I fixed my image links - here’s the results of my experiment (this was testing just a 5 day forecast, not 45):
http://www.atomicshrimp.com/st/content/weather_2
The forecasts changed more than the weather varied
Only rarely was the 5-day forecast correct.
My (perhaps hasty) conclusion is what I posted above in #9 - any idiot can predict the weather as far in advance as you like, as long as he is allowed to keep amending his prediction right up to the last moment.
As promised, here are my results of monitoring Accuweather’s long-term forecast. A screenshot of the forecast from 23 February to 5 April was taken from the Akron, OH Month Outlook page sometime before 23 Feb., so it ended up being a 42-day forecast. I didn’t monitor how the temperature predictions changed from then on, only how the high temps (in degrees Fahrenheit) compared to the original predictions. Note that the spreadsheet has separate sheets for Fahrenheit and Celsius, although the graphs end up looking the same :). Enjoy.
Screenshot of original predictions with high temps noted:
Excel spreadsheet of data with graphs and some math: