Wednesday, March 10, 2010

The Newly-Discovered Jarvykortta Proxy -- II

This post is a rewrite of an essay I published on 28 Nov. 2009, The Newly-Discovered Jarvykortta Proxy: Relevant to the Tiljander Varve Series? It's been edited and reformatted for clarity. I assume that a fictional monastery exists alongside a fictional "Jarvykortta River" (cf. the real Lake Korttajarvi) in fictional Ruritania (cf. Finland). The point is to explore whether a proxy series can be inadvertently used in an upside-down orientation. The example illustrates that Yes, such an error is quite possible. I argue that a data series refers back to one or more physical attributes. Care should be taken to see that such data are used in the orientation that makes logical sense, with respect to what they represent in the physical world.

We begin.

I recently returned from a trip to Ruritania. While there, I met with the Abbott of the Jarvykortta Monastery, who told me that the monks have kept a continuous record of the ice breakup date of the adjacent Jarvykortta River since before 400 AD.

Recalling the The Nenana, Alaska Ice Classic, it struck me that the Monastery's record could be a useful data set for a multi-century paleotemperature reconstruction. Accordingly, I asked if I might have access to these records, and any other information the Abbott might have. He graciously agreed.

Since recordkeeping began in the year 389, the date of the Jarvykortta River breakup has been as early as February 15 (in 1231) and as late as June 20 (in 1967). For graphing purposes, I have taken the 11-year rolling average of this series. In this, I followed the precedent set by Finnish graduate student Mia Tiljander, as far as how she handled the X-Ray Density data of the lakebed sediments that she analyzed from the bottom of Lake Korttajarvi, Finland. The reference to the Tiljander et al peer-reviewed publication from 2003 that describes that work is here.

(In fact, it was the Finnish Lake Korttajarvi XRD series and not this newly-discovered Ruritanian Jarvykortta River series that was used by Mike Mann's research group as one of the long-term temperature proxies in their 2008 peer-reviewed paper in the journal PNAS.)

Clearly, temperature could have an important influence on when the ice breaks up. Still, as with most proxies, there are some complications that must be considered. In that regard, the Abbott related the following three points to me:
  1. There is a natural hot spring that empties into the Jarvykortta River, about 1 kilometer upstream of the Monastery. Its flow is quite constant, year-to-year. This addition of hot water makes the river ice break up earlier than would otherwise be the case.
  2. Around 1720, local farmers began piping some of the hot spring's output to their homes. As the population of the area grew in the 18th and 19th Centuries, this practice became more widespread. This reduction in hot-spring flow into the river likely led to increasing delays in the timing of the Spring breakup of the ice by the monastery.
  3. For much of the 20th Century, the nearby town maintained a skating rink for the winter and spring, right in the river, just upstream of the monastery. Cooling coils were placed in the river to keep the ice solid, well into the spring. In the late 1920s/early 1930s, and again in 1967, hockey playoffs went into late May. Those years, the town kept the cooling coils going into the late Spring.
The year-by-year record of the ice breakup, 389-1985, is graphed below.

Here is the same record, except smoothed by averaging adjacent years' dates. This is the output of an 11-year rolling average (as was shown by Tiljander et al. (2003) in their Figure 5, for the Lake Korttajarvi proxies).

These data are downloadable from BitBucket, as the Excel file Jarvykortta-Ice-Out-Proxy.xls.

So: how should the Ice-Out signal be connected to regional warming or cooling? The Abbott wisely suggested that when the winters are longer and harsher, the ice breakup is later in the year. Conversely, he offered, milder winters lead to an early breakup of the ice.

The Abbott also counseled me to account for the confounding effects of the hot springs piping and the hockey rink on the date of ice breakup.

As we parted, the Abbott offered me a warning on the pitfalls of over-reliance on computers. For instance, he noted, an algorithm might interpret the progressively advancing ice breakup date from 1850 to the present as the consequence of regional warming.  That would lead to a very odd result, orienting the Jarvykortta River Ice-Out signal such that ice breakups later in the year would be correlated with warmer temperatures.

"River ice breaking up in June as an indication of mild winters!," I exclaimed. "That makes no sense at all!"

After saying my farewells, I set out on my trek back to the capital city.  I stole a final look backwards towards the monastery, resplendent in the setting sun.

The Abbott was chuckling.

* * * * * * * * * *

Leaving Ruritania and moving to Finland:  during the course of her graduate studies at the University of Helsinki, Mia Tiljander had a leading role in obtaining cores of the silt at the bottom of Lake Korttajarvi.  Because those sediments are "varved," they can be dated with precision to the year they were deposited.  Tiljander and her advisor preserved the borehole cores and analyzed them, seeking information on the climate of the past 3,000 years in that part of Finland.  These studies were written up in a 2003 peer-reviewed article in Boreas, and in her doctoral dissertation (references here).

