Weather Blog

Why forecasts don't go beyond 7 days

Why forecasts don't go beyond 7 days

Shhh, don't tell anyone, but we actually have a forecasting model that gives a 16-day forecast.

But before you go sending us suggestions on how to squeeze 16 days onto our extended forecast graphic (not to mention grappling another two minutes of air time from the stingy news producers to read through 16 days of weather) there's a reason we stop at seven days.

For one, it takes a lot of computer power to churn through that much data. So much so that once we get beyond seven days, the computer turns into an "express" mode where it uses a lower-resolution terrain so it can hurry up and finish the calculations for the last nine days and still get a forecast out in a timely matter.

But in doing so, we lose some important topographical features around here - namely, the Cascade and Olympic Mountains. It's more an issue in the winter when the models will typically overestimate how far arctic air will move in from the interior of Canada since it's not correctly accounting for the blocking power of the Rockies and Cascades.

Second, if the forecast model has made a mistake in the shorter term, or has some missing data, that mistake gets really amplified as you extrapolate the forecast into the later days.   So the 16 day forecast is more of an initial guess, rather than gospel for beyond a week.

Sometimes 7 Days is Too Long Too

The inspiration from the blog today is that I've received an unusually high number of requests for weather for Labor Day Weekend. Seems it's quite popular for outdoor weddings this year and while over this parched summer, you could toss a dart at a calendar and find a dry day, over Labor Day Weekend the Northwest always seems to be a tractor beam for any rain that could be in the region.

But this year, we face an added challenge in trying to get an early gauge for the long weekend: The dreaded cut-off low.

Forecasting models are semi-consistent up until Tuesday, but then a low pressure center in the Gulf of Alaska drops south and gets cut off, meandering in the Pacific.  Think of groups of people going down the middle of a river in innertubes.  Everyone's going along, following the current, but maybe a group along the edge of the current gets blown toward the shore and behind a rock.  Separated from the main flow of water, you end up just parked there, perhaps drifting aimlessly.

That's sort of how cut-off lows work -- they just spin, but detached from the jet stream, they just meander until something else comes by to pick it up and push it along.

But as far as weather forecasting goes, nothing monkey-wrenches a long range forecast quicker than a cut-off low, which to forecast models is akin to putting differential equations in front of an English major (or, asking science majors to decipher iambic pentameter).

The models don't handle very well the uncertainty of a storm that doesn't move in the regular flow of weather. It's like asking it to divide by zero -- it really messes up the model equations and, in essence, thwarts their ability to reliably predict the weather.

What happens is the models basically start guessing where the low will go and then base a long range forecast on that, but then they keep changing their guesses, which can have a big snowball effect on extended forecast.

Case in point, let's take a look at three forecast model images:



This first one shows a big "arctic" ridge of high pressure near the Alberta/NW Territories/BC border, with an area of low pressure just off Vancouver Island bringing rain toward the Northwest -- this forecast is for showers to develop soon and highs in the mid-upper 70s.


Here is the second one:



This second one shows that arctic high stronger but farther northeast. Note the low off Vancouver Island is completely gone and instead, we have ridging over the Northwest with even a hint of a thermal trough nearby. This forecast is for sunny skies over the Northwest with highs around 78-80ish in Seattle.


and here is No. 3:



This third one has a broad area of low pressure over the Pacific Northwest. The arctic high is nowhere to be found, instead just a big West Coast trough making for raw, rainy day around here with chilly highs probably barely into the low 60s.

Now for the reveal:

Each of these three images are forecasting for the same time: The morning of Saturday, September 5.  The first model was taken from the Wednesday evening model run, the second was from the Wednesday night run, and the third one is from around dawn Thursday -- each run just six hours apart.

So that gives you an idea of how widely variable a computer forecast model can be, and why sometimes an extended forecast is a big challenge.

Hopefully these wide discrepencies get fixed as the day gets closer and they usually do, which goes back to why we don't forecast out all that far, nervous brides and grooms aside :)