EarthScope Forums

Dear Colleagues,

The results of the strain rate comparisons can be found at the following location along with everyone’s data.  I’ll also send a large pdf-file in the next e-mail.

    ftp://topex.ucsd.edu/pub/sandwell/strain

We would like to rotate these images and captions on the EarthScope site so please send corrections.  It is very possible I have made a mistake reprocessing your data so please check the GMT scripts.  This should be a “living” comparison so updates and additions are welcome.  I don’t have a lot of time to spend on this so feel free to make changes yourself.  Currently the models disagree by a factor of 5-8.  I hope that in 5 years we will have strain rate maps that agree to within say 50%.

Dave,

Great, the ppt looks very interesting.

Two, obvious additional tests to disentangle method vs. data effect come to mind: Have everybody use the same dataset (to understand systematic, method-based “uncertainty”), and have a few methods compute strain-rates for different (density, geodetic type, individual errors, etc.) datasets (to understand resolution and regular uncertainty). The feasibility of each test remains to be seen; not all methods will be directly comparable, for example.

It would also be interesting to have something like log(model_1_eps/model_2_eps) amplitude maps, and average ratios of difference (similar to the upper part of your correlation triangle? Not sure what was plotted there).

Perhaps specifics could be discussed at the SCEC meeting?

Cheers

Thorsten

Dave, Thorsten-

  I think chatting at SCEC would be great.  There’s a lot of great diversity (and clustered similarities!) among these approaches.

  In addition to the tests Thorsten suggested I think it would be worthwhile to chat about how to interpret these maps in the context of some end member scenarios such as a creeping fault segment (e.g., San Andreas north of Parkfield).  In this case there would be no interseismic strain accumulation/earthquake cycle effects, yet there would certainly be a finite strain calculated on the basis differential GPS velocities at stations spanning the creeping fault. In this case the mapping between surface strain rate and any sort of mechanical stressing rate would need to be considered differently.  This sort of thing is pretty straightforward but worth chatting about so that we don’t map high calculated strain rates to high mechanical stressing rates in regions where there might be negligible physical strain accumulation.

Brendan

David,

Thank you for putting this together.  This is very informative.

The root of the problem, as we have discussed at the EarthScope meeting, is the correct estimation of the derivatives of a highly non-liner, sampled vector field.  This is compounded by the fact the sampling is uneven.  Our group has been working on estimating the strain field using a simple linear approach directly from the data, and it has allowed us to explore the error in the strain estimates and its dependence on sampling and network orientation relative to the vector field.  We have found that even when examining the simple problem of an infinite, buried, pure strike-slip fault, with a very dense, evenly sampled network of stations, the errors can vary by an order of magnitude using the SAME algorithm.  We will be presenting our results at SCEC.

It may be helpful to do a comparison of the different techniques with the most simple problem (such as the buried infinite strike slip), with a regular grid of stations first, and then with an uneven sample of the field that mimics PBO’s.
Incidentally, other fields, such as atmospheric sciences, deal with similar issues, when estimating the divergence and vorticity of the wind field, for example.  It may be useful to approach colleagues in this area to see how they are treating the problem.

Sharon

We have a powerpoint of the results at the following location.

This will change as models are added and changed.

  ftp://topex.ucsd.edu/pub/sandwell/strain/strain_rate_all.ppt

David

Dear All,

I just want to point out that these strain rate data are available to everyone and we should all explore ideas proposed by Thorsten, Brendan, and Sharon.  The main point is that there is broad disagreement which is due to an inadequate data distribution.  We are working on an InSAR solution to fill in the gaps but this is difficult because of decorrelation.

Brendan’s point about whether the strain in the creeping section north of Parkfield is either zero or very high is critical to understanding the earthquake cycle.  I would not take the correlation matrix among the models (see powerpoint) too seriously yet.  For example the harvard model is not well correlated with any of the other models.  This is probably due to the assumption that the creeping section has zero strain rate and all the other models have high strain rate.

I may not make it to SCEC but we should definitely continue the discussions there.

Again all the data are at the following ftp site.  Bridget Smith and I are building a new model model now using a hybrid approach = dislocation_model + smooth interpolation of residuals.  All updates and model replacements are welcome.

  ftp://topex.ucsd.edu/pub/sandwell/strain

David