I heard someone on TV talking about European spaghetti plots and wondered if they were making a movie or cooking dinner. Turns out they were weather forecasters. (More)
Squirrels use a crude method to forecast the weather: we assume today’s weather will be the same as yesterday’s. That may sound silly, but it’s surprisingly reliable.
For example, if you forecast that the daytime high would be about the same as the day before last month in Tampa, you would have been correct on 18 of 28 days. Only ten times last month did the daytime high change by 4° or more from the day before. If you used that method to forecast sunny or cloudy and rainy days, you would have been correct on 22 of 28 days last month, and that wasn’t a fluke. In January 2013 a same-as-yesterday forecast in Tampa would have been correct on 18 of 31 days for temperature and 22 of 31 days for sunny or cloudy.
In short, more often than not, today’s weather will be about the same as it was yesterday. That persistence forecast is one important benchmark. If weather folk are going to spend time and money crunching data with sophisticated computer models – and not waste that time and money – their forecasts need to be correct more often than the same-as-yesterday persistence forecasts.
Another benchmark is the almanac forecast – that today’s weather will be about the same as it usually is on March 6th – but that’s less precise. For example, historical records for March 6th in Tampa show a 90% chance that today’s high will be between 67° and 85°, and a 50% chance it will be between 71° and 82°. Cloudiness and precipitation on March 6th vary even more widely. As between a same-as-yesterday persistence forecast and a same-as-it-usually-is-on-this-date almanac forecast, you’re much better off with persistence.
That stands to reason. While there are seasonal climate patterns, daily weather is driven more by local atmospheric conditions than by the earth’s position in orbit around the sun. Those local atmospheric conditions tend to persist for several days, and that’s why the same-as-yesterday forecast will be accurate more often than not.
Of course, the same-as-yesterday model isn’t very useful if there’s a major storm coming at you. Those tend to hit quickly. Today it may be basically nice, if cloudy and breezy. But tomorrow a hurricane floods your city, a blizzard buries you, or a severe thunderstorm breeds a tornado that tears off your roof. In fact, you’re probably more worried about whether the weather folk forecast storms accurately than whether today will be 38° and cloudy or 30° and clear. (If not, you should be. Just sayin’.)
And that brings us to the European spaghetti plots that were neither ideas for western movies nor dinner plans. They’re storm forecast models, which makes me glad the weather folk said “plots.” Had Mrs. Squirrel heard some guy say “And here’s a revealing view of our European spaghetti models,” she would have turned off the TV without looking up from her knitting. (No, I won’t explain how I know that.)
The point is, most storm forecast models use spaghetti plots. The European storm forecast model runs 50 simulations, inputting slightly different conditions each time. The model then compares the results of those simulations to estimate the storm’s likely track and intensity. Other models use similar methods, with different time and distance scales and somewhat different initial condition inputs.
Which brings me to your complaint: “Why don’t the weather folk use the actual conditions and get my local forecast right?” (You think I don’t read Campus Chatter?)
The answer, simply, is complexity. I’ll pause and let you soak up that sentence.
Welcome back. The path of a storm, and your local weather, can vary widely based on surprisingly small changes in initial conditions. The European storm forecast model uses a 40-km (roughly 25 mile) horizontal grid, with vertical slices a few thousand feet deep, and hourly time steps. But forecasters can’t gather all of that data for every 40-km-square, few-thousand-feet deep slice of the atmosphere, every hour. The models input the data they have and use fluid dynamics to estimate the missing data in the gaps. They also use a process called parameterization to estimate processes and conditions – like local thunderstorm – that may happen within a single grid cell.
Even with grid cells that big, and even with super computers, each simulation takes awhile to run. The computer runs 50 of those simulations, each with slightly different input data, each generating a storm track that looks like a strand of spaghetti. The weather folk then look at that spaghetti and estimate the storm’s most likely track.
The results aren’t perfect. Sometimes a storm blows out before it reaches you. Sometimes it hits a lot harder than the weather folk predicted. And sometimes you get a storm they didn’t predict at all, because it was driven by local conditions too small for them to measure.
But because storms move quickly, the spaghetti plot forecasts are much more reliable than a persistence forecast. That, not perfection, is their benchmark. The weather folk are more accurate, more often, and for more important weather systems, than if they simply said what any squirrel would tell you: “Today’s weather will be about the same as yesterday’s.”
Not that you’d listen to squirrels.
Good day and good nuts.