tag:blogger.com,1999:blog-7729093380675162051.post7421513049092837902..comments2024-03-16T02:27:38.423+11:00Comments on moyhu: GWPF is wrong, warming has not stopped.Nick Stokeshttp://www.blogger.com/profile/06377413236983002873noreply@blogger.comBlogger127125tag:blogger.com,1999:blog-7729093380675162051.post-80620909140267901312011-11-16T07:35:17.090+11:002011-11-16T07:35:17.090+11:00KevinC, I think this is the underlying problem too...KevinC, I think this is the underlying problem too.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-52199802548351139262011-11-16T05:06:30.138+11:002011-11-16T05:06:30.138+11:00Retrying ...
Kevin C,
I used 1950-1979 as the co...Retrying ...<br /><br />Kevin C,<br /><br />I used 1950-1979 as the common baseline, as this is what was used in the BEST summary data set (not 1950-1980). <br /><br />This brings up GISS imperceptibly, and CRUTEM quite a bit (as expected).Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-14577336466660198482011-11-16T02:50:16.527+11:002011-11-16T02:50:16.527+11:00Deep:
Have you fitted the baseline of all the line...Deep:<br />Have you fitted the baseline of all the lines on 1951-1980? (By mean, not by eye?) I managed to fool myself until I did this.<br /><br />Carrick:<br />I think I may have figured out why CCC doesn't reproduce the GISS-land result (either the one we have, or the one used in the Berkeley plot). It is not sufficient to simply exclude cells which are not land based when calculating the average. You also have to weight land-containing cells according to the percentage land in the 1200km circle. I'm working on my own quick-and-dirty reimplementation of GISS to see if I'm right.<br /><br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-24157373822982397812011-11-15T13:04:44.263+11:002011-11-15T13:04:44.263+11:00Here is my emulation of the BEST chart using publi...Here is my emulation of the BEST chart using published data sets.<br /><br />The CRUTEM nhsh (dotted red line) matches theirs, but the "simple average" (solid red line) comes much closer.<br /><br />GISS Land (solid blue) is much closer than GISS ts used in the original BEST fig 7, but still does not correspond to what BEST showed in their update. <br /><br />http://deepclimate.files.wordpress.com/2011/11/best-comparison-decadal-1950-2009-w-cru-giss-land.jpg<br /><br />I'd say that's a question for your GISS contact: What exactly did they provide BEST, and is that data set published?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-52441744221768377732011-11-13T00:35:14.775+11:002011-11-13T00:35:14.775+11:00OK, here's some more to go on. A step in the 1...OK, here's some more to go on. A step in the 10 year running mean means there must be a delta-function in the values - offset from the step by the length of the averaging window. So the step drop in CRU in 1970 must be due to a big difference between BEST and and CRU in ~1980.<br />And indeed there is. Take the baselined 12 month running means and difference them. CRU is about 0.25 deg below BEST from 1980/1 to 1981/7.<br />The 12 month mean of the difference between the two shows some interesting features. I guess the next step is to look at the grids for the 1980 feature and see if there is anything geographical about it.<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-369865464376678852011-11-12T06:04:54.084+11:002011-11-12T06:04:54.084+11:00Carrick
"I suspect there is more involved wi...Carrick<br /><br />"I suspect there is more involved with GISTEMP's new land only reconstruction than just a land mask. (I suspect they modified their code to more closely emulate BEST's algorithm.)"<br /><br />If I were hansen I would replace his silly reference station method with Tamino's method!<br /><br />It puts it on a more sound statistical footing, you bring more data in. It's an obvious improvement. In fact, when I get time I will do Tamino's world! I will take his algorithm, grid the planet and use his method on every grid cellstevenhttps://www.blogger.com/profile/06920897530071011399noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-42541490794894615742011-11-12T05:58:57.917+11:002011-11-12T05:58:57.917+11:00"Solved it! It's this problem again...
