Model profile
The GEM weather model, explained
Last updated: 16 November 2026
When forecasters talk about “the Canadian model”, “the CMC” or “the GDPS”, they mean the same thing: the global configuration of GEM, the Global Environmental Multiscale model run by Environment and Climate Change Canada. It is the third North American global model alongside the GFS and its AI sibling, and a valuable independent voice in any comparison.
GEM at a glance
| GEM | |
|---|---|
| Full name | Global Environmental Multiscale model |
| Run by | Environment and Climate Change Canada (CMC, Montréal) |
| Global system | GDPS · ~15 km |
| Regional systems | RDPS ~10 km · HRDPS ~2.5 km |
| Global runs | 00 / 12 UTC |
| Forecast range | 10 days |
| Ensemble | GEPS · 21 members |
| Data | Free open data from ECCC |
What GEM is
GEM stands for Global Environmental Multiscale. It was developed in Canada by the research and operations branches of the national weather service, and it is run at the Canadian Meteorological Centre (CMC) in Montréal. That is why the same model turns up under three names: GEM is the model code, CMC is the centre that runs it, and GDPS — the Global Deterministic Prediction System — is the operational global forecast it produces.
“Multiscale” is the important word. The same model code is configured for a global forecast, a continental one and a kilometre-scale one, so Canada’s whole forecast chain shares one set of physics.
The GDPS: Canada’s global forecast
The GDPS covers the whole planet at a horizontal resolution of about 15 km. Instead of a single latitude–longitude grid, which bunches up at the poles, it uses a “Yin-Yang” arrangement: two overlapping grid panels wrapped around the globe like the two halves of a tennis ball. Each panel avoids the poles, which matters for a country with as much Arctic territory as Canada.
The atmosphere is coupled to an ocean and sea-ice model, so the forecast responds to changes in sea-surface temperature and ice cover as it runs. The system is run twice a day, from 00 and 12 UTC, and each run forecasts 10 days ahead.
AI steering for the large-scale flow
In May 2026 the GDPS became one of the first operational physics-based models to be guided by an AI model. Its large-scale temperature and wind patterns are now nudged towards GEML, ECCC’s own machine-learning weather model, while GEM’s physics continues to produce the fine detail, the precipitation and the surface weather.
The idea is to combine the strengths of both approaches: AI models have proved very good at the broad pattern several days ahead, while physics-based models produce sharper, physically consistent detail. ECCC estimated the change would improve predictability by about half a day at day seven.
RDPS and HRDPS: the regional models
Two higher-resolution systems sit inside the global one. The RDPS (Regional Deterministic Prediction System) covers North America and the Arctic at about 10 km, runs four times a day and forecasts roughly three and a half days ahead. It is the Canadian counterpart of a continental model and the main guidance for days one and two.
The HRDPS (High Resolution Deterministic Prediction System) covers Canada and the northern United States at about 2.5 km and forecasts two days ahead. At that resolution it simulates thunderstorms, lake-effect snow bands and mountain weather directly.
ECCC also runs the global model as an ensemble, the GEPS, with 21 members started from slightly different initial conditions.
What GEM is good at
GEM is built and tuned by a service whose day job is forecasting for high latitudes, so it gets heavy use for Canadian and northern United States winter weather, Arctic outbreaks and storms tracking across the Great Lakes and the Atlantic provinces.
Its bigger value is independence. GEM has its own data assimilation, its own grid and its own physics, so it does not share the systematic errors of the GFS or the ECMWF model. When all three agree, that agreement means more than three runs of similar models would.
Where GEM is weaker
Averaged over the globe, GEM has usually verified a little behind ECMWF and close to the other national models. It has only two runs a day, so its guidance can be up to twelve hours old when a fresh GFS or ICON run arrives. And it has a long-standing reputation among forecasters for spinning up tropical storms too readily, so a lone GEM hurricane more than a week out deserves caution.
The 2026 AI-guided upgrade may change the first of those points. It is too recent for long verification records, so treat older rankings with care.
How to use GEM alongside other models
Use GEM as a tie-breaker. If the GFS and ECMWF disagree about a storm five days out, the model GEM sides with is not automatically right, but a two-against-one split that persists across several runs is a real signal.
On the MeteoModels map you can overlay GEM with the other global models and read the consensus, or switch on the certainty mask to see where GEM and the rest agree for your location.
Frequently asked questions
What does GEM stand for?
GEM stands for Global Environmental Multiscale. It is the weather model developed and run by Environment and Climate Change Canada, and it is used for global, regional and kilometre-scale forecasts.
Is the CMC model the same as GEM and GDPS?
Yes. GEM is the model, CMC (Canadian Meteorological Centre) is the centre that runs it, and GDPS (Global Deterministic Prediction System) is the name of the operational global forecast. Weather sites use all three names for the same thing.
How often does the Canadian model update?
The global GDPS runs twice a day, at 00 and 12 UTC. The regional RDPS and the high-resolution HRDPS run four times a day.
How far ahead does the GEM model forecast?
The global GDPS forecasts 10 days ahead. The regional RDPS forecasts about three and a half days ahead and the HRDPS two days.
Is the Canadian model accurate?
It is a capable global model that has generally scored slightly behind ECMWF on global averages and close to other national models. It was upgraded in May 2026 to follow an AI model for the large-scale flow, which ECCC expected to add about half a day of skill at day seven.