Comment on the subsidence adjustment applied to the Kemp et al. proxy of North Carolina relative sea level

posted Oct 10, 2011, 7:34 AM by Aslak Grinsted   [ updated Dec 17, 2013, 2:32 AM ]
Kemp et al. obtained a really nice proxy of relative sea level from North Carolina. North Carolina is subsiding and they therefore apply a 'GIA' correction to obtain a sea level record free of this particular local effect. In our comment we argue that the uncertainty they use in the subsidence correction is too small. The important point is not that we have a better estimate than theirs, but that there is a range of different estimates, and that their uncertainty does not bracket them all. Further the particular estimate they have chosen is an outlier, and cannot explain tide gauge observations (see right column). The effect of a greater subsidence correction and greater uncertainty is shown in this figure (click to zoom):

In their paper they then proceed to calibrate a semi-empirical model of global sea level, to this local sea level record. I believe this is not a suitable calibration target precisely because it is local. Further, our key point is that the uncertainties are too low, and that the proxy record as a consequence should not provide as strong constraints to the fitting procedure (therefore the model confidence intervals are artificially too tight). In my personal opinion the NCRSL record is much better suited as a validation for sea level models (in contrast to a calibration target). 

Links to paper, comment and reply:
  • Kemp et al's PNAS paper: PNAS
  • Our technical commentPNAS comment [PDF]
  • Their response: PNAS reply 
    • In my opinion their response does not address the key issue of uncertainty.

My reaction to their response:
Their response deals almost exclusively with what is the best value to choose as the GIA. This is exemplified in their final conclusive paragraph: "Regional geological data remain the most robust means to estimate late Holocene GIA ...". However, this response does not at all address the key problem we are highlighting: They have an overly optimistic uncertainty interval. We believe that the uncertainty interval should be several times greater than their stated value, and we base this on the work of Engelhart et al. [Disclaimer: Kemp et al. use the Engelhart paper to support the 'optimistic' uncertainty interval. The authors therefore feel that we misrepresent Engelhart. I would argue that we represent Engelhart's data exactly as it is and make our own conclusions. Can that really be called a misrepresentation?]

The figure and text in the column on the right shows that a 1mm/yr subsidence rate cannot explain the regional tide gauge trends. 

Detailed response to their 4 points: 
  1. [ICE5g is a poor choice]: Our comment is regarding the uncertainty interval, not what is the best guess value. It is clear from our text that the reason we chose ICE5G is because it is closer to a mid-range value when considering the various estimates (including their geologic estimate). So, this point is moot. 
  2. [GPS rates are too uncertain to determine regional trends]. This is the relevant quote from Engelhart et al. showing just how different geologic and gps rates are: 
    • "The geological rates of subsidence decline rapidly with distance from Hudson Bay along the U.S. Atlantic coast compared to the GPS observations. The GPS observations suggest that high rates of subsidence from the collapse of the forebulge extend into Virginia and the Carolinas (Sella et al., 2007; Snay et al., 2007). For example, the geological data within Chesapeake Bay, Virginia, estimate subsidence of 0.9 ± 0.3 mm a–1 compared to nearby GPS observations of 3.5 ± 1.6 mm a–1 (Sella et al., 2007) and 2.6 ± 1.2 mm a–1 (Snay et al., 2007)"
    • This quote shows that GPS inferred subsidence is significantly greater, than the geologic estimate. They may be right that it is hard to extrapolate to what it exactly should be in NC, but it is clear that GPS indicates that the forebulge is collapsing significantly faster than geologic estimates allow. 
  3. [Grinsted et al. 'wrongly claim' a too large local tide gauge rate]: Our claim is based on the supplementary material to Engelhart et al. which can be downloaded here. The figure in question is reproduced in the column on the right. look for X=2950 km. If anything our 4mm/yr claim seems too low!
  4. [The larger rate is inconsistent with Mediterranean proxy records]: First, we do not state we have the best estimate, only that the claimed uncertainty is too low. So, again the point is moot. Secondly North Carolina Relative Sea Level is not Global sea level, so it is allowed to not match with other regions. (And perhaps there are other local issues with these records).

Tide gauges show that geologic GIA estimate is too low

The figure below is from the supplementary info to Engelhart et al. The important thing here is that each tide gauge (red dot) has been chosen, so that it should be unaffected by confounding factors like ground water pumping. The distance between Tump Point, NC and Churchill, Canada is ~2950 km. For comparison the Global average rate of rise over the 20th century is 2mm/yr. We can subtract the global rate from the local relative rate of sea level rise to estimate the subsidence. That gives a subsidence of atleast 2 mm/yr.   

It seems clear that before this issue is resolved, it is dangerous to assume global=local and that it therefore is premature to use the record as a calibration target. 

For comparison 
Santamaría-Gómez et al. estimates VLM rates in a novel way. Long-term vertical land motion from double-differenced tide gauge and satellite altimetry data. 

In Chesapeake Bay they have subsidence estimates at three locations. Their estimates are from 3.0, 2.5, and 1.4 mm/yr (for stations 299, 1635, 1636). This is also greater than the geologic subsidence rate of 0.9 mm/yr.