Flow depth is the key HIM for many tsunami fragility curves (Section 1; Koshimura et al., 2009; Suppasri et al., 2013), however, other hydrodynamic HIMs are often cited as being more important (De Risi et al., 2017; J Macabuag et al., 2016; Song et al., 2017). Here we analysed the relative importance of flow depth (interpolated and simulated), flow velocity and hydrodynamic force. Flow depth showed high importance for physical damage to roads and utility poles (Section 4.1). This is not to say that hydrodynamic features of tsunamis are un-important with respect to direct damage impacts, and it is likely these factors contribute to direct damage through other actions. Non-hydrodynamic hazards, such as scour and debris impact, showed high importance for physical damage and were likely have exacerbated damage.
Scour, being the strongest variable in RFM and SRC testing (Figure 3 and Figure 4) for roads (and second highest for utility poles), and debris impact the strongest for utility poles, demonstrates a clear need to further investigate the non-hydrodynamic impacts from tsunamis. The next steps would be to determine when and where scour and debris impact is most important with further investigation into the relationships between hydrodynamic force, soil conditions (scour) and land-use that could lead to scour and debris potential. Previous studies have investigated debris potential for tsunami impact assessment of buildings (J Macabuag et al., 2016; Naito et al., 2014; Shafiei et al., 2016). However, the only example of debris-based fragility assessment for roads is Williams et al. (2020a), which investigates the relationship between debris deposition and road service rather than debris impact and damage level.
This work, for scour and debris impact, could be undertaken with existing tsunami impact datasets (e.g. Tohoku 2011, Illapel 2015, and Sulawesi 2018) and progressed further by subsequent field observations. Ultimately this could progress the tsunami hazard field, beyond the use of traditional hazard maps for impact assessment, by enhancing these with non-hazard parameters (debris potential, land-use, ground conditions etc.). The importance of scour for component damage presented in this study is supported by previous studies (J. Chen et al., 2013; Yeh et al., 2013) that have observed considerable scour-based damage for roads. Notably, the distribution of scour observed in the 2018 Sulawesi tsunami (Paulik et al., 2019; Williams et al., 2020b), in that high concentrations were observed along a coastal esplanade and low concentrations observed >50 m from the shoreline is consistent with that observed in the Illapel event (i.e. along Avineda Costanera), which are comparable in relative size. This may indicate a relative importance of the scour-causing conditions (e.g. ground conditions and land-use) for relatively small tsunami events in contrast to larger subduction zone events (e.g. IOT and Tohoku) where scour is still observed at high concentrations further inland (i.e. >50m).
There is a positive monotonic relationship between flow depth and both flow velocity and hydrodynamic force in the SRC tests (Figure 3). This indicates flow depth could be a fair proxy for flow velocity and hydrodynamic force, and vice versa, in the application of fragility curves. This is consistent with previous studies (Maruyama and Itagaki, 2017; Williams, et al. 2020b; Williams, et al. 2020a) that propose, but do not test, this relationship. However, this is inconsistent with a number of studies (Song et al., 2017; Wang et al., 2020) that claim flow depth is not a reliable metric for assessing tsunami hazard and damage relationships. One implication of this finding is that while velocity and hydrodynamic force are important HIMs, flow depth is of the greatest importance, and could therefore be used as a reliable proxy for tsunami exposure, impact and risk assessment. More work should be done in the future to correlate these network component and HIM relationships. As interpolated flow depth is the most commonly used HIM for fragility assessment (e.g. Koshimura et al., 2009; Maruyama and Itagaki, 2017; Mas et al., 2020). This, and future, suites of empirical and numerical tsunami fragility curves could be applied to refine tsunami impact and risk assessment with the likes of HIM weightings supported by variable correlation testing (i.e. from this study and future studies).
Roads in Coquimbo performed poorly when compared with the 2011 Tohoku event, and performed considerably well when compared with the 2018 Sulawesi event (Figure 9). In the 2011 Tohoku tsunami, the probability of complete damage (i.e. DL3) at 5 m flow depth was 0.10 (Williams et al., 2020a), compared to 0.23 (interpolated depth) and 0.27 (simulated depth) reported in the present study (Figure 9a). It should be noted that these use a GLM method, with manual data binning, as do the Illapel fragility curves from the same study. Using the same dataset, but different fragility function development methods (Section 3.4), Coquimbo roads are shown to perform worse in the present study (e.g. 0.53 probability of exceeding/reaching DL3 at 5m (interpolated depth) and 0.27 (simulated depth)) than indicated by Williams et al. (2020), (e.g. 0.12 probability of reaching or exceeding DL3 at 5m). The CLM method, used in the present study, is considered to be appropriate for fitting curves to datasets of this nature (Lallemant et al., 2015; Williams et al., 2019a; Williams et al., 2020b), implying that this study has refined the event fragility curves in a way that would have considerably higher levels of damage modelled for subsequent impact and risk assessments. This study, therefore, represents a considerable improvement to the field, and a case could be made to similarly improve the comparable road component fragility curves for the 2011 Tohoku tsunami event (Graf et al., 2014; Williams et al., 2020a). The most directly comparable previous study for road fragility curves are from the 2018 Sulawesi tsunami (also CLM method), which has a 0.57 probability of reaching or exceeding DL3 at 5 m flow depth (Williams et al., 2020b). With respect to utility poles, the low threshold for pole damage (e.g. 0.56 of reaching or exceeding DL3 at 5m (interpolated depth) and 0.46 (simulated depth) is lower in the present study, compared to pole fragility analysed for the 2018 Sulawesi tsunami (e.g. 0.86 of reaching or exceeding DL3 at 5m).
It is crucial that the local hazard and network component conditions are considered when applying tsunami fragility curves. The 2015 Illapel tsunami was a relatively small tsunami by historical standards. It was also documented as producing relatively long wave-lengths given its regional subduction zone earthquake origins – meaning a slow deterioration of wave energy inland (Heidarzadeh et al., 2015). Comparably, with the only other available suite of CLM tsunami fragility curves for infrastructure network components for an event of this magnitude, the 2018 Sulawesi tsunami was documented as a notably short wave-length (traveling ~0.3 km inland), given its locally sourced earthquake and landslide induced origins (Aránguiz et al., 2020). Conversely, the 2011 Tohoku tsunami, which has GLM fragility curves for infrastructure network components available (Williams et al., 2020a), was documented as having considerable long wave-lengths (traveling ~11 km inland). The fragility curves developed in this study would not be appropriate to apply for events of higher or smaller wavelengths, as much as the other events’ (Sulawesi and Tohoku) fragility curves would not necessarily be appropriate for an event comparable to the 2015 Illapel tsunami. On a similar note, the levels of shaking experienced in each respective case-study, and therefore it’s bearing on interpolated component damage levels, varies considerably (11.5 – 21.5 g in Miyagi and Iwate Prefectures (Tohoku) and 11.5 – 40.1 g in Palu (Sulawesi) and compared with 0.20–0.29 g in Coquimbo (Illapel)) and should be considered in any application of fragility curves for impact and risk assessment (USGS, 2015). The growing catalogue of post-event empirical component damage, hazard and fragility datasets, provides wider scope for increasingly sophisticated and credible tsunami impact assessment, of which will ultimately inform tsunami risk reduction globally.