As I’m readying a new version, and accompanying materials, I’m interested in examples of use of autolearn and WXSIM-Lite. If you have been running autolearn for at least a year and have been using WXSIM-Lite for at least a few months, I’d be interested in a copy of your correc.txt file, as the graphs autolearn can make could show seasonal variations in correction factors and accuracy, and also the effects of implementing WXSIM-Lite. I’ll assume if you send me a copy, I have your advance permission to publish graphs based on the data, along with your location!
Hi Tom, I’m very glad to give you my correc.txt file! What do you think of it?
The location is “Acquaviva delle Fonti (BA), Italy”
This is the always up-to-date file on my website: http://www.meteoacquaviva.it/correc.txt
Thanks for all these files! I’ve done some analysis, and ALL of the long term ones I’ve checked so far (where I was able to get 3-6 month periods for the same seasons in different years, with and without WXSIM-Lite) show definite improvement. It averages a 14% improvement in the 6 files that met these criteria, and that’s including any WXSIM-Lite mix over 50% (so some of this wasn’t with optimal mix), but that ranges from 7.6% to 29.6% improvement. The good “outlier” (biggest success story) is Silkeborg, Denmark, where the already reasonable good mean absolute error of between 1.5 and 2 Celsius degrees has improved to less than 1 degree! Again, of the 6, that’s the best (both in terms of improvement and actual error) “advertisement” I’ve seen, but all do show definite improvement. You can really see in this one how errors reduced as the average mix percentage of the data sample built up to the final (and probably near-optimal) 65%.