Liquefaction and Landslide ========================== Landslides and liquefaction are well-known perils that accompany earthquakes. Basic models to describe their occurrence have been around for decades and are constantly improving. However, these models have rarely been incorporated into PSHA. The tools presented here are implementations of some of the more common and appropriate secondary perils models. The intention is seamless incorporation of these models into PSH(R)A calculations done through the OpenQuake Engine, though the incorporation is a work in progress. In what follows, we provide a brief overview of the implemented models, preceded by general considerations on the spatial resolution at which these analyses are typically conducted. For more in-depth information on the geospatial models, we recommend referring to the original studies. Additionally, we offer corresponding demonstration analyses, which can be found in the `demos section `_) of our GitHub repository. We encourage users to check them out and and familiarize themselves with the required inputs for performing liquefaction or landslide assessment. We also provide tools to extract relevant information from digital elevation data and its derivatives, which are often given as rasters. General considerations ---------------------- ***************************************************************** Spatial resolution and accuracy of data and site characterization ***************************************************************** Much like traditional seismic hazard analysis, liquefaction analysis may range from low-resolution analysis over broad regions to very high resolution analysis of smaller areas. With advances in computing power, it is possible to run calculations for tens or hundreds of thousands of earthquakes at tens or hundreds of thousands of sites in a short amount of time on a personal computer, giving us the ability to work at a high resolution over a broad area, and considering a very comprehensive suite of earthquake sources. In principle, the methods should be reasonably scale-independent but in practice this isn't always the case. Two of the major issues that can arise are the limited spatial resolutions of key datasets and the spatial misalignments of different datasets. Some datasets, particularly those derived from digital elevation models, must be of a specific resolution or source to be used accurately in these calculations. As we will see in the coming sections of this document, a common proxy to most of the geospatial models is shear wave velocity in the top :math:`30 \, \text{m}`. For example, if :math:`V_{s30}` is calculated from slope following methods developed by `Wald and Allen (2007) `_, the slope should be calculated from a DEM with a resolution of around 1 km. Higher resolution DEMs tend to have higher slopes at a given point because the slope is averaged over smaller areas. The mathematical correspondance between slope and :math:`V_{s30}` was developed for DEMs of about 1 km resolution, so if modern DEMs with resolutions of 90 m or less are used, the resulting :math:`V_{s30}` values will be too high. In and of itself, this is not necessarily a problem. The issues can arise when the average spacing of the sites is much lower than the resolution of the data, or the characteristics of the sites vary over spatial distances much less than the data, so that important variability between sites is lost. Liquefaction models ------------------- Several liquefaction models are implemented in the OpenQuake engine. One of them is the method developed for the HAZUS software by the US Federal Emergency Management Agency. This model involves categorization of sites into liquefaction susceptibility classes based on geotechnical characteristics, and a quanitative probability model for each susceptibility class. The remaining models are the academic geospatial models, i.e., statistical models that uses globally available input variables as first-order proxies to characterise saturation and density properties of the soil. The shaking component is expressed either in terms of Peak Ground Acceleration , :math:`PGA`, or Peak Ground Velocity , :math:`PGV`. These methods are simplified from older, more comprehensive liquefaction evaluations performed at a single site following in-depth geotechnical analysis. ***** HAZUS ***** The HAZUS model (see `HAZUS