Sea Level Rises … Tisdale Falls

A reader asked whether we might take a look at a recent post by Bob Tisdale on WUWT. Let’s do that, shall we? Incidentally, Tisdale has since created another post on WUWT on the same subject, which is: a paper by Sallenger et al. showing that sea level has recently accelerated in a region of the Atlantic coast, a region they call the “Northeast hot spot,” or NEH.

Sallenger et al. looked for acceleration in sea level rise by analyzing tide gauge data over time “windows” of various sizes. Their main method was to compare the rate of sea level rise (SLR) in the first half of the window to that in the second half of the window, mainly using 60-year, 50-year, and 40-year windows ending with 2009. They refer to the difference as “sea level rate difference,” or SLRD. Positive values of SLRD indicate acceleration — i.e, that sea level rise was faster in the 2nd half of the time window than in the 1st half. They even tested whether or not the difference was statistically significant. They even corrected their results for autocorrelation. Good for them!

They then looked for geographical patterns in SLRD along the Atlantic coast. That leads to this graph (which is in the supplemental information but I’d have liked to see in the main paper)

For latitudes from about Cape Hatteras to Boston, SLRD is higher than in other regions. Hence the “hot spot,” or NEH.

Tisdale doesn’t buy it. He acquired data for 8 stations, including 7 which Sallenger et al. include in the NEH. Tisdale uses monthly averages, whereas Sallenger et al. used annual averages. The monthly averages really don’t help you get at the long-term trends that Sallenger et al. are trying to measure but they don’t hurt that goal either — but they do keep all the noise intact so that time series graphs will be harder to interpret visually and statistically.

Tisdale also adds Wilmington NC because, he says, “John Droz was interested in North Carolina” (he’d been talking about the recent attempt by NC legislators to command sea level not to accelerate). Tisdale uses this data to cast doubt on the results of Sallenger et al. But there are problems with this. For one thing, Sallenger et al. include 20 stations in their NEH, they make no secret about which 20 stations, and all the data are easily available. But Tisdale uses only 7 of them. For another thing, Wilmington NC might be interesting — but it’s not in the NEH region, so you’ve got no business including it in any discussion of sea level in that area.

There’s another problem too. Tisdale got his data from KNMI, which is a wonderful website but their tide gauge data only go through 2008. It was ridiculously easy for me to get the data from PSMSL (Permanent Service for Mean Sea Level) like Sallenger et al. did. You don’t even need to google it, Sallenger et al. give the link in their paper and even give a hyperlink in the supplemental information. If you’re going to use a different end-point for time, use more recent data, not less!

Tisdale suggests that Sallenger et al. don’t show any graphs of actual sea level data from hotspot stations because they might be trying to hide something:

Curiously, there were no time-series graphs of the “hotspot” sea level anomalies in Sallenger et al (2012). That raises the question: what didn’t they want to show?

What they didn’t want to show was a graph that wouldn’t convey important results. The fact is, the raw time-series data are noisy enough that they don’t make the acceleration visually obvious — that’s what analysis is for. Besides which, I guess Tisdale didn’t look at the supplemental information, which includes a time-series graph for New York City.

Next, Tisdale averages the tide gauge records to estimate sea level for his “Hotspot+1″ region. But this introduces some bias into the estimate because not all the records are complete — there are missing months, so however you compute anomalies there’s no guarantee that different station records are properly aligned with each other. The records should really be aligned before averaging! And I reiterate, that it’s silly to talk about the behavior of the “hotspot” based on “Hotspot+1″ — Wilmington NC doesn’t belong, you’ve got no business including it.

Tisdale then computes trend rates for different segments of the data. But instead of emulating Sallenger et al. and comparing the 1st and 2nd halves of a time window, he simply splits the data since 1935 into two pieces of unequal length. He does this so he can cast doubt on the conclusion that the change in sea level rate occurred around 1990. Ironically, he first shows that the rate after 1990 actually is higher than that before, using this process:

First things first: Is there a break point around 1990 when the trends shift? There is definitely a shift in trends before and after 1989/90 in the detrended “Hotspot+1” sea level data, Figure 4. There are also comparable shifts in trend around 1976/77 and about 1981/82, Figure 5 and 6. But Sallenger et al (2012) went through a detailed analysis and elected to discuss the shift around 1990.

Tisdale doesn’t get it. All he’s shown is that the trend after various times is higher than the trend before various times — which is exactly what you’d expect from acceleration, whenever the change takes place.

