tag:blogger.com,1999:blog-7729093380675162051.post7859230595043837203..comments2024-03-28T13:56:47.604+11:00Comments on moyhu: Area weighting and a 60 stations global temperature.Nick Stokeshttp://www.blogger.com/profile/06377413236983002873noreply@blogger.comBlogger11125tag:blogger.com,1999:blog-7729093380675162051.post-58025920230032676772013-11-26T22:36:25.004+11:002013-11-26T22:36:25.004+11:00Nick, to me the most interesting is that the low n...Nick, to me the most interesting is that the low number of stations would allow to study their quality and especially to ascertain that they are really rural. Personally, I would not put the method at the foreground here even if I am normally a fan of methodological studies.<br /><br />Two related papers on the number of stations needed are:<br /><br />Jones, P.D., Osborn, T.J. and Briffa, K.R. 1997: Estimating sampling errors in large-scale temperature averages. J. Climate, 10, 2548–2568.<br /><br />Hawkins, E. and Jones, P.D. 2013: On increasing global temperatures: 75 years after Callendar. Q. J. Royal Meteorol. Soc., doi:10.1002/qj.2178 (early view).Victor Venemahttps://www.blogger.com/profile/02842816166712285801noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-66728867569038164212013-11-26T20:43:42.488+11:002013-11-26T20:43:42.488+11:00Victor,
Thanks, for the remarks. I don't think...Victor,<br />Thanks, for the remarks. I don't think this kind of mesh usage has been done before - I haven't come across it. I'll look at it again - I'm handling meshes better now. I think they do make the best use of the data.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-4604026721012356642013-11-26T10:18:16.913+11:002013-11-26T10:18:16.913+11:00That is a really interesting study. 60 stations ar...That is a really interesting study. 60 stations are not much, that would allow one to actually study the station histories of these 60 stations and especially the state they are in now to confirm that they are really rural. May be publishable, if this has not been done before.Victor Venemahttps://www.blogger.com/profile/02842816166712285801noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-59989609573913217122011-08-02T23:10:51.069+10:002011-08-02T23:10:51.069+10:00It didn't work (even with a 60 month moving av...It didn't work (even with a 60 month moving average). You can see a few features of the ITR, but the appearance and disappearance of stations adds a load of spurious features too. It looks like the anomaly calculation is needed even when working with a set of fairly complete stations.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-22458765503302882932011-08-02T22:44:51.827+10:002011-08-02T22:44:51.827+10:00Don't worry unless its useful to you. Using my...Don't worry unless its useful to you. Using my crude code the reconstruction looks pretty much like yours.<br /><br />My next step is to turn the 60 stations into a spreadsheet. Given the list only contains stations which are 90% complete on 1940-2000, it may be possible to calculate a temperature record without worrying about missing data or anomalies, and the uniform sampling makes area weighting rather less important. So it may be possible to make the calculation accessible to anyone who can use a spreadsheet.<br /><br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-58442001712142644922011-08-02T22:37:53.154+10:002011-08-02T22:37:53.154+10:00Thanks, Kevin,
I like the principle that you are u...Thanks, Kevin,<br />I like the principle that you are using. And the result looks promising.<br /><br />I could do a reconctruction with it if you like, for comparison.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-18522956591997277932011-08-02T20:51:05.614+10:002011-08-02T20:51:05.614+10:00OK, I got a 60 station list I'm happy with. Af...OK, I got a 60 station list I'm happy with. After a few attempts I settled on this:<br /><br />1. Calculate area for each station. (I used a cruder method by looping over a 2deg grid allocating the area of each grid cell to the nearest station).<br /><br />2. Calculate the nearest neighbouring station for each station in the current list.<br /><br />3. Sort the list of neighbour distances, lowest first. For each neighbour pair, eliminate the station with the lower area until the desired number of stations has been eliminated in this cycle.<br /><br />Area and distance now both play roles. My area result is different from yours owing to a different initial set of 'long running' stations, but shows the same problem of close pairs. The distance version looks better. However the standard deviation of the resulting areas is about 50% higher.<br /><br />Here's a pic:<br />http://img30.imageshack.us/img30/8835/mappn.png<br /><br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-37650346548492028722011-07-13T09:39:06.386+10:002011-07-13T09:39:06.386+10:00Kevin,
Yes, that might help. generally I haven'...Kevin,<br />Yes, that might help. generally I haven't worried much about details of area shape - the idea is that there should be enough correlation so that there isn't really a bias when you change weighting locally. I'm trying to avoid bias on a larger scale. So locally, once there is an area weighting, the focus shifts trying to spread the weight (trading areas) to keep the variance down.<br /><br />But it's true that the 60-station exercise is pushing to the limit of correlation, so the effect you're describing is more important then. It's still essential to have area consistency - basically so the sums behave like area integrals. But radius of gyration could be used as a smoothing principle while still maintaining area consistency.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-58448896993651890312011-07-13T00:39:44.854+10:002011-07-13T00:39:44.854+10:00I had an idea about this. Instead of eliminating t...I had an idea about this. Instead of eliminating the stations by area, you could weight the area by the ratio of the radius of gyration of the polygon about the station coords.<br /><br />(Possibly raise the area to some power first if it needs a higher weight.)<br /><br />That will tend to disfavour stations which are near-neighbours at the edge of their polygons, for example pairs on the W African coast, Baja California and Venezuela.<br /><br />Kevin CAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-83587342790198832952011-03-22T22:53:56.144+11:002011-03-22T22:53:56.144+11:00Sorry, Anon, I linked the wrong image. It's fi...Sorry, Anon, I linked the wrong image. It's fixed above. It actually makes some difference to the trends.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-69834871650613497752011-03-22T20:45:50.498+11:002011-03-22T20:45:50.498+11:00The last figure stops in 1999. I would like to see...The last figure stops in 1999. I would like to see it until 2009.Anonymousnoreply@blogger.com