manual `_) classifies each site into a liquefaction susceptibility class, :math:`LSC`, based on the geologic and geotechnical characteristics of the site, such as the sedimentological type and the deposition age of the unit. Note that descriptions of the susceptibility classes may not align perfectly with the descriptions of the geologic units. In addition to the :math:`LSC` and the local ground acceleration at each site, the depth to groundwater at the site and the magnitude of the causative earthquake will affect the probability that a given site will experience liquefaction. The equation that describes this probability is: .. math:: P(L) = \frac{P(L | PGA=a) \cdot P_{ml}}{K_m K_w} \\ (1) :math:`P(L|PGA=a)` is the conditional probability that a site will fail based on the :math:`PGA` and the :math:`LSC`. :math:`P_{ml}` is the fraction of the total mapped area that will experience liquefaction if :math:`P(L|PGA=a)` reaches 1. These terms both have :math:`LSC`-specific coefficients; these are shown in Table 1. :math:`K_m` is a magnitude-correction factor that scales :math:`P(L)` for earthquake magnitudes other than :math:`M7.5` potentially to account for the duration of shaking (longer shaking increases liquefaction probability). :math:`K_w` is a groundwater depth correction factor (shallower groundwater increases liquefaction probability). +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | :math:`LSC` | :math:`PGA_{min}` | :math:`PGA_{slope}` | :math:`PGA_{int}` | :math:`P_{ml}` | +================+=======================+=========================+=======================+====================+ | very high | 0.09 | 9.09 | 0.82 | 0.25 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | high | 0.12 | 7.67 | 0.92 | 0.2 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | moderate | 0.15 | 6.67 | 1.0 | 0.1 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | low | 0.21 | 5.57 | 1.18 | 0.05 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | very low | 0.26 | 4.16 | 1.08 | 0.02 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ | none | :math:`\infty` | 0.0 | 0.0 | 0.0 | +----------------+-----------------------+-------------------------+-----------------------+--------------------+ Table 1: Liquefaction values for different liquefaction susceptibility categories, :math:`LSC`. :math:`PGA_{min}` is the minimum ground acceleration required to initiate liquefaction. :math:`PGA_{slope}` is the slope of the liquefaction probability curve as a function of :math:`PGA`, and :math:`PGA_{int}` is the y-intercept of that curve. :math:`P_{ml}` is the Map Area Proportion, which gives the area of liquefaction within each map unit conditional on liquefaction occurring in the map unit. ***************** Geospatial models ***************** ################# Zhu et al. (2015) ################# The model by `Zhu et al. (2015) `_, is a logistic regression model requiring specification of the :math:`V_{s30} \, [\text{m/s}]`, the Compound Topographic Index, :math:`CTI`, a proxy for soil wetness or groundwater depth, :math:`gwd \, [\text{m}]`, the :math:`PGA_{M,SM} \, [\text{g}]` experienced at a site, and the magnitude of the causative earthquake. The model is quite simple. An explanatory variable :math:`X` is calculated as: .. math:: X = 24.1 + 2.067\, ln\, PGA_{M,SM} + 0.355\,CTI - 4.784\, ln\, V_{s30} \\ (2) and the final probability is the logistic function: .. math:: P(L) = \frac{1}{1+e^X} \\ (3) The term :math:`PGA_{M,SM}` is the :math:`PGA` corrected by magnitude scaling factor, :math:`MSF`, that serves as proxy for earthquake duration. The :math:`MSF` is calculated as per `Youd et al. (2001) `_: .. math:: MSF = \frac{10^{2.24}}{M^{2.56}} \\ (4) Both the :math:`CTI` and the :math:`V_{s30}` may be derived from digital elevation data. The :math:`Vs30` may be estimated from the topographic slope through the equations of `Wald and Allen (2007) `_, which uses a very low resolution DEM compared to modern offerings. As topographic slope tends to increase with increased DEM resolution, the estimated :math:`Vs30` does too; therefore a low-resolution DEM (i.e., a 1 km resolution) must be used to calculate :math:`Vs30`, rather than the 30 m DEM that is the current standard. This results in a more accurate :math:`Vs30` for a given slope measurement, but it also means that in an urban setting, sub-km-scale variations in slope are not accounted for. The :math:`CTI` (`Moore et al., 1991 `_) is a proxy for soil wetness that relates the topographic slope of a point to the upstream drainage area of that point, through the relation: .. math:: CTI = \ln (d_a / \tan \delta) \\ (5) where :math:`d_{a}` is the upstream drainage area per unit width through the flow direction (i.e. relating to the DEM resolution). It ranges from :math:`0` to :math:`20`. It was developed for hillslopes, and is not meaningful in certain very flat areas such as the valley floors of major low-gradient rivers, where the upstream drainage areas are very large. Unfortunately, this is exactly where liquefaction is most expected away from coastal settings. Model's prediction can be transformed into binary class (liquefaction occurrence or nonoccurrence) via probability threshold value. The authors proposed a threshold of 0.2. ##################### Bozzoni et al. (2021) ##################### The parametric model developed by `Bozzoni et al. (2021) `_, keeps the same input variables (i.e., :math:`PGA_{M,SM}`, :math:`CTI`, :math:`V_{s30}`) and functional form as in `Zhu et al. (2015) `_. Regression parameters are calibrated based on the liquefaction case histories observed during seismic events in Europe. The implemented model is associated with the ADASYN sampling algorithm. The explanatory variable :math:`X` is computed as: .. math:: X = -11.489 + 3.864\, ln\, PGA_{M} + 2.328\,CTI - 0.091\, ln\, V_{s30} \\ (6) and the probability of liquefaction in calculated using equation (3). The adopted probability threshold of 0.57 converts the probability of liquefaction into binary outcome. ################# Zhu et al. (2017) ################# Two parametric models, a coastal model (Model 1), and a more general model (Model 2) are proposed by `Zhu et al. (2017) `_. A coastal event is defined as one where the liquefaction occurrences are, on average, within 20 km of the coast; or, for earthquakes with insignificant or no liquefaction, epicentral distances less than 50 km.The implemented geospatial models are for global use. An extended set of input parameters is used to describe soil properties (its density and wetness). The ground shaking is characterised by :math:`PGV \, [\text{cm/s}]`. Soil density is described by :math:`V_{s30} \, [\text{m/s}]`. Soil wetness in Model 1 is chatacterised by a set of features: mean annual precipitation, :math:`precip \, [\text{mm}]`, distance to the coast, :math:`d_{c} \, [\text{km}]`, and distance to the river, :math:`d_{r} \, [\text{km}]`. Distance to the coast also indicates the geologic age - younger deposits are found near the coast. Soil wetness in Model 2 is characterised by closest distance to the water body, :math:`d_{w} \, [\text{km}]`, which is determined as :math:`\min(d_{c}, d_{r})`, and the ground water table depth, :math:`gwd \, [\text{m}]`. Mean annual precipitation is from a global layer developed by `Hijmans et al. (2005) `_. Distance to the nearest river is calculated based on the HydroSHEDS database by `Lehner et al. 2008 `_. Water table depth is retreived from a global dataset by `Fan et al (2013) `_. Distance to the nearest coastline data was computed from `Ocean Color `_. The explanatory varibale :math:`X` is calculated as: Model 1: .. math:: X = 12.435 + 0.301 \ln(PGV) - 2.615 \ln(V_{s30}) + 0.0005556 precip \quad .. math:: -0.0287 \sqrt{d_{c}} + 0.0666 d_{r} - 0.0369 \sqrt{d_{c}} \cdot d_{r} \\ (7) Model 2: .. math:: X = 8.801 + 0.334 \ln(PGV) - 1.918 \ln(V_{s30}) + 0.0005408 precip \quad .. math:: -0.2054 d_{w} - 0.0333 wtd \\ (8) and the probability of liquefaction is calculated using equation (3). Zero probability is heuristically assigned if :math:`PGV < 3 \, \text{cm/s}` or :math:`V_{s30} > 620 \, \text{m/s}`. The proposed probability threshold to convert to class outcome is 0.4. Another model's outcome is liquefaction spatial extent, :math:`LSE`. After an earthquake LSE is the spatial area covered by surface manifestations of liquefaction reported as a percentage of liquefied material within that pixel. Logistic regression with the same form was fit for the two models, with only difference in squaring the denominator to improve the fit. The regression coefficients are given in Table 2.: .. math:: L(P) = \frac{a}{\left( 1 + b\,e^{-c\,P} \right)^2} \\ (9) .. raw:: latex \vspace{15pt} +----------------+-------------+-------------+ | **Parameters** | **Model 1** | **Model 