METAR statistics: method, tables and caveats

Tafrob editorial team · Updated · 10 min read

Short answer

This study counts how often each element appears in METARs, using hourly snapshots of the NOAA Aviation Weather Center cache. Identical report texts are counted once, elements are detected with text patterns applied to the report body (before RMK), and each share is the number of matching reports divided by all distinct reports in the sample.

Key points

  • The sample is a set of NOAA cache snapshots collected hourly; a report text seen in several snapshots is counted once.
  • Most elements are searched in the report body, the text before RMK; a short list of remark-level items is searched in the full text.
  • Each share is a count divided by all distinct METARs, so elements overlap and shares do not add up to 100 percent.
  • The sample leans toward North America, so it describes the cache, not worldwide aviation reporting.
  • Detection relies on text patterns, so odd formats can produce false positives or misses; the tables are descriptive, not operational.
Automated weather observation sensors on masts and a platform beside a runway at an airport, with parked aircraft in the distance
Photo: Eric Phelps, CC BY 3.0, via Wikimedia Commons

Dataset: 121,001 METARs from 5,526 stations and 12,960 TAFs, NOAA Aviation Weather Center, 18 November 2026 to 19 November 2026. Generated 2026-11-19.

How often each METAR feature appears
FeatureShareReports
METARs giving wind in knots (KT)95.6%115,652
METARs with an RMK remarks section62.4%75,456
METARs flagged AUTO (no human observer)62%75,045
METARs with altimeter in inches of mercury (A)54.2%65,620
METARs with visibility in statute miles (SM)52.1%63,090
METARs with the AO2 remark43.3%52,367
METARs with QNH in hectopascals (Q)41.8%50,602
METARs with a T-group (tenths of degree) remark34.7%41,968
METARs with CLR30.3%36,690
METARs with an SLP remark27.2%32,874
METARs with visibility 9999 (10 km or more)25.1%30,406
METARs with at least one BKN layer22.4%27,155
METARs with an OVC layer19.3%23,351
METARs with at least one FEW layer18.8%22,692
METARs with any present weather group17.2%20,832
METARs with at least one SCT layer16.6%20,041
METARs with calm wind (00000)13.8%16,644
METARs ending with a NOSIG trend13.6%16,422
METARs reporting gusts (G)11.5%13,862
METARs with a temperature below zero (M)10.9%13,154
METARs ending with the $ maintenance flag9.8%11,875
METARs with a variable direction group (dddVddd)8.6%10,460
METARs containing CAVOK7.2%8,750
METARs with the AO1 remark5.3%6,357
METARs with variable wind (VRB)5.2%6,343
Reports that are SPECI (special) rather than METAR4.4%5,277
METARs giving wind in metres per second (MPS)3.3%3,996
METARs with a BECMG or TEMPO trend3.3%3,964
METARs with NSC2.6%3,150
METARs with a runway visual range (RVR) group2.5%3,001
METARs reporting cumulonimbus (CB)2.5%3,078
METARs with FZRANO (freezing rain sensor out)1.9%2,287
METARs with TSNO (lightning sensor out)1.7%2,067
METARs with NCD1.6%1,889
METARs with PWINO (weather sensor out)1%1,160
METARs reporting towering cumulus (TCU)0.9%1,043
METARs with vertical visibility (VV)0.8%1,010
METARs flagged COR (corrected)0.4%528
METARs with recent weather (RE..)0.4%522
METARs with SKC0.4%505
METARs with a wind shear group (WS)0.1%102

Dataset: 121,001 METARs from 5,526 stations and 12,960 TAFs, NOAA Aviation Weather Center, 18 November 2026 to 19 November 2026. Generated 2026-11-19.