In his second post, Tisdale gets around to accusing Sallenger et al. of cherry-picking their time windows to get their result:

The more I look at Figure 1, the more it appears that I’m presenting a classic example of trend comparisons with cherry-picked start years. It might appear to some that Sallenger et al (2012) cherry-picked 1990 to front load the short-term data with the impact of the eruption of Mount Pinatubo, which would have driven the trend up of the data after 1990, while I picked 1996 to front load the data with the impact of the 1997/98 El Niño, which would drive the trend of the data after 1996 down . I’ll admit to my cherry-picked start year. Will Sallenger et al (2012) admit to theirs and its impact on their claim of accelerated sea levels?

Tisdale really doesn’t get it. Sallenger et al. didn’t cherry-pick 1990, in fact they didn’t “pick” it at all — they determined it. They did so by trying windows starting with all possible start years from 1894 through 1970, ending with 2009, computing SLRD based on 1st-half vs 2nd-half trends. By noting the window which gave the strongest SLRD, they estimated that the rate shift had occurred around 1990.

Tisdale also smooths the data, first with a 13-month running average, then with a 121-month (approximately 10-year) running average, so he can show a graph (with no analysis) and proclaim “obvious” multidecadal variability. That’s fine, but it doesn’t make the long-term trend visually evident, you need to smooth on a time scale longer than 10 years.

You also need to estimate trends on a time scale longer than 10 years. Tisdale doesn’t get the fact that 10-year trends are over too short a time span to show the actual acceleration that has really occurred. But he proudly proclaims

Looking at the trailing decadal (120-month) trends, Figure 9, we can see very clearly that the trend of the most recent decade (through December 2008) is not as high as it has been in the past.

It’s just another example of drawing conclusions based on too short a time span — standard practice for fake skeptics.

It seems to me that Tisdale really doesn’t know what he’s doing, or even what he has done. In part he actually confirmed the acceleration of sea level in the hotspot region by showing trend increase, in spite of having place multiple obstacles in the way of proper analysis. In spite of this, he draws the wrong conclusion. Tisdale didn’t just stumble, he fell flat on his face.

In the next post we’ll leave Tisdale behind (which isn’t difficult) and take a look at actual tide gauge data from the Northeast hotspot. It’s much more interesting than Bob Tisdale’s opinions.

About these ads

13 responses to “Sea Level Rises … Tisdale Falls

  1. You also need to estimate trends on a time scale longer than 10 years. Tisdale doesn’t get the fact that 10-year trends are over too short a time span to show the actual acceleration that has really occurred

    Indeed. Not that I know much about complicated things like tides, but shouldn’t the minimum data span analysed be on the order of 20 years anyway to take account of the Metonic cycle?

  2. What a curious pair of posts from Bob Tisdale.
    I do find Bob is a contratian of a somewhat different stripe than most others so I wonder if his first post was deemed lacking (by normal cortrarian standards) & he quickly cobbled the second together to make good the damage.
    The second post is evidently an after thought but that implies it was produced in less than 48 hours, a conclusion reinforced by the man himself when he writes “Unfortunately, the Sea Level Time Series Wizard wasn’t working this weekend.

    This is not the only section of his second post that suffers missing data. The vast majority of it is all about SST and not about SLR at all!
    Yes, I understand that sea surface temperature and sea level are different datasets. I’m using sea surface temperatures as an example of an ocean climate variable, since model outputs and observed satellite-based sea level data are not available through the KNMI Climate Explorer for the term we’re interested in discussing.
    Of course, even satellite SLR data isn’t what Sallenger et al 2012 were discussing. That was tidal gauge data, something which appears to be well outside Tisdale’s comfort zone. “In other words, (tidal gage-based sea level data) is troublesome to work with, which is one of the reasons I rarely look at sea level data. So I selected the stations with the longest data, 70 years or more, along the Atlantic coast that were included in the “hot-spot” group of Sallenger et al (2012),

    Tisdale’s accusation of cherry-picking (quoted by Tamino above) is more than laughable. It is a strengthened version of “Are the volcanic eruptions of El Chichon in 1982 and of Mount Pinatubo in 1991 the reasons for the low, actually negative, decadal sea level trends around the same time?” in the first post and in his second post “But (Sallenger et al 2012) never attempted to account for them, in the paper, to see how much the catastrophic eruption of Mount Pinatubo in 1991 could have impacted their claim of accelerating sea levels.
    Strangely Tisdale “…never attempted to account for them.” If he had, he would have noted that none of his graphs of various running averages show any important features that could be attributable to events that began in March 1982 or June 1991. Indeed, if anything El Chichon & Pinatubo are associated with increases in sea level not decreases.

