GEM vs GFS vs ICON: which forecast to trust on a lake

The big three

GEM (Global Environmental Multiscale, Environment and Climate Change Canada) is the home-team global model: ~15 km grid running four times a day, with the Canadian Gasussian-grid physics tuned to North American terrain. GFS (NOAA) is the American counterpart, and ICON (German weather service) the European outsider — all three are decent at large-scale patterns days out, but blind to lake-breeze circulations smaller than their grid cells.

HRDPS (High Resolution Deterministic Prediction System, ECCC) is the Canadian lake sailor's friend: a 2.5-km grid updating every hour, explicitly modeling the Great Lakes, terrain, coastlines, and — critically — lake breezes and gusts. Its weakness is range: it only goes out ~48 hours, and beyond that you are back on the globals.

The practical playbook

Days 3–7: trust the globals (GEM/GFS/ICON) for the pattern — is there a gradient wind at all, is a front coming — and ignore exact numbers. Day 2: start comparing models; if they agree within a couple of knots, confidence is high. Today: trust HRDPS, especially for the timing of the afternoon fill-in, gust magnitudes, and thunderstorm initiation. Disagreement between HRDPS and the globals on a summer afternoon at a big lake usually means the globals missed the lake breeze.

How we use them

LakePulse Canada fetches Open-Meteo's blended best-match forecast for every lake's coordinates — it automatically selects the strongest model for the timeframe, mixing HRDPS-class resolution near-term with globals for the week out. When you want second opinions, the hourly forecast pages show the same hours the score was computed from, so you can apply your own model skepticism with full information.