Seismic response of a motorway bridge founded in an active landslide: a case study

A twin girder 7-pier bridge, belonging to the "Egnatia" highway that has been facing numerous challenging geohazards, is built within an active landslide. Its seismic performance is investigated here through a comprehensive analysis of the interaction between bridge, foundation, and the precarious slope, which might affect 4 of the piers. The numerical 3D modeling considers in a realistic way the coupled effects of topography, soil nonlinearity, slope instability, and reinforced-concrete plasticity during seismic loading (kinematic and inertial). Alternative foundation schemes and slope stabilizing techniques are generically compared and evaluated. The aim is to develop a multi-hazard risk assessment platform that could facilitate the long-term management of motorways while shedding some light on the multi-hazard soil-structure interaction (MH-SSI).


Introduction
Transportation networks constitute a vital ingredient of modern, densely populated societies, where the transfer of goods and people bring social and economic prosperity. Motorways are spatially distributed systems comprising bridges, tunnels, culverts, and earthworks that might be affected by multiple hazards. The smooth operation of these network systems after a disruptive event is crucial. Bridges are perhaps the most vulnerable assets of a road network; their damage has been historically considered a significant threat to motorists (Priestley et al. 1996). Earthquakes and geohazards such as landslides, debris flow, and ground movement could be a detrimental combination for highway bridges (Bhattacharya et al. 2018;Kawashima and Buckle 2013;Winter et al. 2013). Bridges often cross streams and slopes, and could be susceptible to erosion, landslides, and liquefaction, all contributing to damage, with direct and indirect losses (Kawashima et al. 2011). Therefore, it is important to assess the effects of topography, local soil conditions, potential ground instabilities, and seismic shaking. This calls for confronting seismic and geotechnical hazards simultaneously (Kiremidjian et al. 2007).
In the last years, the focus on natural hazards and their effects on infrastructure grew significantly, leading to seismic risk assessment studies of road networks and bridges. Probabilistic seismic analyses have been adopted to identify the hazard levels, evaluate the vulnerability of each structural component to each specific hazard, conduct a risk assessment of the assets, and finally assess the resilience of highway networks (Argyroudis et al. 2015; Kilanitis and Sextos 2019;Kiremidjian et al. 2007). But most of these studies have addressed only the dominant hazard, without considering multi-hazard scenarios that may affect the infrastructure. The seismic hazard in particular, with emphasis on soil-foundation-superstructure interaction, has been very extensively investigated (I. Anastasopoulos et al. 2015;Kappos and Sextos 2001;Makris et al. 1994Makris et al. , 1996Mylonakis et al. 1997). Hence, the need for a comprehensive approach that considers the simultaneous occurrence of multiple-hazard actions (Chulahwat and Mahmoud 2017; Gidaris et al. 2017;Yanweerasak et al. 2018). More specifically, the simultaneous seismic shaking and ground-failure actions have been investigated by (Fotopoulou and Pitilakis 2017;Pitilakis and Kalliopi 2013;Tang et al. 2020). In this paper, a comprehensive numerical investigation of the seismic behaviour of a bridge founded on an active landslide has been undertaken to provide insight on the effects of coupling of seismic and ground kinematics.
Among the methodologies developed the to study the multiple-hazard seismic soil-foundation one could mention analytical solutions (Di Laora et al. 2017;Elahi et al. 2010;Gazetas et al. 1993;Wen et al. 2015), numerical studies (Barla 2018;Jin et al. 2010;Kourkoulis et al. 2010;Luo et al. 2019;Uzuoka et al. 2007), and 1-g seismic and centrifuge experiments (Wang and Zhang 2014;Yan et al. 2020). To mitigate landslide risk and protect the affected infrastructure, the use of stabilising piles and deep foundations has been examined by (He et al. 2015;Kanagasabai et al. 2011;Kourkoulis et al. 2011Kourkoulis et al. , 2012Yu et al. 2015). Most of this research investigated soil-structure interaction by decoupling the superstructure from the foundation in a multi-step approach.
In this study we examine a bridge of the Egnatia motorway in Greece, which is supported on pile and caisson foundations on top of an active landslide. A detailed numerical model of the whole structure-soil-foundation system is generated. Rational constitutive models are adopted to represent both soil and concrete under static and dynamic actions. A multi-hazard approach is considered herein, where seismic and ground-instability are imposed simultaneously on the system. The barely acceptable performance obtained from the analyses of the foundation-bridge system prompted the investigation of certain slope stabilisation interventions, in the form of a double series of staggered rows of rigidlycapped piles. It is shown that with such improvements even a smaller size of foundation would have offered an acceptable performance, even with the strongest seismic excitation-a refutation of the fallacy commonly held that by conservatively oversizing foundations we achieve greater safety (Ioannis Anastasopoulos et al. 2010). Of course, other mitigation schemes could also be incorporated for a technically sound solution, avoiding the unnecessary conservatism in foundation design.

