Nothing to See Here, Move Along!

Well it’s been quite a hurricane season in the Atlantic so far, hasn’t it? It’s even gotten the notice of the UK press.

Here are a couple of graphics from Philip Klotzbach, meteorologist at Colorado State University specializing in Atlantic basin seasonal hurricane forecasts:

Source: Philip Klotzbach, via Twitter.

We’ve had six major hurricanes this season (Harvey, Irma, Jose, Lee, Maria, Ophelia), three of them exceptional in the amount of cyclonic energy they’ve accumulated and expended (Irma, Jose and Maria). The UK Express notes that Harvey has had the largest rainfall and Irma has had the longest duration as a Category 5 hurricane, ever (at least since 1851 and possibly since the Eemian).

Source: Philip Klotzbach, via Twitter.

The above seven parameters each have the 2017 Atlantic hurricane season in the top twenty seasons since 1851, six of them in the top ten—although only one in the top five.

Weather Underground’s Dr. Jeff Masters has this to say:

It’s been an exceptionally busy Atlantic hurricane season in 2017.  Rina’s formation in November means that all six months of the official Atlantic hurricane season, June through November, saw formation of a named storm. Since hurricane names began being assigned in 1950, only six other years have had this happen: 1954, 1989, 1996, 2005, 2011, and 2013

Now does this mean anthropogenic global warming’s had anything to do with this?

Well according to Kerry Emanuel, atmospheric scientist at MIT, the 2017 hurricanes are merely a foretaste of the future and that the increased damages from hurricanes due to ever-increasing development of the shorelines is being exacerbated by rising sea levels and changing storm characteristics due to anthropogenic climate change. He also says that the average annual number of hurricanes has increased, that the probabilities of hurricanes like Harvey and Irma have already been changed, and that as the climate warms these storms will continue to gain strength until they reach their inherent limits and their maximum size will begin to level off as well. The UK Express alleges that Prof. Emanuel said that “hurricanes with the magnitude of Hurricane Harvey had gone from a once a century event to a once every 5.5 years event.”  The real reality is that Prof. Emanuel states in his video that in the time period 1981-2000, Harvey would have been a 2,000-year storm, and could become a 100-year storm in 2081-2100. If I recall correctly, this year NOAA said Harvey was a 500-to-1,000-year storm.  More on Harvey: 1% chance of same rainfall for anywhere in Texas in the 90s, now 6% Chance in any one year.

On the other hand, climate scientist and climate forecast consultant Dr. Judith Curry states it’s way too soon to tell, that each individual hurricane should be subjected to diagnostics---instead of using models to create imaginary storms in a world not subject to global warming---to determine if global warming has had a hand in it. She notes that there have been busy hurricane seasons in decades past and even some notable hurricanes in the Nineteenth and early Twentieth Centuries.

So isn’t all this is just the noise from natural variability? Or is it?

You can go to this and this webpage put up by Weather Underground and view the statistics therein, pursue the linked webpages therein (here’s one, another one, and another one), and view the hurricane FAQs and years with the most hurricanes webpages at the NOAA website.  Note, the pages are not updated for 2017 (except the one about Irma).  You can review these pages and decide for yourself.

As far as I’m concerned, our weather has gotten more extreme---and weirder, to boot. Speaking of weird weather, there’s disturbance that might become a named sub-tropical storm, Sean. Near the Azores. In November.

Cato Working Paper No. 35 — How Well Has It Held Up? (Part 3)

Now I will come to the conclusion of this three-part series ad show you how well Michaels and Knappenberger’s Climate Models and Climate Reality: A Closer Look at a Lukewarming World has held up. Because if the trends indicating a gross overprediction in the IPCC CMIP-5 model runs, a slower warming and a smaller climate sensitivity, which are the Cato paper’s conclusions, are broken, then their conclusions are liable to be ill-founded as well, which means we could be right back at the IPCC’s median estimate of 3.0 deg C ECS (Equilibrium Climate Science) sensitivity---give or take 1.5 deg C.

