date: Tue, 20 Nov 2007 16:03:03 +1100 from: David Thompson subject: The volcanoes... to: Phil Jones Hi Phil, I'm really glad your interested. And thanks for the comments. First a general comment about the direction of the paper. Then some specific thoughts on the volcano results. A general comment: The paper has two main parts: 1) clarifying the importance of the step in 1945 and 2) providing new insight into the climatic impacts of volcanic eruptions. I've been emailing with John Kennedy about how to approach the dip in 1945. It's now seems clear the dip is due to uncorrected data issues. But the step contributes substantially to the appearance of global-mean cooling in the middle of the 20th century. So correcting the dip is almost certainly going to impact the appearance of the global-mean time series. It's good to have the data experts on board to help frame the discussion carefully and appropriately! The attached Fig. 4 includes a few new figures you might not have seen. SSTs diverge rapidly from land data in 1945 in the tropics. I'll keep iterating with the Hadley Centre on this and keep you posted ... The volcano results: Thanks for pointing me to the monograph. I thought I was up to date on the literature, but I missed that one. I've ordered the monograph from the library and it will arrive in a couple days. In the meantime: please see the attached Figures 2-3. They clarify some of my thinking on the topic. The left column of Fig 2 shows the composite for the largest 5 tropical eruptions (Krakatoa, Santa Maria, Agung, El Chichon, Pinatubo) for the globe (top), global land (middle) global ocean (bottom). The anomalies are shown with respect to the 4 year period before the eruption (ie, the mean for years -4 to 0 is zero). The middle column shows the exact same results, but for the residual data. The key here is that the volcanic signal is much clearer in the residual data, both in terms of the signal after the eruption date, but also in terms of less volcanic noise during the rest of the composite period. You can see the improvement in the global mean, in the land data, and in the SST data. It's also clear from both the left and middle panels that there is a lingering trend in the composites. For example: if you look at the top left panel (the composite of the raw global mean), the anomalies ~8 years before the eruptions are more pronounced than the anomalies after the eruptions. This doesn't make physical sense, and reflects the fact the eruptions are superposed on the global-mean warming of the past century. So it occurred to us that the global warming of the past century is likely biasing the volcanic signal. And .... The right hand panel shows the residual composite after the trend calculated for the decade preceding the eruptions has been removed from the 20yr composite period. In this case, the prevolcanic period is very quiescent, as you'd expect. But the volcanic signal is very, very pronounced, and seems to linger for more than a decade. That volcanos may have a >10 year timescale is supported by the recent Church et al paper in Nature (attached as a PDF) on changes in ocean heat content after volcanic eruptions. And I think it makes physical sense. See, for example, the attached Fig. 3. Fig. 3 shows the simulated output to a 3 W/m2 forcing imposed over two years. The result is based on the simple Hasselman climate model, that is: c*dT/dt = Forcing - alpha*T where T is temperature, Forcing is the 3 W/m2 for two years, alpha*T is Newtonian cooling, and c is the heat capacity of the atmosphere+the depth of the ocean to which the signal penetrates. The model is extremely simple, but shows that if the signal penetrates the top 100-150m of the ocean mixed layer, then the e-folding damping timescale of the volcanos is at least 6-8 years. If you use a shallower mixed layer (say 10m), then you get a shorter damping timescale, but the amplitude of the response is much too large (~1 K). So the argument is that to get the right amplitude (~0.3 K), then you need a mixed layer response of at least 100m. I think that a reader could argue about how we detrend the data. And so I don't think we can "prove" that the volcanic signal lasts for a decade or more based on observations and our simple model alone. But I do think we can make a compelling case that the data are suggestive of a long time scale, and we can certainly show that a long time scale is consistent with changes in ocean heat content down to 100 m. So the paper could get people chatting about the possibility. What do you think? -Dave ps I understand what you mean about centering about January (to account for the midwinter warming). It turns out that centering matters for a few months during NH midwinter (as you know). But it doesn't change the general appearance of Fig. 3. Attachment Converted: "c:\eudora\attach\Figures_VNov141.pdf" Attachment Converted: "c:\eudora\attach\Church_etal_Nature2006.pdf" On Nov 20, 2007, at 3:47 AM, Phil Jones wrote: Dave, I've had a read of the email now and looked at