When you check the day's temperature or see if it's going to rain on your way to work, the answer can vary widely depending on where you look. Luckily, the website ForecastAdvisor—which grades the accuracy of U.S. forecasting outlets such as The Weather Channel, National Weather Service (NWS), CustomWeather, and AccuWeather—can help you sort the good from the bad. If you head over to their site you can plug in your zip code and see your local numbers, but they were kind enough to provide us with their raw data for 2013, and the results are pretty surprising.
ForecastAdvisor takes one-day, two-day, and three-day forecasts for 752 locations nationwide, and then compares these predictions to the actual weather; the "accuracy" figure for each service is an average of the accuracy of each of its three forecasts. High and low temperature forecasts count as accurate if they're within 3° F of the actual figure.
If a forecast calls for precipitation, ForecastAdvisor counts it as accurate if there's any precipitation, of any type. For some parts of the country this might be a little oversimplified—the distinction between a rain/snow/sleet prediction can be pretty important—but it makes the figure easy to record and understand.
In addition to the four major outlets listed above, ForecastAdvisor also tracks MeteoGroup and Foreca forecasts for most locations. Both are European and not particularly well-known or used in the U.S., but it's worth mentioning that MeteoGroup was outstandingly accurate in 2013 (you can see their forecasts here). Depending on location you can also find scattered data for Weather Underground, WeatherBug, and Dark Sky.
So what places are stuck with the worst predictions? The map below shows—for the four major outlets—how accurate the most accurate forecasting service was. There are 713 ForecastAdvisor cities in the contiguous U.S.; the map is split into regions based on which one of these cities was closest (apologies to Alaska and Hawaii*). The accuracy range is different between the three forecast types, so in order to keep regional variation visible we had to put each map on a slightly different color scale.
The easiest variable to predict was precipitation, with an average of 82.1 percent accuracy. This is largely because precipitation is simplified to a "yes/no" proposition—predicting clear skies every day would net you 70 percent accuracy in many parts of the country—but also because rain and snow are also fairly predictable across large swathes of the U.S. It rarely rains in the southwest, and the outlets had the most difficulty along the Gulf Coast (where intense thunderstorms are hit-or-miss most of the year) and especially around the eastern Great Lakes, where a large portion of the yearly precipitation falls in the form of lake-effect snow.
Why are there such regional variations in temperature accuracy? For one, the time of day when ForecastAdvisor pulls forecasts means that accuracy is generally better for high temps than low temps. Beyond that, the forecasters seem to have a pretty hard time with the central and northern Plains high temperatures. This section of the country is typically in the path of strong low pressure systems as they form coming off the eastern slope of the Rockies. These cyclones, combined with the flat terrain of the region, allow for some pretty big temperature swings – some locations can experience a 50 degree drop in temperatures in just a few hours.
The regional impact on low temperature accuracy seems to fall on a sharply west/east divide. Meteorologists have a difficult time forecasting lows for mountainous areas of the western United States, while low temps in the southeast – which don't vary much during a large chunk of the year – are pretty easy to predict.
So which forecasting service should you use? The map below shows which of the four major forecasters was the most accurate, for ForecastAdvisor's 713 contiguous U.S. locations:
The Weather Channel—the Atlanta-based weather monolith/broadcaster—blew away the competition in all three predictions. It was the most accurate forecaster of high temperature in 93% of all cities surveyed, and also did the best at forecasting low temperature (winning in 69% of cities) and precipitation (64% of cities).
Despite its national dominance, The Weather Channel has some competition in certain regions. If you live in the West, you might consider using AccuWeather or the NWS—the federal service that relies more heavily on experienced local forecasters—-for low temperatures. If you live in the Midwest or Northeast, NWS seems to do better with precipitation, while CustomWeather performs well in the Mountain West and the Southern coast.
*This is for space reasons mostly, but also because Northern Alaska a. has a lack of reliable observation locations, to test accuracy, and b. it avoided by some forecasting services outright. This means that, on the map, somewhere like Barrow would be colored by the accuracy of forecasts in Fairbanks, some 500 miles away.