First, let me show you a very recent graph presented by climate scientist Gavin Schmidt of NASA’s GISS:

Chart G: Global Mean Surface Temperature Anomalies from the 1908-1999 Mean. Anomaly measurements are in degrees C. The CMIP5 Ensemble used has not been blended according to the Hadley Centre’s HadCRUT4 method of accounting for land surface air temperatures and sea surface temperatures. The forcing-adjusted CMIP5 takes into account the updated (historical) forcings to date. The NASA GISS’s GISTemp plotted temperature rise is shown by the red line; the Hadley Centre’s HadCRUT4 by the blue line. Cowtan & Way’s plot and the NOAA NCEI plot are shown by the orange and pale blue lines respectively. The estimated 2017 GISTemp temperature plot and its uncertainty spread is courtesy Gavin Schmidt; the estimated HadCRUT4 plot and its uncertainty spread for 2017 are mine, based on the latest data available from the Hadley Centre website (accessed 11/07/2017). Nota bene that the upper end of this uncertainty spread is hidden by the GISTemp estimated temperature plot for 2017.

The above chart clearly shows that the original CMIP5 model ensemble clearly overpredicts the amount of warming observed; the forcing-adjusted ensemble, less so. The NOAA NCEI and HadCRUT4 temperature plots dip below the 2.5th percentile line of the original ensemble and the original ensemble mean is only hit by the GISTemp and Cowtan & Way temperature plots for 2016, the year of a particularly strong El Nino. The estimated temperatures for 2017 show a drop of approximately 0.09 deg C for both the GISTemp and HadCRUT4 plots.

Clearly, we still have an apparent overprediction problem here even though the planet’s mean temperature rise is no longer in the Pause. To check this we can simply do the same sort of trend runs that Michaels and Knappenberger did to indicate a slow rate of warming in their paper.

Chart H. The trends for the predicted annual average global surface air temperatures taken from Chart G for the CMPI5 ensemble (pale green, purple and pale blue lines), and for the Hadley Centre’s HadCRUT4 (blue line) and NASA GISS’s GISTemp (red line) observed global mean temperatures taken from their latest data for a blend of land and ice surface air temperatures and ocean sea surface temperatures. Observed temperatures for 2017 are estimated as explained under Chart G above. The trends were plotted for all years from 1975 through 2008; the trend lengths run from 10 years (2008-2017) to 43 years (1975-2017).

It appears that in the above chart the CMIP5 Ensemble trends are noticeably higher than for Figure 2 in Part 1. Only the Ensemble Mean trend for 26-year trend length, corresponding to 1992 the year after Mt. Pinatubo blew up, is anywhere close to equal to the Multiple Model-run Mean (MMM) trend shown in Figure 2, as checked in Chart D in Part 2.  The observed trends for longest trend lengths appear at about 0.18 deg C/dec, well below the 2.5th Percentile line at approximately 0.20. After the 37-year trend length, the Hadley and NASA GISS lines begin to diverge. But the shorter observed trend lines after the 26-year trend length---YIKES! They diverge, run an arc and fly above the 97th Percentile line, crashing through the roof! And this is the problem with using short trend lengths; Cato’s figures 2 and 3 exhibit their shorter trend lengths showing trends dropping progressively toward the bottom of the graph, falling into the subbasement.

The above graph shows the influence of the still-warmed global mean temperature(s) of 2017 coming 0.09 deg C off the El Niño-influenced peak temperature(s) in 2016. The trends for trend lengths shorter than 16 years also show the influence of the Pause---the longer into the Pause, the higher the trend. Trend lengths shorter than 10 years would show even higher trends approaching 0.85 deg C per decade.

