One of those commenting on the paper by Shepard et al. in the journal Natural Hazards was Albert Parker. Rather than dissect his comment on Shepard et al., let’s take a look at another paper he recently published in that same journal, Oscillations of Sea Level Rise along the Atlantic Coast of North America north of Cape Hatteras, (2012, Nat Hazards, DOI: 10.1007/s11069-012-0354-7).
Essentially, it’s an attempt to refute Sallenger et al. (2012, Nature Climate Change, doi:10.1038/nclimate1597), and more broadly speaking, it’s an attempt to deny the existence of acceleration in sea level rise, pretty much anywhere, anytime — by showing the acceleration in sea level rise.
The first thing that struck me about the paper is this:
Perhaps they were in a hurry. The next thing that struck me is this graph of sea level at “the Battery” (New York):
Several things are notable. First, the graph is rather low resolution and the lines are thick enough to obscure the details. That’s bad practice, and should have been a red flag for the editors. Perhaps they were in a hurry. Another interesting fact is that the values range from about 1200 to 2000, whereas the values I downloaded range from about 6500 to 7300. Looking at the references to find his pointer to the data shows this (with link):
Permanent Service on Mean Sea Level (2012) Data and station information for the battery, NY. www.psmsl.org/data/obtaining/rlr.monthly.data/1896.rlrdata. Accessed 19 June 2012
The link goes to station data from PSMSL for West Tuas, Singapore.
Must be a misprint. You’d think the editors would have caught that. Perhaps they were somewhat hurried.
Nonetheless, the suffix “.rlrdata” indicates “revised local reference” data. PSMSL states that “For scientific purposes, the RLR dataset is normally superior to the ‘METRIC’, although the latter, which contains the total PSMSL data holdings, can also be analysed bearing in mind the above datum continuity considerations.”
It turns out the data used by Parker are the “metric” data, which lack certain data continuity checks performed by PSMSL. It doesn’t affect the analysis because for this station, the difference between the RLR and metric data sets is a constant. Still — you’d think the editors would have checked this stuff. Maybe they were a bit rushed.
One thing the graph shows which does affect analysis is that the annual cycle of variation has not been removed. Here’s the raw data (same as above only offset by 5276 mm because it’s the RLR data file rather than metric):
If we linearly de-trend, then fold it with a period of 1 year, we can see that the annual cycle is substantial:
Here’s a tip for Parker, and Burton, and Boretti, and all the other hacks who play around with sea level data from PSMSL: use their annual rather than monthly data. It has almost no effect on trend analysis, and saves you the trouble of removing the annual cycle (which seems to be too difficult for you).
Oh well, leaving in the annual cycle doesn’t affect the analysis very much. Still, you’d think that the paper’s reviewers, if they were savvy about time series, would have pointed this out and insisted on a revision. Perhaps the review was on the “fast track.”
Let’s get down to the actual substance of Parker’s paper (finally!). It starts thus:
Sallenger et al. (2012) use in their analysis a short time interval of just 30 years to compute the sea level rise (SLR) to analyse the tide gauges along the North American Atlantic coast north of Cape Hatteras. Because there is a relevant 60-year cycle in the North Atlantic records, the 30-year choice is the worst that can be chosen because it is half the 60-year cycle. Fitting the last 30 years, the real SLR trend is overestimated, and fitting the previous 30 years, the real SLR trend is underestimated. The comparison of these two SLRs is not a proof of the existence of hot spots in the positively accelerating seas as claimed by the authors but only the evidence that the authors apparently do not know of the multi-decadal oscillations in the Atlantic.
I have to wonder … did Parker even read Sallenger et al? They didn’t just pick 30 years as the time spans to compare, and they weren’t attempting to prove acceleration in the NEH (NorthEast Hotspot), although they did so. What they did is demonstrate a spatial pattern of sea level differences, by computing the difference in SLR (seal level rate) between the 1st- and 2nd halves of a time window. They did this using windows with widths 60, 50, and 40 years. Why, they even applied some statistics — to estimate the uncertainties in those differences (called “SLRD” for “sea level rate difference”).
Spatial patterns are expected to exist, and to change with rising sea level according to model results. Their goal, as stated in the abstract, is to look for observational evidence of their existence:
These sea-level variations form unique spatial patterns, yet there are very few observations verifying predicted patterns or fingerprints6.
What they found is that for the most recent data (up to the end of 2009) there is a distinct spatial pattern. For a stretch from about Cape Hatteras to Cape Cod on the Atlantic coast of the U.S., there was statistically significant higher SLR in the 2nd half of the time span (the most recent 30, 25, or 20 years) than in the 1st half of the time span (the preceding 30, 25, or 20 years). Here, for instance, is their map of SLRD using the 60yr time span:
Sallenger et al. then report:
The observed NEH is similar to the modelled NEH projected for the end of the twenty-first century2-4 and later1. A robust prediction across all models is for significantly greater SLR north of Cape Hatteras in agreement with the observed NEH.
It’s difficult to know how Parker failed to realize that Sallenger et al. are perfectly aware of the complex pattern of accelerations and decelerations in tide gauge records from North America and of the dependence of the result on the size of the time span. In fact they even computed SLRD at hotspot stations between the 1st and 2nd halves of all possible time spans from 1894 through 1970, to the present, time spans ranging from 40 to 115 years:
What I find truly mystifying is Parker’s snide remark “evidence that the authors apparently do not know of the multi-decadal oscillations in the Atlantic.” If you want to make snide remarks like that, get yourself a blog. If you’re writing a scientific paper … It’s also deplorable because of its blatant and obvious falsehood. Sallenger et al. are perfectly aware of multidecadal oscillations in the Atlantic, they even computed cross-correlation between SLRD and rate differences of both AMO (Atlantic Multidecadal Oscillation) and NAO (North Atlantic Oscillation).