Summary of design
This bridge ("Panagia" Interchange) of the Egnatia motorway in Greece is a typical T-girder beam bridge built in 2007. It consists of two branches with 37.5 m span and pier height up to 25 m. It is a modern structure with decks supported on elasto-metallic rubber bearings providing seismic isolation. Part of the bridge is founded on a precarious slope suffering from creeping deformations, and vulnerable to permanent sliding in a strong seismic event, as well as after heavy rainfall (Fig. 1).
Seven circular piers support each deck and are founded on deep foundations. A longitudinal section of the bridge is given in Fig. 2. The foundations are either pile groups of 6-8 piles of 1.2 m diameter and 23-30 m length (P1-P4), or massive caissons (P5-P7). The latter are located within the area of potential landslide, not only to ensure sufficient foundation capacity and protect the structure, but also to stabilize the slope. The depth of these caissons varies between 20 and 32 m below ground level and their diameter between 5 and 7 m. Twenty additional 0.8 m-diameter piles were constructed around the caisson periphery perhaps for enhancing the structural capacity of the foundation (or other unknown to us reasons related to construction). Typical foundation cross-sections are shown in Fig. 3, while the geometric parameters are given in Table 1.  Table 2, based on the geotechnical investigation by "Geognosi", a geotechnical engineering company.

Seismicity
The G1 bridge is located in a mountainous region of moderate seismicity (design PGA for 475-year return period according to EC8 is 0.16 g). In this study, the OpenQuake opensource software is used for performing the seismic hazard and deaggregation analyses, by employing the area source model of the SHARE project (Giardini et al. 2013) (Fig. 5a) and the ground motion prediction equation of Boore and Atkinson (2008). The methodology is described in more detail in Loli et al. (2020). The Intensity Measure (IM) considered herein is the average peak spectral acceleration, AvgS a (Vamvatsikos and Cornell 2005), and is calculated using the following equation: We notice that IM is derived by combining n spectral acceleration ordinates (S α ) at periods T Ri . Each ordinate is the geometric mean of the 5%-damped spectral acceleration from the two horizontal components. The period range of [0.3 s, 3.0 s] with an increment of 0.1 s is adopted here. The resulting site-specific seismic hazard curve is depicted in Fig. 5b in terms of AvgS a versus the mean annual frequency (MAF) of exceedance. The conditional mean spectrum (CMS), its 2.5th and 97.5th percentiles (CMS ± 2σ) and the response spectra of the selected records are presented in Fig. 5d.
Eleven (11) acceleration time histories for each seismic intensity level (i.e., weak, moderate, and strong) have been adopted as excitation at the base of the models. The accelerograms and the elastic response spectra of the "strong" scenario only are presented in Fig. 6. A summary of the characteristics of the selected ground records is provided in Table 3.
For this study, a scenario-based approach has been adopted. Seismic motions consistent with three different levels of the seismic hazard (i.e., weak: probability of exceedance 20% in 50 years; moderate: probability of exceedance 10% in 50 years; and strong: probability of exceedance 2% in 50 years) were considered with the current condition of the creeping/ failing slope (as described in detail below). Analysis has been performed for the worstcase scenario of maximum groundwater height reaching the ground surface-a reasonably conservative approximation in view of past observations that found the ground water level above the failure zone even during dry periods, the relatively high precipitation levels in the area, and the low soil permeability.
In this case, the effect of pore pressures in reducing the effective stress and thereby the strength of the weak zone is indirectly taken into by proportionally reducing the effective friction angle (φ') in the calculations. Results are presented for all the investigated hazard scenarios, but more details and discussion are provided for the most detrimental case: intense earthquake (probability of exceedance 2% in 50 years) and water table is at the ground surface.
For the most intense level of shaking, results from analysis imply mobilisation of the failure surface and significant slope movement. Residual displacements of the slope NIN2, within the failure wedge, range from 0.5 m to over 3 m depending on the details of the excitation.

