Using The Models To Predict Rainfall?

[johnholmes]

As the pdf will show I have been running intermittent checks on GFS for some time and have recently added the NAE output.

It has to be said that the results so far are not very impressive, whether the time scale is T+72 or more, or down to T+6 as this morning showed.

The data I show is relevant to my area alone. You may well find different results for your area.

As I mention at the end it would be useful if others would start checks for their areas.

Rainfall forecasts by computer models 1 june 2010.doc

[reef]

They've been very poor here during May. There have been numerous occasions where at least a few mm have been predicted only for hardly anything to materialise.

Today is another example. What was forecast widely across the Models/Metoffice/BBC as some quite heavy rain has given nothing measurable. In fact it has only just about wet the ground.

[johnholmes]

update for here

its still drizzling and the radar still shows echoes around so I may get more but so far its a total of 0.6mm, even with the correction its way below what even the 06z, less than 3 hours from the event showed on both GFS and NAE.

As I commented in the pdf I am really surprised at NAE being so far out.

[The PIT]

update for here

its still drizzling and the radar still shows echoes around so I may get more but so far its a total of 0.6mm, even with the correction its way below what even the 06z, less than 3 hours from the event showed on both GFS and NAE.

As I commented in the pdf I am really surprised at NAE being so far out.

Rainfall predictions have been very poor for a while and there's a long way to go before they get it right. Todays heavy rain has produced 0.2mm so far. Lunchtime forecast was still embarrassingly going on about heavy rain.

One thing that really does effect Sheffield are the hills and it's amazing how many times heavy rain just peters out as soon as it gets over the moors.

[johnholmes]

the shelter is well understood and when making forecasts in the days prior to computers this was taken into account.

That does not seem to be the case now, TV Met O forecasters, let alone weather presenters, have to 'toe the party line'. Fair enough but for the life of me I cannot see how with low and middle level winds from a S-W direction the model cannot be programmed to take this into account.

When a low creates low level flow from east of south, especially for several hours, then falls can be much as the model predicts.

[kold weather]

I know GFS has got an issue with being overkeen in very light rain solutions, I can't quite remember all the details but the bias was generally to do with over-extending the area of say 0.5-1mm an hour, its quite possible the NAE may also have this issue but I don't think its quite as bad as the GFS, which does it quite often in frontal set-ups.

[johnholmes]

update on my rain

the most fell after 18z, presumably on the occluded front which had become part of an extended trough that neither model seemed to have right.

total rainfall was 1.6mm so with the correction 2.1mm so still 0.9mm below the lowest either model predicted; 30% different. At worst, with 8mm then that figure falls to 11%, the forecast at 06z was worse with both models than 24 hours earlier.

Noted your point KW, how long ago was that do you know please?

Again it still fails to explain the NAE problem.

But its only 3 events, I'll monitor each as often as I can.

[Backtrack]

It seems the models are good when it comes to a widespread event, but precipitation on the lighter side, is harder to predict.

It would also seems they do much better on snowfall amounts, than rainfall amounts.

Just my opinion

[johnholmes]

I'm starting a check for the Sunday/Monday period, well started yesterday.

this will be Extra from T+72 with NAE joining at T+48 both down to T+12

just for my location

also just using 1 run, the 12z until T+24 when 00, 06 and 12z will be used from both

[CLH the red dragon]

I'm starting a check for the Sunday/Monday period, well started yesterday.

this will be Extra from T+72 with NAE joining at T+48 both down to T+12

just for my location

also just using 1 run, the 12z until T+24 when 00, 06 and 12z will be used from both

Could I be so bold as to ask about Sunday in Wales. Lots of the sites show as v light precip. Doing a 117 mile bike ride.

Previous experience of such forecasts leads me to believe a passing shower may appear but that it will remain dry.

Thoughts?

[kold weather]

It seems the models are good when it comes to a widespread event, but precipitation on the lighter side, is harder to predict.

It would also seems they do much better on snowfall amounts, than rainfall amounts.

Just my opinion

That would somewhat also agree with the idea that the models do overdo very light precip levels, the GFS has a proven bias with that on the NOAA/NCEP page (I'll hunt it down because it has other biases that were found to occur in the US which can be applicable here, including convective feedback issues.) and I see no reason why it won't affect over models to some degree.

[The PIT]

It'll be interesting too see what the results are over this weekend.

[johnholmes]

for the two models ideas see my last post in the model discussion thread, I'll monitor what I get and adjust as I always do for the shelter/problems here.

[johnholmes]

Well this morning and the 00z runs show NAE and Extra even further apart.

NAE suggests 19mm by 00z tonight with GFS/Extra on 7mm.

NAE showed 5mm by 06z I had 1.8mm so say 2.4mm which is still way below its forecast.

