Ground Motion Model Selection¶

GMM Selection Overview¶

Tectonic classification of earthquakes is a key component of the USGS Global ShakeMap system because it serves as the basis for the selection of ground motion prediction equations (GMPEs). GMPE selection has a significant impact on the estimated ground motion intensities. The revised ground motion selection algorithm provides a probability that the earthquake occurred in all possible seismotectonic classifications. These classifications are configured to be associated with an appropriate GMPE (or weighted set of GMPEs, see Ground-Motion Prediction) and the GMPEs associated with each classification are then combined based on the probability associated with each classification.

The select module reads an event’s event.xml file for origin information and then constructs a GMPE set for the event based on the event’s residence within, and proximity to, a set of predefined tectonic regions and user-defined geographic areas. The GMPE set, and the selection of that GMPE set for use in processing, are written to model_select.conf in the event’s current directory. Similarly, the configuration may select an intensity prediction equation module (IPE), a ground-motion to intensity conversion equation module (GMICE), and a cross-correlation function (CCF) for the region, and these, too, will be written to model_select.conf. Any modules that are not specified for a particular region will default back to the modules set in the global model.conf.

The behavior of select is controlled by the select.conf configuration file. See the documentation in select.conf for more on customizing select.

The tectonic regions, and additional geographic layers, that the event may fall within are defined by the STREC configuration. See the STREC documentation for information on adding additional layers, then use select.conf to customize the ground motion modules that the new layers will use.

Tectonic Regions¶

Fig. 1.13 shows a global map of the first layer of information, which is a set of four mutually exclusive and globally exhaustive classes, which we refer to as tectonic regions: active crustal region (ACR), stable continental region (SCR), subduction (SUB), and volcanic (VOL).

../_images/select_regions.png — Fig. 1.13 Map of main tectonic regions.¶

These regions are a simplification of the seismotectonic domains developed by Garcia and others (2012). The approximate mapping between the new tectonic regions and the seismotectonic domains is:

Tectonic Region	Subtypes	Seismotectonic domains
Active crustal region (ACR)	Extensional, Compressional	ACR (deep)
		ACR (shallow)
		ACR (oceanic boundary)
Stable continental region (SCR)	Marginal, Not marginal	ACR (oceanic boundary)
Stable continental region (SCR)	Marginal, Not marginal	SOR (generic)
Subduction (SUB)	Crustal, Interface, Intraslab	ACR deep (above slab)
		ACR oceanic boundary (above slab)
		ACR shallow (above slab)
		SCR (above slab)
		SOR (above slab)
		SZ (inland/back-arc)
		SZ (on-shore)
		SZ (outer-trench)
		SZ (generic)
Volcanic (VOL)	–	ACR (hot spot)
Volcanic (VOL)	–	ACR (hot spot)

Note that currently the only subtypes that are made use of are the subduction zone subtypes.

There are a number of configuration options for how the tectonic regions are modeled in select.conf. Here is an example of for ACR:

[tectonic_regions]
    [[acr]]
        horizontal_buffer = 100
        vertical_buffer = 5
        gmpe = active_crustal_nshmp2014, active_crustal_deep
        min_depth = -Inf, 30
        max_depth = 30, Inf
        ipe = VirtualIPE
        gmice = WGRW12
        ccf = LB13

where:

horizontal_buffer - The buffer distance (km) that extends into neighboring regions across which the GMPEs are blended.
vertical_buffer - The buffer distance (km) that blends the depth dependence of the GMPEs within this tectonic region.
gmpe - A list of one or more GMPE sets found in gmpe_sets.conf.
min_depth - A list of one or more minimum depths (km) corresponding to the GMPEs listed under gmpe.
max_depth - A list of one or more maximum depths (km) corresponding to the GMPEs listed under gmpe.
ipe - An intensity prediction module; must be found in the collection of ipe_modules in modules.conf.
gmice - A ground motion to intensity module; must be found in the collection of gmice_modules in modules.conf.
ccf - A cross-correlation module; must be found in the collection of ccf_modules in modules.conf.

The process by which select builds a GMPE set is somewhat complicated. STREC reports the tectonic region the earthquake lies within, as well as the distance to the closest polygon of the other tectonic region types. For example, for an earthquake in California STREC would report that the event was zero distance from region ‘acr’ (which is to say that it lies within the active crustal region), but STREC would also report distances to regions ‘scr’ (stable continental), ‘volcanic’, and ‘subduction’. Each non-subduction region is also configured with a “horizontal buffer.” The buffer determines how far the region extends into neighboring regions. The buffer for subduction regions is always zero. If the event happens within the buffer of a neighboring region, the distance and buffer are used to build a weighted combination of the GMPE sets representing the regions in question.

For example, if an earthquake occurred within the ‘scr’ region, but was 40 km from the “acr” region, and the ‘acr’ region’s horizontal buffer was 100 km, then the ‘scr’ region would be given a weight of 1.0, and the ‘acr’ region would be given (100 - 40) / 100 = 0.6. Normalizing by the total, the final weights would be 0.625 ‘scr’ and 0.375 ‘acr’.

Fig. 1.14 maps the probabilities for the main tectonic regions in the northwest US. This illustrates how the horizontal buffer smoothly transitions between the regions.