ht..."Solved it! It's this problem again...<br />http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch3s3-2-2.html<br />CRUTEM3 is not a global mean, it's the mean of the land averages for the two hemispheres. By using the weighted sum of the hemispheres, or better, by doing a global average over all the cells, CRUTEM3 is brought back into agreement with BEST and the other land only products.<br />Kevin C"<br /><br />Yup. This same issue bit me in the ass when I tried to match their SST figures. I wrote them and they explained that their global SST is an average of NH and SHstevenhttps://www.blogger.com/profile/06920897530071011399noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-84490339646907050682011-11-12T04:39:47.603+11:002011-11-12T04:39:47.603+11:00Well, to a layperson like me what is odd is it loo...Well, to a layperson like me what is odd is it looks like BEST is showing something is rotten in England, only this time it's not the food.<br /><br />Loos like graphing this is a good exercise?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-30472160412084163942011-11-12T03:54:50.861+11:002011-11-12T03:54:50.861+11:00OK, here's another attempt at reproducing the ...OK, here's another attempt at reproducing the comparison graph. The weighted average of hemispheres is definitely what they have done with the CRUTEM3 data, using the formula given in AR4. I've omitted the GISS line completely, given that it's wrong.<br />Can you see any more gotchas with this version?<br /><a href="http://postimage.org/image/jbljqnq4h/" rel="nofollow">[GRAPH HERE]</a><br />NOAA now agrees very nicely with BEST, except for a startling excursion from 1970-1980.<br />CRUTEM3 agrees nicely to 1970, and then takes a downward step. Offsetting the CRU series upwards brings 1970-2000 into line, and then it starts sagging. Something is very odd!<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-22841708811856728782011-11-12T03:28:52.461+11:002011-11-12T03:28:52.461+11:00Yes, you're right, it's still wrong.
I thi...Yes, you're right, it's still wrong.<br />I think the BEST and NOAA lines are right. The GISS line is completely wrong - we still don't have the right data for this.<br />Once the baseline is fixed, the CRUTEM3 line is close, but not right, especially in the 19thC. I'm guessing they used the weighted average of hemispheres rather than global cell average. I'll have another go at reproducing their comparison.<br />I'm making a right mess of this!Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-19691113476672124952011-11-12T01:39:12.347+11:002011-11-12T01:39:12.347+11:00KevinC, I noticed the series don't agree with ...KevinC, I noticed the series don't agree with each other very well before 1970. Even NCDC (NOAA) and BEST, which use the same exact underlying data set currently,<br /><br />As I recall, there isn't that much difference in the number of stations between 1950-2000. So I wonder what gives.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-84566238347431842382011-11-11T22:48:12.046+11:002011-11-11T22:48:12.046+11:00JCH: Here you go.
[GRAPH HERE]
The exact appearanc...JCH: Here you go.<br /><a href="http://postimage.org/image/86y3q2izj/" rel="nofollow">[GRAPH HERE]</a><br />The exact appearance depends on the baseline because there seems to be a little bit of drift. I did it by eye on 1980-1990.<br />You can see a bit of a sag in CRUTEM3 at the end, but it's nothing compared to the BEST version of this plot.<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-55644620454654438042011-11-11T04:57:15.270+11:002011-11-11T04:57:15.270+11:00KevinC: Yes, weighting in latitude bands sounds li...KevinC: <i>Yes, weighting in latitude bands sounds like a good idea to me. It certainly makes far more sense than calculating hemispheres and then averaging those.</i><br /><br />I think the argument is, hemispherical averaging reduces some of the bias associated with most of the land-mass being in the Northern hemisphere.<br /><br />I would argue hat these "land-only" <i>global</i>reconstructions don't mean very much because of the many different ways it can get measured. There isn't a "land-only" Earth, so it's an artificial construct.<br /><br />(That said, I think it makes sense to divide it into land-only cells, and even talk about which variables are more explanatory of the difference in trend between cells, such as latitude and elevation, as is claimed by BEST.)