cc: Phil Jones date: Tue, 18 Mar 2008 10:41:03 +0000 from: Ian Harris subject: Re: further CRUTS2.1 vs 3.0 comparisons to: Tim Osborn On 18 Mar 2008, at 10:25, Tim Osborn wrote: > Hope this is useful and looks ok -- I'd like to pass on the TMP > data to the QUEST partners to try to appease them... Phil and Harry > do you have confidence that TMP is ready for this? > > Harry -- I'd now like to also check DTR, TMN and TMX... I see there > are files in the same place (/cru/cruts/version_3_0/primaries/) for > these variables too... are these files ready for me to use? VAP > isn't there, but perhaps it is ready in .../secondaries/, but that > isn't readable by me. If VAP is in there and ready to use then > please change the permisssions. T-related parameters are 'ready', station data notwithstanding. I wouldn't look at secondaries until we've solved the precip problem ;-) > Finally, have you found the factor-of-10 problem yet with the > creation of the PRE data? My results attached here do indicate > that the problem is likely to be as simple as that, so presumably > easy to find just by inspecting a few of the intermediate > calculations? And then WETDAYS can be recalculated and made > available to me too. How long? Looking now. Not simple at all. Nonstandard (to my eyes) percentage anomalising at the beginning! And reverse engineering that at the end apparently hasn't worked, nor did regular 'de-anomalising' before that. To show you what I mean (from the CRUTS READ_ME diary): > I started off using a 'conventional' calculation: > > absgrid(ilon(i),ilat(i)) = nint(normals(i,imo) + > * anoms(ilon(i),ilat > (i)) * normals(i,imo) / 100) > which is: V = N + AN/100 > > This was shown to be delivering unrealistic values, so I went back > to anomdtb to see how the anomalies were contructed in the first > place, and found this: > > DataA(XAYear,XMonth,XAStn) = nint(1000.0*((real(DataA > (XAYear,XMonth,XAStn)) / & > real(NormMean(XMonth,XAStn)))-1.0)) > which is: A = 1000((V/N)-1) > > So, I reverse engineered that to get this: V = N(A+1000)/1000 > > And that is apparently also delivering incorrect values. You see, knowing the conventional approach doesn't help here, at least I don't think so. I am now checking through the intermediate stages and programs with just one sample year (1980). Looking at embedded scale factors, missing value codes, etc. I want it done today. Cheers Harry Ian "Harry" Harris Climatic Research Unit School of Environmental Sciences University of East Anglia Norwich NR4 7TJ United Kingdom