The IPCC CMIP5 Model Ensemble still overpredicts the amount of surface warming as can also be clearly seen in Chart G. This is because the Ensemble does not take into account the blend of land and ice surface air temperatures and ocean sea surface temperatures reflected in the observed data, and because the forcings under the RCP 8.5 (also RCPs 6.0, 4.5 and 2.6) scenario post-2006 were higher than the observed historical forcings after that year. Apparently the historical forcings between 1992 and 2006 were also erroneous and had to be updated, as shown in Figure 4 of the Cowtan et al 2015 paper, “Robust comparison of climate models with observations using blended land air and ocean sea surface temperatures,” accessible here.

And now I will show you the trends for the up-to-date forcing-adjusted CMIP5 as shown in Chart G versus the trends from the observed data:

Chart I The trends for the predicted annual average global surface air temperatures taken from Chart G for the forcing-adjusted (supplied with updated historical forcings from 1992 to date) CMPI5 ensemble (pale green, purple and pale blue lines), and for the UK Met Office Hadley Centre’s HadCRUT4 (blue line) and NASA GISS’s GISTemp (red line) observed global mean temperatures taken from their latest data for a blend of land and ice surface air temperatures and ocean sea surface temperatures. Observed temperatures for 2017 are estimated as explained under Chart G above. The trends were plotted for all years from 1975 through 2008; the trend lengths run from 10 years (2008-2017) to 43 years (1975-2017).

The observed temperature trend plots are still low in the long-term trend lengths when compared to the Ensemble Mean, but this time they are within the 95-percent spread, i.e., at or above the 2.5th percentile line. One possible explanation for this is that this CMIP-5 model, as the original one, does not take into account the blend of ocean sea surface and land air surface temperatures. If it did, Chart G would have stated as such. As in Chart H, the observed trends start at 0.18 deg C / decade, there is a peak at the 26-year trend length, corresponding to 1992 the year after Mt. Pinatubo blew up, and after that the observed temperature trends swing in an arc and fly through the roof of the 97th Percentile line, indicative of the twin influences of the Pause and the 2016 El Niño heat spike which the climate is still in, despite the Pacific Ocean presently exhibiting a La Niña. One last interesting tidbit: the Ensemble Mean and the 97.5th Percentile lines converge at the 13-year trend length, a curious coincidence.

It should be emphasized that in both Charts H and I the observed temperature trends diverge after the 37-year trend length, with the 2017 NASA GISS line roughly exhibiting a trend of 0.20 deg C / decade and the UK Met HadCRUT4 roughly showing a trend of 0.15 deg C / decade, before both trend lines rocket skyward under the influence of both the 2016 El Niño heat spike and the 2002-2014 Pause.

The one thing the Cato got right was that the original and blended CMIP5 Model Ensembles with historical forcings to 2006 overpredicted the amount of warming compared to the amount already observed. Cowtan et al also noted this with the original Ensemble and they made a double comparison using 84 of the model runs: all surface air temperatures vs land surface air temperatures, and historical to 2006 then RCP 8.5 vs updated historical forcings and presented their findings in Figure 4 of their paper:

Chart J (Cowtan et al 2015 Figure 4):  “Comparison of 84 RCP8.5 simulations against HadCRUT4 observations (black), using either air temperatures (red line and shading) or blended temperatures using the HadCRUT4 method (blue line and shading). The shaded regions represent the 90% range (i.e., from 5% to 95%) of the model simulations, with the corresponding lines representing the multimodel mean. (a) Anomalies derived from the unmodified RCP8.5 results and (b) the results adjusted to include the effect of updated forcings from Schmidt et al. [2014]. Temperature anomalies are relative to 1961–1990.” The purple shading is the overlap between the red and blue shadings.

Please note in the above chart that the multiple model set of 84 simulations that had the best fit to the observed post-1991 (post-Pinatubo) temperatures was the set that was run with (a) updated blended temperature anomalies (HadCRUT4 method) and (b) updated forcings to 2013. The observed temperatures run through 2014. Neither the updated forcings nor the observed temperatures include the years 2015 through 2017 which have the El Niño warming spike that brought an end to the Pause.  For those years’ observed temperatures catching up to the mean multimodel prediction, even with all surface air temperatures, see Chart G at the top.

