In 2010 and 2011 Christchurch, New Zealand experienced a series of earthquakes that caused extensive damage across the city, but primarily to the Central Business District (CBD) and eastern suburbs. A major feature of the observed damage was extensive and severe soil liquefaction and associated ground damage, affecting buildings and infrastructure. The behaviour of soil during earthquake loading is a complex phenomena that can be most comprehensively analysed through advanced numerical simulations to aid engineers in the design of important buildings and critical facilities. These numerical simulations are highly dependent on the capabilities of the constitutive soil model to replicate the salient features of sand behaviour during cyclic loading, including liquefaction and cyclic mobility, such as the Stress-Density model. For robust analyses advanced soil models require extensive testing to derive engineering parameters under varying loading conditions for calibration. Prior to this research project little testing on Christchurch sands had been completed, and none from natural samples containing important features such as fabric and structure of the sand that may be influenced by the unique stress-history of the deposit. This research programme is focussed on the characterisation of Christchurch sands, as typically found in the CBD, to facilitate advanced soil modelling in both res earch and engineering practice - to simulate earthquake loading on proposed foundation design solutions including expensive ground improvement treatments. This has involved the use of a new Gel Push (GP) sampler to obtain undisturbed samples from below the ground-water table. Due to the variable nature of fluvial deposition, samples with a wide range of soil gradations, and accordingly soil index properties, were obtained from the sampling sites. The quality of the samples is comprehensively examined using available data from the ground investigation and laboratory testing. A meta-quality assessment was considered whereby a each method of evaluation contributed to the final quality index assigned to the specimen. The sampling sites were characterised with available geotechnical field-based test data, primarily the Cone Penetrometer Test (CPT), supported by borehole sampling and shear-wave velocity testing. This characterisation provides a geo- logical context to the sampling sites and samples obtained for element testing. It also facilitated the evaluation of sample quality. The sampling sites were evaluated for liquefaction hazard using the industry standard empirical procedures, and showed good correlation to observations made following the 22 February 2011 earthquake. However, the empirical method over-predicted liquefaction occurrence during the preceding 4 September 2010 event, and under-predicted for the subsequent 13 June 2011 event. The reasons for these discrepancies are discussed. The response of the GP samples to monotonic and cyclic loading was measured in the laboratory through triaxial testing at the University of Canterbury geomechanics laboratory. The undisturbed samples were compared to reconstituted specimens formed in the lab in an attempt to quantify the effect of fabric and structure in the Christchurch sands. Further testing of moist tamped re- constituted specimens (MT) was conducted to define important state parameters and state-dependent properties including the Critical State Line (CSL), and the stress-strain curve for varying state index. To account for the wide-ranging soil gradations, selected representative specimens were used to define four distinct CSL. The input parameters for the Stress-Density Model (S-D) were derived from a suite of tests performed on each representative soil, and with reference to available GP sample data. The results of testing were scrutinised by comparing the data against expected trends. The influence of fabric and structure of the GP samples was observed to result in similar cyclic strength curves at 5 % Double Amplitude (DA) strain criteria, however on close inspection of the test data, clear differences emerged. The natural samples exhibited higher compressibility during initial loading cycles, but thereafter typically exhibited steady growth of plastic strain and excess pore water pressure towards and beyond the strain criteria and initial liquefaction, and no flow was observed. By contrast the reconstituted specimens exhibited a stiffer response during initial loading cycles, but exponential growth in strains and associated excess pore water pressure beyond phase-transformation, and particularly after initial liquefaction where large strains were mobilised in subsequent cycles. These behavioural differences were not well characterised by the cyclic strength curve at 5 % DA strain level, which showed a similar strength for both GP samples and MT specimens. A preliminary calibration of the S-D model for a range of soil gradations is derived from the suite of laboratory test data. Issues encountered include the influence of natural structure on the peak-strength–state index relationship, resulting in much higher peak strengths than typically observed for sands in the literature. For the S-D model this resulted in excessive stiffness to be modelled during cyclic mobility, when the state index becomes large momentarily, causing strain development to halt. This behaviour prevented modelling the observed re- sponse of silty sands to large strains, synonymous with “liquefaction”. Efforts to reduce this effect within the current formulation are proposed as well as future research to address this issue.