the pictures over the weekend. It seems very interesting and I like to be involved. The volcano response looks clear, but this may be a result of the way you've done superposed epoch analysis - and maybe on the residuals (i.e. without ENSO and COWL). Anyway have you seen this paper? Jones, P.D., Moberg, A., Osborn, T.J. and Briffa, K.R., 2003: Surface climate responses to explosive volcanic eruptions seen in long European temperature records and mid-to-high latitude tree-ring density around the Northern Hemisphere, In (A. Robock and C. Oppenheimer, Eds.) Volcanism and the Earths Atmosphere. American Geophysical Union, Washington D.C. 239-254. I would normally send you a pdf, but for some reason when AGU allowed us to get them in 2004 they were enormous - 40Mb, so too big to email. It is worth you getting hold of this and comparing to what we've al;ready done there. We used longer European records, but for the NH we used 5 large tropical eruptions in the 20th century. I don't think your 10 years before/after (as opposed to our 5) is the cause of the difference. It could be your extraction of trends, or could just be your working with residual T. The paper will show you how to calculate significance levels for the volcanic signal. I hope you've always used the January of the eruption year to designate the 4 volcanoes you've chosen. I can't see in your volcano plot the greater cooling in summers (NH mainly) cf NH winters. This was a clear signal in the work we did. Some other thoughts: 1. The COWL series looks to have smaller variance before about 1925. Also COWL variance looks more one-sided in recent decades. 2. As you or David said, the 1945 drop could be that 1940-44 are too warm. Still think that it doesn't look natural. Cheers Phil At 06:44 14/11/2007, David Thompson wrote: Dear David, Phil and John, (This is a bit of a long email, so you might want to grab a cup of coffee - or tea - before reading on...) Thanks again for the quick and helpful responses last week. Mike and I would be happy to include John as a coauthor on our paper. And David, Phil: we understand if you are too busy to join another project. But if you are interested in joining the paper, too, that would be great. The goal of the paper is to clarify some key aspects of 20th century temperature variability, and the study would certainly benefit from your expertise. Before I get too far ahead of myself, let me review the main points of the paper as it currently stands. I've attached 3 pages of figures (the figures will evolve as the writing evolves, but as of now it appears the paper will end up being short and punchy) Figure 1 includes 2 panels. The top time series in the top panel shows the global-mean temperature time series. The next two time series show the linear fit of ENSO and the COWL (cold-ocean/warm- land) time series to the global-mean. The ENSO time series is found as a damped thermal response to variations in the cold-tongue (this gives a marked improvement in the representation of ENSO in global- mean temperatures). The COWL time series represents the effects of random fluctuations in climate acting on the different heat capacities of the ocean and land (eg: periods of warm advection over land/cold advection over the ocean lead to warmer than normal global- mean temperatures by virtue of the fact that the continents have a lower heat capacity). I'll provide more details of the ENSO and COWL methodologies in a future email, but the main point is that a lot of the high frequency 'noise' in global-mean temperatures can be accounted for on the basis of two simple, physically based time series. The bottom panel in Fig. 1 includes a reproduction of the global-mean time series (top) and also shows the residual global mean time series, which is found by removing the ENSO and COWL time series from the global-mean time series. The bottom panel of Fig. 1 is the key figure in the paper. We think it's remarkable how well the fitting 'cleans up' the global-mean time series. If you look closely, you'll see that the major volcanoes of the past century (marked by solid vertical lines) are much, much clearer in the residual time series. But the fitting not only isolates the volcanoes, it also isolates the very large drop in Aug 1945. The Aug 1945 drop is about 0.3 K, almost twice as large as the response to Pinatubo. The residual time series also suggests a slightly different view of 20th century temperature variability. The canonical view is that temperatures warmed in the 20s, settled from the 40s-70s, and warmed from the 70s-the present. But if you stare at the residual time series, you get the impression that global-mean temperatures have actually risen steadily over the past century, but that the warming has been disturbed by several discrete and abrupt drops in temperature. As for the volcanos: In figure 2 we're exploiting the fitting procedure to provide a 'cleaned up' version of the volcanic response in surface temperatures. The figure shows the composite temperature response for the 4 