Now what does this mean for the Cato Working Paper No. 35? Despite the paper being correct about the CMIP5 Model Ensemble with its original forcings grossly overstating the predicted warming, it does not follow that the trend lines are levelling off toward zero or even 0.10 deg C / decade. If anything, it shows that taking trends that are shorter than 30 years and extrapolating a lower Charney / ECS climate sensitivity toward a doubling of CO2 from them is a fool’s errand and that the longer trend lengths with their observed trend values of about 0.18 deg C / decade are to be preferred in roughly predicting any future warming. And what was the lower sensitivity favored by the Cato paper? 2 degrees Celsius for a doubling of CO2, which is more than enough to bring on the coming global superstorms that would wreck our civilization. A higher sensitivity, say 3 degrees Celsius, is more likely given the observed longer-term trend values and would be even worse for us because the storms would arrive at a lower CO2 content, that is, sooner.

Furthermore, the difference in the surface warming versus the mid-troposphere warming is creating a higher heat gradient between the surface and the mid-troposphere, meaning that more water vapor is going up into the atmosphere, creating more clouds and weirder and more extreme weather in diverse places. After all, we do know our weather has gotten weirder and some climate scientists are on record stating that it has become more extreme.

Cato’s Working Paper No. 35 has not held up well at all.

Cato Working Paper No. 35 — How Well Has It Held Up? (Part 2)

Note: Graphs and texts in blue are from the Cato Working Paper No. 35.

Note: Texts in blue are from the Cato Working Paper No. 35.

Now I am going to compare Michaels and Knappenberger’s historical findings in their Cato Working Paper No. 35 with the 2017 NASA-GISS observational data graph I found at Robert Fanney’s blog “Robertscribbler,” presented below. You will be able to find it at the NASA website, located here (click on "Global Annual Mean Surface Air Temperature Change").

Chart A: The black squares are the annual global mean average estimated temperatures since 1880. Obviously NASA didn’t exist all the way back to 1880 so the older averages were computed from data that were taken from various sources such as the US and British Imperial weather services. The red line represents the “Lowess Smoothing,” I haven’t the faintest clue what that means. The Blue I’s represent uncertainty bars: 0.25 C spread (+/- 12.5 C) in the late 19th Century (1892), 0.17 C spread in the mid-Twentieth (1948) and 0.10 C spread in the late 20th / early 21st Centuries (2009). The 2017 global mean temperature (not shown) is estimated to be approximately 0.92 deg C above the 20th Century average.

This graph shows a cooling from 1880 to 1909, the a slight warming ‘til about 1945, another slight cooling until 1976 ad then a consistent rise at roughly 0.2 deg C per decade through 2016. There is a low point at 1992, the year after Mount Pinatubo blew up. The 2002-2014 Pause is hidden in the annual means’ noise but the Lowess Smoothing detects a shorter pause or at the very least a massive deceleration.

Well we can use these temperatures to check Michaels and Knappenberger’s plot of the NASA GISS temperature data and I’ve done so by printing the above on an 8½ by 11, drawing gridlines straight across and scaling to the nearest millimeter (half-millimeter if the square falls right between two lines) to the nearest gridline. The temperatures are plotted not in 10th of a degree Celsius, but in hundredths of a degree. So I scaled off from -0.5 to +1.0 and came up with a scale converter. I did the same thing from Cato’s Figure 2 for the GISS multiyear trends, plugged everything into an Excel spreadsheet, used the LINEST command to construct a trend, and down below are the plots.

Chart B: The red line is the plot line of the multiyear decadal trends I generated for the reference year 2015 from the 2017 NASA GISS temperature records graph downloaded ‎Sunday, ‎October ‎01, ‎2017. The blue line represents the plot line of the multiyear decadal trends. Note the red line is noticeably forward a year beginning about the trend length of 40 years. Otherwise it is in reasonable close conformity to my red line graph.