../_images/select_example_tect_regions.png — Fig. 1.14 Maps of the probability of the four main tectonic regions in the northwest US. Top left: active crustal region (ACR); Top right: stable continental region(SCR); Bottom left: subduction (SUB); Bottom right: volcanic (VOL).¶

Each region’s GMPE set is in turn comprised of a weighted set of other GMPE sets, based on the earthquake’s depth. For each of the non-subduction regions, select builds a weighted combination of the configured GMPE sets based on the event’s depth. If the earthquake falls within a subduction region, STREC reports the probabilities that the earthquake is crustal, on the subduction interface, or within the subducting slab. select combines the GMPE sets for each of these regimes, weighted by their probabilities, into a subduction GMPE set that is specific to the earthquake’s location.

Subduction Subtypes¶

Within subduction zones, we distribute the probability given to the subduction zone tectonic region between its three subtypes. By default, this primarily relies on the Hayes (2018) Slab2 model.

Although this rarely occurs, events that are located in a subduction zone but the slab model is not defined, we compute the probability of the interface subtype as a function of depth and magnitude. The relevant section of select.conf is:

[subduction]
    default_slab_depth = 36.0
    [[p_int_mag]]
        x1 = 7.0
        p1 = 0.0
        x2 = 8.5
        p2 = 1.0
    [[p_int_dep_no_slab_upper]]
        x1 = 17.0
        p1 = 0.0
        x2 = 27.0
        p2 = 1.0
    [[p_int_dep_no_slab_lower]]
        x1 = 45.0
        p1 = 0.0
        x2 = 55.0
        p2 = -1.0

These parameters define taper functions that give more probability to interface for larger magnitudes, crustal for shallow events, interface for intermediate depth events, and slab for deeper events, as illustrated in Fig. 1.15.

../_images/sub_no_slab.png — Fig. 1.15 Profiles showing the probability of crustal, interface, and interslab subuction subtypes (assuming the probability of subduction is 1.0) with depth for a magnitude of 7 (left), 8 (center), and 9 (right).¶

The slab model is defined for most locations in subduction zones. When available, we distribute the probability the subtypes with a series of heuristic steps using the following parameters:

The distance between the interface in the slab model and the hypocentral depth; see the p_int_hypo section of select.conf.
The angle of rotation between the plane tangent to the slab at the location of the earthquake and the focal mechanism; see the p_int_kagan section of select.conf (the angle is sometimes called the “Kagan angle”).
The position of the hypocenter relative to the maximum depth of the seismogenic zone, as given by the slab mode; see the p_int_sz section of select.conf.
The position of the hypocenter relative to interface in the slab model; see the p_crust_slab section of select.conf.
The absolute depth of the hypocenter; see the p_crust_hypo section of select.conf.

Because of the unique treatment of the tectonic subtypes for subduction zones, its section in select.conf includes some additional settings:

[[subduction]]
    horizontal_buffer = 100
    vertical_buffer = 5
    gmpe = subduction_crustal, subduction_interface_nshmp2014, subduction_slab_nshmp2014
    min_depth = -Inf, 15, 70
    max_depth = 15, 70, Inf
    ipe = VirtualIPE
    gmice = WGRW12
    ccf = LB13
    use_slab = True
    [[[crustal]]]
        gmpe = subduction_crustal
    [[[interface]]]
        gmpe = subduction_interface_nshmp2014
    [[[intraslab]]]
        gmpe = subduction_slab_nshmp2014

Note that the additionl subsections (e.g., [[[crustal]]]) and their associated GMPEs after the use_slab key are used when the slab model is being used to distribute the subduction probabilities. In this case, the results based on the previously listed gmpe, min_depth and max_depth are overwritten. If use_slab is set to False, then subtype subsections are ignored and the subduction zone is treated like all the other tectonic regions (e.g., configurable with the gmpe, min_depth, and max_depth keys).

Geographic Regionalization¶

The select module also considers the earthquake’s presence within, or distance from, any number of user-defined geographic layers. If the earthquake is within a layer, that layer’s parameters (as configured in select.conf) replace any or all of the parameters of the corresponding tectonic regions, and the calculation of a weighted GMPE set proceeds as before. For example, the layer section of select.conf might contain:

[layers]
    [[california]]
        horizontal_buffer = 50
        [[[scr]]
            horizontal_buffer = 25
        [[[acr]]]
            horizontal_buffer = 25
            gmpe = Special_California_GMPE
            min_depth = -Inf
            max_depth = Inf
            ipe = Allen2012
            gmice = WRGW12

If an earthquake falls within the ‘california’ layer, the tectonc regions ‘scr’ and ‘acr’ would have their horizontal buffers reset to 25 km and, in addition, the ‘acr’ region would have its GMPE selection reset to the GMPE set ‘Special_California_GMPE’ for earthquakes of all depths. Similarly, the IPE would be set to “Allen2012” and the GMICE to “WGRW12”.

If the earthquake is not inside a custom geographic layer, but within the horizontal buffer distance of one, the GMPE sets for the modified and unmodified tectonic regions are each determined separately and a weighted combination of the two is computed (where the weights are based on the distance and the horizontal buffer, as described above).

Unlike the tectonic regions, the geographic layers consider only the nearest layer. If an earthquake falls within more than one layer (possible if layers are nested), the first one encountered in select.conf is used and any other(s) will be ignored.

The polygons for geograhpic regionalization are located in the data/layers subdirectory of the current shakemap profile install directory. The files are associated to the key name within the [layers] section of select.conf. Using the example above, when select sees the [[california]] key name, it will look for a file called california.wkt in <install_dir>/data/layers.