Can we develop a more targeted approach to mitigating seismic risk?

The study addresses how to target seismic risk mitigation more efficiently by identifying regional high-risk zones where potential earthquake damage is most severe. Recent catastrophic earthquakes and rapid urbanization have increased exposure and structural vulnerability, stressing the importance of pre-disaster measures amidst budget constraints. The research question asks whether a robust, fine-resolution delineation of high-risk zones can be achieved despite uncertainties—particularly in seismic hazard—so that limited resources can be prioritized to maximize loss reduction and resilience.

Since the 1960s, probabilistic seismic hazard maps guide new building codes, yet systematic strategies for upgrading existing building stock are less developed. Prior work shows pre-disaster mitigation is cost-effective versus post-event response. In seismic risk modelling, uncertainty arises from hazard, exposure, and vulnerability, with multiple studies (e.g., Cologne, Istanbul, Thessaloniki) indicating hazard variability typically dominates total risk uncertainty. Empirical GMPEs are widely used for ground motion prediction but assume ergodicity, handle spatial correlation imperfectly, and are weakly constrained near sources due to sparse data. Physics-based simulations (PBS) directly model rupture and wave propagation, capture 3D complexities and spatial correlation, and provide higher-resolution heterogeneity, but depend on robust input models (fault geometry, media, friction) and computational limits. Integrative efforts (e.g., CyberShake, nonergodic GMPE development) aim to improve hazard estimates. A remaining gap is how to combine different risk model outputs to consistently identify high-risk zones for prioritizing mitigation.

Case study and scenarios: Four Mw 7.5 deterministic earthquake scenarios were defined on the Jiaocheng fault (Shanxi Rift System, China), differing by nucleation location but sharing hypocentral depth (~8.66 km) and nucleation patch size (radius 2 km). The Jiaocheng fault is a normal fault (dip 60°) near Taiyuan.

Hazard modelling:

• Physics-based simulations (PBS): Dynamic rupture and 3D wave propagation were computed via a curved-grid finite-difference method (CG-FDM), validated in dynamic rupture code verification exercises. PGA maps at 250 m grid spacing were generated for each scenario.
• Empirical GMPEs: Median PGA maps (250 m) were produced using four NGA-West models—BA08, CB08, BSSA14, CB14—considering appropriate magnitude, distance, and model terms (e.g., style-of-faulting, hanging wall, depth, dip, attenuation in CB14). GMPE-based PGA spatial patterns are scenario-invariant, while PBS-based maps vary with nucleation, offering enhanced spatial heterogeneity.

Site amplification rectification: Detailed Vs30 was unavailable. The study adopted site classes compiled for the Shanxi Rift System (based on 3,106 engineering boreholes, 31 strong-motion, 65 seismic stations) and applied site adjustment coefficients from GB 18306-2015 to rectify both PBS- and GMPE-based PGA maps from their reference site conditions to mapped site classes.

Exposure model: A 1 km × 1 km residential building stock model for mainland China was downscaled to the study area and updated with localized 2022 unit construction prices from the Shanxi provincial housing authority for 17 residential subtypes (by structure and storey class). Total residential replacement value in the study area is approximately 911 billion RMB.

Vulnerability model: Because building-level code information was not available in the stock model, a separate building-by-building survey of downtown Taiyuan (52,739 extant buildings) provided distributions of construction periods by structure and storey class. Construction periods were aligned with national seismic zonation map issuance years (1977, 1990, 2001) to infer seismic code levels (pre-, low-, moderate-, high-code) for vulnerability assignment. Vulnerability curves were selected for three representative structural classes (brick-wood/other, mixed masonry, steel-RC) stratified by code level.

Loss estimation and validation: For each hazard input (PBS and each GMPE), rectified PGA grids were combined with the exposure and vulnerability models to compute residential direct financial losses. Modeled losses were compared against losses estimated using an empirical intensity-based loss ratio model calibrated from historical Chinese earthquake damage (reference model), via intensity-converted maps, to evaluate consistency across intensity bins and totals.