largest tropical volcanoes of the 20th century. The composite is done such that the 10 year period before each volcano has a trend of zero and mean of zero. (If you don't remove the trend then the long term warming biases the composite). The response in the residual data is surprisingly coherent (much more so than in raw data). But we think what's most striking is that surface temperatures do not appear to fully recover for up to a decade after each volcano (the composite is limited to 9 years after the eruptions since the eruption of Pinatubo occurred 9 years after the eruption of El Chichon). You can see the long timescale of the volcanoes in the residual global-mean time series: if you follow the temperature time series before, say, Agung or Pinatubo, you can see that it takes a long, long time for global-mean temperatures to catch up to where they would have been, assuming they would have continued to warm... And as for the dip in 1945: Fig 2 shows the residual (ie with the COWL or ENSO time series removed) global-mean land and ocean time series. The point here is that the large dip in Aug 1945 does not show up in the land data. My impression from your emails last week is that the dip is almost certainly due to changes in instrumentation during the war. But it's also my impression that the specific reasons for the dip are not yet known. It's possible that the dip is offset by spurious rises in SSTs at the start of the war. But this isn't certain. And even so, there is a large drop in SSTs between the period before the war (1939) and the period after the war (1945). To my eye, the residual ocean time series in Fig. 3 suggests temperatures ratcheted downwards spuriously in 1945. In our view, the dip in Aug 1945 is very important and warrants being highlighted in the literature. In fact, once you know it's there, it's hard to view any time series of global-mean temperatures and not wonder how different it would appear if the dip was not there. For example, Fig. 4 shows the raw and residual global-mean temperature time series assuming the 0.3 K drop in Aug 1945 is spurious. I realize the figure is crude, and it might not make it into the paper. But the point is that you would get a very, very different impression of 20th century temperature trends if the dip proves to be an artifact of the end of the war. So our main point regarding the drop in Aug 1945 is this: if we assume the dip is spurious, then the global-warming of the past century would be at least ~0.3 K larger than currently thought, and global-mean temperatures would have risen steadily throughout the past century. That's enough for now. I'm working on the writing and our goal is to submit something by Xmas. Please let me know what you think, and if you are interested in continuing to interact with Mike and I on the paper. And again: thanks for your time and interest. -Dave  -------------------------------------------------------------------- -------------------------------------------------------------------- David W. J. Thompson [1]www.atmos.colostate.edu/~davet Dept of Atmospheric Science Colorado State University Fort Collins, CO 80523 USA Phone: 970-491-3338 Fax: 970-491-8449 Dear David, Phil and John, (This is a bit of a long email, so you might want to grab a cup of coffee - or tea - before reading on...) Thanks again for the quick and helpful responses last week. Mike and I would be happy to include John as a coauthor on our paper. And David, Phil: we understand if you are too busy to join another project. But if you are interested in joining the paper, too, that would be great. The goal of the paper is to clarify some key aspects of 20th century temperature variability, and the study would certainly benefit from your expertise. Before I get too far ahead of myself, let me review the main points of the paper as it currently stands. I've attached 3 pages of figures (the figures will evolve as the writing evolves, but as of now it appears the paper will end up being short and punchy) Figure 1 includes 2 panels. The top time series in the top panel shows the global-mean temperature time series. The next two time series show the linear fit of ENSO and the COWL (cold-ocean/warm-land) time series to the global-mean. The ENSO time series is found as a damped thermal response to variations in the cold-tongue (this gives a marked improvement in the representation of ENSO in global-mean temperatures). The COWL time series represents the effects of random fluctuations in climate acting on the different heat capacities of the ocean and land (eg: periods of warm advection over land/cold advection over the ocean lead to warmer than normal global-mean temperatures by virtue of the fact that the continents have a lower heat capacity). I'll provide more details of the ENSO and COWL methodologies in a future email, but the main point is that a lot of the high frequency 'noise' in global-mean temperatures can be accounted for on the basis of two simple, physically based time series. The bottom panel in Fig. 1 