January 30, 2012

The Problem with Proxies

Filed under: Droughts, Precipitation

There is a new paper (in press in the journal Geophysical Research Letters) that presents a lesson that we all should keep in mind—results based on reconstructions of climate phenomena that are based on once or twice removed “proxy” indicators, may not be as reliable as they appear (or as they are presented) to be. If this brings hockey sticks and salacious emails to mind, you are not alone.

“Proxies” are putative indicators of climate for which there are no direct measurements. Tree rings, for example, are wider when the summer is wet and not too hot. The actual “explained variance” between them and, say, annual temperature is complex to derive and not all that high. The same is true for most other types of proxies (e.g., corals, ice cores, lake sediments, stalagmites, boreholes, etc.). Therefore, the uncertainties in using proxies to “reconstruct” some aspect of the climate are typically large (certainly larger than typically portrayed) and making (robust) conclusions from such analyses becomes a bit tricky. Such problems are among the reasons that many people jumped all over Michael Mann’s infamous “hockey stick” reconstruction of climate, which claims to accurately represent annual temperatures on a year-to-year basis back some 1000 years. Lesson: be very careful with proxy climate data.

Today’s example involves wildfire occurrence in the western U.S., and the climate patterns that may influence it.

Several years ago we ran an article that highlighted the results of research done by Thomas Kitzberger and colleagues that was published in the Proceedings of the National Academy of Sciences reporting a relationship between wildfires in the western U.S. and semi-oscillatory conditions in the Pacific Ocean, namely the multi-year rhythm of El Niño/La Niña (ENSO) and the multi-decadal variability of the Pacific Decadal Oscillation (PDO).

ENSO is the well-documented (back to the 19th century) oscillation of temperatures in the tropical Pacific Ocean when the normal cold pool of water that upwells west of South America is suppressed and then recovers. When there’s an El Niño on, global temperatures are above any mean or long-term trend line for a year or so. When La Niña takes over, global temperatures drop below normal (or the trend) for a little while.

As it name implies, the PDO is much a much longer-lived temperature oscillation, and one that is reflected more in the temperate Pacific Ocean than in the tropics. It’s been related to fish distributions, ocean currents and, yes, western wildfires.

We highlighted the Kitzberger work a few years back because, at the time, there was a lot of talk about how global warming was leading to a dramatic increase in forest fires in the western U.S. Kitzberger showed that natural variability also had a large role to play in the wildfire climate of region, and documented a relationship between ENSO, PDO, Atlantic Ocean temperature anomalies, and wildfires for several centuries into the past. Here is a quote from our WCR article:

“…one might be surprised to learn that most large-scale fire events in the western United States over the last five hundred years have been fundamentally the result of natural ocean climate cycles, and not global warming.”

Now, a new study led by Kurt Kipfmueller of the University of Minnesota suggests that the linkage between PDO and western fire occurrence identified by Kitzberger and colleagues over the past several hundred years may not be robust signal.

In order to investigate whether there is a long-term relationship between fire and the PDO, you need two things: a long-term record of fire occurrence and (surprise), a long-term record of the behavior of the PDO. The occurrence of fires can be established directly from fire scar records preserved in tree rings cores, but the state of the PDO can only be inferred indirectly using some sort of a PDO proxy.

And to make matters worse, a PDO proxy does not directly reflect the PDO, but rather some aspect of the weather that may itself be impacted by the state of the PDO, such as seasonal precipitation or soil moisture. So reconstructing the strength of the PDO prior to there being direct measures of the Pacific sea surface temperatures can be a bit tricky (see our Figure 1).

However, there have been several different attempts to do just that. What Kipfmueller and colleagues found was that the different reconstructions of the PDO yielded different, and often times opposite, relationships with western U.S. fire occurrence (again, Figure 1). So, any conclusions as to how the PDO effects western U.S. wildfires depend strongly on which PDO reconstruction that was used in the study. This is the problem with proxies!


Figure 1. Several different proxy reconstructions of the Pacific Decadal Oscillation (PDO) highlighted in the Kipfmueller et al. article. Notice how the reconstructed values of the PDO (the solid black lines) from each research effort match up pretty well with the observations during the period over overlap with direct observations (black dotted lines), but also how the different sets of reconstructions vastly differ in the period prior to observations. The different colored triangles represent fire occurrences in different regions of the wester U.S. (Figure adapted from Kipfmueller et al., 2012). For our friends over at Skeptical Science, please note that a reconstruction of ENSO has been deleted from the original figure as we feel that it is irrelevant to our present discussion. Please feel free to throw another tantrum.

Here is how Kipfmueller sum it up:

These discrepancies have major implications for any attempts to use paleoecological data to examine potential connections between the pre-instrumental behavior of the PDO and regional fire activity. To put it bluntly, any conclusion that extensive wildfires are more or less common when the Pacific Decadal Oscillation is in one phase or the other depends to some extent on the choice of PDO reconstruction. Using one set of reconstructions, region-wide fire activity appears to be higher when the PDO is positive; in another set, fires are more frequent when the PDO is negative. Until the paleoclimate community can reach a consensus regarding the temporal evolution and spatial structure of low-frequency behavior in the north Pacific prior to 1900, any attempts to use proxy data to determine whether or not the state of the PDO exerts a consistent influence on wildfire activity in the western United States (or elsewhere) will produce ambiguous results.

This is not to say that the PDO has no influence the frequency of wildfires in the western U.S., just that what that influence is cannot be robustly established using existing data sets, and the results from previous efforts to do so (such as Kitzberger et al., 2006 and several other research teams which have undertaken similar investigations) are not overly reliable. Thankfully, for Kitzberger and colleagues, their original conclusions concerning fires and climate variability did not solely depend on the PDO, but also ENSO, and even a strong influence from the Atlantic Ocean as well.

As for our original WCR article titled “Wildfires: The Results of Cyclic Oceanic Variations or Global Warming?”, well, we don’t think that the new Kipfmueller et al. results have much impact on our original take home message from Kitzberger’s research—that wildfires and droughts are linked in the western U.S., and that elements of the natural climate interact to influence the occurrence of wildfires there.

But Kipfmueller’s results do give plenty of room for thought about just how much faith to place on results derived from reconstructions of climate phenomenon based on once or twice removed proxy indicators.
References:

Kipfmueller, K., et al., 2012. Does proxy uncertainty affect the relations inferred between the Pacific Decadal Oscillation and wildfire activity in the western United States? Geophysical Research Letters, in press.

Kitzberger, T., Brown, P.M., Heyerdahl, E.K., Swetnam, T.W. and T.T. Veblen, 2007. Contingent Pacific-Atlantic Ocean Influence on multicentury wildfire synchrony over western North America. Proceedings of the National Academy of Sciences, 104 (2), 543-548.




No Comments

No comments yet.

RSS feed for comments on this post.

Sorry, the comment form is closed at this time.

Powered by WordPress