If the Model is Wrong, What's Next?

Suppose you decide that the analysis or short-range forecast of the computer model is wrong: what do you do about it?

Usually, the model won't be TOTALLY wrong, so it's only necessary to make slight modifications to the model forecast. If you suspect that the model errors will be significant, though, you're better off throwing out the model completely and making your forecast based on what you expect the weather to do.

Errors fall into two categories: errors of position and errors of magnitude. Examples will be shown below.

In an error of position, life is simple. A position error is frequently equivalent to an error in timing. In other words, if the cold front is 200 miles ahead of its forecasted position, that means it's moving too quickly, and the model is too slow. In that case, just drag the time line out a bit, and accept that what the model forecasts to occur will actually happen, just at a different time. For forecasts of specific weather elements at specific times, do a space adjustment: if the weather feature of interest is forecasted to be too far east by 100 miles, look at what the model is forecasting 100 miles east of your forecast city, and use that as guidance. Basically, the weather for your location is being forecast by the model, but in the wrong place or at the wrong time.

Errors in magnitude can be more difficult. If the system in question is an entire storm system, you will need a good working knowledge of how various aspects of the storm are affected by changes in intensity. If it's just a matter of the temperatures being too hot or too cold, or something like that, a good starting point is to assume that the errors which are present now will still be present tomorrow. Sometimes this works, and sometimes it doesn't, and it takes lots of experience to be able to predict how a model's forecasts will evolve.

Most models also have known biases. For example, you can generally expect the NGM to locate storm systems too far to the southeast. Things like this only come from experience, and these characteristics change as the models change. If you're forecasting professionally, you'll need to stay on top of things like this, but if you learned this stuff now, it would probably change in four years anyway as the models change.

Suppose you're forecasting for Wichita, and according to the model, the rain is supposed to end in Wichita around 10Z. A check of the latest radar map shows that the back edge of the rain is already about 150 miles farther east than the model forecasted it to be. Should you deviate from the model?

Suppose the Wichita forecast is for light showers, and you check the radar and discover an intense squall line forming in western Kansas. Should you deviate from the model?