Current techniques used to predict liquefaction-induced lateral spreading are mostly empirical (Bray et al. (2010) [
10]). Observations from recent earthquakes have shown that these models become inaccurate when extrapolated beyond their limits, such as large-magnitude events or different fault types. In this section, the phenomena of liquefaction and lateral spreading and the issues with current prediction models, when applied to subduction earthquakes, are described.
2.1. Liquefaction and Lateral Spreading
The word liquefaction was first used after the 1964 Niigata Mw7.6 earthquake (Kawata et al. 2018 [
11]). This phenomenon is defined as a change of the soil phase from a solid to a liquid state due to pore water pressure increment, and the corresponding loss of effective stress, during an earthquake (
Figure 1). As Youd (2018) [
12] indicated, when an earthquake occurs, waves propagate through the soil, shear strains increase, pore water pressure goes up, and the intergranular forces get reduced. As pore water pressures reach a critical level, and the intergranular stresses approach zero, the soil behavior goes from a solid to a viscous liquid state.
Liquefaction is a relevant soil phenomenon for geotechnical design as it may cause local or global failures of foundations and even the collapse of complete structures (Jia (2017) [
13]). Among the potential consequences of soil liquefaction, one of the most dangerous ones is lateral spreading. Youd (2018) [
12] defined this phenomenon as the horizontal displacement of a soil layer riding on liquefied soil either down a gentle slope or toward a free face like a river channel (
Figure 2). When the underlying soil layer liquefies, the non-liquefied upper soil crust continues moving down until it reaches a new equilibrium position.
Figure 3 shows two recent examples of liquefaction-induced lateral spreading: (1) west levee of the Westside Main Canal in the 2010 Sierra El Mayor Mw 7.1 earthquake, where a cumulative horizontal displacement of more than 1 m was observed (
Figure 3a), and (2) the Muzoi Bridge in the 2005 Nias Island Mw 8.6 earthquake, where a lateral movement of more than 4 m towards the river on both sides of the bridge was reported (
Figure 3b).
Prediction of lateral spreading is essential because it can cause damage to the overlying and subsurface infrastructure, and the amount of displacement may influence the design of the infrastructure concerning the decision to, for instance, perform soil improvement in the area affected by this phenomenon (Bray et al. (2017) [
10]).
2.3. Current Models and Large-Magnitude Subduction Earthquakes
Tryon (2014) [
7] evaluated six empirical models used in practice (Youd et al. (2002) [
5]; Bartlett and Youd (1995) [
2], Faris et al. (2006) [
3], Zhang et al. (2012) [
4], and Zhang et al. (2004) [
17]) with three case-histories from the 2010 Maule Mw 8.8 subduction earthquake. He found that site-to-source distances are difficult to define accurately for large subduction zone earthquakes. They can vary significantly between seismic regions, making it difficult to recommend a method for calculating such an “R” value.
Figure 4 shows a summary of different distance terms that can be considered: D1 = hypocentral distance, D2 = epicentral distance, D3 = closest distance to high-stress zone, D4 = closet distance to the edge of the fault rupture, D5 = closest distance to the surface projection of the rupture (Joyner Boore distance). In large subduction earthquakes, although there is a small area where the earthquake begins (hypocenter), there are multiple zones on the contact between plates (“patches”) where energy is released at different times and with different intensities. Hence, although distances D1, D2, and D3 could be defined, they do not necessarily have a reasonable correlation with the intensity of the ground motion at the site of interest. Additionally, for seismically active countries, such as Chile and Peru, D4 and D5 are very small or even zero. From a design point of view, estimating these distances before an earthquake occurs is very difficult.
Similarly, Williams (2015) [
8] used two case-histories from the 2010 Maule Mw 8.8 earthquake to evaluate the empirical methods developed by Youd et al. (2002) [
5] and by Bartlett and Youd (1995) [
2], concluding that they are extremely sensitive to the distance term, R, and that the current definition of R for these two methods (the Joyner Boore distance) resulted in predictions that were more than two times the measured values. The semi-empirical models by Zhang et al. (2004) [
17] and Faris et al. (2006) [
3] also over predicted the displacement but in these cases due to the depth weighting factor of their models. In particular, the empirical model of Zhang et al. (2004) [
4] predicted displacements roughly six to eight times larger than the measured displacements. On the other hand, De la Maza et al. (2017) [
22] studied one case history (Caleta Lo Rojas) from the 2010 Maule Mw8.8 earthquake. They used the Youd et al. (2002) [
5] methodology with different distances, finding that the distance to the zone that bounds 10% of the largest slips resulted in satisfactory values when compared against in-situ post-earthquake measurements.
In this study, we analyzed 13 lateral spread cases from six sites affected by the 2010 Maule Mw 8.8 earthquake, where lateral spreading took place (
Figure 5).
Figure 6 shows a comparison between observed and calculated lateral spreading using Youd et al.’s (2002) [
5] methodology with three R-value definitions. The first one is the original R from Youd et al.’s (2002) [
5] methodology, the second one is the distance to the maximum observed coastal uplift, and the third one is the distance used by De la Maza et al. (2017) [
22], which is defined as the distance to the zone that bounds 10% of the largest slips. The measured lateral displacements at the selected sites were between 1 and 2 m.
In all cases, the conclusion was similar to those of Tryon (2014) [
7], Williams (2015) [
8], and De la Maza et al. (2017) [
22], namely in that the Youd et al. (2002) [
5] model, for large-magnitude subduction earthquakes, overestimates the liquefaction-induced lateral displacements by a factor of more than two.
Figure 6c shows, however, that there are a few sites where the predictions were close to the measurements. Those sites were those where the R-value was that of De la Maza et al. (2017) [
22] and where the average fines content in the cumulative thickness of the saturated granular layer was less than 5%. This is only an initial observation, and much more case-histories need to be studied before generalizing, or not generalizing, this conclusion.