From: P.Jones@uea.ac.uk To: "C G Kilsby" Subject: RE: Outputs from WG Date: Mon, 16 Apr 2007 13:55:37 +0100 (BST) Cc: david.sexton@metoffice.gov.uk, "Phil Jones" , "Colin Harpham" , "H J Fowler" Chris et al, I'll sedn some more thoughts on Thursday when back from the EGU. It is too hot in Vienna to sit through too many talks ! I suspect we need a subset of indices. The program will calculate all those recommended in various programs. One possibility is to keep them all and let users decide. We do need to make a series of checks though at some stage to make sure they are OK. I think you'll have some fruitful discussions on some of these on April 24. I hope you can come to closure on a few things. Cheers Phil > All: > > Indices > > I had a session with UKCIP last week, and we did get on to dicsussing > what outputs might come out of WG (as well as DDP etc.) and the issue of > indices derived from daily data (i.e. requiring time series) came up, > with the distinct possibility of confusion/inconsistency as David > mentions! > > I would be happy to produce indices only from WG, as long as we can > check they are sensible first of course! > E.g. heatwave duration (various thresholds), drought duration, various > accumulations of rainfall ? > Less clear cut might be gale days (definition?), snow days, proportion > of days above temp threshold etc. > > I think we will need to consider the list in detail, as far as what is > included (STARDEX list?), how they are calculated/validated and also > whether they can be calcualted from some other source and found to be > inconsistent. > E.g. is it planned to take the (17?) RCM runs and analyse/release these > indices as well ? > > > Rainfall stats - pdfs > > I think (hope?) lag1-ac and skewness will actually be quite well behaved > (if not realistic) even when you convert/downscale. The more > validation/analysis we do of these fields the better anyway. > > > Separate topic: measures of reliability > > May be a can of worms, but I think we need to address it sooner rather > than later: UKCIP02 had subjective measures of reliability attached to > different variables/predicted changes. We must do better, and a case in > point is the WG where we sidestep the bias issue by using change > factors. We therefore need to provide some measure (per grid square, per > varaible?) of reliability. > > For example: if control annual rainfall is more than (say) 10% biased, > reduce reliability measure and inform the user when generating. > Problem 1: which model runs to use for this check? > Problem 2: how to assess more complex measures e.g. annual cycle in > rainfall/temperature? > Problem 3: need a common, easily understood scale of reliability > Furthermore - WG procedure introduces more uncertinty, e.g. for wind > > Thoughts? > > Cheers, Chris > > > >>-----Original Message----- >>From: david.sexton@metoffice.gov.uk >>[mailto:david.sexton@metoffice.gov.uk] >>Sent: 16 April 2007 08:07 >>To: Phil Jones >>Cc: david.sexton@metoffice.gov.uk; C G Kilsby; Colin Harpham >>Subject: RE: Outputs from WG >> >>Hi, >> >>we will try for lag-1 correlation and skewness but an issue >>for us is whether something doesn't work when we convert the >>equilibrium pdfs to time-dependent ones or we downscale to 25km. >> >>As Phil has said that you can do all the derived indices >>except gale days, if we could get a decision from the project >>management team to cut those variables from MOHC list of >>outputs without making any extra work for you, then that would >>free up some time for us to investigate this further. >> >>Looking forward to seeing Colin's results on 24th. >> >>Cheers, David >> >> >> >> >>On Fri, 2007-04-13 at 17:16 +0100, Phil Jones wrote: >>> Some more thoughts - keep in on the loop in case i get a chance >>> to respond from Vienna or next Thursday. >>> >>> Phil >>> >>> >>> At 16:32 13/04/2007, david.sexton@metoffice.gov.uk wrote: >>> >Hi, >>> > >>> >On Fri, 2007-04-13 at 16:00 +0100, C G Kilsby wrote: >>> > > Phil, David >>> > > >>> > > Briefly, and