2** | +================+=============+=============+ | a | 42.08 | 49.15 | +----------------+-------------+-------------+ | b | 62.59 | 42.40 | +----------------+-------------+-------------+ | c | 11.43 | 9.165 | +----------------+-------------+-------------+ Table 2: Parameters for relating probabilities to areal liquefaction percent. ########################## Rashidian and Baise (2020) ########################## The model proposed by `Rashidian and Baise (2020) `_ keeps the same functional form as the general model (Model 2) proposed by `Zhu et al. (2017) `_; however, introducing two constraints to address the overestimation of liquefaction extent. The mean annual precipitation has been capped to :math:`1700 \, \text{mm}`. No liquefaction is heuristically assign when :math:`PGA < 0.1 \, \text{g}` as an additional measure to decrease the overestimation of liquefaction. Additional novelty introduced in this model is the magnitude scaling factor, :math:`MSF`, to multiply the :math:`PGV` to mitigate the potential over-prediction in earthquake with low magnitude. The explanatory variable :math:`X` is evaluated using the equation (8) that corresponds to the general model of `Zhu et al. (2017) `_. The spatial extent is evaluated identically using the equation (9). The proposed probability threshold to convert to class outcome is 0.4. ##################### Akhlagi et al. (2021) ##################### Expanding the liquefaction inventory to include 51 earthquake, `Akhlagi et al. (2021) `_ proposed two candidate models to predict probability of liquefaction. Shaking is expressed in terms of :math:`PGV \, [\text{cm/s}]`. Soil saturation is characterised using the set of proxies: distance to the nearest coastline, :math:`d_{c} \, [\text{km}]`, distance to the closest river, :math:`d_{r} \, [\text{km}]`, elevation from the closest water body, :math:`Z_{wb} \, [\text{m}]`. Soil density is characterised either by :math:`V_{s30} \, [\text{m/s}]` or topographic roughness index, :math:`TRI` which is defined as the mean difference between a central pixel and its eight surrounding cells. The explanatory variables of two candidate models are: Model 1: .. math:: X = 4.925 + 0.694 \ln(PGV) - 0.459 \sqrt{TRI} - 0.403 \ln(d_{c} + 1) \quad .. math:: -0.309 \ln(d_{r} + 1) - 0.164 \sqrt{Z_{wb}} \\ (10) Model 2: .. math:: X = 9.504 + 0.706 \ln(PGV) - 0.994 \ln(V_{s30}) - 0.389 \ln(d_{c} + 1) \quad .. math:: -0.291 \ln(d_{r} + 1) - 0.205 \sqrt{Z_{wb}} \\ (11) and the probability of liquefaction is calculated using equation (3). Zero probability is heuristically assigned if :math:`PGV < 3 \, \text{cm/s}` or :math:`V_{s30} > 620 \, \text{m/s}`. The proposed probability threshold to convert to class outcome is 0.4. ####################### Allstadt et al. (2022) ####################### The model proposed by `Allstadth et al. (2022) `_ uses the model proposed by `Rashidian et al. (2020) `_ as a base with slight changes to limit unrealistic extrapolations. The authors proposed capping the mean annual precipitation at :math:`2500 \, \text{mm}`, and :math:`PGV = 150 \, \text{cm/s}`. The magnitude scaling factor :math:`MSF`, explanatory variables, :math:`X`, probability of liquefaction, :math:`P(L)`, and liquefaction spatial extent, :math:`LSE` are calculated using the set of equations previously shown. The proposed probability threshold to convert to class outcome is 0.4. ####################### Todorovic et al. (2022) ####################### A non-parametric model was proposed to predict liquefaction occurrence and the associated probabilities. The general model was trained on the dataset including inventories from over 40 events. A set of candidate variables were considered and the ones that correlate the best with liquefaction occurrence are identified as: strain proxy, a ratio between :math:`pgv \, [\text{cm/s}]` and :math:`V_{s30} \, [\text{m/s}]`; distance to the closest water body, :math:`d_{w} \, [\text{km}]`, ground water table depth and :math:`gwd \, [\text{m}]`, average precipitation, :math:`precip \, [\text{mm}]`. ************************************************** Permanent ground displacements due to liquefaction ************************************************** Evaluation of the liquefaction induced permanent ground deformation is conducted using the methodology developed for the HAZUS software by the US Federal Emergency Management Agency. Lateral