Wednesday, May 02, 2012

Final Comparison of Accuweather's 25 Day Forecast To Reality

Last week ended the first 25 days of Accuweather's first 25 day forecast...and after a bit of delay on my end, the results were finally calculated.   The forecast results were mixed for the most part but there were some useful indicators that came out of it.

First, the final data -- the pink and navy colored lines are the actual high and low for each day, with orange and lighter blue the "norms" for each day.  The bolder brown and orange lines were the forecast highs and lows from Accuweather's first forecast.   It's important to note that forecast accuracy is highest closer to the forecast date so it's not a surprise that Accuweather was accurate in the early stages.


Where Accuweather missed the mark was on the magnitude of the mid month warmup -- it was off by over 20 degrees on the forecast high for Saturday and Sunday, April 14th and 15th, as the predictions of 51 and 60 ended up with a reality of 73 and 82.  That said, the forecast methodology did get the idea that a warmup was forthcoming but was slow to catch onto when it was to take place...and it was conservative in how warm it would get.   The conservative part is not surprising to me -- I think part of the formula in their forecast is climatology in order to "mute" the potential effects of an anomalous warm up or cold snap.

Also, the forecast did catch the idea that later April will be cooler than average and did so with pretty good accuracy...but keep in mind that this is one snapshot in time and that the forecast changed rather markedly down the line...not for the better as the coastal storm aftermath day was later projected to have a high near 80 with plenty of sunshine.


Factoring in each day's high temperature error without regard to whether it was colder or warmer than average and their average error came out to 6.3 degrees/day.  The average forecast was three degrees colder than reality if colder/warmer were factored in...much of that buoyed by underforecasting the mid month warm spell.

I stated last week that I felt that this type of forecast can provide a decent "big picture" idea of what will happen in a timeframe two to three weeks away...I don't believe the specifics are necessarily useful as they can be prone to some massive errors if a pattern is mistimed by a day or two.  If users are looking for specifics, the accuracy is sometimes off by quite a bit.  However, if you're looking for a general idea if the weather in two weeks or so will be warmer or colder then this forecast has some value.   That said, the mass market will not look at things quite that way and will instead look at the specific and get hung up on the 22 degree miss...or any other big miss that comes forth down the line.