Tuesday, May 22, 2012

Five Miles, Two Locations, Vastly Different Results

Inspiration for tonight comes from a tweet I received earlier today asking why a forecast from wunderground.com was calling for the mid 90's for this weekend?  We're not going that aggressive at Phillyweather.net, nor is the National Weather Service, nor are any other forecasters that we know of although a run at or into the 90's is definitely possible (more on that tomorrow).  This isn't the first time we've run into this question from readers and tweeters, wondering why a forecast from wunderground is running uber aggressive  and wondering how much validity there is to it?

If you're not familiar with Wunderground, it generally is one of the best weather information sites out there for getting real time data, conditions, graphics, and looking for computer modeling.  However, the forecast algorithm it uses relies on local weather observation sites in people's backyards, schools, on rooftops, wherever someone has a thermometer and has decided to plug into Wunderground.com's observer network.  It relies on the observer's temperature climo, location, and some computer modeling to spit out a forecast for any given zip code.  These forecasts aren't written by human but are a classic case of what goes in must come out.

In many instances, if the data is reliable, the forecast should follow suit.   However, if you have a faulty site or one that historically runs warm or cold...guess which way the forecast will ultimately lean.

To illustrate this point, I used two different sites not too far away from each other.  One, the Airport, is the "official" site for the city.  The other is a personal weather station in South Philly at Broad and Washington.  As the crow flies, probably a five mile or so difference between the two locations.  The temperature at the South Philly site runs around three degrees warmer than the Airport site does in this particular instance...granted, the South Philly site is in "real time" and updates more frequently than the Airport and given it's almost into Center City I would expect the site to run a degree or two warmer than the Airport.  That's fine...and an acceptable "cushion" on temperature given location and concrete influence.  However, the algorithm that goes from temperature reality to forecast future really spreads things out in a much wider fashion.

I took the forecasts for those sites -- the top is the zip code for the Airport (19153) and the bottom is a zip code near Broad and Washington (19147).  The difference in temperature forecasts between the two zip codes, only a few miles apart, are a bit high.

Granted, this is one site.  If you were to go to weather.gov for the two zip codes, the differences in temperatures are much smaller...only on the order of a degree or two at most, which is more in line with reality than five to nine degrees in spread.

The moral of the story?  If you want an official forecast, go to weather.gov for a forecast by zip code OR go here and scroll down for your county to get a county-wide forecast.   These forecasts are based less on temperature histories from personal weather stations and generally man-influenced to a much greater degree than one that relies on computer modeling and calculus to get the intended result.  Note we're not slamming wunderground as a whole here...but note that when you get a forecast from another website that isn't the National Weather Service you may very well be getting a forecast that's spit out from a computer.