I’ve been having problems with my temperature forecast for the last month or so. Today as an example the forecast hi from the 9 am run, actually 9:22 so it should have had 12Z data was 71 deg and it’s already 78 deg. With that said the NWS was forecasting 72 deg, so they are off as much as I was. AccuWeather is calling for 79. Any suggestions ? BTW these forecast were auto runs. Would upgrading to the Pro version and allowing the temp biases to be adjusted after a period of time help ?
Is it the case that WXSIM consistently forecasts highs too low? Is the problem only with the highs?
By the way, 9 AM can be a fairly bad time to run the program, as temperatures are changing rapidly at that time of day. How are the afternoon, or pre-dawn forecasts?
The pro mode would probably help with this, though it’s usually good to characterize the nature of the problem a bit first.
It appears to be the Highs that are off the most. Looking a the forecast in Wret the average is about 8 deg lower than actual. I went back to July and it was a little less but still low by about 6 deg. The afternoon forecast seems to be better as it has better local station data and the current temp is well above what was forecast in the morning. I’ll move my forecast back earlier and later and see what happens. I’ve always felt the 00Z and 12Z GFS was the more accurate.What would you recommend for the best forecast time ? Thanks
6:00 am forecast called for a high of 64 deg. Actual high was 75. 3:00 pm forecast nailed it, but it should have. Overnight low 54.4 and forecast was 57. The forecast have never been this far off and no one else seems to be having the problem. Strange. :?
Ok I think I’m starting to get a clue here. In the local station data the “estimated clouds” are 88 % when it’s been clear all day and the solar reading is 97.5 % right now. Think I’ll just take the WxSim forecast off of the web site til I can sort this out. MIght have something to do with the log files getting goofed up the beginning of the month. Anyone know where Wxsimate gets it’s solar data ?
Were does WxSim get it’s data for the calibration run ? The calibration run is showing temps that are -9 deg F from the WD temps. I checked the 92011lg.txt file and can’t find any error there. I’m just about ready to give up.
Any help is GREATLY appreciated.
The calibration run is not supposed to use actual temperatures. It starts out with a climatological normal temperature for the date, always beginning at the last “midpoint” of the day (maybe 10:37 AM or 8:54 PM, for example - it varies with date, time, and conditions). Make sure first that you really are talking about the calibration run.
I’m suspecting bad cloud forecasts, as these can affect highs a lot. Look carefully to see if the forecasts have been too cloudy. If they have been, try unchecking FOUS from the data import form. I’ve sometimes had reports of erroneous cloud forecasts with FOUS in use. Actually, the one other I had was in Massachusetts - not too close to you, but maybe there’s some regional problem.\
The clouds in the calibration run come either from initial METAR data or solar sensor data from your weather station.
Thanks for the info on the calibration runs. Just for clarification is has been my 3 am run forecast high temperatures have been off by 8 - 10 deg low. I unchecked the FOCUS and I’ll see what happens. I did check my some June temps and the forecast highs were with 1 - 3 degrees of the actual on the early morning run. It seems like the 8 - 10 lower forecast than the actual high temps has been for the last 6 weeks or so.
Unchecking the FOCUS in WxSimate seems to have really helped the forecast higs in the forecast. I’ll watch the forecast and WRET over the next few days and let you know fore sure. Thanks
I think it’s the cloud cover in the FOUS (not “FOCUS” by the way - “FOUS” is one of those awkward government acronyms, meaning something like “FOrecast for U.S.”, I think). That’s helpful information for me. I think I’ve seen this problem more in the summer. For now, at least, just go with the GFS data.
I think I have a problem also. WXSim is forecasting very low lows overnight compared to Weather Underground. They differ by up to 13c !
Unless, of course, WU is miles out and the forecast becomes reality, in which case they’ve made a huge error in their calculations.
It will be interesting to see which proves correct, but I would have thought WXSim was a tad too low; the lowest temp we’ve had in the last month is over 7c - although the temperature is dropping. It’s unusual to see such large WXsim / WU differences.