Parker wants us to believe that sea level is oscillating about a constant trend, so there’s no “real” acceleration (his new version of the “no acceleration” thing), since the implication is clear that any upswing will be followed by an inevitable downswing.
Even if Sallenger et al. (2012) apparently do not, somebody else at NOAA knows that the sea levels are oscillating, with the multi-decadal oscillations up to the 60 years requesting proper consideration to infer trends from recorded data. The “Executive Summary” of the report NOAA (2009) says: “…50–60 years of data are required to obtain a trend with a 95% confidence interval of +/- 0.5 mm/yr. This dependence on record length is caused by the inter-annual variability in the observations.”
Newsflash: neither the length of time required for a given trend precision, nor the presence of inter-annual variability, are evidence of oscillatory behavior. Shortly after, Parker says this:
A minimum of 60 years of recording are needed to compute a sea level longer-term trend cleared of all the shorter-term oscillations up to the Atlantic multi-decadal oscillation of period 60 years (some other references in the literature to the approximately 60-year cycle are provided by Burton 2012 and a detailed dissection of Sallenger et al. 2012 is provided by Tisdale 2012).
Wow. Just … wow.
Burton 2012 and Tisdale 2012 are not “the literature.” They’re blog posts. Not very good ones. If you look at Parker’s reference list, you’ll find technical reports from NOAA, PSMSL as a data source, blog posts by Burton and Tisdale, but only one actual peer-reviewed scientific paper: Sallenger et al.
Parker charges ahead, computing the trend rate at New York (the Battery) for moving 20- 30-, and 60-year time spans. Here’s his result for the 30-year spans
The plotted time is the endpoint of the 30-year time window. I can’t imagine why his time axis starts at 1852 when the plotted data don’t start until 1923 … seems kind of a waste of space. Maybe he was in a hurry. He does mention that “In the specific location of The Battery, NY, the value of the December 1953 30-year SLR is larger than the December 2009 value, but obviously Sallenger et al. (2012) do not use this comparison to prove that the sea levels are decelerating.” So, at least Parker got in another “shot” and was able to show a graph indicating that SLR was higher in the past than it is as present. Or … does it?
Don’t get the wrong impression, that the plotted values represent the average SLR at the time plotted. Actually they represent the average for a time window ending at that time, so they better represent the SLR at the midpoint of the time span — the value plotted at 2012 represents the estimated SLR around 1997.
I recently modified my smoothing program so that in addition to computing the estimated smoothed value from data, it would also compute the estimated smoothed velocity at each moment of time. With the right choice of time constant, the estimates are very close to those computed from linear regression on a 30-year window. This makes it possible to estimate the 30-yr rate up to the present. The uncertainty on that estimate is higher than for a moment of time with data both before and after … but it’s still a more realistic estimate than taking the 1997 value to represent 2012.
Here’s the SLR computed from linear trends on 30-year time spans, with the plotted time being the midpoint of the time span (I didn’t feel it necessary to do so for every possible month, just for each year, I did feel the need to add 2-sigma error bars, and yes I de-seasonalized the data first):
When we superimpose the estimate from the smoothing program we find that they agree quite well, although the 30-year time spans don’t provide estimates after 15 years ago:
The latest values are higher than previous values. The uncertainty is large enough that we can’t say with confidence that SLR at the Battery is faster now than at any time during the 20th century … but we certainly can’t agree with Parker that is isn’t.
Next, Parker tries to show the lack of acceleration, and even the existence of deceleration, in North American tide gauge data by comparing linear trends of PSMSL data up to 1999 (computed by NOAA), to those from data up to 2006. After all that harping about how you need at least 60 years due to oscillatory behavior, he now wants us to believe in deceleration based on a difference of just 7 years between data sets? Why, yes, he does.
And need I remind you that Sallenger et al. isn’t about 20th-century North American SLR acceleration anyway? It’s about the spatial pattern of recent changes in SLR in North America.
What about that spatial pattern? I took the 20 “hotspot” stations listed in table S1 of the supplementary information to Sallenger et al., and the 36 North American “not hotspot” stations, and compared their changes in SLR. To do this, I first removed the annual cycle from each. Then I linearly detrended each because some of them show dramatically different overall rates due to glacial isostatic adjustment, but changes in the rate aren’t much affected by GIA which has been reasonably constant throughout. Then I aligned the linearly de-trended de-seasonalized records, separately for “hotspot” and “non-hotspot” stations. Here’s the result for “hotspot” stations:
Here it is for “non-hotspot” stations:
Estimating the SLR departure from its overall mean value, gives this comparison between the two groups:
Perhaps you notice the big difference. It’s that the hotspot stations show much faster sea level rise after about 1990 than the non-hotspot stations. Which is exactly what Sallenger et al. established.
Parker then says this:
More than the Table 1, it is the visual scanning of the monthly departures from a smooth, long-term linear trend for all the 128 stations of the United States (as well as for all the 195 other global locations proposed by the National Oceanic and Atmospheric Administration NOAA, 2012b) that shows no sign of sharp departures and rules out the acceleration claim.
In my opinion, the most bizarre aspect of this is that Parker actually believes it. Truth is stranger than fiction.