Where the data comes from — and where it doesn't
Every other site in this network reads its own country's weather service. Forecast Norway pulls MET Norway's 1 km grid off THREDDS; Forecast Finland asks FMI's open download service for whatever raster it likes; Forecast Britain reads the Met Office's UKV from an AWS open bucket. We started this site expecting to do the same thing, and went looking.
Middle Eastern national meteorological services publish gridded model output that anyone may reuse. Not one, across seventeen countries and roughly four hundred million people. Checked live from this server in September 2026; the full scan is below.
This is not a complaint about the services themselves — several of them run serious operational modelling. Saudi Arabia's NCM runs an integrated WRF-based forecasting system substantial enough to have been written up in Weather and Forecasting. The UAE's NCM operates a radar network and a research aircraft fleet. Türkiye's MGM has a dense automatic station network. The output simply is not published under a licence that permits reuse, and mostly is not published in machine-readable form at all.
The consequence is that every high-resolution weather forecast on the public internet for this region comes from a European or American model. That is the gap this site is pointed at, and it is why running our own mesoscale model is a question worth asking rather than a vanity project.
The scan
Every national service in and adjacent to the frame, checked for a machine-readable open data service. "Gridded" means model output on a grid that a third party may download and republish.
| Country | Service | Open API | Gridded model output | What is actually there |
|---|---|---|---|---|
| Israel | IMS | Partial | No | A real, documented, free API at api.ims.gov.il — the only proper one in
the region. About 85 automatic stations at 10-minute resolution. Token issued by email
on request. Station observations only; the INCA nowcast grid is not public. |
| Türkiye | MGM | Undocumented | No | servis.mgm.gov.tr answers JSON when sent an Origin header —
it is what mgm.gov.tr's own pages call. Undocumented and unlicensed, so this site does
not use it. The official "açık veri" page redirects away. |
| Saudi Arabia | NCM | No | No | Runs a substantial operational WRF suite; publishes none of it. Some historical station summaries appear on the national open-data portal. |
| UAE | NCM | Registered | No | api.ncm.gov.ae is an "e-Services Gateway" behind registration, not open
data. Radar and warnings are rendered images on their own site and app. |
| Qatar | QMD | No | No | qweather.gov.qa did not answer from Europe when tested. |
| Oman | DGMET | No | No | Web pages and a mobile app. |
| Kuwait | KMD | No | No | Web pages. Dust and visibility warnings are published as text. |
| Bahrain | BMD | No | No | Web pages. |
| Iraq | IMOS | No | No | No machine-readable service found. |
| Iran | IRIMO | No | No | Did not answer from Europe when tested. |
| Jordan | JMD | No | No | Web pages. |
| Lebanon | LMS | No | No | Airport-based service; web pages. |
| Syria | DGMS | No | No | No machine-readable service found. |
| Yemen | CAMA | No | No | No machine-readable service found. |
| Egypt | EMA | No | No | ema.gov.eg answers with a 118-byte stub page. |
| Cyprus | DOM | No | No | Web pages. (Cyprus is in the EU, and some of its data reaches European aggregators that way.) |
| Palestine | PMD | No | No | Web pages. |
Observations from most of these countries do reach the world, via the WMO Global Telecommunication System — the hourly METARs and six-hourly SYNOPs that airports and principal stations have exchanged internationally for decades. That is how global models see this region at all, and it is a few hundred points for an area the size of the continental United States and a half.
What this site uses instead
Global models
| Source | What | Resolution | Cadence | Licence |
|---|---|---|---|---|
| ECMWF open data | IFS operational forecastThe highest-scoring global model in the world | 0.25° | 4 ×/day, T+240 | CC BY 4.0 |
| ECMWF open data | AIFS single, data-drivenMachine-learned, now beating IFS on many upper-air scores | 0.25° | 4 ×/day, T+240 | CC BY 4.0 |
| DWD open data | ICON globalThe finest free grid there is over Arabia. Read from our own mirror rather than a second time from Offenbach — see the note below | 13 km | 4 ×/day, T+120 | CC BY 4.0 |
| NOAA on AWS | NCEP GFSThe only one with an hourly axis and a surface visibility field | 0.25° | 4 ×/day, T+384 | US public domain |
Regional products worth more here than anywhere else
- WMO Barcelona Dust Regional Centre — the SDS-WAS node for North Africa, the Middle East and Europe. A ten-model dust ensemble (MONARCH, ICON-ART, SILAM, NASA GEOS, MetOffice-UM, MOCAGE, LOTOS-EUROS, WRF-Chem, NOA, EMA-RegCM4) plus a multi-model median, on open THREDDS, covering this domain exactly. There is nothing else like it for any other region.
- Meteosat IODC — EUMETSAT's Indian Ocean satellite sits at 45.5°E, almost directly over Riyadh. Its 15-minute dust RGB is the single best real-time picture of a Shamal there is, and it is free through EUMETView's WMS.
- Copernicus CAMS — global atmospheric composition forecasts including dust optical depth and PM10, four-day range, free with registration.
- Iowa State IEM — decades of archived METARs for every reporting airport in the region, which is the only verification dataset available to anyone outside a national service.
How the data is fetched
Everything is pulled by HTTP byte range. Each of these feeds publishes an index
beside every GRIB file — ECMWF a JSON-lines .index with offsets and lengths,
NCEP a wgrib2 .idx — so a cycle costs the fifteen records this site draws
rather than the whole global file. Measured on this server: an IFS step is 94 MB whole
and about 10 MB as the records used; a whole IFS cycle is 440 MB instead of 4.5 GB. DWD's
ICON has no index, so its files are taken whole and regridded from the icosahedral mesh
through a nearest-neighbour index built once and cached.
ICON is the exception, and it is the more interesting case. DWD publishes ICON on an unstructured icosahedral mesh — one compressed file per variable per forecast hour, no index, no byte ranges — so a cycle costs 3.3 GB however little of it you want, and the values need regridding before they are a map at all. But a sister site of ours, Forecast Germany, has been pulling the whole ICON feed onto one of our own machines for months, merged into one file per hour and already on a 0.125° latitude/longitude grid. So this site reads it there, and cuts the Middle East window on that machine before anything crosses the network: 155 MB a cycle instead of 3,276 MB, a twenty-one-fold saving, and DWD's public bandwidth is spent once for the whole network rather than once per site. It is the same data under the same licence; it simply crosses the Atlantic-facing internet one time fewer.
Then each record is cropped to the Middle East window before anything else happens — 24–65°E, 11–43°N is 23,701 points out of a global grid of 1,038,240 — warped onto the frame's Web Mercator raster, and written as a palettised PNG whose 128-colour table is published alongside it. That last detail is what lets the Model Explorer read a value back out of a pixel under your cursor without downloading anything extra.
Attribution and licence
Generated products on this site are derived works of the sources above and are shared under the terms each requires. ECMWF IFS and AIFS data © ECMWF, licensed CC BY 4.0. DWD ICON data © Deutscher Wetterdienst, CC BY 4.0. NCEP GFS is US Government work in the public domain. Country outlines, coastlines and rivers from Natural Earth, public domain.