can respond fully next week when I have >>some more time! >>> > > >>> > > Some crucial points here, >>> > > 1. the one re 90%ile of one variable not same as for >>other variables. >>> > > Some simple restrictions need considering before diving off into >>> > > full joint pdfs etc. >>> > > Also, another dimension emerges with seasons, e.g. 90%ile winter >>> > > rainfall, or 90%ile summer rainfall? >>> > >>> >Joint pdfs are just an issue for me in that I am giving you several >>> >inputs to WG and they have to be consistent. For example, we are >>> >finding we only get wetter summers for lower end of temperature >>> >increases. Plus we already intend to provide sets of sampled values >>> >for lots of variables that are consistent for any given point in >>> >model parameter space. >>> >>> The joint pdfs are an issue for the WG as well. Not so much for >>> Chris, but for us we have to reproduce the statistics for >>> the other variables. Colin >>> has solved the double counting issue for the means (for T etc), >>> but we've yet to look at the variance. >>> >>> Colin should be able to show some of the results on the 24th >>> as to how well the WG works. This fits the WG (with our rainfall >>> component) to HadRM3 and then applies our modification >>> technique to an A2 future (for comparison with the true RCM >>> future for the 2070s). Sunshine is the only real problem. >>> >>> I don't think we need to repeat this with the NS rainfall, >>> but discuss that once you've seen some preliminary results >>> on the 24tjh. >>> >>> >>> > > >>> > > 2. Bit concerned to hear David talking of some precip >>stats being >>> > > secondary or optional - I would say mean, var and pdry days are >>> > > all >>> > > essential: from our experience autocorrelation and skewness are >>> > > also pretty well behaved and we would rather have them >>if at all possible! >>> > >>> > >>> >Good. This discussion is throwing up a few discrepancies which need >>> >clarifying. That some precip stats are of secondary >>importance, is an >>> >impression I was getting from Phil's earlier emails last month. >>> >>> >>> I think there is some misunderstanding here. What I said earlier >>> confirms what Chris has said - if they are available then Chris >>> would like them. Chris will need to consider is they may be >>> fully relevant due to the scale issue (25km squares vs points). >>> Could be an issue for skew and r1. >>> >>> Checking this out a la fitting directly to HadRCM3 control >>> data might be useful here. See Colin's plots though before >>> deciding. >>> >>> >>> >>> >I look forward to the fuller response next week. I will be mainly >>> >away then which is why I raise these issues now. It would >>be good to >>> >have a good chat about them on the 24th. >>> > >>> >Cheers, David >>> > >>> > > >>> > > >>> > > Cheers, >>> > > Chris >>> > > >>> > > >-----Original Message----- >>> > > >From: Phil Jones [mailto:p.jones@uea.ac.uk] >>> > > >Sent: 13 April 2007 15:46 >>> > > >To: david.sexton@metoffice.gov.uk >>> > > >Cc: david.sexton@metoffice.gov.uk; C G Kilsby; Colin Harpham >>> > > >Subject: Re: Outputs from WG >>> > > > >>> > > > >>> > > > David, >>> > > > More thoughts embedded. >>> > > > >>> > > > Phil >>> > > > >>> > > >At 15:12 13/04/2007, david.sexton@metoffice.gov.uk wrote: >>> > > >>Hi, >>> > > >> >>> > > >>I think we have clarified or converged on most of my points. I >>> > > >>have some comments on points 2 and 4. >>> > > >> >>> > > >>Cheers, David >>> > > >> >>> > > >> >>> > > >> >>> > > >> >>> > > >> >>> > > >>On Fri, 2007-04-13 at 14:42 +0100, Phil Jones wrote: >>> > > >> > >2. WG will produce 100 versions of 30-yr sequences for >>> > > >all (or just >>> > > >> > >one?) WG variables for all months for a given combination >>> > > >of 30-yr >>> > > >> > >period, emissions scenario and location. >>> > > >> > > >>> > > >> > >I am still not clear how to generate the 100. >>Percentiles of >>> > > >> > >PDFs is confusing me. I think Ag needs a clear procedure >>> > > >outlined by us >>> > > >> > >for 