I’ve been carefully studying data and forecasts for a couple of sites (Matangi NZ and Duggleby UK), where WXSIM (new version at least) seems “thunder happy”. I’m not sure yet what the solutions for these particular sites are, BUT I have discovered something which affects thunder forecasts considerably: the upper air temperature bias correction (for correcting systematic errors in GFS data, for example). I have just done some experiments, using Duggleby as the site. I find that as the upper air bias correction goes from -1 to +2, lifted indices become more negative by about 2 degrees at the beginning of a forecast, and by as much as 7 degrees after about 4 days. The difference does seem to increase with time fairly consistently, as the “corrected” GFS data gets more entrenched into the forecast.
I also checked to see about how much this might affect thunder forecasts. Based on LI, at least, this difference of 7 degrees could make a difference in thunder indices of almost three categories.
This is the whole range I’m talking about here, but some of you have probably pushed this correction to one end of the other, away from the default of 0.5 which I generally have set. If you set it to 2, this would generally lower LI by 1 in the first few hours of the forecast, increasing to 3 or maybe even 4 degrees late in a several-day forecast. This would cause thunder forecasts to be slightly overdone at the very beginning, and then by almost 2 categories late in the forecast. That would necessitate toning down the thunder just about all the way currently allowed.
If you raised the GFS bias in an effort to correct surface temperatures, you might consider returning that to the default value, and looking at other ideas for surface temperature adjustment. The upper air temperature bias was not intended for adjusting surfae temperatures.
The reasons for all this are complex, but my experiments to confirm that changing that bias factor plays havoc with the thunder forecasts.
Please let me know if you have indeed been using a significantly different setting there from the default, and what you had found about thunder forecasts.
I was using Upper Temperatures value 2, which is the maximum. I used it because my high temperatures were too low but thanks to Tom :), I was able to raise my high temperatures without Upper Temperatures so I can use the default (0.5) value now. As Tom expected, it had a huge impact to my thunder forecast. It lowered my Severe Index by approximately two. If I now use -2 sensitivity setting, it’s just too low and I need to raise clearly :).
I did some testing with the thunder sensitivity setting and actually the default value (0) is quite good now, maybe just a little bit too aggressive. I think the best value is around -0.5.
This was very good notice Tom because I think that other people have also too big Upper Temperatures value.
Hi,
very interesting analisys from Tom. I’m running now wxsim with sensivity to zero and gfs upper bias to zero (instead of 0,5 default value). My first impression is that things going better.
But the question is, this historycal default GFS adjustment setted by Tom a long time ago may be obsolete now?
So is better to set all the bias GFS setting parameters to zero and watch what will happens?
Sante
I have just checked my upper bias and found it to be 0.5 (default) but I still need the -1.9 set for convective sensitivity in order to get rid of excessive thunder forecasts.
Could you tell me your values for the six various indices (Boyden, Lifted, Showatler, Total Totals, K, and KO) at a time when storms are being overforecast? You could even list a day or so of them, from lastret.txt, after specifically choosing these items and doing a manual View Text (you could also just send me a .wxf file). I’d like to see which of these is the biggest ‘culprit’ in the overprediction.
I’ve been reading an extensive article on the relative merits of these indices in different places, in addition to the articles I’ve already read. It’s from 1988, but is very comprehensive:
Tom when I get a chance I’ll have a play after putting back the convective sensitivity to normal. As an aside I have found that a sensitivity of -1.7 seems to predict ‘some thunder possible’ for Saturday which is about right and no thunder later in the forecast which is also probably correct when looking at UKMO.
