Epstein: How accurate are long-range weather forecasts?

You can now get weather forecasts over a month in advance, but be careful about making plans based on them.

Essdras M. Suarez / Globe Staff

There’s nothing more frustrating than making plans around a weather forecast and then having that forecast change: A rained-out party or a day at the beach that turns chilly can make one feel like there’s no point in even paying attention to the weather.

Recently, some organizations are touting the ability to forecast as far out as 30 or even 45 days. But should you even bother with a long range forecast that far into the future? Just how accurate is a forecast, anyway?

There’s a lot of subjectivity to the answer, but it’s worth exploring and it may help set expectations for what to expect from your local meteorologist.


The short answer to the question: a 30-day specific forecast is almost worthless, however, I hope you’ll read the rest of the article because there’s much more to the story.

I’ll show you two images in this blog which really make this point powerfully. Each image shows the correlation between the forecast of the upper level pattern at about 18,000 feet and the actual observed winds. If the forecast is correct, the score is a “1,” and as the models become less accurate, the score falls. A “zero” would be no better than a complete random guess.

Image one is a five-day forecast while the second image shows a 10-day forecast. Notice how the accuracy fluctuated dramatically and how much less accurate at 10-day forecast can be. You could argue you’d be better just using the average temperatures and sky condition for a given day once you get beyond day five.


Recent 5-day forecast accuracy

Recent five-day forecast accuracy.

Recent 10-day forecast accuracy

Recent 10-day forecast accuracy.

Patterns Matter

Sometimes, the forecast seems to change on a dime. One day you hear it’s going to be sunny in three days, then the following day the forecast changes to “cool with rain.” How can this happen so fast? Certain weather phenomena—like a small fluctuation in the jet stream—can be tough for the computer models to resolve. Think of it like trying to see a type of tree in a forest from 100 yards away; you can tell there are trees, but whether it’s an oak or maple can be challenging to know.


As you walk closer and closer to the forest the tree becomes clearer, you begin to narrow down the type and finally, and standing just a few feet away you can readily identify it. This is similar to how some weather patterns work.

We know the basic features of the weather days in advance, but how exactly those features will end up is often unknown three or four days into the future.

In patterns where there is a huge temperature gradient over a short distance the forecast can change dramatically. You might see a high temperature of 90 on the fifth day of a long-range image lower by 20 degrees three days later.

What Can You Rely On?

About 80% of the time a three-day forecast should be accurate enough for planning purposes; the accuracy will increase, obviously, as the day itself approaches. Days four and five of a medium-range forecast are likely to change and forecasts further into the future can have dramatic shifts.

Here in New England we have mountains, the ocean, a huge cold landmass to the north and warm and humid air sitting to our south much of the year. All of these factors and more can make forecasting not only a challenge, but frustrating. I would always recheck the weather forecast each day and not rely on older forecasts as they can quickly become obsolete.


This discussion has ended. Please join elsewhere on