GFS/Extra gave 1mm between 06 and 09z, nothing to 06z.

I will show the full set of data tomorrow, from T+72 down to T+6 and what I recorded in that forecast period.

[johnholmes]

Just a quick update now with the full update tomorrow sometime.

I set off at 0900 for Bakewell to lead a walk with 1.8mm showing on my rainguage. Along the M1 one of the heaviest downpours, no lightning seen, developed, visibility 200-300m and much much less at times due to spray, it lasted about 15-20 minutes and I assumed would head over this area. It seems it did with my next door neighbour saying how hard it rained, although my weather station shows the heaviest rate was in the early hours in under another cell. The rainguage now reads 11.6mm, so add 25-30%, say 15-16mm and the latter stages of NAE showing 19mm with Extra 15mm turned out just about spot on. However the earlier runs were nothing like as good.

Full update and I hope some thoughtful comment tomorrow.

[reef]

Interesting figures John, in my case here both Extra and the NMM actually under-read by some way today.

The NMM predicted 8mm in 24 hours to 11pm today and NW Extra rainfall accumulation is showing around 9mm during the last 24 hours. In reality we have had 12.8mm which means both were around 30% too low.

[johnholmes]

Checking how GFS/Extra and NAE performed for the weekend of 5-6 June 2010

(for my area)

GFS at T+72-96 showed less than 1mm for the Saturday to Sunday night

At T+48-72 it gave 11mm

By T+24-48 Extra showed 15mm with NAE (valid out to 12z Sunday) giving 3mm

The T+12-24 showed GFS (did not use Extra) with 10-15mm and NAE up to 25mm by 12z Sunday

The final run before events was the 00z on Sunday with Extra showing 7mm and NAE 19mm by midnight Sun/Mon.

The exact amounts were

Extra

1mm to 09

1 mm to 12

3 mm to 15

2 mm to 18

NAE

5mm to 06

12mm to12

18mm to 18

19 mm to 00

1.8mm fell before 06z and 10.0mm from 09-24z, both figures uncorrected.

The set up was not an easy one for the models. A destabilising system being pushed east by an advancing upper trough with the flow largely coming from the south as it developed. For whatever reason all models struggle with this set up. Trying to predict any particular unstable cell moving over the area, intensity and timing is difficult at times scales much beyond what we can see on the radar.

That said GFS/Extra by T+48 gave a reasonable indication and this got nearer what actually fell here down to the T+12-24. I’m not sure why its 00z output, only 9 hours away from the main cell was so poor.

NAE did much better on this time scale as was its T+12-24 issue. The T+48 was at the extreme range of its time scale, only going out to Sunday 12z but that issue was poor.

Two main cells affected this area, one late Saturday into the early hours of Sunday. Another one during Sunday morning with a much more minor cell in the evening.

Overall, starting from a rather cynical view I feel both models in their time scales with a very difficult situation did fairly well. Taking GFS/Extra as a base for early indications, say from T+48 it gave reasonable guidance. It does show that at greater time scales beyond T+48 amounts are very much less likely to be correct. Something my data shows over the past 2-3 years. The ‘score’ rate beyond T+48 is about 4/10.

NAE, as it should, was better at closer time scales.

Thus, in this case, using a mix of models one was given a reasonable idea of both timing and intensity/amounts for this area.

It was as said above a destabilising situation rather than a warm frontal approach. I will keep my eyes open for other systems on similar time scales, unstable types and more usual warm frontal approaches, especially when the upper flow ahead of the developing area is shown as southerly.

[johnholmes]

Local rainfall today, Tuesday 26 November 2010

I’ve recorded 2.3mm which correct gives about 3mm.

Once again the various models at quite close time frames had very variable outputs.

12z Monday

Extra=7mm

GFS=3 south to 18 north (my spot location is mid way between the two)

NAE=5mm

NMM=2mm

00z Tuesday

Extra=16mm

GFS=10 (s) 12 (n)

NAE=18mm

NMM=12mm

Why the predicted higher totals from the 00z run? I simply cannot understand that as by the 06z radar outputs there was nothing even approaching that pattern. Looking back at the Net Wx radar from 00z that too suggested nothing like some of the higher totals the models went for.

Again it’s a front moving in from the west. Peak and Pennine shelter?

I try to be positive about model outputs but continue to be totally baffled as to why all the models, at times, and close in, are so far out with the rainfall totals.

The 12z Monday is far superior and quite close to what I recorded (with the correction) and other places around me that report.

Scampton and Waddington=6mm in the 12 hour period to 1700z

Cranwell=5mm

Coningsby=4mm

Wainfleet=3mm

Nottingham=2mm

These are the nearest stations I can find to me.