Present weather phenomena and descriptors, share of all METARs
CodeMeaningShare of METARsReports
RArain7.4%8,947
BRmist4.8%5,755
SNsnow2.6%3,168
SHshowers1.8%2,196
HZhaze1.5%1,815
FGfog1.4%1,640
DZdrizzle1%1,165
VCin the vicinity (VC)0.6%726
TSthunderstorm0.4%439
FUsmoke0.3%391
FZfreezing0.2%226
UPunknown precipitation0.2%213
DRlow drifting0.2%190
BCpatches0.1%158
BLblowing0.1%117
MIshallow0.1%103
SGsnow grains0%59
ICice crystals0%53
PRpartial0%48
PLice pellets0%29
DUwidespread dust0%22
GRhail0%7
GSsmall hail or snow pellets0%6
SAsand0%5
DSduststorm0%4
SQsqualls0%2
VAvolcanic ash0%1
PYspray0%0
POdust or sand whirls0%0
FCfunnel cloud0%0
SSsandstorm0%0
+FCtornado or waterspout (+FC)0%0
Most frequent complete weather groups
GroupShare of METARsOccurrences
-RA4.9%5,942
BR4.8%5,756
-SN2%2,458
HZ1.5%1,815
RA1%1,257
FG0.9%1,131
-SHRA0.9%1,088
-DZ0.8%949
VCSH0.5%572
FU0.3%391
-SHSN0.3%351
+RA0.2%290
UP0.2%213
DRSN0.2%190
DZ0.1%178
FZFG0.1%176
BCFG0.1%158
SHRA0.1%153
TS0.1%144
SN0.1%140
VCTS0.1%115
-TSRA0.1%109
MIFG0.1%103
BLSN0.1%100
IC0%53
TSRA0%51
PRFG0%48
SHSN0%39
VCFG0%32
-FZRA0%31
+SN0%31
-PL0%29
+SG0%23
+DZ0%23
+TSRA0%20
SG0%19
-SG0%17
-FZDZ0%15
+SHRA0%14
DU0%12

Dataset: 121,001 METARs from 5,526 stations and 12,960 TAFs, NOAA Aviation Weather Center, 18 November 2026 to 19 November 2026. Generated 2026-11-19.

Reporting stations by first letter of the ICAO location indicator
PrefixRegion (approximate)StationsShare
KContiguous United States2,43344%
CCanada4928.9%
LSouthern Europe, Israel, Turkey4548.2%
ENorthern Europe3877%
SSouth America3536.4%
PAlaska, Hawaii and US Pacific1903.4%
VSouth and Southeast Asia (V)1813.3%
URussia and former USSR1492.7%
MMexico, Central America, Caribbean (M)1432.6%
RJapan, Korea, Philippines, Taiwan1082%
FSouthern and Central Africa1071.9%
OMiddle East, Pakistan, Afghanistan771.4%
DWest Africa (D)681.2%
HEast and North-East Africa671.2%
GWest Africa and Atlantic islands (G)651.2%
YAustralia561%
WMaritime Southeast Asia561%
NSouth Pacific and New Zealand470.9%
ZChina, Mongolia, North Korea410.7%
TEastern Caribbean310.6%
BIceland, Greenland, Kosovo190.3%
AWestern South Pacific20%

What does the METAR study measure?

The METAR study measures how often specific elements, such as CAVOK, gusts, runway visual range or a given weather code, appear in a sample of METAR texts. It does not forecast anything and it does not rank airports: it counts features in strings. The tables on this page are generated from the same dataset, so they always match the method described here and are updated when the dataset grows.

Each table answers a different question. The first lists every tracked element with its share of reports and the raw count. The second lists the present weather codes, then the most frequent complete weather groups. The third counts reporting stations by the first letter of their ICAO location indicator. Together they describe what a typical report in the sample looks like and who contributes to it. For a decoding of each group, start from how to read a METAR.

Where do the reports come from?

The reports come from the METAR cache file of the NOAA Aviation Weather Center, which its documentation describes as all current METARs in CSV form, refreshed once a minute. A collection script downloads that file once an hour and saves each download unchanged, with a timestamp in its name. A second file, the TAF cache, is downloaded on every sixth hourly iteration. The script identifies itself with a custom user agent, as the Weather Center documentation asks of automated clients.

If a download fails, the failure is logged and the script moves on to the next hour; there is no retry. The raw files are never edited, so every figure can be recomputed from them. The collection period and the number of snapshots appear in the note above each table. The data source is a US Government product, and the page on METAR and TAF data sources and APIs explains the other ways to obtain the same reports.

How are duplicate reports removed?

Duplicates are removed by comparing the full text of the report. The analysis script reads all saved METAR snapshots in file name order and stores each report in a map whose key is the raw text with surrounding spaces trimmed. When the same text appears in a later snapshot, it simply overwrites the earlier entry, so it is counted once.

This has three consequences worth knowing:

  • A station that issues the same text twice in a row counts once, which is rare because the observation time is part of the text.
  • Routine METARs and special reports (SPECI) are mixed in one pool. The key includes the METAR or SPECI prefix when the cache provides one, so a SPECI and a routine report are never merged. The SPECI page explains why special reports exist.
  • The count is a count of distinct reports, not of station-hours. A station that reports every 30 minutes contributes twice as many reports as one that reports hourly.

Reports with an empty text field are skipped. The station of a report is read from the CSV station column for the station and region counts, and from the first token of the raw text for the unit table described on the units page.

How is each element detected?