spreading and vertical settlements can have detrimental effects on the built environement. ######################### Lateral spreading (Hazus) ######################### The expected permanent displacement due to lateral spreading given the susceptibility category can be determined as: .. math:: E[PGD_{sc}] = K_{\Delta} \, E[PGD|(PGA/PL_{sc}) = a] \\ (12) Where: :math:`E[PGD|(PGA/PL_{SC})=a]` is the expected ground displacement given the susceptibility category under a specified level of normalised shaking, and is calculated as: .. math:: 12\, x - 12\ for\ 1 < PGA/PGA(t) < 2 \\ (13) .. math:: 18\, x - 24\ for\ 2 < PGA/PGA(t) < 3 \\ (14) .. math:: 70\, x - 180\ for\ 3 < PGA/PGA(t) < 4 \\ (15) :math:`PGA(t)` is theminimum shaking level to induce liquefaction (see Table 1) :math:`K_{\Delta}` is the displacement correction factor given that modify the displacement term for magnitudes other than :math:`M7.5`: .. math:: K_{Delta} = 0.0086M^3\ - 0.0914M^2\ + 0.4698M\ - 0.9835 \\ (16) ############################ Vertical settlements (Hazus) ############################ Ground settlements are assumed to be related to the area's susceptibility category. The ground settlement amplitudes are given in Table 3 for the portion of a soil deposit estimated to experience liquefaction at a given ground motion level. The expected settlements at the site is the product of the probability of liquefaction (equation 1) and the characteristic settlement amplitude corresponding to the liquefaction susceptibility category, :math:`LSC`. +---------------+--------------------------+ | **LSC** | **Settlements (inches)** | +===============+==========================+ | very high | 12 | +---------------+--------------------------+ | high | 6 | +---------------+--------------------------+ | moderate | 2 | +---------------+--------------------------+ | low | 1 | +---------------+--------------------------+ | very low | 0 | +---------------+--------------------------+ | none | 0 | +---------------+--------------------------+ Table 3: Ground settlements amplitudes for liquefaction susceptibility categories. Landslide models ---------------- Landslides are considered as one of the most damaging secondary perils associated with earthquakes. Earthquake-induced landslides occurs when the static and inertia forces within the sliding mass reduces the factor of safety below 1. Factors contributing to a slope failure are rather complex. The permanent displacement analysis developed by `Newmark (1965) `_ is used to model the dynamic performance of slopes (`Jibson et al., 2000 `_, `Jibson 2007 `_). It considers a slope as a rigid block resting on an inclined plane at an angle :math:`\alpha` (derived from Digital Elevation Model, DEM). When the input motion which is expressed in terms of acceleration exceeds the critical acceleration, :math:`a_{c}`, the block starts to move. The crtical acceleration accounts for the shear strength and geometrical characteristics of the sliding surface, and is calculated as: .. math:: a_{c} =(F_{s}-1)\ - \sin(\alpha)\cdot g \\ (17) The lower bound of :math:`a_{c}` is set to 0.05 to avoid unrealistically large displacements. The static factor of safety is calculated as: .. math:: F_s = \frac{c'}{\gamma t \sin(\alpha)} + \frac{\tan(\phi')}{\tan(\alpha)} - \frac{m \gamma_{w} \tan(\phi')}{\gamma \tan(\alpha)} \\(18) where: :math:`c \, [\text{Pa}]` is the effective cohession with typical values ranging from :math:`20 \text{kPa}` for soils up to :math:`20 \, {MPa}` for unfaulted rocks. :math:`\alpha^\circ` is the slope angle. :math:`\phi'^\circ` is the effective friction angle with typical values ranging from :math:`30^\circ` to :math:`40^\circ`. :math:`\gamma \, [\text{kg/m^3}]` is the dry density of the soil or rock. It ranges from :math:`1500 \, \text{kg/m^3}` for soils to :math:`2500 - 3200 \, \text{kg/m^3}`. The slope-normal thickness of a failure slab is set to its default value of :math:`t=2.5 \, \text{m}`. :math:`m` is the proportion of slab thickness that is saturated with default value of 0.1. :math:`\gamma_{w} \, [\text{kg/m^3}]` is the unit weight of water which equals to :math:`1000 \, \text{kg/m^3}`. Note that the units of the input parameters reported in this document corresponds to the format required by the Engine to produce correct results. The first and second term of the the equation corresponds to the cohesive and frictional