Potential landslide actions
Monitoring the slope movements around the site began in 2002. Several inclinometers and a few piezometers were installed at different locations in the slope (Fig. 7a). Unfortunately, most of the instruments installed in 2002 stopped giving readings when construction of the bridge started in 2007. Additional instruments were installed in boreholes N1-N6 in 2007 and on the bridge IM4-IM6 in 2014. Horizontal displacements recorded by the latter instruments are plotted in Fig. 7b. Accumulated displacements reaching up to 8 cm were measured in location N2, 46 m, uphill from Pier 5. Similar displacements were measured downhill, especially of N5 and N6. But, N6 stopped working in 2007 (shortly after installation), signifying an abrupt increase of shear deformations (revealed by the high displacement rate). Analysis of inclinometer readings led to the identification of a landslide area that comprises the central piers P4, P5, P6. Due to its steeply sloping terrain, the crosssection near P5 pictured in Fig. 7a is recognized as the most critical. The unstable soil layer is between about 8 and 11 m thick, and the distributions of horizontal displacements with depth, as per Fig. 7b, indicate strain accumulation in a thin zone (slip plane).

3
The stability of piers P5(L) and P5(R) depends on their massive foundations, which were intended to also act as a slope stabilizing measure. The foundations consist of a 32 m deep reinforced concrete circular caisson diameter 7 m in diameter. Their capacity is enhanced by 20 Φ80 cm piles installed in the periphery at a distance from the caisson, with a common rigid Accelerograms and elastic (ξ = 5%) response spectra of the ansemble of 11 excitations used as input in the numerical analyses, that are consistent with the seismic hazard with probability of exceedance 2% in 50 years (Loli et al. 2020) cap. Critical for the design was the seismic excitation. Note that the aforementioned recorded displacements were solely creeping deformations along the "failure" zone (slip-plane), probably a result of seasonal movement of the water table (Loli et al. 2020). In addition, recent information indicated that the abrupt increase of the rate of slope deformation during 2013 (Figs. 7b and 12) may be attributed to the fact that excavation material and debris were placed on top of the slope, increasing detrimentally the weight of the moving mass. Naturally, even larger displacements might be expected in case of an earthquake.

