An official website of the United States government
The abundant post-earthquake data from the Canterbury, New Zealand (NZ) area is poised for use with machine learning (ML) to further advance our ability to better predict and understand the effects of liquefaction. Liquefaction manifestation is one of the identifiable effects of liquefaction, a nonlinear phenomenon that is still not well understood. ML algorithms are often termed as “black-box” models that have little to no explainability for the resultant predictions, making them difficult for use in practice. With the SHapley Additive exPlanations (SHAP) algorithm wrapper, mathematically backed explanations can be fit to the model to track input feature influences on the final prediction. In this paper, Random Forest (RF) is chosen as the ML model to be utilized as it is a powerful non-parametric classification model, then SHAP is applied to calculate explanations for the predictions at a global and local feature scale. The RF model hyperparameters are optimized with a two-step grid search and a five-fold cross-validation to avoid overfitting. The overall model accuracy is 71% over six ordinal categories predicting the Canterbury Earthquake Sequence measurements from 2010, 2011, and 2016. Insights from the SHAP application onto the RF model include the influences of PGA, GWT depths, and SBTs for each ordinal class prediction. This preliminary exploration using SHAP can pave the way for both reinforcing the performance of current ML models by comparing to previous knowledge and using it as a discovery tool for identifying which research areas are pertinent to unlocking more understanding of liquefaction mechanics.
more »
« less
Earthquake-Induced Liquefaction Manifestation Multiclass Prediction Utilizing Random Forests for the Canterbury Earthquake Sequence
The abundance of post-earthquake data from the Canterbury, New Zealand (NZ), area can be leveraged for exploring machine learning (ML) opportunities for geotechnical earthquake engineering. Herein, random forest (RF) is chosen as the ML model to be utilized as it is a powerful non-parametric classification model that can also calculate global feature importance post-model building. The results and procedure are presented of building a multiclass liquefaction manifestation classification RF model with features engineered to preserve special relationships. The RF model hyperparameters are optimized with a two-step fivefold crossvalidation grid search to avoid overfitting. The overall model accuracy is 96% over six ordinal categories predicting over the Canterbury earthquake sequence measurements from 2010, 2011, and 2016. The resultant RF model can serve as a blueprint for incorporation of other sources of physical data such as geological maps to widen the bounds of model usability.
more »
« less
- Award ID(s):
- 2047838
- PAR ID:
- 10553632
- Publisher / Repository:
- American Society of Civil Engineers
- Date Published:
- ISBN:
- 9780784485316
- Page Range / eLocation ID:
- 222 to 231
- Format(s):
- Medium: X
- Location:
- Vancouver, British Columbia, Canada
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Remote reconnaissance missions are promising solutions for the assessment of earthquake-induced structural damage and cascading geological hazards. Space-borne remote sensing can complement in-field missions when safety and accessibility concerns limit post-earthquake operations on the ground. However, the implementation of remote sensing techniques in post-disaster missions is limited by the lack of methods that combine different techniques and integrate them with field survey data. This paper presents a new approach for rapid post-earthquake building damage assessment and landslide mapping, based on Synthetic Aperture Radar (SAR) data. The proposed texture-based building damage classification approach exploits very high resolution post-earthquake SAR data integrated with building survey data. For landslide mapping, a backscatter intensity-based landslide detection approach, which also includes the separation between landslides and flooded areas, is combined with optical-based manual inventories. The approach was implemented during the joint Structural Extreme Event Reconnaissance, GeoHazards International and Earthquake Engineering Field Investigation Team mission that followed the 2021 Haiti Earthquake and Tropical Cyclone Grace.more » « less
-
On 14th August 2021, a magnitude 7.2 earthquake struck the Tiburon Peninsula in the Caribbean nation of Haiti, approximately 150 km west of the capital Port-au-Prince. Aftershocks up to moment magnitude 5.7 followed and over 1,000 landslides were triggered. These events led to over 2,000 fatalities, 15,000 injuries and more than 137,000 structural failures. The economic impact is of the order of US$1.6 billion. The on-going Covid pandemic and a complex political and security situation in Haiti meant that deploying earthquake engineers from the UK to assess structural damage and identify lessons for future building construction was impractical. Instead, the Earthquake Engineering Field Investigation Team (EEFIT) carried out a hybrid mission, modelled on the previous EEFIT Aegean Mission of 2020. The objectives were: to use open-source information, particularly remote sensing data such as InSAR and Optical/Multispectral imagery, to characterise the earthquake and associated hazards; to understand the observed strong ground motions and compare these to existing seismic codes; to undertake remote structural damage assessments, and to evaluate the applicability of the techniques used for future post-disaster assessments. Remote structural damage assessments were conducted in collaboration with the Structural Extreme Events Reconnaissance (StEER) team, who mobilised a group of local non-experts to rapidly record building damage. The EEFIT team undertook damage assessment for over 2,000 buildings comprising schools, hospitals, churches and housing to investigate the impact of the earthquake on building typologies in Haiti. This paper summarises the mission setup and findings, and discusses the benefits, and difficulties, encountered during this hybrid reconnaissance mission.more » « less