24th. I think the easiest way to make WG >>consistent with >>> > > >> > >MOHC pdfs is the following (assuming I am correct so far): >>> > > >> > > >>> > > >> > > a. User selects WG, 30-yr period, emissions >>scenario and >>> > > >> > > location >>> > > >> > (up >>> > > >> > >to 1000km^2). >>> > > >> > > b. Work out which 25km x 25km box over UK is closest to >>> > > >> > >this >>> > > >> > multi- >>> > > >> > >site location. >>> > > >> > > c. For the 30-yr period, emissions scenario and >>location >>> > > >> > >in b), >>> > > >> > DDP >>> > > >> > >internally produces a table of changes in mean T, % >>> > > >changes in mean >>> > > >> > P, >>> > > >> > >and changes in variance of P for each month for >>100 randomly >>> > > >> > >sampled different model variants. DDP ALREADY needs this >>> > > >capability. >>> > > >> > > d. So we have an internal matrix with 3*12=36 >>columns and >>> > > >> > > 100 >>> > > >> > rows. WG >>> > > >> > >loops through 100 rows, using each set of 36 >>numbers to drive WG. >>> > > >> > User >>> > > >> > >gets 100 WG's. Does what they like with it. >>> > > >> > >>> > > >> > >>> > > >> > Sort of. The 100 versions of the WG I was talking >>> > > >about will all >>> > > >> > have >>> > > >> > the same statistics. I thought these 100 would be from >>> > > >one point >>> > > >> > within >>> > > >> > the pdf (or the joint pdf) - say the 10, 50 or 90th >>> > > >percentile. We >>> > > >> > could make >>> > > >> > this percentile selectable. >>> > > >> > >>> > > >> > The 100 (or 1 or whatever) are representative of some >>> > > >> > future 30-year period. >>> > > >> > Your a) and b) are fine. >>> > > >> > >>> > > >> > Another option is like yours. There is a pdf >>(or joint pdf). >>> > > >> > The 100 could be >>> > > >> > from each of the 100 percentiles? Does this make sense? >>> > > >Or the 100 >>> > > >> > could >>> > > >> > come from sampling the percentile space assuming a normal >>> > > >> > distribution? >>> > > >> > >>> > > >> > Your 2) is an important aspect to sort out on the 24th. >>> > > >> >>> > > >> >>> > > >>I agree that we need to discuss this but it would be good to >>> > > >thrash it >>> > > >>out a bit more before 24th. UKCIP08 needs the WG pdf to be >>> > > >>consistent with the MOHC pdf. Your solution tries to >>do this but >>> > > >>a problem with selecting a percentile is that a model variant >>> > > >>that is the 90th percentile for temperature is not >>90th percentile for other variables. >>> > > >>There is also a related issue about how you chose a model >>> > > >variant near >>> > > >>a given percentile. The solution I propose means these are >>> > > >not issues. >>> > > >>So we could sample M model variants and run N WGs for >>each model >>> > > >>variant. M has to be a good size to make sample >>> > > >representative of MOHC >>> > > >>pdf but N does not have to be large as internal variability >>> > > >is already >>> > > >>generated by using a different set of parameters and a >>> > > >different seed for each WG. >>> > > >>I think this solution is simpler than the percentile-based >>> > > >solution. Do >>> > > >>you agree? >>> > > > >>> > > > >>> > > > Sounds OK. Let's see what Chris thinks. >>> > > > >>> > > > >>> > > > >>> > > > >>> > > >> > >4. Phil has mentioned in the past that EARWIG produces some >>> > > >> > diagnostics >>> > > >> > >e.g. consecutive dry days, frost days etc. from WG. Will >>> > > >> > >this be done for UKCIP08? >>> > > >> > >>> > > >> > >>> > > >> > The plan is yes for this. Colin has the software >>for this. >>> > > >> > It just needs to be set >>> > > >> > up carefully, as the base for all the diagnostics >>(for the >>> > > >> > future >>> > > >> > runs) has to be >>> > > >> > based on median run of the WG for the present (61-90). >>> > > >We shouldn't >>> > > >> > allow users to change the 61-90 base period (or the >>> > > >choice of the >>> > > >> > median). >>> > > >> > >>> > > >> > >>> > > >> > >>> > > >> >>> > > >>Good. I would like your opinion on a problem I am having with >>> > > >>some