Initial "As-Built" model
An analysis framework was developed to obtain the slope movement and pier performance of the bridge. The finite element code ABAQUS is employed in a series of non-linear seismic analyses. Pier 5L is the most vulnerable to the combined slope and seismic loading, due to its location within the landslide zone (Fig. 2). The axis-to-axis distance of the caissons of the two adjacent piers (belonging to the twin branches of the bridge) is about 3.5 diameters. For dynamic inertial loading at their top under elastic conditions in homogeneous soil the interaction between the two caissons would be significant (Dobry and Gazetas 1988). However, soil nonlinearity that develops in this case, drastically reduces such interaction Kanellopoulos and Gazetas 2020;Radhima et al. 2021). Moreover, the "kinematic" effects are prominent in our case, because in addition to the direct seismic wave loading there is indirect loading from the lateral soil movement. As is well known, interaction between piles due to kinematic loading is much less significant even in elastic soil (Gazetas and Mylonakis 1998;Kaynia and Kausel 1982). Therefore, we do not expect any substantial interaction effects between the two adjacent caissons of the two branches of the bridge, especially in view of intense soil nonlinearity. The threedimensional FE modelling realistically captures the soil-foundation interaction, accounting for the exact geometry of soil and superstructure, kinematic boundaries, and pile groupcaisson response.
Using a detailed 3D model simulating the caisson-pile P5 foundation we look for the damage due to a potential slope instability originating during strong earthquake shaking (Figs. 8 and 9). Inelastic behaviour of soil, foundation, and column are modelled explicitly while the deck remains elastic. The model encompasses 640 m of ground in the longitudinal direction (of seismic shaking) and 45 m deep ground below the caisson toe. The adopted model comprises the full geometry of the slope/valley, and displacement is allowed only in the longitudinal direction of y axis. Half the model is shown in Fig. 8 for clarity. Dynamic analyses are performed for all seismic scenarios. The complete foundation-soil-superstructure is modelled for the most critical cross-section. In the analysis, acceleration time histories are applied at the base. The foundation is modelled with its steel reinforcement (Fig. 9) and the soil with nonlinear 8-nodal continuum elements having a Von-Mises failure criterion (described subsequently), while the pier column is nonlinear beam with moment-curvature (M-C s ) relationship for actual reinforced concrete sections, and derived with the RC model of Chang and Mander (1994). The caisson-and-piles foundation is modelled with 3D solid elements and the concrete-damage-plasticity [CDP] model. The M-C s relationship of this solid Sect. (3D) is calculated with a pushover analysis, while the ultimate capacity M u is Fig. 9 Details of finite element modelling, including soil, foundation, and superstructure components calculated using RC cross-section analysis (Chang and Mander 1994). Results are shown in Fig. 10. Particular care is taken for the simulation of the failing zone, highlighted in dark grey in Fig. 8. Soil-foundation Interface behaviour allows sliding (with friction coefficient μ = 0.5) and detachment.
Linear elastic connector elements and dashpots model the shear stiffness (K s ), the vertical stiffness (K v ), and damping (C) of the isolation bearings inserted between piers (or abutments) and deck. Appropriate gap elements, allowing adjacent nodes to be in contact (gap closed) or separated (gap open), replicate a δ c = 20 cm clearance between deck and stoppers (shear keys), as shown in Fig. 9. This clearance is relatively small compared to the elastic range of the bearings; hence, the assumption that the bearings remain elastic appears reasonable (Kalfas et al. 2017), as also verified by our analyses. Calculation of the bearing properties is undertaken using the following equations (Koh and Kelly 1988): where A = 0.25 m 2 is the plan area of the bearing (m 2 ); t = 0.015 mm is the thickness (m) of each elastomeric layer (Fig. 9); n = 7 is the number of individual elastomeric layers; G = 1 MPa is the elastic shear modulus of the elastomeric material; E c = 5 MPa the vertical stiffness of the elastomeric material; ξ = 10%t is the damping factor of the bearing; and ω = the angular (circular) frequency in rad/s, which in this study is taken equal to the fundamental natural frequency of the bridge.
The deck is modeled with elastic hexahedral continuum elements having the properties of RC (E = 30 GPa, γ = 25 kN/m 3 ). We properly simulate the kinematic constraints imposed Appropriate "free-field" boundaries are used at the lateral edges: dashpots are placed at the base ( C b = V s A ) to simulate the halfspace under the 45 m of the soil (ρ = 2.1 t/m 3 , V s the shear wave velocity, taken equal to 800 m/s 2 , and A = the effective area of each dashpot, function of the element size). In addition, the symmetric counterpart of the model (Fig. 8) is considered, and "node-to node" kinematic constraints are applied along its lateral edges (boundaries), forcing two nodes to have identical displacements. This aims to simulate the response of a plane-strain lateral box subjected to in-plane vertically propagating waves.

Soil
Apart from the narrow zone of elements to model the slip plane, the stress-strain behavior of the soil layers (namely, III, IV, V, VI, as per VuceƟc and Dobry, 1994 (PI = 15) VuceƟc and Dobry, 1994 (PI = 30) This Study γ τ G Fig. 11 Results of simple shear tests for the calibration of the soil constitutive parameters: comparison with the experimental curves of Vucetic and Dobry (1991) for low plasticity clays shown in Fig. 11 Due to lack of borehole and test data, the parameters of the stiff ophiolite layer (Unit VI) were taken equal to those of the stiff flysch layer (Unit IV). This constitutive law has been extensively validated previously against physical model and numerical tests, demonstrating its effectiveness in capturing the response of surface and slightly embedded foundations to cyclic and seismic shaking, the performance of piles and caissons against lateral loading, and the sliding response of slopes and retaining systems (I. Anastasopoulos et al. 2015;Garini and Gazetas 2021;Nikos Gerolymos et al. 2015;Giannakos et al. 2012;Ntritsos et al. 2015).

Slip surface
An elastoplastic constitutive law, with Mohr-Coulomb failure criterion and isotropic strain softening, models the soil in the predefined sliding surface (I. Anastasopoulos et al. 2007)marked in dark grey in Fig. 8. For this zone, the initial, pre-yielding, soil behaviour is assumed to be linearly elastic with stiffness calculated from the secant modulus G s as derived from soil element testing (Soil V, in Table 2). For the post-peak softening response, a reduction of the friction angle (φ') and dilation angle (ψ) has been calibrated against the recorded slope displacements measured in-situ since 2007. Simulation of gradual degradation from c = 5 kPa, φ peak = 33°, ψ peak = 3° in 2007 to c = 0 kPa, φ res = 18°, r est = 0° in 2019 results in numerically computed displacements that compare relatively well with the actual measurements (Fig. 12). Subsequently, the calculated state reached by the slope after creeping is used as the initial condition to calculate the seismic response through dynamic time-history analysis (Sect. 5).

