GTL
Geotechnical Layer (GTL) refers to material properties from surface to about 350m depth. GTL is known to have significant impact on local strong ground motions:
Contents
Definition of Terms
- Base Model - CVM-S4 material properties as delivered by CVM-S4
- Starting Model for Inversion - CVM-S4 material properties with min Vs=1000m/s
- Final Model with Inversion - CVM-S4 with Perturbations
Integrating CVM-SI4.26 perturbations into Final Model
Description of algorithm:
The TJ model was first described by Tom Jordan. In this model, we set the CVM-S4 value as the floor for the perturbations in areas where Vs <1000m/s.
So if we have CVM-S4 Vs = 500m/s and perturbations of -100m/s, then we'd use CVM-S4.26 Vs = 500m/s. On the other hand, if the perturbations were +100m/s, we'd use CVM-S4.26 Vs = 600m/s.
The net effect is that we recover the CVM-S GTL but more smoothly add in positive perturbations. This is in contrast to the 2.2.1 idea in which we don't use base Vs + positive perturbation but rather starting Vs + positive perturbation.
Difference plots are: tj_base_[depth]_0m_cs.png (TJ model - base model) and also horizontal slices are just tj_[depth]_cs.png.
Powerpoint and online plots are located at:
Proposed Work Order
(Please note, the numbers marked in bold are the e-trees that will be delivered first.)
A simplified workflow would go like this:
Evaluate base (reference) model:
1. Run CH with base CVM-S. This is done already, I have those results from my paper with Jacobo.
Intermediate reference model (with inversion but without Vs30-based GTL):
2. Run CH with perturbations, that is: CVM-SI.23, but without GTL. This, ideally, should be done for two basic cases:
2.0 A model that is Po's inversion starting model (CVM-S4 with min Vs = 1000m/s).
2.1. A model that does not reverses from the starting model to the original model, but simply applies the perturbations and interpolates for arbitrary resolution. (This model, as it does not have any information about soft deposits, will only be used as reference for later runs.) This is the version bundled with UCVM 13.9.0.
2.2. A model that reverses the staring model to the original CVM-S and while applying the perturbations from the inversion. Applying the perturbations while reversing implies applying perturbations directly wherever the starting model is the same as the original CVM-S but using a particular rule in those places where the starting model is stiffer than the original CVM-S. Cush a rule can be:
2.2.1. Applying negative perturbations only if outside the basin and always applying positive perturbations even if inside the basin. (This is the so-called RE idea.)
2.2.2. Same as 2.2.1. but disregard positive perturbations inside the basin. This occurs to me now as I think again about Tom's comments during the telecon.
2.2.3. Same as the TJ model described above.
Out of the box, both 2.2. models should be better than 2.1. That is because, as I said before, 2.1. here is built only as a base model for the ones that will have the Vs30-based GTL next. On the other hand, the two 2.2. models do have some kind of GTL information as both recover that information from the original CVM-S when reversing the staring model Between 2.2.1. and 2.2.2., I have a sense that 2.2.1 will be better, but that is to be seen. Doing a comparison between 2.2.1 and 2.2.2 at this point is totally fair game.
Geotechnical models:
Now that we have the full framework we can work on adding the Vs30-based GTL.
3.1. A model that is based on 2.1. and incorporates the Vs30-based GTL. This, ideally, in two possible ways.
3.1.1. Applying the Vs30-based GTL as is available today, that is, using a fixed depth as reference.
3.1.2. Applying the Vs30-based GTL with a new implementation that uses a fixed Vs value at bedrock as reference (an idea suggested before by both Harold Magistrale and myself.)
These two models simply disregard the need for recuperating the basin structure out of the original CVM-S and assume that applying the Vs30-based GTL is sufficient to represent soft-soil deposits. Then...
3.2. A model that is based on 2.2., that is, one that reverses the staring model to the original model while applying the perturbations from the inversion and "adding" information from the Vs30-based GTL as well, following two options:
3.2.1. Applying the Vs30-based GTL using a fixed depth as reference.
3.2.2. Applying the Vs30-based GTL using a fixed Vs value at bedrock as reference.
Here we would be doing direct comparisons between 3.1 and 3.2 for each corresponding pair. I believe that one of the two 3.2. options will prevail as the better one given what we have with us today, but this is purely intuitive and speculative, so the runs would have to be done in order to see.
Previous GTL approaches
For reference, results from earlier approaches are posted here:
- Profiles - CVM-S4.23 + Vs30 GTL
- Profiles - Ricardo, En-Jui idea
We note that the Salton Sea area looks weird in the RE GTL then Vs30 GTL variation. The reason for this is that the starting model was harder than the original CVM-S4 material properties, as such we use the original CVM-S4 Vs, Vp which are a lot firmer than in the Vs30 data (we could correct this by making an exception to the rule for the Salton Sea area).
We note how in the "RE then Vs30 GTL" idea, the main CVM-S GTL area in the LA basin is nicely preserved, but interestingly when you combine the RE + Vs30 GTL ideas some elements of the CVM-S GTL look like they have been almost recovered.
GTL for CVM-S4 with Perturbations
We plot standard maps and profiles for each CVM implementation. Two new versions are as follows:
- Profiles - This is the version that adds the Vs30 GTL on top of the Ricardo, En-Jui GTL idea.
- Profiles - This is the version that does the Ricardo, En-Jui GTL, but if that isn't applicable, it does the Vs30 GTL.
Proposed Method for selecting GTL
First, we need to have the complete reference framework before applying the GTL. We already have runs for CVM-S in its original form. And your email suggests to do runs for one or the two models David worked on. We need, however, the middle step run(s). That is, before trying runs with models that incorporate GTL, we need to do runs with just incorporating the inversion results. That will provide the actual reference for the GTL. Otherwise we won't be able to know if the improvement was due to applying the GTL or applying the perturbations from the inversion.
Second, we need to have two ways of applying the GTL. One that uses a fixed depth as reference, and if possible and easy to implement, a second one that applies the GTL using a fixed Vs value as reference. (David, I don't know how easy could this be to implement in UCVM.) We may skip this alternative implementation of the GTL if time is of the essence, but I do think---based on some figures for vertical cuts I've made---that using the GTL as currently available in UCVM carries the potential of altering the basins' shapes, and then, once that happens, we are no longer comparing apples with apples.
The last step is then doing the simulations with the GTL. Depending on whether developing the fixed-Vs reference GTL is easy or not, I would have preferences on the two current models. If we are limited to use the current implementation for "adding" GTL, then I prefer the one that only applies GTL outside the basins. That is, the if-not-then second option in your email. And, if implementing the fixed-Vs reference implementation for "adding" the GTL is easy and ready soon, then I will try both, and not just one of the two approaches prepared by David.