Among her studies, Tiljander measured the ability of each year's deposits to absorb X-Rays.  Low X-Ray Density (measured in greyscale values) indicates a high organic content of that year's sediment. The XRD values range from a low of 46 to a high of 172.  Tiljander interprets a high organic content to mean that the summer was warm and the growing season long.  Boreas (2003), page 570:
The thickness of the organic lamina is the sum of autochthonous production within the lake and the amount of allochthonous material and is related to the primary production during the open water season. The layer above the mineral matter is defined as organic, because of the less dense structure in X-ray images and low grey-scale values. A thick organic lamina probably indicates a warm summer and a relatively long growing season.
The allegorical Ruritanian "Jarvykortti River Monastery" proxy is comprised of a set of figures  that is identical to the actual set of Lake Korttajarvi varve X-Ray Density greyscale values.  Excel translated the numbers from greyscale values into days-of-the-year.  

For instance, the X-Ray Density of the varve that was deposited in 1220 was 54. So, in 1220, I had the ice on the Jarvykortta River break up on February 23--the 54th day of the year. In the decade of the 1220s, the ice broke up as early as Feb. 17 (1229), and as late as Feb. 28 (1225).

By contrast, when 170 years had passed, the ice in front of the mythical Jarvykortti Monsastery broke up about 5 weeks later in the season. For the 1390s, the earliest year of the ice breakup was 1395 (March 20), while the latest was 1391 (April 15).

In her paper's Table 2, Mia Tiljander noted 8 special periods of low X-Ray density (much organic matter; long growing seasons) and 10 special periods of high XRD (much mineral matter, short growing seasons). Among them:

Years Duration Varve structures
AD 1380–1420 40 yr Mineral matter
AD 980-1250 270 yr Organic matter

Thus, according to Tiljander's interpretation, the low X-Ray Densities of the 1220s corresponded to a time of warmer summers and longer growing seasons.  The higher XRD values of the 1390s indicated a time of cooler, shorter summers.

These trends can be seen in the Lake Korttajarvi X-Ray density trace.  Here it is, oriented in the same fashion as the days-of-the-year Korttajarvi River Monastery record, earlier.

But it makes more sense to orient the Lake Korttajarvi varve record conventionally, such that higher temperatures are nearer the top of the graph.

The grey-colored box to the right of the figure (1750-1995) serves as a reminder of Tiljander's warning that the Lake Korttajarvi record became progressively more dominated by local non-climate signals during the 18th, 19th, and 20th centuries.  Boreas, page 572:
In the case of Lake Korttajarvi it is a demanding task to calibrate the physical varve data we have collected against meteorological data, because human impacts have distorted the natural signal to varying extents during the past 280 years and the meteorological data in the Jyvaskyla area are only available since 1881.
And Boreas, page 575:
In the 20th century the Lake Korttajarvi record was strongly affected by human activities. The average varve thickness is 1.2 mm from AD 1900 to 1929, 1.9 mm from AD 1930 to 1962 and 3.5 mm from AD 1963 to 1985. There are two exceptionally thick clay-silt layers caused by man. The thick layer of AD 1930 resulted from peat ditching and forest clearance (information from a local farmer in 1999) and the thick layer of AD 1967 originated due to the rebuilding of the bridge in the vicinity of the lake's southern corner (information from the Finnish Road Administration). Varves since AD 1963 towards the present time thicken because of the higher water content in the top of the sediment column. However, the gradually increasing varve thickness during the whole 20th century probably originates from the accelerating agricultural use of the area around the lake.
To recap, the mythical Jarvykortta Monastery records are unreliable from the mid-1700s on, thanks to hot-spring diversion and hockey rink operation. The effect of both these artifacts was to progressively increase the date of ice breakup through the 19th and 20th Centuries -- a phony "getting cooler" signal.

In like manner, The Lake Korttajarvi X-Ray Density records are unreliable from the mid-1700s on, thanks to agricultural activities and road-building. The effect of both these artifacts was to progressively increase varve X-Ray Density through the 19th and 20th Centuries -- a phony "getting cooler" signal.

In their search for suitable new temperature proxies, Mike Mann's lab group developed an algorithm that would identify records that showed statistically-significant changes during the 1850 to 1995 period of warming. The Lake Korttajarvi Varve record and the Jarvykorrta River Monastery record both show a pattern of rising X-Ray Densities or ice-out dates over that time. The rises coincide nicely with the computed Southeast Finland regional temperature trend.

Rising Temperatures.
Rising Values (XRDs or ice-out dates).

Is this statistically-significant linkage of the Lake Korttajarvi XRD record and the computed regional temperature trend reflective of a real-world connection?

Even though the suggested relationship defies common sense?

Even when contamination of the record has flipped the relationship Upside-Down from what it should be?

