New Paper: Sedimentation rates across space and time in Eastern North America

We’ve got a new paper out in the August issue of Quaternary Science Reviews looking at sedimentation rates across the northeastern United States. The paper is co-authored with Jack Williams, Jessica Blois, Stephen Jackson, Jennifer Marlon, Chris Paciorek, Bob Booth, Maarten Blaauw and Andrés Christen (incidentally, a really great group of co-authors, you should all write papers with them).  We were primarily interested in looking at sedimentation rates since Bayesian age-depth models such as Bacon require well informed priors to produce accurate estimates of deposition times.  I talked about this paper earlier on this blog in the context of providing open data along with attachments.  Having said that, the publication itself is not open.

Our three main findings were that:

1. Sedimentation rates differ significantly between lacustrine and palustrine environments
2. Sedimentation rates show no spatial structure across the region
3. Sedimentation rates show variability through time, most strongly in the last 500 years, but also appear to show some periodicity throughout the Holocene.

Figure 1. The first axis of a functional Principal Components Analysis (fPCA) ordination of US Historical Climate Network Precipitation data plotted, showing a clear partition of historic precipitation across the Appalachians.

I’m happy to note that Maarten Blaauw has already let me know that, as a result of the results of the paper, he’s updated the priors for accumulation rates in Bacon, so that’s a fairly immediate success.

There were some interesting patterns in the data.  We knew that local basin effects on sedimentation (available sediment, number of inputs, basin depth) would likely play a large role in determining overall deposition rates, but to some degree I had expected that there may have been some spatial effect, particularly since the US Historical Climatology Network seems to show that there is a very real partitioning of climate, or at least precipitation, across the region, centered on the western edge of the Appalachians (Figure 1).  This partitioning comes from using functional data analysis to generate functional principal components ordinations of the annual precipitation data across the USHCN network. The code to do this also uses some cool steps to generate the first derivative of the fPCA layers so we can see that the boundary is really very sharp between the eastern and western systems.  I’ll try to fix it up a bit and then post something about this soon.

Given that precipitation seems to be partitioned along a north/south boundary (and that the boundary is so sharp!), it was somewhat surprising that spatial models fit to the sedimentation data did no better than the random effects models (results are in the paper).  This would imply to me that the basin level effects on sedimentation are simply so big that any sort of regional pattern is unidentifiable.  One possible method to test this more explicitly would be to either (1) build a better dataset that incorporates a large number of morphometric characteristics for the basins, or to (2) subdivide the dataset into basins with very similar characteristics.  Thinking about it a bit further, it may also be useful to use a categorical variable (east/west) to stand in for the spatial component.  Is it okay to say that on reflection there are some things I could have done better?

Figure 2. Changes in deposition times over time from the eastern North American dataset, with confidence intervals based on a moving average with a 500 year window.

Beyond this there are patterns of deposition through time that appear to be very interesting.  Lacustrine basins seem to show little change over time, with the exception of rapidly declining deposition times (deposition times are measured in years / cm of sediment, so declining rates mean faster accumulations of sediment depths) over the last 500 years or so.  This is a well known phenomenon, and we make the argument that this is largely a function of sediment compression, although people differ in this opinion.  The palustrine dataset shows much more variability through time, but it is hard to say whether this is a function of data size or of greater sensitivity to climatic variability through time.

I think that our study really does two things, one it will help improve researcher skill in defining age-depth models for new (and redefining age-depth models for older) cores using software such as Bacon.  The second important function of the paper is looking at sedimentation rates in a regional context using updated age-depth models (based on Blois et al., 2011).  This sets out a number of questions that I’ve partly identified here, including:  Why is there no (apparent) regional control on sedimentation in lakes?  What causes the regional signal in palustrine sediments, is it autogenic control on basin productivity?  One outstanding question with regards to priors for Bacon is, how do we independently estimate the autocorrelation of deposition times if all of our age-depth models inherently impose some form of autocorrelation based on the technique used.  Gah.

Almost forgot to add a track for this post:

Marvin Gaye:  Trouble Man (live), I’m in a Bayesian workshop right now and I’ve had one of the instrumental tracks from the Trouble Man soundtrack stuck in my head all day.

Advertisement