Following the 22nd February 2011, Mw 6.2 earthquake located along a previously unknown fault beneath the Port Hills of Christchurch, surface cracking was identified in contour parallel locations within fill material at Quarry Road on the lower slopes of Mount Pleasant. GNS Science, in the role of advisor to the Christchurch City Council, concluded that these cracks were a part of a potential rotational mass movement (named zone 11A) within the fill and airfall loess material present. However, a lack of field evidence for slope instability and an absence of laboratory geotechnical data on which slope stability analysis was based, suggested this conclusion is potentially incorrect. It was hypothesised that ground cracking was in fact due to earthquake shaking, and not mass movement within the slope, thus forming the basis of this study. Three soil units were identified during surface and subsurface investigations at Quarry Road: fill derived from quarry operations in the adjacent St. Andrews Quarry (between 1893 and 1913), a buried topsoil, and underlying in-situ airfall loess. The fill material was identified by the presence of organic-rich topsoil “clods” that were irregular in both size (∼10 – 200 mm) and shape, with variable thicknesses of 1 – 10 m. Maximum thickness, as indicated by drill holes and geophysical survey lines, was identified below 6 Quarry Road and 7 The Brae where it is thought to infill a pre-existing gully formed in the underlying airfall loess. Bearing strength of the fill consistently exceeded 300 kPa ultimate below ∼500 mm depth. The buried topsoil was 200 – 300 mm thick, and normally displayed a lower bearing strength when encountered, but not below 300 kPa ultimate (3 – 11 blows per 100mm or ≥100 kPa allowable). In-situ airfall loess stood vertically in outcrop due to its characteristic high dry strength and also showed Scala penetrometer values of 6 – 20+ blows per 100 mm (450 – ≥1000 kPa ultimate). All soils were described as being moist to dry during subsurface investigations, with no groundwater table identified during any investigation into volcanic bedrock. In-situ moisture contents were established using bulk disturbed samples from hand augers and test pitting. Average moisture contents were low at 9% within the fill, 11 % within the buried topsoil, and 8% within the airfall loess: all were below the associated average plastic limit of 17, 15, and 16, respectively, determined during Atterberg limit analysis. Particle size distributions, identified using the sieve and pipette method, were similar between the three soil units with 11 – 20 % clay, 62 – 78 % silt, and 11 – 20 % fine sand. Using these results and the NZGS soil classification, the loess derived fill and in-situ airfall loess are termed SILT with some clay and sand, and the buried topsoil is SILT with minor clay and sand. Dispersivity of the units was found using the Emerson crumb test, which established that the fill can be non- to completely dispersive (score 0 – 4). The buried topsoil was always non-dispersive (score 0), and airfall loess completely dispersive (score 4). Values for cohesion (c) and internal friction angle (φ) of the three soil units were established using the direct shear box at field moisture contents. Results showed all soil units had high shear strengths at the moisture contents tested (c = 18 – 24 kPa and φ = 42 – 50°), with samples behaving in a brittle fashion. Moisture content was artificially increased to 16% within the buried topsoil, which reduced the shear strength (c = 10 kPa, φ = 18°) and allowed it to behave plastically. Observational information indicating stability at Quarry Road included: shallow, discontinuous, cracks that do not display vertical offset; no scarp features or compressional zones typical of landsliding; no tilted or deformed structures; no movement in inclinometers; no basal shear zone identified in logged core to 20 m depth; low field moisture contents; no groundwater table; and high soil strength using Scala penetrometers. Limit equilibrium analysis of the slope was conducted using Rocscience software Slide 5.0 to verify the slope stability identified by observational methods. Friction, cohesion, and density values determined during laboratory were input into the two slope models investigated. Results gave minimum static factor of safety values for translational (along buried topsoil) and rotational (in the fill) slides of 2.4 – 4.2. Sensitivity of the slope to reduced shear strength parameters was analysed using c = 10 kPa and φ = 18° for the translational buried topsoil plane, and a cohesion of 0 kPa within the fill for the rotational plane. The only situation that gave a factor of safety <1.0 was in nonengineered fill at 0.5 m depth. Pseudostatic analysis based on previous peak ground acceleration (PGA) values for the Canterbury Earthquake Sequence, and predicted PGAs for future Alpine Fault and Hope Fault earthquakes established minimum factor of safety values between 1.2 and 3.3. Yield acceleration PGAs were computed to be between 0.8g and 1.6g. Based on all information gathered, the cracking at Quarry Road is considered to be shallow deformation in response to earthquake shaking, and not due to deep-seated landsliding. It is recommended that the currently bare site be managed by smoothing the land, installing contour drainage, and bioremediation of the surface soils to reduce surface water infiltration and runoff. Extensive earthworks, including removal of the fill, are considered unnecessary. Any future replacement of housing would be subject to site-specific investigations, and careful foundation design based on those results.
