Why models disagree
Even at the same resolution, models start from different data, use different physics and different settings.
The same grid spacing is not the same model
Grid spacing says how finely the atmosphere is divided. It says nothing about what is done inside each box, how the model is started or how the output is produced. Even two models of exactly 9 km differ in all of the following.
| Difference | What happens | Effect on the forecast |
|---|---|---|
| Starting point (the analysis) | Different observations are used, quality-checked differently and blended by different methods | The two start from slightly different atmospheres; the gap grows with time |
| Physics rules (parametrisation) | Convection, clouds, turbulence and surface exchange are approximated differently | Different rain amounts, gust strength, sea breeze timing, fog |
| The grid itself | Different grid layouts and vertical layers; different land-sea mask and smoothed terrain | Coastal boxes are land in one model, sea in another; the wind at the shore differs |
| Numerical methods | Different ways to calculate the equations, different time steps | Small differences in how fast systems move and how sharp fronts are |
| Output processing | Different ways to compute 10 m wind, gusts and rain, and to interpolate to a point | The wind number can differ even when the underlying flow is similar |
| Run time | One model’s latest run may be several hours older than another’s | You may be comparing different information, not different models |
Why small differences grow
The atmosphere is chaotic: tiny differences in the starting state are amplified as the forecast proceeds. After one or two days two good models usually agree on the broad picture; after four or five days they can place a low or a front clearly differently. This is not a failure of one model. It is how far into the future the available information can reach.
What to do when they disagree
- Do not average blindly. The mean of two different forecasts is often a weather that neither predicts, such as a front that never arrives or a wind that no model shows. Look at what each says and why.
- Ask how large the difference matters. A difference of 2 kn is noise; a difference of 15 kn or six hours in the arrival of a front is a real decision.
- Look at an ensemble. It shows whether the disagreement is within the range the atmosphere allows: Ensembles and probabilities.
- Compare the latest runs, not old ones. A newer run has more recent observations.
- Check which model did better recently in your area, with your own logbook or with nearby observations: Checking a forecast against reality.