We’ve been looking closely at the written testimony from Judith Curry before a recent meeting of the Environment and Public Works committee of the U.S. Senate. What we’ve seen so far argues against relying on Curry to give accurate and relevant information.
One of her main evidences that the IPCC AR5 (5th assessment report) is wrong about expressing greater confidence than the AR4, is her discussion of sea level. We’ve already looked at a small part of her discussion, a statement so misleading that I was amazed she would actually say it.
But her main argument regarding sea level rise is this:
It is seen that the rate of rise during 1930-1950 was comparable to, if not larger than, the value in recent years. Hence the data does not seem to support the IPCC’s conclusion of a substantial contribution from anthropogenic forcings to the global mean sea level rise since the 1970s.
She’s referring to (and reproduces) this graph from the IPCC AR5:
It shows trend estimates (i.e., the rate of sea level rise) over time, based on linear regression of 18-year time spans from three global sea level data sets (reconstructions based on tide gauge data), and the trend over the last 18 years based on satellite data (labelled “altimeter”), which at the time of writing only covered 18 years (which is why, I believe, they chose 18 years as their time scale). The times which are plotted are the beginning of each 18-year time span.
I have three complaints about Curry’s argument:
Let’s take each in turn.
1) Curry fails to mention why IPCC AR5 shows this graph or what they say about it
Here’s what they say:
A long time-scale is needed because significant multidecadal variability appears in numerous tide gauge records during the 20th century (Holgate, 2007; Woodworth et al., 2009; Mitchum et al., 2010; Woodworth et al., 2011; Chambers et al., 2012). The multidecadal variability is marked by an increasing trend starting in 1910–1920, a downward trend (i.e., leveling of sea level if a long-term trend is not removed) starting around 1950, and an increasing trend starting around 1980. The pattern can be seen in New York, Mumbai, and Fremantle records, for instance (Figure 3.12), as well as 14 other gauges representing all ocean basins (Chambers et al., 2012), and in all reconstructions (Figure 3.14). It is also seen in an analysis of upper 400 m temperature (Gouretski et al., 2012; Section 3.3.2). Although the calculations of 18-year rates of GMSL rise based on the different reconstruction methods disagree by as much as 2 mm yr–1 before 1950 and on details of the variability (Figure 3.14), all do indicate 18-year trends that were significantly higher than the 20th century average at certain times (1920–1950, 1990–present) and lower at other periods (1910–1920, 1955–1980), likely related to multidecadal variability. Several studies have suggested these variations may be linked to climate fluctuations like the Atlantic Multidecadal Oscillation (AMO) and/or Pacific Decadal Oscillation (PDO, Box 2.5) (Holgate, 2007; Jevrejeva et al., 2008; Chambers et al., 2012), but these results are not conclusive.
While technically correct that these multidecadal changes represent acceleration/deceleration of sea level, they should not be interpreted as change in the longer-term rate of sea level rise, as a time series longer than the variability is required to detect those trends. Using data extending from 1900 to after 2000, the quadratic term computed from both individual tide gauge records and GMSL reconstructions is significantly positive (Jevrejeva et al., 2008; Church and White, 2011; Rahmstorf and Vermeer, 2011; Woodworth et al., 2011). Church and White (2006) report that the estimated acceleration term in GMSL (twice the quadratic parameter) is 0.009 [0.006 to 0.012] mm yr-2 (1 standard deviation) from 1880 to 2009, which is consistent with the other published estimates (e.g., Jevrejeva et al., 2008; Woodworth et al., 2009) that use records longer than 100 years. Chambers et al. (2012) find that modelling a period near 60 years removes much of the multidecadal variability of the 20th century in the tide gauge reconstruction time series. When a 60-year oscillation is modeled along with an acceleration term, the estimated acceleration in GMSL since 1900 ranges from: 0.000 [–0.002 to 0.002] mm yr–2 in the Ray and Douglas (2011) record, 0.013 [0.007 to 0.019] mm yr–2 in the Jevrejeva et al. (2008) record, and 0.012 [0.009 to 0.015] mm yr–2 in the Church and White (2011) record. Thus, while there is more disagreement on the value of a 20th century acceleration in GMSL when accounting for multi-decadal fluctuations, two out of three records still indicate a significant positive value. The trend in GMSL observed since 1993, however, is not significantly larger than the estimate of 18-year trends in previous decades (e.g., 1920-1950).
The whole point of this discussion, and of their figure 3.14, is to show that if you want to estimate changes in the rate of sea level rise which are climatically relevant, “A long time-scale is needed“. The IPCC report didn’t ignore or downplay multi-decadal variability (although Curry seems to think that it does about everything, not just sea level rise). It doesn’t pretend that multidecadal variability isn’t natural variation, in fact it gives at least one possible root cause which is “natural.” The IPCC report didn’t ignore either the existence or the possible causes of multidecadal variability, they decided instead to deal with it. I think the one who really needs to “deal with it” is Judith Curry.
The main point of the IPCC discussion is that those short-term, decadal to multi-decadal, possibly natural variations should not be interpreted as change in the longer-term rate of sea level rise, so don’t use them to draw conclusions about climatically induced acceleration or its absence. I thought their statement was pretty clear. Apparently Judith Curry either didn’t get it, or didn’t want to, because she has done exactly what the IPCC report warns you should not do. My opinion: classic Curry.
As for the truth or falsehood of the claim that “the rate of rise during 1930-1950 was comparable to, if not larger than, the value in recent years,” let’s take a look at this graph of the rate of sea level rise:
The thick red line shows the estimated rate based on linear regression applied to 10-year time spans. The dashed gray line shows the long-term rate, i.e. the one that’s relevant to climate change. Note that the decadal rate in the 1940s is comparable to, if not higher than, the most recent rate, but for the long-term rate it is not.
