Is there a meteorologist in the house?
July 4, 2010 8:25 PM   Subscribe

What's the best way to predict future weather? and Please make sense of the harsh winter and harsh summer we're experiencing in the mid-atlantic....

Have studies of the Farmer's almanac been done to verify whether its predictions are accurate? Are there other websites that predict weather (either accurately or not) a few months to a year out? My friends and I were sitting around tonight fretting about yet another extreme winter of snows (for the midatlantic what we had last winter was extreme). We're now having extreme heat. How do we make sense of these extreme weather temperatures? How do I find out whats in store this coming year?
posted by dmbfan93 to Science & Nature (8 answers total) 1 user marked this as a favorite
My understanding is that there is, period, a five day window for accurate weather prediction.

Or are you talking about generalities?
posted by dirtynumbangelboy at 8:31 PM on July 4, 2010 [1 favorite]

The best we can do is use models based on large-scale phenomena like El Nino, Pacific Decadal Oscillation, Madden-Julian Oscillation and so forth. These give fairly general guides as to whether, for example, there will be above-average rainfall or below-average rainfall. They can't, of course, predict rare, extreme events. I'm familiar with Almanacs, and our local one has, for years, employed some guy who uses little more than mysticism to predict the coming year.

Weather is a hard problem.
posted by Jimbob at 8:39 PM on July 4, 2010

Almanacs are pretty much soothsaying.

The location and intensity of the jetstream has the greatest impact on what kind of winter youre going to have in the mid-Atlantic and afaik there is no predicting that ahead of time. If it moves south and sits there you can reasonably assume more storms to come in the short term but that's about it.
posted by fshgrl at 9:14 PM on July 4, 2010

Forecasts out several months or seasons are tricky.

Short-range forecasts, out to several days, are deterministic. Observations of temperature, humidity, pressure, wind speed and direction around the globe, can be projected forward for several days before even the smallest of measurement errors grows and the forecast becomes useless. Theoretically, these deterministic forecasts are possible for out to 2-3 weeks in the future but the usable skill level (for you or I but not necessary for an energy utility) is pretty minimal after 7-10 days. The theoretical limits, btw, come out of calculations made by Edward Lorenz in the early 1960s that eventually led to chaos theory.

Long-term (years to decades to geologic time scales) climate forecasts are probabilistic in nature. In this case climate scientists are concerned with external forcings, most famously greenhouse gases, but also variations in solar output, wobbling of the earth's axis, continental drift, etc. These things vary too slowly to affect the atmosphere on a monthly or seasonal basis.

Seasonal forecasts involve using those things that Jimbob mentioned El Nino, PDO, etc. that affect the weather and that vary over the seasonal time scale. For example we know the El Nino (abnormally warm tropical eastern Pacific waters) of the past few months is transitioning to what appears to be a mild La Nina (abnormally cool tropical eastern Pacific waters). There's been enough El Nino/La Nina over the years that we have some confidence that a certain set of weather patterns will result during an El Nino and another set will result during a La Nina. Same is true for other seasonal phenomena like the PDO and the Arctic Oscillation.

The seasonal outlooks that result are typically stated in terms of likelihood of above/below/near normal temperature and precipitation over the forecast period. What they won't, and can't, do is give any indication of specific weather events occurring on specific days.

Almanacs like to make specific predictions months in advance, The Old Farmer's Almanac says this week in the Mid-Atlantic states will have thunderstorms to start and sunny skies to finish. That may be entertaining but there is no scientific basis to support such a forecast.
posted by plastic_animals at 9:45 PM on July 4, 2010 [1 favorite]

Agreed with dnab, 5 days is about as far out as it gets. I'm a follower, and I don't even know why, because I always click on the 10-day and it ALWAYS changes by the time those ten days get within the realm of 3-4 days away. Some of it, I think, is confirmation bias. If the Farmer's Almanac says it's going to be an exceptionally hot July, for instance, and there are 3 or 4 or even 10 really super-hot days in July that year, well, then, they were right!
posted by deep thought sunstar at 11:02 PM on July 4, 2010

The problem is what has become known as the "butterfly effect". That's a term from chaos theory and it refers to systems which exhibit a property known as "extreme sensitivity to initial conditions".

When they began early efforts to use computers to predict weather, what they found was that small errors tended to grow exponentially, and beyond a certain point they swamped the signal, so that the model bore no resemblance at all to the real system it was simulating. This is an inherent problem with all "brute force" simulations and it puts a horizon on ability to make long term predictions.

A "brute force" simulation is one that defines an increment of time and starts with a description of the system state at time T(0). Using that, it calculates what the state will be at time T(1). It then uses the description of T(1) to create T(2), and in turn uses T(2) to create T(3). Such simulations invariably spiral out of control because the description of T(0) is never (cannot be, in fact) precisely correct. So why use them? Because there is no known alternative for a lot of kinds of systems, including the weather.

The weather bureau has extremely elaborate brute-force simulations which try for a detailed description of the weather, and those tend to have a useful horizon of about five days. Beyond that their predictions are no better than chance. They have other simulationss which are much less granular, and those can see further out -- but they see less well. They can see months out, but about all they can see is "wetter than usual" or "cooler than average".

What's the best way to predict future weather?

When you're talking about predicting months ahead, the best way is to roll dice. It's just as accurate as the other ways, and it's a lot less work.

By the way, that's what the Farmer's Almanac does, more or less. They take historical norms and use those as brackets for random number generation.
posted by Chocolate Pickle at 11:30 PM on July 4, 2010 [1 favorite]

Here's an example of chaos theory, using a magnetic pendulum. It shows how a slight change in the starting state produces a different outcome, which is why weather is so hard to predict. Even if, for example, January 1st this year was just like January 1st some other year, there's going to be some slight difference which means that January 2nd this year may be entirely different from January 2nd that other year.

Another chaotic system, this time a double pendulum, you can see that the interaction between the two pendulum arms means that an arm may have a steady motion for a while and then suddenly have a more extreme motion. This is similar to the extremes in our weather, which are caused by the interaction of a large number of chaotic systems. (The fluid motions inside the sun, the interaction of the orbits of a very large number of solar bodies, the fluid motions in the atmosphere....)

The best we can do is a probabilistic forecast which is based off of systems that are chaotic but relatively low energy, like the Earth's orbit (the other planets and so on do have a slight effect, but a small one since the Earth is fairly massive...otoh if you consider something small like the asteroids, they don't have nice, consistent orbits because they do get pushed around by Jupiter). So, next summer? Very likely warmer than the winter.

I hope that makes you feel a little better about the state of the planet, even if you still can't predict it.
posted by anaelith at 4:38 AM on July 5, 2010

Late to the party, and there have been a bunch of good answers already. In general, the computer models that we use to forecast the weather lose their skill steadily with time, so that by 5-7 days out the forecasts usually are not very accurate and by 14 days the forecast isn't much better than chance. There are different methods for the long range seasonal forecasting that involves stuff like El Nino and other events that can effect the global winds (jet stream) such as the Madden-Julian Oscillation, North Atlantic Oscillation, etc. This is what CPC (plastic_animals linked to their page above) uses, but there is still a lot that is not understood, and even the vague seasonal forecasts are often not that great.

As for verifying the Old Farmer's Almanac, there haven't been any formal studies that I know of, but a few people did some informal studies:
Focusing on California
General Verification discussion from Weather Underground Founder

Once again, the Almanac seems to do just about as well as flipping a coin.
posted by weathergal at 9:13 PM on July 5, 2010

« Older Looking for the right grade school in Rochester...   |   What are these types of jobs called? Newer »
This thread is closed to new comments.