Structural elements
The nonlinear concrete response is simulated with the concrete damage plasticity (CDP) model (Lee and Fenves 1998;Lubliner et al. 1989), built-in the finite-element code ABAQUS, used for both monotonic and cyclic loading (Antoniou et al. 2020;Behnam et al. 2018;Krahl et al. 2018;Zoubek et al. 2013). The adopted model captures the effects of irreversible damage associated with the two prevailing failure mechanisms of concrete: (a) tensile cracking and (b) compressive crushing. More details of the model constitutive laws and application are provided in SIMULIA (2014).
The parameters ψ, ϵ, σ bo /σ co (or f bo /f co ) and K c of the CDP model are defined as follows: • The dilation angle of the material (ψ) was assumed equal to ψ = 15°, which is close to a low value for concrete (it may be as high as 50°), since a less stiff response is expected herein. • The parameter ϵ, the "flow potential eccentricity", defines the rate at which the hyperbolic flow potential approaches its asymptote. • The ratio σ bo /σ co of the initial biaxial compressive yield stress to the initial uniaxial compressive yield stress and, • The ratio K c of the second stress invariant in the tensile meridian q(TM) to that of the compressive meridian q (CM) at the initial yield (0.5 < K c ≤ 1.0).
In this research, these parameters, calculated according to Alfarah et al. (2017), are listed in Table 4.
The concrete body of the caisson and periphery piles are modelled with quadrilateral continuum solid elements and their inelastic behaviour is described with CDP. The characteristic concrete compressive strength f ck is 35 MPa. The tensile strength is calculated according to EC2: while the Young's modulus is estimated according to Chang and Mander's (1994): The post-yielding of concrete is defined by the stress-strain relationship of concrete under uniaxial compression and uniaxial tension. In addition, the relationship of the damage parameters d t and d c , defining the unloading-reloading behaviour of concrete, with the tensile and compressive strain respectively, must be defined. These parameters are calibrated against the relationships of Chang and Mander (1994); the corresponding curves are plotted in Fig. 13. Truss elements are employed to model the S500 steel reinforcement in both longitudinal and transverse direction. Perfect bonding between the reinforcement and concrete is assumed: neither sliding nor separation are allowed. The reinforcement configuration for each pier and caisson piles is summarized in Fig. 9. The caisson longitudinal steel reinforcement ratio ρ 1 = 0.8% and the transverse t comprises Ø40 mm/15 cm. The periphery piles have longitudinal reinforcement 12Ø22 steel bars, and transverse Ø30/15. Steel mechanical behavior is described with an elastic-perfectly-plastic law with typical properties for S500 steel; yield strength f y = 500 MPa, ultimate strain ε u = 15%, elastic Young's Modulus E = 210 GPa, Poisson's ratio ν = 0.20.