Each element is detected with a regular expression applied to the report text. The pattern always requires whitespace, or the start or end of the string, around the token, so that a group is not matched inside a longer one. A wind in knots is a digit followed by KT and a space, wind in metres per second is a digit followed by MPS, an altimeter setting in inches is a space, the letter A and four digits, and a QNH in hectopascals is a space, the letter Q and four digits.

The list below describes the main families of patterns used in the first table.

Family What a match means Where it is searched
CAVOK, NOSIG, AUTO, COR The token appears between spaces Report body
Wind KT or MPS suffix, gusts (G followed by a speed), VRB, calm 00000, variable direction (dddVddd) Report body
Visibility Statute miles (digits or a fraction followed by SM, with optional P or M prefix) or the value 9999 Report body
Runway visual range The letter R, two digits, optional L, C or R, a slash and four digits Report body
Clouds FEW, SCT, BKN or OVC followed by three digits, CB or TCU suffixes, VV, SKC, CLR, NSC, NCD Report body
Temperature A negative temperature, an M before the first value Report body
Trend and wind shear BECMG or TEMPO as a separate token, WS followed by a runway or ALL Report body
Remark-level items RMK, AO1, AO2, SLP, the T group, the maintenance flag, PWINO, TSNO, FZRANO and the SPECI prefix Full text

The details of each group are on separate pages, for example remarks (RMK) for the remark-level items.

Why does the analysis separate the body from the remarks?

The analysis separates the body from the remarks because remarks can contain strings that look like body groups. A report is split at the first occurrence of RMK surrounded by spaces, and only the text before that point, the body, is used for most elements. Without this step, a remark such as a cloud type, a visibility note or a pressure tendency could be mistaken for a coded group.

A few elements are meant to be found in the remarks, so they are searched in the full text: the presence of RMK itself, the AO1 and AO2 augmentation indicators, the sea-level pressure group (SLP), the hourly temperature T group, the maintenance indicator and the sensor-status remarks. These are mostly US conventions, which is one reason the shares of those rows are small outside the United States.

How are present weather codes counted?

Present weather codes are counted from the weather field that NOAA extracts into its own column, not from a pattern on the raw text. Each token of that field is split into an optional intensity sign, an optional VC for vicinity, descriptors (MI, BC, PR, DR, BL, SH, TS, FZ) and phenomena (DZ, RA, SN, SG, IC, PL, GR, GS, UP, BR, FG, FU, VA, DU, SA, HZ, PY, PO, SQ, FC, SS, DS).

For the code table, a code is counted once per report even if it appears in several groups of the same report. A report with -SHRA and TSRA therefore adds one to SH, RA and TS, not two to RA. The share is the number of reports containing the code divided by all distinct reports, including those with no weather at all. A present weather group has the form described in the METAR weather codes list.

The second table counts complete groups instead, such as -RA or BR, exactly as written. It lists the 40 most frequent ones, again as a share of all reports. A group is counted each time it occurs.

How are regions assigned?

Regions are assigned from the first letter of the station identifier, using a small table of 22 prefixes. This is an approximation: ICAO location indicators are allocated by region, but a single letter does not always match a country or a homogeneous area. The letter K covers the contiguous United States, C covers Canada, P covers Alaska, Hawaii and the US Pacific, E covers Northern Europe, L Southern Europe with Israel and Turkey, U Russia and former Soviet states, and so on.

The stations table counts distinct station identifiers in the sample for each prefix and shows each prefix as a share of all distinct stations. Prefixes that are not in the lookup table are shown with the bare letter. For an explanation of how location indicators are built and how they differ from IATA codes, see ICAO airport codes.

How should the tables be read?

The tables should be read as frequencies within the sample, not as rules. A share is the count of matching reports divided by all distinct reports. Three habits help:

  1. Compare complementary rows only when they are mutually exclusive. Knots and metres per second can be compared, but gusts and variable wind cannot.
  2. Treat small shares as order-of-magnitude indications. A pattern can miss a rare format or catch an unexpected one.
  3. Read the dataset note above each table. A short collection period captures one weather situation and one time of day, which can bias weather rows.

The share column may show a dash or zero for an element that is not coded in the sample, which does not prove that the element never exists.

What are the main caveats?

The main caveat is coverage. The cache is a feed that NOAA maintains for its own users, and the number of US stations is high compared with the rest of the world. Stations in the K prefix weigh heavily in all shares, and US-only items such as AO2, the T group or the maintenance flag are frequent for that reason.

Other limits:

  • Observation times differ from one region to another, so a snapshot taken at one hour captures reports issued at different local times.
  • Special reports are mixed with routine ones, which slightly inflates elements typical of changing weather.
  • Text patterns cannot detect a malformed group and may match a station-specific coding.
  • A prefix of one letter can hide big differences between countries, for example inside the E, L or O prefixes.

