The other day I upgraded WXSIM to version 12.6 Beta 6. Since then I have noticed that the high temperature appear to be, on the average, about 10 degrees lower than actual. Prior to this they were a bit closer, 60 per cent of the time within a degree or two.
Did I miss something in setup? I have not changed anything.
Thanks!
Are you running manual models? I noticed a large variation b/w auto/manual after 12.6, w/ auto being the more accurate model.
Always auto.
Al…
Wow, I can actually answer this one because I’m nearby, and have seen a bit of the same thing!
I think it’s a cloud cover problem. We have this tropical depression thing that passed west of us, and the GFS (and probably NAM FOUS, too) have suggested lots of clouds, which WXSIM has believed. The cloud cover in the forecast is keeping temperatures down a lot, but in reality (at least here) the sun has popped out for maybe up to an hour at a time, which has kicked the temperatures well above the forecast values.
This is actually one of the outstanding issues in WXSIM, for which I don’t have a totally satisfactory solution. On average, the cloud cover forecasts are pretty good, but these breaks aren’t quite properly modeled in the program (though I do have a statistical variation incorporated, which helps some).
I do have some suggestions that should help:
(1) Make sure the ‘shower’ option on the Interrupt Planner form is checked. I probably had it checked by default in your customization, but make sure. This at least allows breaks between the rain, if not the clouds.
(2) The ‘Stop AM Rain’ refinement on the main (data entry) form. I’m not using it right now, but I probably should be. What it does is to limit shower activity to the afternoon and early evening, under the assumption that it’s convective and requires daytime heating to get going - even if external model data throws some in earlier or later. This will allow more warming through about 2 PM, and then the showers come along.
It’s interesting that here in Georgia (and probably other places) a forecast in summer of overcast skies in summer is rarely actually true, except with an east wind (“the wedge”). It’s almost like the temperature is obligated to make it into the 80’s F, and the sun will break through the clouds just to make sure!
One effect of this is that in a summer with a mix of sunny and mostly cloudy days, the forecast is almost perfect on average, but the cloudier days are forecast too cool and the sunny days just a bit too warm. I’m still looking for a way for WXSIM to “know” that the sun is bound to break out!
By the way, this problem may exists outside the southeastern U.S. as well. Any comments from elsewhere?
BTW - I think it’s a coincidence that you just got this problem with version 12.6. The weather pattern changed just about that time.
Tom
Tom,
That’s saweet. Nooner models have increased temperature quite nicely. I should reword…naaaa.
Maybe a tropical depression setup option?
Aug23 low shows a chilly 62F…dang butterflies are buzzing around too, along w/ the Canadian ‘gooses’…eh.
I am curious to hear feedback on this thread. For my area (southeast U.S.) it’s actually the biggest source of error I know of remaining in WXSIM’s forecasts, and I didn’t get around to doing a careful study this summer on it.
I think activating ‘Stop AM Rain’ would largely take care of the problem, but I’m not sure. Now the weather has changed here in ATL and it’s not good for testing it at the moment.
Something similar might happen in other parts of the world as well, at least in the warm season. I don’t want to lay the blame entirely on the imported model data, because overall the cloud forecasts are usually pretty good. It might be more a matter of how to handle the data in WXSIM.
Again, the problem is the in mostly cloudy (but not quite overcast) summer conditions, there are just a few too many clouds or showers coming in through the model data, and WXSIM’s temperature modeling is not getting enough sunny breaks to realitiscally catch the high temperatures (it’s forecasting too low in these cases).
This summer, WXSIM has been extremely accurate (about 1 degree F errors, on average) on the lows, but highs have averaged about 4 degrees off - somethimes a little too high on sunny days, but often considerably (5-8 degrees F) too low on mostly cloudy, showery days, when it failed to get enough sun, so to speak. I think I might have cut that error down a good bit if I had the ‘Stop AM Rain’ on, but I didn’t do that! 
In winter here, this isn’t much of a problem, as the predicted cloudy days really do stay cloudy! (And shower action is not very sun-dependent, as it is here in the summer with convective activity the rule).
Any other experiences with this?