Response of the G1 bridge under seismic and landslide actions
Our analyses are based on the fundamental simplifying assumption of a synchronous seismic base excitation (see a discussion on this approximation in Sextos et al. (2003)). The analysis accounts for both land-sliding and ground shaking. The numerical analyses are divided into the following steps: • Initially, a static step the gravitational forces apply to the whole model. • Next, the angle of friction (φ') and cohesion (c) of the pre-defined slipping surface is reduced to • Residual values and consequently, displacement of the slope is induced (Fig. 12). • As a last step, dynamic time history analyses are performed, with an initial deformation and stress • State from the previous step.
In other words, the calculated state reached by the slope after displacement of the pre-defined slipping surface is used as a new initial condition to calculate the earthquake response. Analysis is performed by performing a static step without modelling the creeping. Then, dynamic analysis of the system is performed, in which slippage at the failure surface is taking place during shaking. This is a reasonable approximation of what is expected years after the original creep has taken place.
The eleven (11) accelerograms adopted for each of the three (3) hazard scenarios, i.e., weak, moderate, and strong excitation levels, are used as input to the FE analysis. The response of the bridge to both slope deformation and seismic shaking of the H-CMP015 record (strong excitation) are presented in Fig. 14. Soil displacement reaches values of more than 3.5 m, indicating full mobilization of the landslide, as seen in the contours of Fig. 14. The horizontal displacement (U 1 ) of the caisson foundation reaches nearly 0.5 m while its rotation (θ) reaches 0.32 degrees. Evidently, mobilization of the soil deformation and substantial horizontal displacement and rotation of the footing are realized at t = 6.5 s, when excitation is strongest. At the same time, the maximum overturning moment on the foundation is about 70% of its ultimate capacity. Moment-curvature plots have been evaluated for all the bridge piers (P1-P7). Evidently, Pier P5 suffers the most. It experiences a ductility demand to ultimate capacity ratio of about 0.3. The other piers, unaffected by simultaneous soil movement, are less vulnerable to earthquake excitation. The bearings of Pier 3 displace Δu = 12 cm, less than the clearance between elastomers and shear keys.
A summary of the results from all dynamic analyses for moderate and strong excitation levels is presented in Figs. 15 and 16, respectively. Results for low amplitude earthquakes are not presented. The two figures compare for each of the 11 motions: the maximum displacements at the crest of the slope; the residual deformations of the caisson foundation; the ratios between applied moment on the caisson and its ultimate capacity; shear strains of the bearing; curvature-ductility demand over the ductility capacity.
The study showed that P5 is the most vulnerable pier for all seismic scenarios due to the additional stressing imposed by soil-foundation deformations. Results for the moderate shaking accelerograms that correspond to the design excitation level (Fig. 15) indicate minimal damage to the bridge. Although for a few seismic cases soil displacements reach significant values (of the order of over 1 m), foundation deformations remain limited, and the applied moment on the caisson footing for all cases is well below its ultimate capacity; this is indicative of elastic foundation response: thereby no damage is likely of concrete and steel reinforcement. Ductility demand for pier 5 remains below 10% of its ductility capacity.
The corresponding results for the strong excitations are presented in Fig. 16. Now, crest deformation reaches values of more than 3.5 m (with the H-CMP015 excitation, illustrated in Fig. 14). Ductility demand on Pier 5 is higher but still within acceptable levels, while the moment applied to the foundation is substantial, indicating concrete cracking. The above results show the potential of severe damage to some bridge components if a relatively strong earthquake excitation is accompanied by full landslide mobilization.
Overall, the "as-built" bridge indicates that the foundation remains within its elastic structural regime, with minimal damage even for the worst-case scenarios, suggesting that a more robust foundation would be seismically unnecessary. In addition, the deck bearings remain within their elastic limit, and displace less than the clearance limit (of 20 cm).
As a consequence, the shear keys are not engaged. Notice that excitations H-CMP015 and SSE330 are the most pernicious as they contain a large number of excitation cycles.
However, on the contrary, the simultaneous mobilisation of the landslide leads to permanent footing rotation and increased damage of the pier and superstructure. Hence enhancing the robustness of the foundation would be necessary. To this end, an alternative scheme to improve the performance of the system by installing "nailing" pile rows uphill of the foundation (in place of the questionable peripheral pile ring). If effective, apart from their stabilizing effect, such piles would also act as sacrificial members, protecting the caisson foundation in case of extreme shaking. The response of such an alternative is analysed below.

The need for mitigation
To further elucidate the negative role of slope instability on the performance of the system, we decouple the effect of land-sliding from that of the terrain topography. To this end, we ignore the presence of slippage interface, i.e., the upper clayey-silty-sand layer (Unit V) lies directly onto the flysch-and-ophiolite layer (Unit III, Unit IV and Unit VI). We consider the initial ("as-built") scenario only. Comparison of foundation response of the original analysis and this new no-landsliding analysis is presented in Fig. 17.
The comparison reveals the significance of slope instability on the response of the bridge. It appears that seismic shaking alone-i.e., kinematic effects, inertial loading, and even "topographic amplification" (aggravation of amplification from the non-horizontal (a) (b) (c) (d) Fig. 17 a Contours of horizontal soil displacement due to GYN000 excitation where the predefined slip surface is neglected, and b of the original case where slippage takes place; c horizontal displacement and d rotation of the foundation top for the above two scenarios ground surface, see Assimaki and Gazetas (2004))-lead to only about one-half of the total deformation of the system. Hence, stabilizing the upper layer by nailing the slippage zone is justified to reduce soil movement. This would improve the overall response, and perhaps could have allowed a reduction of foundation size, if it had been adopted before construction.