My plot shows the jump in trend from the 11-year trend length to the 10-year trend length. This is the first year that the signal from 2015 has made itself known to a much higher amplitude within the trends. All subsequent years with their shorter trend lengths are probably excessively noisy.

Michaels and Knappenberger’s plot has that one year deviation that I had noticed before, starting at about the 40-year trend length—it is ahead of my plot by exactly one year. This is a fat-fingered error: either in the authors’ Figure 2 graph, or in the stated year runs in its caption. The global cooling from Mount Pinatubo was removed from 1992 to 1993! Any way let the record show that the NASA GISS 2006-2015 trend increased from the authors’ value of 0.12 deg C per decade to the true value of almost 0.20. With errors such as these and the others noted in Part 1, errors that I, a layman have discovered, it really doesn’t inspire any confidence in me that they are doing rigorous enough science---hard science---for it to pass peer review or auditing!

Now we are going to check the proper plotline against the IPCC’s CMIP-5 forecast, as shown in the Cato paper. The multiple model runs were based on historical data up to 2006 and the RCP4.5 scenario after that year.

Chart C: Cato’s CMIP5 model and NASA GISS 2015 data set runs showing trends from 1951 through 2006 with reference year to 2015, with my NASA GISS 2017 data set runs for and referenced to the same years. The purple line is the multiple model-run mean, the cyan and spring green lines are the model-runs’ 97.5% and 2.5% percentiles respectively. The blue line is Cato’s trend plot from the 2015 NASA GISS data set and the red line is my trend plot from the 2017 NASA GISS data set. 

Again, this graph shows that the one-year discrepancy that I had noticed in Figure 2 (Part 1) is proven. The peaks in the trends as computed by Michaels and Knappenberger are in the 23-years’ trend length (1993-2015), instead of the 24-years’ length (1992-2015) as my NASA GISS plot (red line) shows and is reflective of the actual global mean temperature low point in 1992 (see Chart A). The authors’ GISS plot shows a continued low warming trend of about 0.12 deg C per decade whereas mine increases to 0.200 deg C per decade and close to the multiple model-run mean (MMM).

Now to find out the source of the Cato paper’s discrepancy I took their plotted NASA GISS values from Figure 2, and compared them to my computed trends for the NASA GISS temperature records from 1950 through 2005 and referenced to 2014. When I made a plot, my numbers and the paper’s numbers became widely variant when running into the shorter trend lengths, so I won’t belabor the point here.

So then what I did was reset their end years, 1951 and 2006, to 1950 and 2005 respectively, and kept my plot the same as it was in Chart B, with an added year 1950 and the 2006 year dropped. The referenced year remained the same throughout: 2015. And this is what I found in Chart D below:

Chart D: The Cato paper’s CMIP5 model and NASA GISS 2017 data set runs showing trends reset to beginning in 1950 and ending in 2005 with reference year to 2015, with my NASA GISS 2017 data set runs for and referenced to the same years. “Annual years” are the years for the starting points of the respective trends. End points for all trends is 2015. The trend lengths now vary from 66 years (1950-2015) to 11 years (2005-2015). Explication of colored lines are the same as in Chart C.

This adjustment achieves a much better fit for the Cato paper’s model runs and GISS data set plot lines. The peak trends resulting from the 1992 global mean temperature low point are now where they should be. The Cato NASA GISS trend plot (blue line) and mine (red line) are now in close conformity to each other up to the trend length of 13 years and only diverge after that. This is probably because Michaels and Knappenberger came up with an estimate for the year 2015:

We have also updated our AGU presentation [in December 2014] with our best guess for 2015 average temperatures.

The actual NASA GISS global mean average temperature for 2015 was apparently higher than the authors’ estimate, and the other actual data sets would probably show higher 2015 averages as well. The gross errors that were introduced in Figure 2 (Part 1) and the discrepancies between Figures 2 and 3 are probably due to the author’s stated updating without checking to make sure that their trend lengths in the graph and stated run years in Figure 2’s caption were correct, as they are not.