<br /><br />KevinC: <i>Nick's area weighting also handles this nicely.</i><br /><br />I think NCDC may also, but since they haven't released their code AFAIK, it is difficult to see what they are actually doing (as opposed to the explanation they offer for what they are doing).Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-53275388971178233232011-11-11T04:37:44.757+11:002011-11-11T04:37:44.757+11:00Kevin C - can you do a graph with your new result ...Kevin C - can you do a graph with your new result that has NOAA, GisTemp, CRU, and BEST side by side. I know what agreement looks like, but I want to see it, if possible. Thanks in advance.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-52388296123677936002011-11-11T03:24:32.323+11:002011-11-11T03:24:32.323+11:00Yes, weighting in latitude bands sounds like a goo...Yes, weighting in latitude bands sounds like a good idea to me. It certainly makes far more sense than calculating hemispheres and then averaging those.<br /><br />Nick's area weighting also handles this nicely.<br /><br />As a check, I had a look at the HADCRUT data and compared global and hemispheric averaged results. The differences were minimal since 1920. For the land-ocean products, the number of empty cells is not the dominant factor.<br /><br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-21417162248130603142011-11-11T02:25:34.994+11:002011-11-11T02:25:34.994+11:00It was just easier to write this out as a LaTeX do...It was just easier to write this out as a LaTeX document than try and type it in here.<br /><br /><a href="http://dl.dropbox.com/u/4520911/Climate/land_average.pdf" rel="nofollow">LINK.</a><br /><br />I welcome comments and corrections of course...Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-41447219507044519822011-11-11T02:18:54.340+11:002011-11-11T02:18:54.340+11:00Actually for my method "B" it looks like...Actually for my method "B" it looks like you need a normalization factor:<br /><br />sum over latitude weighted by cos(latitude)) * total_land_cells(latitude)/totaL_cells(latitude)/norm<br /><br />where norm = sum over cos(latitude) * total_land_cells(latitude)/totaL_cells(latitude)Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-2677975181115711232011-11-11T01:44:15.474+11:002011-11-11T01:44:15.474+11:00KevinC, yep you have to be very careful what you m...KevinC, yep you have to be very careful what you mean by land-only averages. Summing the north and southern hemispheres, some would argue is better, because it reduces the bias associated with most of the land-mass being in the Northern hemisphere.<br /><br />The bias effect I was referring to occurs when you have empty 5°x5° cells, and don't account for their potential effect on temperature. That's why prior to 1950 , when surface station coverage started to really get sparse, you end up with a bias towards high latitudes.<br /><br />Of course it's easy to confirm this (<a href="http://treesfortheforest.wordpress.com/2010/05/19/better-late-than-never/" rel="nofollow">Chad redid this calculation here.</a>)<br /><br />The problem with only averaging over nonempty cells and having a variation in the geographic centroid of the data <a href="http://dl.dropbox.com/u/4520911/Climate/temperature_trends.jpg" rel="nofollow">is because of the inherent greater response of high northern latitudes to climate change.</a><br /><br />The "best way" I can think of is to fill in missing cells using an extrapolation method from adjacent cells. (Linear interpolation for example.)<br /><br />The "second best" is to average over longitude first (sum of non-zero cells divided by number of none-zero cells) then average over latitude (sum over latitude weighted by cos(latitude)) * total_land_cells(latitude)/totaL_cells(latitude)<br /><br />To make it clear total_cells(latitude) = 360/5 = 72 if you have 5°x5° cells.<br /><br />total_land_cells(latitude)/totaL_cells(latitude) should look something like <a href="http://farm5.static.flickr.com/4018/4310158555_e6a9b9d824.jpg" rel="nofollow">the blue curve in this figure.</a><br /><br />Unfortunately I don't have the time to play with this, but it would be interesting to see whether you saw a difference for the data previous to 1950 using these different methods of accounting for empty grid cells.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-43699178540910524632011-11-11T00:24:20.895+11:002011-11-11T00:24:20.895+11:00Solved it! It's this problem again...
