How we predict the weather

Mausam Online is built on open, professional-grade weather data. Every temperature, rainfall, wind, and air-quality number you see on the site comes from one of two world-class sources, served through the open-data project Open-Meteo. There is no in-house forecasting model and no AI guess — we surface raw outputs from the same systems that national met departments and airlines rely on.

Weather forecast: ECMWF + GFS via Open-Meteo

Our hourly and 7-day forecasts come from two of the world's leading numerical weather prediction (NWP) models:

• ECMWF IFS (Integrated Forecasting System) — operated by the European Centre for Medium-Range Weather Forecasts. Globally regarded as the most accurate medium-range model. Updated 4× daily.
• NOAA GFS (Global Forecast System) — the US National Weather Service's global model. Open data, refreshed 4× daily.

Both models are blended and served by Open-Meteo, a non-profit Swiss project that aggregates and re-distributes open weather data. We refresh each city's reading every 60 minutes.

Air quality: the national scale for each country

An AQI number means nothing without the scale it was computed on, so every badge on this site names its scale. For India, Pakistan, Bangladesh, Nepal and Sri Lanka we compute the CPCB National AQI (0–500), the scale those countries' own agencies publish. Elsewhere we use the US EPA AQI. The label next to the number says which: CPCB National AQI in India, PAK-AQI (CPCB scale) in Pakistan, DoE AQI (CPCB scale) in Bangladesh.

The badge also names the dominant pollutant — the one whose sub-index is highest and therefore the one that produced the number. “PM10 leading” after a dust event and “PM2.5 leading” on a winter evening are different situations, and a single figure hides that.

How the number is computed, and when we refuse to compute it

CPCB breakpoints are defined on 24-hour means for PM2.5, PM10, NO₂, SO₂ and NH₃, and on 8-hour means for O₃ and CO. Feeding an instantaneous hourly reading into those breakpoints produces a number that looks official and is not: on a Delhi winter evening the hourly PM2.5 can be double the 24-hour mean, which is two whole bands. So we fetch the preceding 24 hours and average, rather than using the current-hour snapshot.

Three conditions have to hold or no badge is shown at all:

• at least 18 of the 24 hours must be present. CPCB itself allows a 24-hour mean from 16 hours; we are stricter.
• at least three pollutants must be available.
• PM2.5 or PM10 must be among them. In South Asia particulates are almost always the binding pollutant, and an index computed without either is not describing the air people are breathing.

A missing badge is recoverable. A confident wrong number is not, which is why the failure is silent absence rather than a fallback guess.

Why not a single scale everywhere?

Because the scales disagree, and the reader's own authorities use the national one. A PM2.5 reading that is “Moderate” on the European index can be “Poor” on the Indian one; publishing the European verdict under an Indian number, or the reverse, produces a page that contradicts the local advisory a reader has already seen. We show the underlying PM2.5 / PM10 µg/m³ values alongside the index in every case, so the figure can be checked against any scale.

This section previously described the European AQI as our scale. That was accurate when written and stopped being accurate when the national scales were implemented; the pages had moved and this page had not. It is recorded here rather than quietly edited, because a methodology page that lags the product is worse than no methodology page.

Official warnings take precedence over the model

Everything on this site is model output. It is not an official warning, and where the two disagree the official one is the one to act on. The national meteorological agencies for our main markets are:

• India Meteorological Department (IMD) — cyclone and heatwave warnings, monsoon onset declarations
• Pakistan Meteorological Department (PMD)
• Bangladesh Meteorological Department (BMD)

Two specific cases where we defer to them rather than to our own derivation:

• Monsoon onset. IMD declares onset for Kerala and for each subdivision using its own criteria and station network. Our per-city onset dates are derived from reanalysis rainfall (see where the climate records come from) and are not IMD declarations; where the derivation is unreliable we say so instead of publishing a date. For Ahmedabad, for example, our criterion identifies an onset in only 28 of 34 years, so no derived date is published for it at all.
• Named storms and heat warnings. These are issued, not computed. We do not restate them, and a red temperature band on this site is a model reading, not a declared heatwave.

How accurate are the forecasts?

Forecast accuracy depends on the location, weather variable and forecast horizon. Local showers and extreme weather can differ from the model output. We have not published a station-by-station validation of this service, so we do not claim a fixed temperature or rainfall-timing error. A 7-day forecast is guidance, not a guarantee.

Update cadence — what refreshes when

• Current conditions: every 60 minutes via Open-Meteo. Browser cache TTL is 60 min.
• Hourly forecast (next 24 h): 4× daily (00, 06, 12, 18 UTC) on the ECMWF cycle.
• 7-day forecast: 4× daily, same as hourly.
• Climate guides: historical monthly statistics from each city's reanalysis record (see below); these are not a live 15-day forecast.
• Air quality (national AQI + pollutant concentrations): hourly from CAMS.
• Static city/state pages: regenerated on every site deploy (≈daily, manual cadence).

Where the climate records come from

The monthly averages on city pages, the monsoon onset pages and the climate guides are worked out from each city's own daily record for 1991–2024 (34 years), taken from Open-Meteo's historical weather archive for the grid point nearest the city. Each record comes from one of two versions of that archive, and every page names the one its own city's record came from:

• Earlier records: Open-Meteo's historical weather archive, whose default model blends ERA5 and ERA5-Land reanalysis with ECMWF IFS HRES from 2017. Open-Meteo offers no rainfall from ERA5-Land, so the rain figures come from ERA5 and IFS.
• Newer records: ERA5-Land (temperature) and ERA5 (rainfall) reanalysis by Copernicus/ECMWF, via Open-Meteo.

Temperatures are monthly averages of daily highs and lows; rainfall is the average monthly total. A grid estimate can differ from a local weather station.

This page and the climate pages previously described every record as ERA5-Land reanalysis. For the records then published that was incomplete, and for rainfall it was wrong. They came from the first of the two versions above, in which ERA5-Land is one of three sources, and Open-Meteo offers no rainfall from ERA5-Land, so neither the rain figures nor the onset dates derived from them came from it. It is recorded here rather than quietly edited, for the same reason as the air-quality correction above.

What we don't do

• No proprietary "secret" model — every number can be cross-checked against the source APIs.
• No AI-generated text in the forecast itself. Our short summaries (e.g. "Perfect chai weather") are picked from a curated phrase library by deterministic rules.
• We use Google Analytics to understand site usage. See our privacy policy for details.

Open data, open code

Mausam Online's static-site code is open source and the data sources above are all public APIs. If you want to verify a number, point any HTTP client at the same Open-Meteo endpoint and you'll get the same JSON we did.

Why we exist

Most weather apps for South Asia are either bloated international products (slow on low-end Android, no Tamil/Bengali/Urdu UI) or thin scrapers that mirror the same data without context. We try to be neither: designed for fast loading, localised (5 languages with native-script city names), and honest about where the numbers come from.

Who runs Mausam Online

Mausam Online is built and operated by Spolek Minizoo.cz z.s., a Czech-Republic-registered nonprofit association (IČO listed on the parent site at minizoo.cz). The association also operates weather services for several European countries; they are listed on the about page rather than linked from here, because they cover different countries in different languages and share no content with this site.

Operational responsibility for forecast quality, AQI methodology, and city/state geo-data sits with the engineering team. Data anomalies, accuracy disputes and methodology questions can be raised via the contact page — we aim to respond within 5 working days.

This page was last reviewed on 27 September 2026 and is updated whenever a methodology change ships (e.g. switching AQI provider, adding a new model, or revising onset heuristics).