The Canterbury Earthquake Sequence (CES) of 2010-2011 produced large seismic moments up to Mw 7.1. These large, near-to-surface (<15 km) ruptures triggered >6,000 rockfall boulders on the Port Hills of Christchurch, many of which impacted houses and affected the livelihoods of people within the impacted area. From these disastrous and unpredicted natural events a need arose to be able to assess the areas affected by rockfall events in the future, where it is known that a rockfall is possible from a specific source outcrop but the potential boulder runout and dynamics are not understood. The distribution of rockfall deposits is largely constrained by the physical properties and processes of the boulder and its motion such as block density, shape and size, block velocity, bounce height, impact and rebound angle, as well as the properties of the substrate. Numerical rockfall models go some way to accounting for all the complex factors in an algorithm, commonly parameterised in a user interface where site-specific effects can be calibrated. Calibration of these algorithms requires thorough field checks and often experimental practises. The purpose of this project, which began immediately following the most destructive rupture of the CES (February 22, 2011), is to collate data to characterise boulder falls, and to use this information, supplemented by a set of anthropogenic boulder fall data, to perform an in-depth calibration of the three-dimensional numerical rockfall model RAMMS::Rockfall. The thesis covers the following topics: • Use of field data to calibrate RAMMS. Boulder impact trails in the loess-colluvium soils at Rapaki Bay have been used to estimate ranges of boulder velocities and bounce heights. RAMMS results replicate field data closely; it is concluded that the model is appropriate for analysing the earthquake-triggered boulder trails at Rapaki Bay, and that it can be usefully applied to rockfall trajectory and hazard assessment at this and similar sites elsewhere. • Detailed analysis of dynamic rockfall processes, interpreted from recorded boulder rolling experiments, and compared to RAMMS simulated results at the same site. Recorded rotational and translational velocities of a particular boulder show that the boulder behaves logically and dynamically on impact with different substrate types. Simulations show that seasonal changes in soil moisture alter rockfall dynamics and runout predictions within RAMMS, and adjustments are made to the calibration to reflect this; suggesting that in hazard analysis a rockfall model should be calibrated to dry rather than wet soil conditions to anticipate the most serious outcome. • Verifying the model calibration for a separate site on the Port Hills. The results of the RAMMS simulations show the effectiveness of calibration against a real data set, as well as the effectiveness of vegetation as a rockfall barrier/retardant. The results of simulations are compared using hazard maps, where the maximum runouts match well the mapped CES fallen boulder maximum runouts. The results of the simulations in terms of frequency distribution of deposit locations on the slope are also compared with those of the CES data, using the shadow angle tool to apportion slope zones. These results also replicate real field data well. Results show that a maximum runout envelope can be mapped, as well as frequency distribution of deposited boulders for hazard (and thus risk) analysis purposes. The accuracy of the rockfall runout envelope and frequency distribution can be improved by comprehensive vegetation and substrate mapping. The topics above define the scope of the project, limiting the focus to rockfall processes on the Port Hills, and implications for model calibration for the wider scientific community. The results provide a useful rockfall analysis methodology with a defensible and replicable calibration process, that has the potential to be applied to other lithologies and substrates. Its applications include a method of analysis for the selection and positioning of rockfall countermeasure design; site safety assessment for scaling and demolition works; and risk analysis and land planning for future construction in Christchurch.
This dissertation addresses a diverse range of applied aspects in ground motion simulation validation via the response of complex structures. In particular, the following topics are addressed: (i) the investigation of similarity between recorded and simulated ground motions using code-based 3D irregular structural response analysis, (ii) the development of a framework for ground motion simulations validation to identify the cause of differences between paired observed and simulated dataset, and (iii) the illustration of the process of using simulations for seismic performance-based assessment. The application of simulated ground motions is evaluated for utilisation in engineering practice by considering responses of 3D irregular structures. Validation is performed in a code-based context when the NZS1170.5 (NZS1170.5:2004, 2004) provisions are followed for response history analysis. Two real buildings designed by engineers and physically constructed in Christchurch before the 2010-2011 Canterbury earthquake sequence are considered. The responses are compared when the buildings are subjected to 40 scaled recorded and their subsequent simulated ground motions selected from 22 February 2011 Christchurch. The similarity of recorded and simulated responses is examined using statistical methods such as bootstrapping and hypothesis testing to determine whether the differences are statistically significant. The findings demonstrate the applicability of simulated ground motion when the code-based approach is followed in response history analysis. A conceptual framework is developed to link the differences between the structural response subjected to simulated and recorded ground motions to the differences in their corresponding intensity measures. This framework allows the variability to be partitioned into the proportion that can be “explained” by the differences in ground motion intensity measures and the remaining “unexplained” variability that can be attributed to different complexities such as dynamic phasing of multi-mode response, nonlinearity, and torsion. The application of this framework is examined through a hierarchy of structures reflecting a range of complexity from single-degree-of-freedom to 3D multi-degree-of-freedom systems with different materials, dynamic properties, and structural systems. The study results suggest the areas that ground motion simulation should focus on to improve simulations by prioritising the ground motion intensity measures that most clearly account for the discrepancies in simple to complex structural responses. Three approaches are presented to consider recorded or simulated ground motions within the seismic performance-based assessment framework. Considering the applications of ground motions in hazard and response history analyses, different pathways in utilising ground motions in both areas are explored. Recorded ground motions are drawn from a global database (i.e., NGA-West2 Ancheta et al., 2014). The NZ CyberShake dataset is used to obtain simulations. Advanced ground motion selection techniques (i.e., generalized conditional intensity measure, GCIM) are used for ground motion selection at a few intensity levels. The comparison is performed by investigating the response of an example structure (i.e., 12-storey reinforced concrete special moment frame) located in South Island, NZ. Results are compared and contrasted in terms of hazard, groundmotion selection, structural responses, demand hazard, and collapse risk, then, the probable reasons for differences are discussed. The findings from this study highlight the present opportunities and shortcomings in using simulations in risk assessment. i