-
On February 6, 2023, a major earthquake of 7.8 magnitude and its aftershocks caused widespread destruction in Turkey and Syria, causing more than 55,000 deaths, displacing 3 million people in Turkey and 2.9 million in Syria, and destroying or damaging at least 230,000 buildings. Our research presents detailed city-scale maps of landslides, liquefaction, and building damage from this earthquake, utilizing a novel variational causal Bayesian network. This network integrates InSAR-derived change detection with new empirical ground failure models and building footprints, enabling us to (1) rapidly estimate large-scale building damage, landslides, and liquefaction from remote sensing data, (2) jointly attribute building damage to landslides, liquefaction, and shaking, (3) improve regional landslide and liquefaction predictions impacting infrastructure, and (4) simultaneously identify damage degrees in thousands of buildings. For city-scale, building-by-building damage assessments, we use building footprints and satellite imagery with a spatial resolution of approximately 30 meters. This allows us to achieve a high resolution in damage assessment, both in timeliness and scale, enabling damage classification at the individual building level within days of the earthquake. Our findings detail the extent of building damage, including collapses, in Hatay, Osmaniye, Adıyaman, Gaziantep, and Kahramanmaras. We classified building damages into five categories: no damage, slight, moderate, partial collapse, and collapse. We evaluated damage estimates against preliminary ground-truth data reported by the civil authorities. Our results demonstrate the accuracy of our classification system, as evidenced by the area under the curve (AUC) scores on the receiver operating characteristic (ROC) curve, which ranged from 0.9588 to 0.9931 across different damage categories and regions. Specifically, our model achieved an AUC of 0.9931 for collapsed buildings in the Hatay/Osmaniye area, indicating a 99.31% probability that the model will rank a randomly chosen collapsed building higher than a randomly chosen non-collapsed building. These accurate, building-specific damage estimates, with greater than 95% classification accuracy across all categories, are crucial for disaster response and can aid agencies in effectively allocating resources and coordinating efforts during disaster recovery.more » « less
-
Tens of thousands of landslides were generated over 10,000 km2 of North Canterbury and Marlborough as a consequence of the 14 November 2016, Mw7.8 Kaikōura Earthquake. The most intense landslide damage was concentrated in 3500 km2 around the areas of fault rupture. Given the sparsely populated area affected by landslides, only a few homes were impacted and there were no recorded deaths due to landslides. Landslides caused major disruption with all road and rail links with Kaikōura being severed. The landslides affecting State Highway 1 (the main road link in the South Island of New Zealand) and the South Island main trunk railway extended from Ward in Marlborough all the way to the south of Oaro in North Canterbury. The majority of landslides occurred in two geological and geotechnically distinct materials reflective of the dominant rock types in the affected area. In the Neogene sedimentary rocks (sandstones, limestones and siltstones) of the Hurunui District, North Canterbury and around Cape Campbell in Marlborough, first-time and reactivated rock-slides and rock-block slides were the dominant landslide type. These rocks also tend to have rock material strength values in the range of 5-20 MPa. In the Torlesse ‘basement’ rocks (greywacke sandstones and argillite) of the Kaikōura Ranges, first-time rock and debris avalanches were the dominant landslide type. These rocks tend to have material strength values in the range of 20-50 MPa. A feature of this earthquake is the large number (more than 200) of valley blocking landslides it generated. This was partly due to the steep and confined slopes in the area and the widely distributed strong ground shaking. The largest landslide dam has an approximate volume of 12(±2) M m3 and the debris from this travelled about 2.7 km2 downslope where it formed a dam blocking the Hapuku River. The long-term stability of cracked slopes and landslide dams from future strong earthquakes and large rainstorms are an ongoing concern to central and local government agencies responsible for rebuilding homes and infrastructure. A particular concern is the potential for debris floods to affect downstream assets and infrastructure should some of the landslide dams breach catastrophically. At least twenty-one faults ruptured to the ground surface or sea floor, with these surface ruptures extending from the Emu Plain in North Canterbury to offshore of Cape Campbell in Marlborough. The mapped landslide distribution reflects the complexity of the earthquake rupture. Landslides are distributed across a broad area of intense ground shaking reflective of the elongate area affected by fault rupture, and are not clustered around the earthquake epicentre. The largest landslides triggered by the earthquake are located either on or adjacent to faults that ruptured to the ground surface. Surface faults may provide a plane of weakness or hydrological discontinuity and adversely oriented surface faults may be indicative of the location of future large landslides. Their location appears to have a strong structural geological control. Initial results from our landslide investigations suggest predictive models relying only on ground-shaking estimates underestimate the number and size of the largest landslides that occurred.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1061/9780784485316.024
