Guest post by Hank
Many readers know that the blogosphere is a buzz with alarmists excited over the latest 73 proxy, 11,280 year climate reconstruction by Marcott et al. published March 7th in the journal Science. Here’s a link to the abstract.
And why are alarmists so excited?
Link to larger image
Do you see it to the very right of the reconstruction? It’s the famous hockey stick. To coin Frankenstein’s famous words when his monster was brought back to life – “it’s alive!”
Alarmists are high-fiving themselves because the Marcott study purports to have found the same blade of the hockey stick in the 73 proxies used in the study, lending further proof that anthropogenic CO2 has warmed the earth in an unprecedented way in the past 150 years.
Most who read suyts space know that I enjoy working with paleo climate data to generate my own charts like here and here. It happens that Marcott et al. made their proxy database available to the public so I downloaded it from here to create my own graphs. I created the above graph from the study’s “Global Temperature Stack” dataset.
The raw proxy datasets were included in the Marcott database. Studying the proxies, I discovered that only nine of the 73 proxies contained data that extended to 1950. Of those nine, only two contained data that extended to 2000. My first impulse was to think there was a bias introduced when the number of proxies fell off to nine. Perhaps one or two of the proxies were strongly biased towards warm.
Eyeballing the data in the nine proxies of interest I started looking at the latest years presented in the datasets for anomalous data points. I didn’t see any. What I saw in my cursory review was a slight cooling trend as they approached present dates.
All of the proxy datasets in the Marcott study span greater than 6,500 years and use the years 4,500 to 5,500 as a common reference and calibration period. The temporal resolution of each proxy varied with the largest resolution being 300 years per measurement. In some proxies the temperature was calibrated to the local temperature or offset of it. In others, the temperature was expressed as an anomaly from a reference period mean.
I needed to turn apples and oranges into all apples. To do this I first normalized each proxy to its 4,500 to 5,500 year mean in the same manner used in the study. Using the reference mean, all absolute temperature scales were converted to anomalies (difference from the mean). Those proxies that were already using anomaly measurements were simply normalized without conversion.
I used a data infilling technique known as regularized expectation maximization to create a new 10 year matrix of the data series for each proxy. I chose a 10 year resolution in order to minimize high frequency filtering. Running statistical tests in SPSS I confirmed that the difference in variance between the raw and the regularized datasets was insignificant.
I now had a dataset for each proxy that could be aligned on 10 year boundaries with all temperatures expressed as anomalies from their respective reference period mean. I have all apples. Let the charting begin!
I output all nine proxies as CSV files and imported them into Excel to finally visualize the blade of the hockey stick. Starting at 1,500 before present (BP), I graphed the nine proxy datasets. And here’s what I got:
Link to larger image
Hang on a second… Where’s the blade? It was supposed to look like this:
Link to larger image
Before I tackle the question of the blade I have some explaining to do. The black line on each graph is a running mean of the proxies used. You’ll notice that the X axis of the two graphs are slightly different. The proxy data uses 1950 as their “Present.” For this reason, zero on the X axis of the nine proxy graph represents 1950. Negative numbers are years following 1950. The value -40 equates to 1990. The plot ends at the year 2000 (-50). The Marcott global temperature series graph uses the year 2000 as its zero point.
Before I get called out on using only nine proxies for the full 1,500 years in my reconstruction, I’ll caution you to look at only the last 50 to 100 years of each graph. That’s where the action is and where the Marcott graph falls off to nine proxies also. That’s where both graphs should agree on what is shown.
Notice the little downtick at the end of the black line in the nine proxy graph above. Look at this graph from Ljungqvist’s 30 proxy reconstruction:
Do you see the downtick at the very right of the plot? I circled it in red. Although both plots are on a different Y axis scale, Ljungqvist’s and my nine proxy chart agree.
Back to the question of what happened to the blade? Marcott gives us a hint of how it wound up in his reconstruction. Remember “Mike’s Nature Trick?” Marcott has a trick of his own. Take a look at this:
The above was excerpted from Marcott’s study. Note that he’s calling attention to the break in the Y-axis at 25. You can see it as a white vertical line in the lower left legend. Notice also that Marcott is attributing the break to the Mann et al. dataset.
This new 73 proxy study has alarmists convinced that this is an independent verification and vindication of Mann’s hockey stick. It isn’t. The hockey stick blade at the end of the reconstruction is resulting from an adjustment of the proxy data to agree with Mann’s treemometer study. That, or it is an outright splice of Mann’s data directly.
Allow me to leave you with a parting image I’m particularly fond of. I think it tells the true story of catastrophic global warming.
Update: In my original run of the nine proxies, I wasn’t concerned with a number of issues, mostly because I was more interested in finding a hockey stick signal. In my later post using 24 proxies here, I used the 1,500 – 6,000 BP period for calculating the weighting of each proxy. Subsequently, I compared my results to the Ljugzvist F.C. 2010 30 proxy reconstruction and validated that my weighting of the proxies based on that time period produced results more comparable to that of Ljugzvist. As such, I applied the same weighting to my first 9 proxy analysis.
I also centered the means of each proxy, not that it mattered for this analysis. The graph of the nine proxies was updated to reflect these changes and improvements. While none of this changes the results of my initial analysis, the updated graph presents a higher quality visualization of the analysis.