components of the strength, while the third component accounts for the strength reduction due to pore pressure. A variety of regression equations can be used to estimate the Newmark displacements, and within the engine, Newmark displacement as a function of critical acceleration ratio and moment magnitude is implemented. The displacement is in units of meters. The computed displacements do not necessarily correspond directly to measurable slope movements in the field, but the modeled displacements provide an index to correlate with field performance. `Jibson et al. (2000) `_ compared the predicted displacements with observations from 1994 Northridge earthquake and fit the data with Weilbull curve. The following equation can be used to estimate the probability of slope failure as a function of Newmark displacement, :math:`D`. .. math:: P(f) = 0.335\ [1 - e^{-0.048 \cdot D^{1.565}}] \\ (19) The rock-slope failures are the other common effect observed in earthquakes. The methodology proposed by `Grant et al., (2016) `_ captures the brittle behavior associated with rock-slope failures and discontinuities common in rock masses. The static factor of safety is computed as: .. math:: F_s = \frac{2 c \sin(\beta)}{\gamma h (\beta-\alpha) \sin(\alpha)} + \frac{\tan(\phi)}{\tan(\alpha)} \\ (20) where: :math:`c \, [\text{Pa}]` is the cohession with typical values ranging from :math:`20 \, {kPa}` for soils up to :math:`20 \, {MPa}` for unfaulted rocks. The cohesion provided by the root systems of vegetated hillslopes, :math:`c_{r}`, is adopted as 0. :math:`\alpha^\circ` is the slope angle. :math:`\gamma \, [\text{kg/m^3}]` is the dry density of the soil or rock. It ranges from :math:`1500 \, \text{kg/m^3}` for soils to :math:`2500 \text{ to } 3200 \, \text{kg/m^3}`. :math:`[m]` is the vertical height of the failure mass and it corresponds to 1/4 of the local relief :math:`H` calculated based on the moving window analysis. :math:`\phi^\circ` is the effective friction angle with typical values ranging from :math:`30^\circ` to :math:`40^\circ`. :math:`\beta` is the slope's critical angle calculated as: .. math:: \beta = \frac{\alpha + \phi}{0.5} \\ (21) The critical acceleration is computed similarly to equation (17). For rock- slope failures, the :math:`\alpha` term is replaced with :math:`\beta`. Finally, the coseismic displacements are estimated using the sliding block displacement regression equation proposed by `Jibson (2007) `_. **************************** Nowicki Jessee et al. (2018) **************************** A geospatial model used to predict probability of landsliding using globally available geospatial variables was proposed by `Nowicki Jessee et al. (2018) `_. The level of shaking is characterised by Peak Ground Velocity , :math:`PGV`. Slope steepness affects slope stability, and here, the topographic slope, :math:`slope`, has been derived from the median elevation value from the 7.5 arc sec Global Multi-resolution Terrain Elevation Data (`Danielson and Gesch, 2011 `_). The model uses lithology, as a proxy for the strength of the shaken material. The global lithology map is available in `Hartman and Moosdort, 2012 `_. Slope stability is further controlled by the composite strength of the soil-vegetation root matrix. The Globcover 2009 data, available at 300-m resolution and separated into 20 classes has been used. More details on this database is available in `Arino et al. (2012) `_. Finally, the Compound Topographic Index, :math:`CTI`, has been used to characterise the wetness of the material. Explanatory variable :math:`X` is calculated as: .. math:: X = -6.30 + 1.65 \ln(PGV) - 0.06 Slope + \alpha \cdot lithology \quad .. math:: + \beta \cdot landcover + 0.03 CTI - 0.01 \ln(PGV) \cdot Slope \\ (22) Coefficients \alpha and \beta values are estimated for several rock and landcover classes. The reader is reffered to the original study by `Nowicki Jessee et al. (2018) `_, where the coefficient values are reported in Table 3. Probability of landsliding is then evaluated using logistic regression. .. math:: P(L) = \frac{1}{1+e^X} \\ (23) These probabilities are converted to areal percentages to unbias the predictions. .. math:: L_{P}(P) = e^{-7.592 + 5.237 \cdot P - 3.042 \cdot P^2 + 4.035 \cdot P^3} \\ (24) Furthermore, we introduced modifications by the USGS, capping the peak ground velocity at :math:`PGV = 211 \, \text{cm/s}`, and compound