High-risk zone delineation: For each loss map, the top 10% of high-loss grids were extracted. Intersections were computed across PBS scenarios and across GMPE maps, and finally combined to a consensus set of intersected high-loss grids. Statistics on the area proportion and captured total loss were calculated. Administrative units containing these grids were identified. A retrofit scenario was also analyzed by assuming all buildings upgraded to high-code vulnerability, to estimate potential loss reduction by structure type.

• The consensus high seismic risk zone (intersection of top 10% loss grids across PBS scenarios and across GMPE models) comprises about 7% of the regional land area but concentrates approximately 85%–88% of the total modeled residential financial loss, indicating strong spatial concentration of risk and robustness to hazard model choice and nucleation location.
• Total residential exposure in the study area is about 911 billion RMB.
• Modeled loss distributions by intensity generally align with an empirical, intensity-based loss ratio model derived from historical Chinese earthquakes: within each intensity bin (IV–XI), loss-to-exposure ratios are comparable, and ratios of total loss between this study and the empirical model are on the order of ~1.3–1.5 depending on scenario/GMPE, indicating consistency in patterns with some systematic differences in magnitude.
• The delineated high-risk grids are concentrated around Taiyuan and neighboring jurisdictions; 30 counties/districts contain intersected high-risk grids, with the largest counts in Qingxu County (240), Wenshui County (227), Xiaodian District (218), Fenyang City (206), Xiaoyi City (159), and Jinyuan District (152), among others.
• Retrofitting analysis suggests substantial potential loss reductions if buildings are upgraded to high-code vulnerability levels, with reductions varying by structural class (brick-wood/other, mixed masonry, steel-RC).

By explicitly addressing the dominant source of risk model uncertainty—hazard—through parallel use of PBS and multiple GMPEs while holding exposure and vulnerability constant, the strategy robustly identifies a small set of locations where losses concentrate under severe but plausible scenarios. The high overlap of top-loss areas across disparate hazard methods and differing scenario nucleations demonstrates that the delineated priority zone is method-agnostic and scenario-resilient. This directly answers the research question by providing a fine-resolution, high-confidence map to target pre-disaster mitigation. The approach supports efficient allocation of limited resources (e.g., retrofitting, preparedness, response planning) and is particularly relevant for dense urban areas like Taiyuan within active fault systems. Consistency with empirical loss models lends further credibility, while the intensity-based analyses help interpret differences and guide code-based retrofit priorities.

The paper introduces and demonstrates a strategy to delineate regional high seismic risk zones at fine spatial resolution and with high confidence by intersecting high-loss areas derived from both physics-based and empirical hazard models combined with calibrated exposure and vulnerability. Applied to Mw 7.5 scenarios on the Jiaocheng fault, the consensus zone captures roughly 85% of total residential financial loss within only ~7% of land area, offering a practical target for prioritized mitigation in Taiyuan and neighboring regions. Future work will: (i) extend PBS bandwidth (e.g., via ANN-enhanced simulations) and consider broader magnitude ranges and nucleation positions; (ii) include multiple faults where relevant; (iii) expand exposure and vulnerability to non-residential buildings and infrastructure and use higher-resolution, building-level inventories as they become available; and (iv) incorporate additional risk metrics (casualties, indirect/economic business interruption) to support comprehensive risk management and intervention planning.

• Physics-based simulations are limited in frequency content (typically up to ~1.5 Hz), constrained by computational resources and absence of a high-resolution velocity model.
• Site amplification is rectified using an empirical site class map and code-based adjustment coefficients due to lack of detailed Vs30; this is the best available dataset but introduces uncertainty.
• Residential replacement values are modeled at 1 km × 1 km resolution rather than building-by-building, reflecting current data limitations.
• Only Mw ~7.5 scenarios on a single fault and residential buildings are considered; the delineated zone should be refined for operational use by including a wider magnitude range, additional building types, and potentially multiple faults.
• Scenarios have not occurred; modeled losses cannot be validated against observed event losses in the study area.
• A comprehensive risk assessment would also include social losses (fatalities, injuries) and indirect economic impacts, which were outside the present scope.