http://w...Solved it! It's this problem again...<br />http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch3s3-2-2.html<br />CRUTEM3 is not a global mean, it's the mean of the land averages for the two hemispheres. By using the weighted sum of the hemispheres, or better, by doing a global average over all the cells, CRUTEM3 is brought back into agreement with BEST and the other land only products.<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-10238743508161891872011-11-10T22:48:58.719+11:002011-11-10T22:48:58.719+11:00Kevin,
I have run TempLS on CRUTEM3 - I wrote a po...Kevin,<br />I have run TempLS on CRUTEM3 - I wrote a post <a href="http://moyhu.blogspot.com/2011/08/first-templs-crutem3-reconstructions.html" rel="nofollow">here</a>. I thought the agreement was fairly good. Steve Mosher in the comments gave a link to a similar analysis that he ahd done.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-17919498679632353722011-11-10T22:08:02.651+11:002011-11-10T22:08:02.651+11:00Clarifications:
My code works on GHCN or CRU:
- Ke...Clarifications:<br />My code works on GHCN or CRU:<br />- Kevin(GHCN) is my code with equal area averaging using GHCN3,<br />- Kevin(CRU) is my code with equal area averaging using CRU data.<br />- NASA(GHCN) is GIS-land,<br />- CRU(CRU) is CRUTEM3.<br />I can also run with equal angle averaging - i.e. using the same grid as CRU. But that doesn't explain the difference.<br />I don't do extrapolation. But I can get a very good match to GIS-land without it. So that doesn't explain the difference either.<br />That also suggests that GIS-land is an average over the land cells (because that is what I do), rather than equal-weighting the hemispheres, so I think I can confirm that the new GIS-land data has avoided that issue.<br />If I run my code in CRU mode on the CRU data, I get this:<br /><a href="http://postimage.org/image/b7lpao7qf/" rel="nofollow">[GRAPH HERE]</a><br />That's a big difference. I could be doing something stupid. But the fact that I can duplicate GIS-land and BEST so well suggests not. The alternative seems to be that something is wrong with the CRU method.<br />I'm digging on the problem. I suspect a lot of others are too, now that CRU is so obviously an outlier. I'm guessing we'll see an answer in a month or two.<br />Nick: You can run TempLS on CRU can't you? Do you see similar results?<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-27240794621012215752011-11-10T09:35:05.361+11:002011-11-10T09:35:05.361+11:00By the way, KevinC, which series are you calling G...By the way, KevinC, which series are you calling GHCN?Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-86485366938987931062011-11-10T09:28:14.801+11:002011-11-10T09:28:14.801+11:00Kevin C, IMO there is an issue with CRUTEM3 for da...Kevin C, IMO there is an issue with CRUTEM3 for data prior to 1950, due to the simplistic way they do the surface averaging. I believe it leads to an anomalous amount of warming, <a href="http://farm5.static.flickr.com/4011/4304398335_64f644e14b.jpg" rel="nofollow">which is just associated with a change in mean latitude</a> of the stations (this is geographically weighted by 5°x5° cells, btw):<br /><br /><a href="http://farm5.static.flickr.com/4007/4305143202_4125127e96.jpg" rel="nofollow">See this.</a><br /><br />The 7.5% bias is "real" ... it is associated with there being more land mass in the Northern hemisphere than in the Southern. Whether you have that bias in trend depends on how you define your "land-only" Earth (for example, if I understand it correctly--and I may not---the traditional GISTEMP method, I believe they weight the SH equally, so their numbers should be biased lower than CRUTEM3).<br /><br /><br />GISTEMP may not have it right—when you don't have data, extrapolation doesn't really solve that problem—but their central value for that period at least has a chance of being unbiased, even if more uncertain than it would have been were the data available.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-51454728463918217472011-11-10T01:34:34.608+11:002011-11-10T01:34:34.608+11:00Carrick:
I'm beginning to think NASA have writ...Carrick:<br />I'm beginning to think NASA have written a new very simple code to produce the land-only data.<br />Why? Because I've just upgraded my own half-arsed temperature code, and I can produce an almost indistinguishable result back to 1920. It's a 60 month running mean, which hides differences in the monthlies of course, but it's still a scary good fit.<br /><a href="http://postimage.org/image/zc63okafl/" rel="nofollow">GRAPH HERE</a><br />More interestingly, I get a similar result if I use the CRU data instead of GHCN. However CRUTEM3 is <i>way</i> different. I've tried Nick's two griding modes, and that doesn't explain the difference. I'm beginning to think CRUTEM3 has a bug.<br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-45508847818950436532011-11-06T15:21:40.256+11:002011-11-06T15:21:40.256+11:00Thanks Kevin, now we just need to know how they ge...Thanks Kevin, now we just need to know how they generated it.<br /><br />Anonymous: <i>You don't need a model to explain why increasing CO2 suddenly stops warming the Earth, you need a model that includes all exogenous factors...</i><br /><br />Good luck on that, if the warming has really "stopped" while CO2 is increasing.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.com