Alternative mitigation measures
We propose "nailing" the sliding surface with rows of piles, while at the same time removing (as unnecessary) the periphery piles of the caisson-piled foundation of P5. The effective total diameter of the footing reduces from 9 to 7 m. We explore the effectiveness of two different pile position scenarios, each with two additional double-pile frames: (a) the two pile frames are installed at 47 m and 74 m distance from the center of the caisson, and (b) the two pile frames at 17 m and 47 m distance from the center of the caisson (Fig. 19). Each frame consists of 1.2 m-diameter piles, heavily reinforced longitudinally (ratio of ρ 1 = 2% of the pile cross-section). The piles, spaced 3 and 4 diameters apart as shown in Fig. 19, have a length of about 35 m, ensuring adequate penetration into the stable layer below the slip surface (approximately 10 m depth), and thereby allowing full flexural mobilization of the pile-group capacity as well as activation of the ultimate reaction to the stabilising ground. They are modelled with non-linear beam elements the nodes of which are connected to the surrounding toroidal solid element nodes. The heads of the piles are connected rigidly, simulating the effect of an RC pile cap. The piles are installed in a staggered configuration that minimizes the shadow effect created from the front row of piles (Poulos and Davis 1980), consequently offering a substantially increased total reaction force to the system . Figure 18 illustrates the slope deformation contours used as a guide for selecting the proper mitigation measures, and Fig. 19 the proposed pile-row positions for the two alternative configurations. Configuration A is placed in the area of the maximum soil deformations. The location of Configuration B is meant to intercept the soil movement at the middle the failing slope and the toe of the landslide, close to the caisson. The effectiveness of each scheme is evaluated in Figs. 20 and 21 for the worst-case scenario (GYN000 excitation). The results unequivocally show that the performance of the system with both mitigation measures, A and B, has improved despite the reduced foundation size: ground displacements have been reduced substantially and the overall performance of superstructure and foundation is enhanced. Location B is superior for the caisson, location A for the slope.
With configuration A, soil movement at the crest substantially decreases to only 0.6 m, from 1.6 m of the "as-built" case. Caisson deformation, however, remains at similar levels with the initial case; hence, no significant improvement of the bridge behaviour is achieved. This is because land-sliding is restrained at the upper part of the slope, but localized soil failure at the toe is barely reduced. Hence, the caisson behaves in a similar way with the unimproved case.
With Configuration B, displacement of the crest is 0.8 m, larger than in A, but the foundation behaviour is improved in terms of both rotation and horizontal displacement. This stems from the interception of landslide mobilisation just before the caisson.
With both pile locations (A and B) the distress of the pier (P5) and the moment onto the caisson decrease by approximately 20% with respect to the initial system (Fig. 21). The piles acting as a remediation feature reach their ultimate capacity during the strong event, hence providing the maximum possible reaction force. Thus, these piled-frames serve as sacrificial members that protect the structure from seismic and landslideinduced actions. Pile systems that reduce the slope movement at the toe (B), while allowing full mobilization of slippage near the crest (type B), are the most beneficial. The installation of periphery pile is not helpful.

Conclusions
The behaviour of the G1 bridge under multi-hazard loading (earthquake shaking and landsliding actions) has been evaluated numerically. The results-including slope residual movement, foundation displacements, and superstructure response-show that ground motions of an exceedance probability of 2% in 50 years would trigger land-sliding deformations, thereby generating significant residual deformation of the foundation and structural damage to the piers. Decoupling the effects of sliding and vibration reveals that the effect of the former is critically important, and its mitigation should be a priority.
An alternative bridge foundation strategy is developed to reduce the land-sliding, rather than to make the foundation more robust. The piles along the periphery of the caisson are removed (reducing the total foundation diameter from 9 to 7 m), and staggered-pile-frames are installed to minimise the potential slope movement. Two possible locations of such pile-frames are studied, one emphasizing the "arrest" of the crest displacement (location A), and the other the ground displacements in the vicinity of the foundation (location B). Both interventions are successful. Unsurprisingly, location A leads to smaller slope displacement at the crest, but it is location B that offers the largest benefit to the foundation and the structure.
Funding Open access funding provided by HEAL-Link Greece. This research has been financed by the European Commission through the Horizon 2020 program "PANOPTIS-Development of a decision support system for increasing the resilience of transportation infrastructure based on the combined use of Fig. 21 Comparison of the two interventions on the response of the slope-caisson system to GYN000 excitation: a Moment-curvature relationship for the column of P5; b Time history of maximum moment in the caisson normalized by its moment capacity