A service that tells you how weird your weather is today. Simple, right?
IMWW uses daily temperature and precipitation (rain & snow) forecasts from The Weather Company/IBM, rated as the most accurate forecast by ForecastWatch. The forecast locations correspond to the historical weather records used in step 2. Some of the 5,000 weather stations used to calculate weird weather.
The 5,000 locations were chosen because they have have plenty of daily temperature and precipitation data going back to at least 1950. The data comes from the Global Historical Climatology Network, the largest, most widely used, and most highly analyzed source of historical weather data. Only the most reliable records were used for IMWW, and in addition to the quality controls that come with this data we did a comprehensive review to ensure a fair comparison with today's weather.
An important question is, Weird Compared to What? We look at historical data before 1990, which gives us enough data from the 1900s to make statistically useful conclusions for many locations, and it's recent enough that many of us might think of it as 'normal' weather (more on that idea later). Example of historical daily high temperatures for Tulsa, OK.
There are some tricky aspects to this step, such as choosing the right statistical distributions to fit to the data; for instance, note that we care much more about what's happening at the extremes of the data than in the middle 'normal' section.
The weather weirdness levels are define as:
|Weirdness||Description||How often expected (roughly)|
|Unusual||Used to happen less than 10% of the time||Every 10 years|
|Odd||Used to happen less than 5% of the time||Every 10 years|
|Strange||Used to happen less than 3% of the time||Every 30 years|
|Crazy||Used to happen less than 2% of the time||Every 50 years|
|Bizarre||Used to happen less than 1% of the time||Every 100 years|
|Bonkers||Pretty much never happened before||Never|
One important idea for the weirdness calcuations is that the weather on, say, March 14 is not going to be that much different than on March 13 or March 16, on average. So to determine what's normal or weird for March 14, we include historical data from a number of days before and after that date. Specifically, 7 days before and after are used, giving 15 days of data for each year of weather history. Since we require 40 years of historical data for every weather station, this comes to at least 600 days of weather data to help us learn what was weird before 1990.
Weather Weirdness is then calculated for every day of the year, which looks something like this for each weather station and each weather type (daily high temperature, daily low temperature, and precipitation).
Weird weather levels for daily low temperature for every day of the year in Tulsa, OK.
We don't want to be limited to just the 5,000 locations, and if the weather is weird at one spot, it's probably weird nearby as well. In fact, the data shows us this is true. So we can estimate weather weirdness in between the individual locations, accounting for things like distance between stations.
Points are weather stations, colored hexagons show weird weather near and between stations