Now on to checking Cato’s Figure 3.

Chart D: The Cato paper’s CMIP5 modified model set and the 2014 Hadley HadCRUT4 data set (blue line) compared with the 2017 NASA GISS data set (red line) for all trends with end year reference of 2014. Trend lengths begin with 65 years (1950-2014) and end with 10 years (2010-2014). The Model set has been modified according to the Hadley HadCRUT4 Method to reflect a blend of land surface air temperatures and sea surface temperatures, and obtained from the University of York website (http://www-users.york.ac.uk/~kdc3/papers/robust2015/index.html). Explication of the model set’s colored lines are similar to those in Charts C & D.

In the above Chart, the NASA GISS trend plot line roughly parallels the HadCRUT4 line until about the 45-years’ trend length, then follows in close conformity thereto until the 27-years’ trend length, whereupon the two lines diverge. Both lines hit a peak at the 23 years’ (1992-2014) trend length—right after Mt. Pinatubo blew up, after which they slowly descend roughly parallel to each other until they hit a low point at 2005. This descent is indicative of a steadily but noisily increasing mean of global temperature until 2002, where it levels off, going into the Pause which ended in 2014. The parallel descent into very low decadal trends indicates that Michaels and Knappenberger had referenced their Hadley data set trends in their Figure 3 for the end year of 2014, and is not indicative of fat-fingered errors at the Cato institute. It also means that Figure 3 is, essentially, correct.

I am unable to check Figure 4 owing to the fact that Christy’s data in Figure 1 went by annual five-year global mean temperatures and his corresponding data in Figure 4 went by annual means as indicated in the two figures and their captions. So I will compare the model forecasts with the observed temperatures in Figure 1 instead.

Chart E. The Cato paper’s CMIP5 Model set for the tropical mid-troposphere compared with the 2015 Christy balloon data sets (red line) for all trends with end year reference of 2015. Trend lengths begin with 40 years (1976-2015) and end with 10 years (2006-2015). The temperature data from the Model set and the balloon data sets have been provided to the Cato Institute by Dr. John Christy of the University of Alabama in Huntsville.

The results are similar to what is seen in Cato’s Figure 4. The data from Dr. Christy comport with the Figure 4 trends that run significantly lower than the mean of the multiple model runs (MMM) and even below the 2.5th percentile line. The plot line shown here also runs below the minimum for all runs from the 33-year to 23-year trend lengths. The plot line also confirms the sense of Figure 4 in that the trends are running almost uniformly at 0.1 deg C per decade. A caution noted here is that Christy’s observational data are from the five-year running averages shown in Figure 1, so some noise has been eliminated from the data. A comparison of this graph with Charts C and D shows that through 2015, when the trends end outside the Pause, the tropical mid-troposphere appears to be warming at a slower rate than is the surface, possibly giving a greater differential between surface and mid-troposphere temperatures and thereby allowing for more moisture-laden air to arise from our lands and seas to supercharge our weather, although a caveatshould be noted that the tropical mid-troposphere temperatures should be compared with the tropical surface temperatures and that temperatures at altitude are expected to warm much less than the surface at higher latitudes.

Conclusion on Checking Cato’s Figures.

Although Michaels and Knappenberger’s Figures 1, 2 and 3 appear to be correct, and their discussion is for the almost the whole very well written, I cannot recommend this paper to anyone without serious reservations because of the gross errors they introduced into Figure 2 when they updated it for release of the paper at the end of 2015. It would have been better they left the plot lines remaining as they were in 2014, as they did for their Figure 3. They have also failed to do another check, which is to compile uniform multiple-year trends, for example, 30-year trends going back from 1985-2014 to 1921-1950.

Well I’m up to 2,000 words now. I will leave it for Part 3 to see if this paper has held up under the increased temperatures of 2016 and 2017.

Filed under: Global Warming, Cato Institute