includes a reproduction of the global-mean time series (top) and also shows the residual global mean time series, which is found by removing the ENSO and COWL time series from the global-mean time series. The bottom panel of Fig. 1 is the key figure in the paper. We think it's remarkable how well the fitting 'cleans up' the global-mean time series. If you look closely, you'll see that the major volcanoes of the past century (marked by solid vertical lines) are much, much clearer in the residual time series. But the fitting not only isolates the volcanoes, it also isolates the very large drop in Aug 1945. The Aug 1945 drop is about 0.3 K, almost twice as large as the response to Pinatubo. The residual time series also suggests a slightly different view of 20th century temperature variability. The canonical view is that temperatures warmed in the 20s, settled from the 40s-70s, and warmed from the 70s-the present. But if you stare at the residual time series, you get the impression that global-mean temperatures have actually risen steadily over the past century, but that the warming has been disturbed by several discrete and abrupt drops in temperature. As for the volcanos: In figure 2 we're exploiting the fitting procedure to provide a 'cleaned up' version of the volcanic response in surface temperatures. The figure shows the composite temperature response for the 4 largest tropical volcanoes of the 20th century. The composite is done such that the 10 year period before each volcano has a trend of zero and mean of zero. (If you don't remove the trend then the long term warming biases the composite). The response in the residual data is surprisingly coherent (much more so than in raw data). But we think what's most striking is that surface temperatures do not appear to fully recover for up to a decade after each volcano (the composite is limited to 9 years after the eruptions since the eruption of Pinatubo occurred 9 years after the eruption of El Chichon). You can see the long timescale of the volcanoes in the residual global-mean time series: if you follow the temperature time series before, say, Agung or Pinatubo, you can see that it takes a long, long time for global-mean temperatures to catch up to where they would have been, assuming they would have continued to warm... And as for the dip in 1945: Fig 2 shows the residual (ie with the COWL or ENSO time series removed) global-mean land and ocean time series. The point here is that the large dip in Aug 1945 does not show up in the land data. My impression from your emails last week is that the dip is almost certainly due to changes in instrumentation during the war. But it's also my impression that the specific reasons for the dip are not yet known. It's possible that the dip is offset by spurious rises in SSTs at the start of the war. But this isn't certain. And even so, there is a large drop in SSTs between the period before the war (1939) and the period after the war (1945). To my eye, the residual ocean time series in Fig. 3 suggests temperatures ratcheted downwards spuriously in 1945. In our view, the dip in Aug 1945 is very important and warrants being highlighted in the literature. In fact, once you know it's there, it's hard to view any time series of global-mean temperatures and not wonder how different it would appear if the dip was not there. For example, Fig. 4 shows the raw and residual global-mean temperature time series assuming the 0.3 K drop in Aug 1945 is spurious. I realize the figure is crude, and it might not make it into the paper. But the point is that you would get a very, very different impression of 20th century temperature trends if the dip proves to be an artifact of the end of the war. So our main point regarding the drop in Aug 1945 is this: if we assume the dip is spurious, then the global-warming of the past century would be at least ~0.3 K larger than currently thought, and global-mean temperatures would have risen steadily throughout the past century. That's enough for now. I'm working on the writing and our goal is to submit something by Xmas. Please let me know what you think, and if you are interested in continuing to interact with Mike and I on the paper. And again: thanks for your time and interest. -Dave -------------------------------------------------------------------- -------------------------------------------------------------------- David W. J. Thompson [2]www.atmos.colostate.edu/~davet Dept of Atmospheric Science Colorado State University Fort Collins, CO 80523 USA Phone: 970-491-3338 Fax: 970-491-8449 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 [3]p.jones@uea.ac.uk NR4 7TJ UK ---------------------------------------------------------------------------- -------------------------------------------------------------------- -------------------------------------------------------------------- David W. J. Thompson www.atmos.colostate.edu/~davet Dept of Atmospheric Science Colorado State University Fort Collins, CO 80523 USA Phone: 970-491-3338 Fax: 970-491-8449