The minimum temperature yesterday was 13.1 so Weather Underground was just about spot on. WXSim was miles out with it’s low. Interestingly the high was pretty close though.
Today it’s forecasting 0c for the next few nights whereas the BBC / Met Office and WU are saying 14c. Clearly it’s going to be woefully wrong again.
I’ve reset some settings to the default and re-ran the forecast several times this morning but it’s still forecasting very low temperatures over the next few nights.
I’m sure I read something in an old post about this but I can’t find it now. Any help much appreciated.
Thanks, Simon
UPDATE - according to my GrADS maps very few places in Western Europe will be anywhere near to zero C in the next 5 days.
Definitely something out of whack here. Can you send me wdata.txt, localdat.txt, localcal.txt, custinit.txt, latest.txt, and latest.wxf? I’d like to see what’s going on by trying the forecast myself.
I’ve experimented some with Simon’s data, and found he had “enhanced nocturnal decoupling” and “supplemental maritime effect” checked. The first can strongly cool the nights off and should be used with care, if at all (it’s mainly for winter nights in New Zealand, for some reason I’ve never quite figured out). The second tends to cool afternoon highs. Neither is really very appropriate except for island (and maybe some coastal) locations.
Even after changing this, though, it looks like temperatures are still somewhat too cool, especially at night. A customer in southern France sent me a similar report, and WXSIM seems to be forecasting 3-4 degrees too cool there, too (even without those two features activated).
It looks like much of western Europe is under an unusually warm area of high pressure, and WXSIM may not be quite believing the warmth! Is anybody else seeing this?
turning off the enhanced decoupling has certainly lifted the forecasted night time lows; however, it’s very cold in the Winter here and using it until very recently (a week or so) had been producing quite accurate forecasts. I will turn it back on when the temperature drops over Winter I think.
WXSim was out by about 11c and it’s now about 4c too low. This is a very long way out imho, and it’s unusual that it doesn’t get a lot closer than this.
It’s unseasonably warm at the moment, but WU etc all seem to be getting it right.
I’m seeing similar but not as extreme differences.
I have run Wret for last four days and max is about 2.5c higher than forecast, min about 3.0c higher than forecast. Predicted daily average is about 4c lower than actual
Thanks for this feedback. I’m still going to look into it some more. It’s odd, because WXSIM in principle should not be too prone to stay close to normal in abnormal weather (in fact, I like to think its ability to handle unusual conditions is one of its strengths). I’m going to try some “retrocasts”, where I use actual upper air conditions instead of just the GFS forecast (which may be just fine, but I need to check).
I’ve experimented with a site in France and one in England, comparing “forecasts” (made after the fact, using actual upper air data to control for any problems that might be GFS data’s “fault” (overall, GFS actually did pretty well on this). I used all the appropriate inputs, like recent temperatures and precipitation (these just reasonable estimates) and realistic advection, along with actual sky conditions and initial temepratures.
The result was that WXSIM was still too cool in this European warm spell. The forecasts for France were about a degree C too cool for lows and about 2-3 degrees too cool for highs. The English site was also too cool, by at least 3 degrees for lows and 1-2 degrees for highs. In summary, even after correcting for everything else, WXSIM was still about 2 degrees too cool. That’s not awful, but it’s more than I’d like, and other forecast sources generally did better (perhaps using persistence as a factor).
I have found by directly manipulating correction factors (please don’t do this - it’s very hard to explain their behavior and you might make things worse), I can get very close to the right forecasts. There’s still some question as to why, for those using the pro version and the correction factors, wret may not have picked up on appropriate values of these factors. Some of it may be sample size and the fact that the really warm weather started very recently.
I’d be interested to hear how forecasts went lately for those actually using wret in this way on a regular basis. I have another question as well - how did WXSIM do during the exceptionally cold weather from the last winter? I’m trying to figure out whether my customizations are too “shy” about extremes in Europe.