Thanks,
Tom
Hi Tom,
Sorry I did not respond sooner. I just made the changes you recommended the other day. I think it may be to soon to tell if forecast are improving. I have noticed that the lows were right on but the highs were 4 to 5 degrees off. This was before the “upgrade”
You can see the comparisons at http://www.barrowcountyweather.com/wxforecast_compare.php
Thanks!
Al
Hello Tom,
Been watching the forecasts closely and find WXSIM to be missing more often than not. Forecasts (highs and lows) were substantially better prior to the upgrade. You can see the numbers at http://www.barrowcountyweather.com/wxforecast_compare.php
I also went through this last Sunday and did some manual forecast to make sure I had the various options checked appropriately. I did make some changes which gave a slight improvement in forecasting the highs, the lows are rarely near actual. Any suggestions?
Many thanks!
Al
Winder, GA USA
Hi Tom,
This summer, WXSIM has been extremely accurate (about 1 degree F errors, on average) on the lows, but highs have averaged about 4 degrees off - somethimes a little too high on sunny days, but often considerably (5-8 degrees F) too low on mostly cloudy, showery days, when it failed to get enough sun, so to speak.
This is exactly what I have seen here in Finland too. WXSIM gets lows quite, if not really, good but highs are off. In almost all cases has WXSIM forecasted lower than it really was. In the comparetable i see for last month WXSIM’s goals for forcastday high is the worst with just around 30% using ± 2C as goal while day + 3 has about 60%. For lows I have been as best in upper 80%'s. This shceme has been quite same whole summer.
Henkka
For those w/ WRET Auto-learn
Wouldn’t it make more sense to select only the same model run times each day to build on? ie. Morning/afternn forecasts are way off, so select only those models each day for the software to learn from.
There’s nothing different internally about 12.6’s modeling from any of the last several (few dozen, perhaps) versions. Using the GFS advection would be probably be the only cause of differences, and they would be small. I think there’s random variation with the weather regime in place. Some users have found it more accurate now.
It’s possible some setting may have gotten changed, and that’s worth checking into. In my running it here, it’s doing about the same as before. I think the high temperature misses may be due to errors in clouds and precip.
Please keep following it, though! 
Tom
Hi Tom and all of you,
I have been slow in updating so I still run 12.5.4 for both my sites.
I get very accurate temperature forecasts, although I notice a difference between the plaintext.txt and the graph produced. The lows are precise down to +/- .2 frequently. The highs differ a bit between plaintext and graph, where the graph is often around 2-3 C lower than the plaintext. And, actual temp is often around 1C above what plaintext suggests. Please note that I do ONLY do manual forecasts (mainly because lack of valid METAR and SYNOP stations updates, so I have to make decisions on which set to use, constantly…). Those frequent changes in what data is used and the low number of sites may explain the high temp offs (at least that is what I have thought…
Given this and that WXSIM is better than official Swedish and/or Norwegian forcasts, and is so up to three days after a forecast (has improved by WXSIM’s ability to “learning”), it is really good. And because it is so good, I can base decisions on roadwork for our dirt roads, and for hunting teams (mainly very good wind direction), I am very happy about what WXSIM achieves.
So, Tom, you you my climate type, and that it seems to work good here, and that may be a bit of indication that some problems seen are local, just as you described)
Cheers,
Ingemar
I think it’s important to use a mixture of model run times, in fairly equal amounts distributed through the day and night, so that everything can get brought better in line. If you were going to do just two a day, they should be around sunrise and mid-afternoon. The combination is important so that the “learning” process can work on BOTH the overall temperature and the diurnal range. Focusing on one or the other could produce skewed results - kind of like aiming way left if you’re a right-handed golfer with a bad slice! (Been trying to teach my son golf lately, which I’m not so good at myself!). It’s better to just get the whole thing right!
Also, in the same vein, it’s important to find WHY the forecasts are off. The high temperature errors are likely due in part to cloud and rain forecast problems. I’ve already had the problem that “learning” from forecasts that were off in this regard (forecasting too cool with too low a diurnal range) can make the sunny days too warm with too much diurnal range. In that sense, it can be helpful to analyze a subset of mostly clear-day forecasts (which were forecast clear and really turned out that way) as a kind of control experiment.
Tom
Thanks Tom.
I’ll also, using Forecast Comparison, try a point forecast match and run auto-learn to see overall adjustments.