  3. I find his lack of errorbars disturbing.

  4. “Tisdale really doesn’t get it. Sallenger et al. didn’t cherry-pick 1990, in fact they didn’t “pick” it at all — they determined it. They did so by trying windows starting with all possible start years from 1894 through 1970, ending with 2009, computing SLRD based on 1st-half vs 2nd-half trends. By noting the window which gave the strongest SLRD, they estimated that the rate shift had occurred around 1990.”

    While it’s pretty rich for anyone at WUWT that to complain about cherry picking without mentioning the skeptic community’s obsession with 1998, it seems like there’s some merit to criticizing this methodology. If there is a time window in which a sea-level rate shift results from some historical anomaly, this sort of thing will pick it out. The data is obviously very noisy so the calculated SLRD will be higher or lower than the true value for any time window because of measurement error or fluctuations unrelated to medium-term trends. At a minimum, in a world with a true SLRD that’s uniformly zero plus some random noise, this method would be guaranteed to find a window with positive SLRD. I think so, anyway — please correct me if I’m reading this wrong.

    However, they aren’t measuring the average SLRD but are instead interested in geographic variation (presumably because sea-level rise and acceleration isn’t really under debate). Your figure shows that the hotspot finding is pretty robust against the start date, although it would be nice to have more data points outside of the hotspot. I wonder whether this might have an anthropomorphic cause in addition to GHG release, since tidal phase and magnitude are sensitive to coastline shape/height and the hot spot corresponds to an area of dense coastal developments.

    [Response: Of course any statistic -- including SLRD -- will fluctuate given random fluctuations in the data. That's why Sallenger et al. applied statistical significance testing to their results.]

    • Yeah looked through the paper and it’s fine in that regard; Fig. 3 shows SLRD for this station subset for all possible time windows so it’s not like they’re cherry picking that time window. They may be reading too much into the few 1990 data points, though. If it’s a result of a few above-trend years for whatever reason, it’ll show the same behavior the peak in Fig. 3 at 1950 (which results from the temporary drop below trend in 1980), but the peak will be higher and shorter in duration because of the shorter time window. Peaks in Fig. 3 (near 1918, 1942, 1950, 1970) correspond to drops below trend at the respective half-years (1964, 1976, 1980, 1990). Note that I’m basing this on Tisdale’s Fig. 2 to the extent that this is accurate.

      This is all sort of beside the point, though. There’s convincing evidence that sea level rise is accelerating and that it’s disproportionately affecting this region so folks living there ought to be concerned about sea level rise. I just wouldn’t read too much into a rapid rise in this plot. If you draw out a quadratic curve with some sort of temporary dip and return to trend, use points before, in the middle of, and after the dip as half-years and do this analysis, you’ll see that it has a similar peak even though the SLRD should actually decrease with shorter time windows for a quadratic curve.

  5. NotRogerPielkeSr

    Bob Tisdale is a world leading blog scientist and I would stake my reputation on his important research findings.

    • Blog Scientism, which he did at Blog-University and Blog-Labs ;? Virtual Reality.

    • Horatio Algeranon

      “The Blognitive Scientists”
      – by Horatio Algeranon

      The Blognitive scientists
      Do blognitive studies –
      Lame brain thinking
      By blogging buddies.

  6. NotRogerPielkeSr: The question is should you be more worried about staking your reputation on Tisdale’s findings, or should Tisdale be worried about the effect on his reputation of being linked with you? After all, there was that little incident with simple anomaly math… (


  7. Tisdale as world leading blog scientist? No! Dr Inferno tops in that department.

  8. The eastern coast of the US is basically a series of low sand spits and beaches. Now these beaches have been washing into the sea since the year dot, and folks with dirt movers have been pouring sand onto the beaches to keep them full of folk with dollars while other folks are busy pumping out water from underneath the beaches for ag and industry. And, of course, Canute like the sea also rises.

    What chance does a poor tidal gauge have against any of that

  9. Question:

    So is the rate of GW increasing significantly now?
    Must be as sea level rise is accelerating by quite a bit, but there must be a lage.

    So what warming years are responsible for the increase in slr rate happening now?
    Is it a 2 5 or 10yr lag?

  10. Speaking of North Carolina, the new (extremely conservative) LNP state government of Queesland is voting to play copycat: