Hi Tom,
i don’t remember if i already requested but from about one week Wxsim forecasted min temp too high from about 2/3 degreees.
Could you tell me how does it fix it?
Can Auto-learn fix it or restrict the temp gap automatically?
Thanks in advance,
Alessandro
i have the opposite issue my forcecasts are too low for the actuals
http://hayesweatherstation.info/wxquality.php
I suspect there is a specific setting which is throwing this out.
I have the pro version with autolearn running
The autolearn should be trying to catch up with and correct this problem. Perhaps you could post correc.txt and details.txt so I can see what’s going on.
Tom
Sure 
In the meantime i’ve flagged “Allow enhanced nocturnal decoupling” to adjust low temp but i don’t know if it’s correct.
After that low temps are one/two degrees less than before.
Thanks,
Alessandro
correc.txt (15.7 KB)
details.txt (27.3 KB)
I have just looked at the correc.txt and details.txt files for both of you. In Alessandro’s case, the learned bias corrections resulting in an mean net error very close to zero for minimum temperatures. In Growlingteger’s case, the min temps are just a bit too cool (half a degree or so). However, in BOTH cases, there’s a large standard deviation in the min temp error, whereas the max temps are pretty consistent. You are getting short spells of one kind of error, followed by short spells of the opposite error. On average (over the last two hundred forecasts or so), the forecasts have been about right, but on a day-to-day basis, significant errors are occurring.
What’s interesting is that you are both having bigger errors in min temperatures than in max temepratures. This is exactly the opposite fo my experience here. I have to suspect wind or cloud forecast errors.
Be careful about the enhanced nocturnal decoupling. I really made that to correct a problem rather unique to New Zealand, and perhaps it is best not to use it elsewhere. Also, if you use it, and then don’t use it, autolearn can get confused. If you like, however, you can leave it on and see if that reduces the stabdeard deviation of error. If you do this it may take some time for autolearn to respond, and maybe you should shorten the analysis period to a couple of weeks or so.
Hope that helps!
Tom
Hi Tom,
thanks for reply!
So, no change should i do on Wxsim?
Because this problem come back when there are some clear nights without wind and with a strong nocturnal radiation.
In this case Wxsim forecasts min temp two/three degrees higher than normal while cloud and wind are correct!
I’ve unflagged enhanced nocturnal decoupling because min temp was always too low than normal.
Regards,
Alessandro
Hi Tom
Thanks for the feedback …looks like I might have been fiddling with the cloud settings previously :oops:
Latest run looks much better after I reinstalled back to a known setup.
Regards
Graham
Alessandro,
Please tell me if I understand correctly:
WXSIM forecasts too WARM on clear, calm nights (with correctly forecast cloud cover and wind), but perhaps too COLD on cloudy or windy nights?
If that is true, I do have an idea. You might turn the enhanced nocturnal decoupling back on, BUT make it weaker. There’s a strength setting on tab 4 under Preferences/Settings. I think you have pro mode, so you can set that. It may be set to 100%. Maybe you can try it at about 60%. The cooling effect is stronger on clear, calm nights anyway, so I don’t think it will hurt the cloudy or windy nights, especially with the autolearn running (though it may take a couple weeks or more to adjust fully to the change).
This probably won’t completely fix the problem, but it’s likely to help some.
Tom
Hi Tom,
thanks for your reply.
Yes, partly correct
Min temp are too high on calm wind without any clouds on sky.
Generally ok on cloudy sky and/or windy night.
I’ve set “…nocturnal decoupling” as your advive but this morning min temp was 3 degrees lower than wxsim forecasted (-1 vs. +2)
Should i wait a couple of week to have any positive resuts?
Thanks in advance,
Alessandro
Hi Alessandro,
Yes, wait a week or two to see if it gets better, and to see if autolearn starts to catch on. Too many sudden changes will just confuse things.
Tom