In this case, it is most certainly the long-term rate which is correct while all those fluctuations are just plain wrong.
You might be thinking, “Who died and made Tamino the arbiter of what’s “right” and “wrong” in estimates of the rate of sea level rise?” Or maybe “Tamino is just calling the fluctuations “wrong” because he doesn’t like them!”
This is one of those cases in which I can be sure, because the data which produce this figure are artificial data. Therefore the answer is already known — with certainty. The dashed gray line is the right answer, the true sea level rise signal. The thick red line results from the fact that Rahmstorf et al. (2012, Clim. Dyn. 39, 861–875, DOI 10.1007/s00382-011-1226-7) added noise to the artificial signal, noise which in fact emulates that found in the Church & White data set.
Rahmstorf et al. also point out that much of the noise in sea level data isn’t just “ordinary” noise. It’s not entirely (or perhaps even predominantly) due to “natural variability,” it’s in large part due to coverage bias. Since it’s bias and not just stationary noise, that means that even the “error bars” we compute by treating it as noise can exclude the true value. They also point out that satellite altimetry shows less fluctuation than reconstructions based on tide gauge data, arguing that some if not much of the observed decadal and multidecadal variability may be an expression of noise, not a reflection of signal.
It’s also worth mentioning that of the three global data sets, the one which shows the highest rate in the first half of the 20th century is the Jevrejeva et al. data, which uses an unusual method of averaging tide gauge records that ends up giving the northern hemisphere oceans greater statistical weight than the southern hemisphere oceans in spite of the fact that the area of the southern hemisphere oceans is much greater than that of the northern hemisphere oceans. That’s exactly the kind of treatment which can lead to coverage bias, not just extra noise.
There’s another aspect which should be mentioned. Computing trend rates based on sliding 18-year windows is the application of a “moving velocity filter” to the data. That means it really represents the estimated rate at the mid-point in time of the observation window. Like with moving averages, in most cases we insist that the window is complete, so our very first estimate applies to half a window width later than the start of the data while the final estimate applies to half a window width before the end of the data. We lose half a window width at each end of the time series.
Here, for example, are the 18-year trends (moving velocities) of the Church & White data, with times plotted being the midpoints of the individual windows:
Note that it barely goes past the year 2000, but even at that the final value is still the highest — although not by much so the result is consistent with “comparable to”. But, notice also that at the very end the rate has been increasing, so it may well have increased further after 2000. To estimate the rate from the Church & White data all the way up to the year 2010 (when the data end), we need to a better way to estimate the rate than the standard “moving velocity” filter.
I, and others (Jevrejeva et al., Moore et al., Rahmstorf et al.) have argued that a good way to do that is with nonlinear smoothing. Those other authors have used SSA (singular spectrum analysis) to accomplish this, while I’ve tended to use lowess smoothing and pick out the linear coefficient at each moment as the trend estimate for that moment. Let’s see whether or not this is consistent with the results of the moving-velocity filter at the same time scale (nonlinear smoothing result in red):
Yes. Yes, it is.
What does it suggest happened after 2000? This:
Apparently the rate did keep increasing (in this data set), so the final estimated rate turns out to be the largest in the entire time span, and just about the same as the rate indicated by satellite altimetry.
Bottom line: even the claim that “the rate of rise during 1930-1950 was comparable to, if not larger than, the value in recent years” is by no means established as surely as Curry believes. Or, in my opinion, as the surely as the IPCC report states.
2) Even if true, Curry draws the wrong (and unjustified) conclusion
Suppose for the sake of argument that “the rate of rise during 1930-1950 was comparable to, if not larger than, the value in recent years.” Might even be true. How do you get from that to “Hence the data does not seem to support the IPCC’s conclusion of a substantial contribution from anthropogenic forcings to the global mean sea level rise since the 1970s,” especially if you don’t even mention the rate of rise on climatically relevant time scales? Logic fail.
3) The data show strong evidence of acceleration in the 20th century.
Jevrejeva et al. (2008, GRL, 35, L08715, doi:10.1029/2008GL033611) applied SSA nonlinear smoothing to estimate the time variations of sea level rise rate throughout their data set, getting this (the black line is global, the blue line for the northeast Atlantic region):
I did the same thing using lowess smoothing, getting this:
Clear result: in addition to multi-decadal variations there is also a consistent increase in the rate of sea level rise throughout the time span. That’s called “acceleration.”
Rahmstorf et al. did a similar analysis on the Church & White 2006 data, the Church & White 2011 data, and the Jevrejeva data (top: Church & White 2006; center: Church & White 2011; bottom: Jevrejeva et al. 2008). The colored lines are what we’re interested in, the estimated rise rates:
In all three cases there is a consistent increase in the rate (a.k.a. “acceleration”) superimposed on multidecadal variation, and in all three cases the estimated rate at the end is the highest of all.
Here’s how I see it: Judith Curry really did nothing more nor less than to scour the IPCC AR5 looking for stuff she could claim weakens the case for dangerous man-made climate change. In so doing, she was willing to ignore what the IPCC report actually says in favor of her preferred interpretation of things. She demonstrated more than once that she doesn’t have sufficient knowledge of what the data have to say, or of what the peer-reviewed literature says, to know what she’s talking about.
It’s rather disappointing, really, because if you’re determined to find fault that’s usually ridiculously easy in any report as lengthy as the IPCC AR5, but she still managed to botch the job. Dismally. She also utterly failed to mention, perhaps even to notice, anything in the IPCC report which strengthens the case. Seriously — are we actually to believe that there isn’t anything like that at all?
I also expect that regarding her testimony, Judith Curry will staunchly refuse to learn anything from the many critics (I’m far from the only one) who have found serious faults in her testimony.
But, that’s just my opinion.