topographic index at :math:`CTI = 19`. To exclude high probabilities of landsliding in nearly flat areas due to the combination of other predictor variables, areas with slopes less than :math:`2^\circ` are excluded. Zero probability is heuristically assigned if :math:`PGA = 0.02 \, \text{g}`. Finally, we adopted the USGS recommendation for modifying the regression coefficient for unconsolidated sediments. The new proposed value is set to :math:`-1.36`. Reference --------- [1] HAZUS-MH MR5 Earthquake Model Technical Manual (https://www.hsdl.org/?view&did=12760) [2] Youd, T. L., & Idriss, I. M. (2001). Liquefaction Resistance of Soils: Summary Report from the 1996 NCEER and 1998 NCEER/NSF Workshops on Evaluation of Liquefaction Resistance of Soils. Journal of Geotechnical and Geoenvironmental Engineering, 127(4), 297–313. https://doi.org/10.1061/(asce)1090-0241(2001)127:4(297) [3] I. D. Moore, R. B. Grayson & A. R. Ladson (1991). Digital terrain modelling: A review of hydrological, geomorphological, and biological applications. Journal of Hydrological Processes, 5(1), 3-30. https://doi.org/10.1002/hyp.3360050103 [4] Wald, D.J., Allen, T.I., (2007). Topographic Slope as a Proxy for Seismic Site Conditions and Amplification. Bull. Seism. Soc. Am. 97 (5), 1379–1395. [5] Zhu et al., 2015, 'A Geospatial Liquefaction Model for Rapid Response and Loss Estimation', Earthquake Spectra, 31(3), 1813-1837. [6] Bozzoni, F., Bonì, R., Conca, D., Lai, C. G., Zuccolo, E., & Meisina, C. (2021). Megazonation of earthquake-induced soil liquefaction hazard in continental Europe. Bulletin of Earthquake Engineering, 19(10), 4059–4082. https://doi.org/10.1007/s10518-020-01008-6 [7] Zhu, J., Baise, L. G., & Thompson, E. M. (2017). An updated geospatial liquefaction model for global application. Bulletin of the Seismological Society of America, 107(3), 1365–1385. https://doi.org/10.1785/0120160198 [8] Rashidian, V., & Baise, L. G. (2020). Regional efficacy of a global geospatial liquefaction model. Engineering Geology, 272, 105644. https://doi.org/10.1016/j.enggeo.2020.105644 [9] Allstadt, K. E., Thompson, E. M., Jibson, R. W., Wald, D. J., Hearne, M., Hunter, E. J., Fee, J., Schovanec, H., Slosky, D., & Haynie, K. L. (2022). The US Geological Survey ground failure product: Near-real-time estimates of earthquake-triggered landslides and liquefaction. Earthquake Spectra, 38(1), 5–36. https://doi.org/10.1177/87552930211032685 [10] Baise, L. G., Akhlaghi, A., Chansky, A., Meyer, M., & Moeveni, B. (2021). USGS Award #G20AP00029. Updating the Geospatial Liquefaction Database and Model. Tufts University. Medford, Massachusetts, United States. [11] Todorovic, L., Silva, V. (2022). A liquefaction occurrence model for regional analysis. Soil Dynamics and Earthquake Engineering, 161, 1–12. https://doi.org/10.1016/j.soildyn.2022.107430 [12] Newmark, N.M., 1965. Effects of earthquakes on dams and embankments. Geotechnique 15, 139–159. [13] Jibson, R.W., Harp, E.L., & Michael, J.A. (2000). A method for producing digital probabilistic seismic landslide hazard maps. Engineering Geology, 58(3-4), 271-289. https://doi.org/10.1016/S0013-7952(00)00039-9 [14] Jibson, R.W. (2007). Regression models for estimating coseismic landslide displacement. Engineering Geology, 91(2-4), 209-218. https://doi.org/10.1016/j.enggeo.2007.01.013 [15] Grant, A., Wartman, J., & Grace, A.J. (2016). Multimodal method for coseismic landslide hazard assessment. Engineering Geology, 212, 146-160. https://doi.org/10.1016/j.enggeo.2016.08.005 [16] Nowicki Jessee, M. A., Hamburger, M. W., Allstadt, K., Wald, D. J., Robeson, S. M., Tanyas, H., et al. (2018). A global empirical model for near-real-time assessment of seismically induced landslides. Journal of Geophysical Research: Earth Surface, 123, 1835–1859. https://doi.org/10.1029/2017JF004494 [17] Danielson, J.J., and Gesch, D.B., 2011, Global multi-resolution terrain elevation data 2010 (GMTED2010): U.S. Geological Survey Open-File Report 2011–1073, 26 p. [18] Hartmann, J., and N. Moosdorf (2012), The new global lithological map database GLiM: A representation of rock properties atthe Earth surface, Geochem. Geophys. Geosyst., 13, Q12004, doi:10.1029/2012GC004370. [19] Arino, O., Ramos Perez, J.J., Kalogirou, V., Bontemps, S., Defourny, P., Van Bogaert, E. (2012): Global Land Cover Map for 2009 (GlobCover 2009), https://doi.org/10.1594/PANGAEA.787668