer — and JMA's own terms page 404s at the URL NOAA links to. Every other source in this series is cited with its licence artifact quoted verbatim; this one has none to quote, so it was replaced by this dataset rather than built on an assumption
Would have supported the standard marine-heatwave definition: five or more consecutive days above the 90th percentile for that day of the year — Same source, same licence, rejected on request budget. The daily files cap at roughly ninety days per OPeNDAP request, so the full record would be about 180 requests against a free public server for a metric this page's questions do not need. What is computed instead is stated exactly as what it is — a month whose mean exceeds the 90th percentile for that calendar month in 1982–2011 — and is never called a marine heatwave
Would have turned a temperature map into a map of which Philippine reefs are actually at risk — Linked here is the licence rather than a dataset, because the problem is the licence and not any one source: the reef-extent and bleaching-observation datasets that would support that analysis carry this attribution condition, which this site does not accept. So the page reports sea temperature and its trend, and does not claim to say which reefs bleached
I wrote a companion post — “The Sea Around The Philippines Has Warmed A Degree, Everywhere” — in simple, everyday words.