A schematic report

A schematic METAR shows which parts of the text each analysis step reads. The example below is fictitious and written only to illustrate the detection rules; the station identifier is a real one, the values are not.

METAR KJFK 191851Z 28012G22KT 10SM FEW035 BKN250 08/M02 A3012 RMK AO2 SLP203 T00831022

For this string, the body is everything before RMK. The wind matches the knots pattern and the gust pattern, 10SM matches statute-mile visibility, FEW035 and BKN250 match the cloud patterns, 08/M02 matches the temperature pattern for a negative dew point, and A3012 matches the altimeter setting in inches. The part after RMK is searched only for the remark-level items: RMK itself, AO2, SLP and the T group. A European report looks different: EDDF 191850Z 24008KT 9999 SCT040 12/06 Q1018 NOSIG would match the knots, 9999, cloud, QNH and NOSIG rows, and none of the remark rows.

In real reports

In real reports, the elements counted here are not independent: they cluster by region and by reporting practice. A report from a US station usually has statute miles, an altimeter setting in inches, an AO1 or AO2 indicator and remarks, while a report from a European station usually has metres, a QNH in hectopascals and often a TREND such as NOSIG or BECMG. Reading the first table together with the units table on the units page makes those clusters visible.

Real reports also contain unusual content that the patterns were not written for: free-text remarks, national groups, runway state groups, automated missing-data markers (a slash in place of a digit) and corrections. The tables show how common the standard elements are. They do not describe how often a given weather situation occurs, because the sample covers a limited period.

Method and limits

This section summarizes the method and the known limits so that the tables can be checked and reproduced.

  • Source: METAR cache of the NOAA Aviation Weather Center (CSV, gzip), downloaded hourly and kept unchanged. The cache is described in the Weather Center data API documentation as all current METARs, updated once a minute.
  • Reproduction: run the collection script (default 48 hours, one download per hour), then the analysis script, which reads every saved file and writes the data file that the tables use. Both are in the repository of this site.
  • Deduplication: by raw report text, trimmed; a later snapshot overwrites an earlier identical text.
  • Detection: regular expressions on the body (text before RMK) or on the full text for a short list of remark-level items. Present weather comes from NOAA's weather field.
  • Regions: first letter of the station identifier, 22 prefixes, approximate.
  • Not represented: the sample is not statistically representative of worldwide reporting; it favors North America, is limited to the collection period, mixes routine and special reports and cannot capture reports that NOAA does not receive.
  • Sources read for the definitions: FMH-1 (2019) for US practice, WMO-No. 782 (2025) and UK CAA CAP 746 (Issue 6) for ICAO-based practice, and the Weather Center data API page for the cache files.

This page is a description of a data sample. It is not an operational source and must not be used for flight planning or for any safety decision.

Training and reference only. For flight planning, always use official briefing sources (your national weather service, AIS or approved provider).

Frequently asked questions

Where does the METAR data on this site come from?

The reports come from the METAR cache file published by the NOAA Aviation Weather Center, downloaded once an hour by a script and stored unchanged. NOAA describes this file as all current METARs, updated once a minute. US Government works of this kind are in the public domain, so the raw files can be reused and analyzed freely.

Are the percentages in the tables representative of all airports in the world?

No. The cache covers stations that NOAA receives, and a large part of them are in North America, so US practices weigh heavily. The shares describe this sample only. The regional table shows how stations are distributed by the first letter of the ICAO code, which helps judge how much weight each part of the world has.

Why do the shares not add up to 100 percent?

Each row answers its own question: what share of reports contain this element. A single METAR can contain wind in knots, a visibility in metres, a cloud group, a pressure group and a trend at once, so it is counted in several rows. Only mutually exclusive choices, such as knots versus metres per second, behave like parts of a whole.

Can I use these figures to decide whether a weather product is available?

No. The tables describe how often elements appeared in a sample of reports. They are not an operational source, they are not updated in real time and they do not replace the official METAR, TAF and briefing services. For flight decisions, use current official reports and the rules of the relevant authority.

Sources

  1. NOAA Aviation Weather Center. Data API and cache files (METAR, TAF and other products)
  2. Office of the Federal Coordinator for Meteorology (2019). Federal Meteorological Handbook No. 1, Surface Weather Observations and Reports (FCM-H1-2019)
  3. World Meteorological Organization (2025). Aerodrome Reports and Forecasts: A Users' Handbook to the Codes (WMO-No. 782), 2025 edition
  4. UK Civil Aviation Authority (2023). CAP 746 Requirements for meteorological observations at aerodromes, Issue 6