of the variables we are providing pdfs for. Some >>quantities >>> > > >>are indices derived from daily model data e.g frost days but I >>> > > >>think >>> > > >there are two >>> > > >>problems with this: >>> > > >> >>> > > >>1. Model bias e.g. a model that is too warm may have very few >>> > > >>frost days and therefore the change looks small. >>Effect will be >>> > > >>a nonlinear function of bias based on shape of >>distribution of daily data. >>> > > >> >>> > > >>2. WG and pdfs could provide two alternative routes to same >>> > > >answer and >>> > > >>they will obviously conflict for reasons we understand e.g. >>> > > >model bias >>> > > >>but the users won't understand. >>> > > >> >>> > > >>To avoid confusing user and potentially reducing their >>> > > >>confidence in UKCIP products, I think it makes sense >>for WGs to >>> > > >>be the sole route towards a prediction of derived >>indices. BTW, >>> > > >>I have a handful of derived indices to do (hot days, wet days, >>> > > >>gale days, heating and cooling degree days and frost >>days) and I >>> > > >>think you cover >>> > > >some of these >>> > > >>already. What do you think? >>> > > >> >>> > > >>Geoff wants to discuss issues connected to the three strands >>> > > >of output >>> > > >>(pdfs, WG, RCM) on the 24th. >>> > > > >>> > > > >>> > > > Model biases will only be a problem with their data >>used directly. >>> > > > So this could be a problem with the larger regions >>where the WG >>> > > > won't work well. The WG won't have biases as it is based on >>> > > > 61-90 as the base period. We will be perturbing these >>with the >>> > > > RCM-based pdfs. >>> > > > >>> > > > Maybe we need to show that the following will/should/must be >>> > > > the same >>> > > > >>> > > > Model-based scenario for 2070s minus model present >>(61-90) equals >>> > > > WG scenarios for the 2070s minus WG present (61-90). >>> > > > >>> > > > Geoff will need to get this across as this is how the three >>> > > >strands will >>> > > > produce the same answers. >>> > > > >>> > > > The WG and the extremes software will do all the temp/precip >>> > > > indices but won't do gale days. >>> > > > >>> > > > >>> > > >>Cheers, David >>> > > >> >>> > > >> >>> > > >> >>> > > >> >>> > > >> >>> > > >>-- >>> > > >>______________________________________________________ >>> > > >>David Sexton PhD Climate Research Scientist Met Office >>> > > >Hadley Centre >>> > > >>for Climate Prediction and Research FitzRoy >>> > > >>Road Exeter EX1 3PB United Kingdom >>> > > >>Tel: +44 (0)1392 886524 Fax: +44 (0)1392 885681 >>> > > >>E-mail: david.sexton@metoffice.gov.uk >>http://www.metoffice.com >>> > > > >>> > > >Prof. Phil Jones >>> > > >Climatic Research Unit Telephone +44 (0) 1603 592090 >>> > > >School of Environmental Sciences Fax +44 (0) 1603 507784 >>> > > >University of East Anglia >>> > > >Norwich Email p.jones@uea.ac.uk >>> > > >NR4 7TJ >>> > > >UK >>> > > >--------------------------------------------------------------- >>> > > >------------- >>> > > > >>> > > > >>> > > > >>> >-- >>> >______________________________________________________ >>> >David Sexton PhD Climate Research Scientist Met Office Hadley >>> >Centre for Climate Prediction and Research FitzRoy >>> >Road Exeter EX1 3PB United Kingdom >>> >Tel: +44 (0)1392 886524 Fax: +44 (0)1392 885681 >>> >E-mail: david.sexton@metoffice.gov.uk http://www.metoffice.com >>> >>> Prof. Phil Jones >>> Climatic Research Unit Telephone +44 (0) 1603 592090 >>> School of Environmental Sciences Fax +44 (0) 1603 507784 >>> University of East Anglia >>> Norwich Email p.jones@uea.ac.uk >>> NR4 7TJ >>> UK >>> >>--------------------------------------------------------------- >>------------- >> >>-- >>______________________________________________________ >>David Sexton PhD Climate Research Scientist Met Office >>Hadley Centre for Climate Prediction and Research FitzRoy >>Road Exeter EX1 3PB United Kingdom >>Tel: +44 (0)1392 886524 Fax: +44 (0)1392 885681 >>E-mail: david.sexton@metoffice.gov.uk http://www.metoffice.com >> >