* * * * * * * * * *

Here's a comparison of the actual Lake Korttajarvi XRD proxy and the Ruritanian Jarvykortta River Monastery one:

SeriesLake Korttajarvi Varve XRDJarvykortta River Ice Breakup
CharacteristicX-Ray Densities of Lakebed SedimentsDate of Ice Breakup of River
SourceTiljander et al (2003)Monastary Abbott
Low value1231: 46 (greyscale)1231: Feb 15 (46th day of year)
High value1967: 172 (greyscale)1967: Jun 20 (172nd day of year)
Influence of temperature on series?YesYes
Source’s interpretation of temp. effectsHarsher winters lead to mineral-rich varves and higher XRDsHarsher winters lead to later date of ice breakup
Non-temperature natural influences on series?YesYes
Source’s interpretation of non-temp. influencesHigher precipitation increases XRD valueLesser flow of hot springs upriver increases date of ice breakup
Human influences?YesYes
Source’s interpretation of human influences1720 on, farming increased
sedimentation and XRD values
1720 on, diversion of hot springs increased ice breakup dates
Known artifactsLate 1920s: Peat cutting increased XRDs
1967: Bridge reconstruction increased XRD
Late 1920s: Skating rink increased ice breakup date
1967: Hockey final increased ice breakup date
Source’s summaryHigher XRDs mean lower temps, unreliable post-1720Later ice breakups mean lower temps, unreliable post-1720
Mann et al (2008)
Computer screening algorithm
1850-1995 shows that higher XRDs
must mean higher temps
Common sense interpretationHigher XRDs mean lower temps. Unreliable post-1720Later ice breakups mean lower temps. Unreliable post-1720

* * * * * * * * * *

It sounds plausible when defenders of the Upside-Down usage of the Lake Korttajarvi X-Ray Density proxy claim, "The higher the varve XRD, the warmer the average temperature." After all, who goes through the day thinking about varve densities?

But this is the same as claiming, "The later in the season that river ice breaks up, the warmer the average temperature."

Everybody knows that when the winter is warm, ice breaks up early--not late!

The identical logic lets us know what Mia Tiljander's varve data is telling us about the climate in the vicinity of Lake Korttajarvi. For instance, it was warmer in the 1220s than it was in the 1390s.

But the 1800s and 1900s? Who knows? The record's no good.

* * * * * * * * * *

The controversy over the use of the Tiljander varve proxies is confusing, at first. "Top-notch scientist Mann publishes in a prestige journal, skeptic McIntyre complains, Mann responds. AGW Consensus vs. Denialists, versus, bla bla bla. Who can tell?"

But on closer inspection, it's simple story. The Mann group's calibration algorithm inadvertently flipped the X-Ray Density Tiljander proxy (and at least one other) so that Cold was interpreted as Warm, and vice versa.

Whoops. Mike Mann made a mistake, Steve McIntyre was correct.

In most other fields, this would have ended with Mann's group promptly acknowledging the error, re-doing the affected calculations, and submitting a formal Correction to the journal.

It's going on two years, and there is no sign of that happening.

Instead, a kaleidoscope of explanations, excuses, and Who's-On-First? routines has been set off by Mann's Delphic response in PNAS: that McIntyre's claim of an error was "bizarre." The spectacle continues to this day.

I've read that: Mann was right--Proxies can't be upside-down--It's too complex for you to understand--Mann interpreted the proxies correctly, as did Tiljander and Kaufman too--There is no single "right" interpretation--The alleged mistake doesn't matter anyway--Look over there, a pony!--You have to be super-knowledgeable about varves to earn the right to an opinion--The proxy data are interpreted by computers, so orientation doesn't matter--Supplemental Figure S8a proves that Mann was correct--Ah, grasshopper, what is this notion of "correct," anyway?--Lots of other studies agree with the conclusions of Mann et al (2008)--Don't you have anything better to do with your life?--Mann's use of the proxies is consistent with Tiljander's interpretation--McIntyre's a fossil-fuel shill!--Mann should be free to ignore Tiljander's interpretation; maybe he did and maybe he didn't--Your simplistic attempts to negate the power of statistical algorithms don't fool anyone.

And so on.

But in science, this sound and fury signifies nothing. At the end of the day, the Mann lab's misplaced trust in their computational algorithms have fooled them into mixing up colder and warmer. It's as simple as that.

Don't change the data to fit the theory--that's backwards!

[Update 12 Mar. 2010 -- minor text edits for clarity.]


  1. It seems a pity that you take the time to put this work up, the comparison between ice & silt is very good, and no bugger reads it!

  2. An excellent summary. If it is confirmed that Mann 2008 did use the XRD proxy with greater density = warmer orientation, then all his defenders should be ashamed of themselves.

    Unless they can accept errors when they can be shown to have occurred, they are not competent to comment on the field. That includes a definite statement on the potential competence of well known commentators such as Gavin Schmidt, Grant Foster (Tamino), & William Connelly, all of who have steadfastly refused to look at the actual facts.

  3. Ed Snack,

    Jeff Id (the Air Vent) recently pointed out the correlations for the four Tiljander proxies to me. Here they are, as recorded by Prof. Mann in the file "1209proxynames.xls", downloadable from his PSU website. Column U, "pass-screening over 1850-1995 (r)"

    0.3066 Darksum
    0.2714 Lightsum
    0.2987 Thickness
    0.1232 X-Ray Density

    As you can see, XRD had the lowest correlation with gridcell temperature anomaly. I believe that 0.1232 is below the passing cutoff.

    This finding does not affect the evidence put forward in this post. The same reasoning is applicable to Mann08's treatment of Lightsum, which did pass this test.