Per-event boxplots

How to cite

The question: how variable are the field measurements within each earthquake — and does the model's range cover them?

This is the inverse of Model vs reality. Instead of scattering individual field points over the simulated cloud, it summarises the spread of measurements event by event, with the distribution from the distinct element method (DEM) simulations alongside for comparison.

Unfamiliar terms?

Median, interquartile range, whiskers and log scale are defined under statistics; FDHI, SURE, vertical separation and principal rupture under the field counterparts and the datasets.

Every data panel is a box-and-whisker plot. The box spans the interquartile range, the line inside is the median, and the whiskers reach the most extreme measurement still within 1.5 × IQR of the box. That is the default in both Tableau and the plotting library used in the original analysis, which is what lets these boxes be compared with the paper's.²

On the field panels each box is one earthquake. On the DEM panels each box is one scarp class instead — that is how the model's range is broken down.² The six are Monoclinal, Pressure Ridge, Simple and a … Collapse variant of each, shown side by side in Figure 2. The two quantities compared here — width and height — are defined on the model geometry in Figure 5.

Two dashboards divide the material.

Model vs field

This dashboard stacks each DEM distribution above the comparable field measure, on a shared axis, for the two quantities the paper compares: deformation zone width (in the field data, fault zone width) and scarp height.

Open full-size on Tableau Public How to cite

Why the width axis is logarithmic

The width panels use a fixed logarithmic axis spanning 0.01–2,000 m, and a toggle on the dashboard decides how much of it is populated.

The paper applies a 50 m upper limit when selecting field measurements for this comparison. That limit is not a plotting choice: it "reflects the maximum of our DEM model bounds and seeks to exclude distributed deformation in natural events that may have occurred across multiple, widely spaced fault strands."³ In other words, the paper restricts the comparison to the regime the model was built to represent — comfortably inside the model's own envelope, a reported DZW range of 0–40.76 m across the experiments.⁵

By default, this page applies the same criterion.⁴ The Show widths > 50 m control on the dashboard adds the rest of the field record: measurements reaching 1,450 m, overwhelmingly from Kaikoura, whose 448 width measurements all lie beyond 50 m — from 50 m to 1,450 m, with a median near 250 m — so that event enters the panel only when the toggle is on.² The axis is fixed, and shared with the DEM panel above, so the model and field spreads stay directly comparable in either state; and only a log scale keeps sub-metre boxes readable on an axis that must also hold a 1,450 m whisker.

The two toggle states answer different questions

With the toggle off, you are reading the comparison as the paper frames it: the field's localised deformation, box against box with the model's range. Switched on, most of what appears beyond the 50 m mark is exactly the distributed, multi-strand deformation the paper deliberately set aside — not evidence that the model under-predicts. The wider view serves a different purpose: seeing how much of the observed record lies outside the modelled regime, and how far.

Vertical separation and SURE

The second dashboard carries measures that have no DEM counterpart on the same axis, so they stand alone: vertical separation (the vertical offset across the rupture, which the field compilation treats as comparable to scarp height⁶) and two measures from the SURE compilation — fault-normal component (the horizontal displacement measured perpendicular to the fault) and its own scarp height.

Coverage differs sharply between them. After the per-sheet filters described below, vertical separation is the richest panel at 2,106 measurements across 23 events, while SURE's fault-normal component covers 9 events (185 measurements) and its scarp height 4 events (74 measurements).²

An event map sits alongside the boxplots on this dashboard, locating the FDHI events, and the vertical-separation panel carries its own event filter so you can narrow the 23 events to a comparable subset.

Open full-size on Tableau Public How to cite

How each panel is filtered, and where the magnitudes come from

The FDHI panels keep only measurements the compilation marks as principal ruptures — the main fault trace, as opposed to distributed or secondary rupture — and only those with a positive value for the measure in question. The counts above are after those filters.² The SURE panels apply no row filter.

Event labels carry a moment magnitude. For the FDHI measurements that value travels with the data; for SURE it does not — the SURE release records no magnitude column, so those values come from a small lookup table curated inside this project, every entry of which is sourced from the SURE 2.0 data descriptor.⁷

Where this comes from

This is chart family 5 in the project's inventory, which maps it to the boxplot panels of the paper's Figure 13.¹ The measurements come from the FDHI flatfile and the SURE database, and the model context from the DEM experiment set — all described on the Data page.

Where to go next

• Model vs reality — the same comparison as a scatter of individual measurements rather than summaries.
• Response curves — what drives the model's own range in the first place.
• Slip regression — inferring the slip behind a measured displacement.
• Data — where the FDHI and SURE measurements come from.

How to cite

Please cite as:

Chiama, K., Bednarz, W., Moss, R., Plesch, A., and Shaw, J. H. (2025). "Quantifying relationships between fault parameters and rupture characteristics associated with thrust and reverse fault earthquakes." Earthquake Spectra, 41(5), 3977–4014. DOI: 10.1177/87552930251346434

If you use the underlying data itself, cite the archives it comes from as well — the DEM experiments are deposited open-access on DesignSafe-CI, and the field compilations carry their own citations. The full list, with DOIs, is under How to cite this data.

Any DOI issued for this site or its source code identifies the software and the website, and does not replace the citations above.

Related work, for the wider project's 3D models — not a source for anything shown here:

Chiama, K., Plesch, A., and Shaw, J. H. (2025). "Along-Strike Variability of Surface Deformation on Thrust and Reverse Fault Ruptures: Insights from 3D Distinct Element Method Models." Seismological Research Letters 96(6), 3473–3489. DOI: 10.1785/0220250173

---

1. notes/chart-families.md in the source repository. ↩

2. notes/dashboard-3-build-spec.md in the source repository, which records the per-sheet filters and populations, the axis decisions, and the history of the width-limit default. ↩↩↩↩↩

3. Chiama et al. (2025), section comparing DEM results with the FDHI dataset. ↩

4. The restricted view is the default so that the comparison shown first is the one the paper makes. The toggle exists because the measurements beyond 50 m are still informative — they show how far real surface deformation can extend once it is no longer confined to a single fault strand. ↩

5. Chiama et al. (2025): "The DZW has a wide range across all the experiments (0–40.76 m)." The shipped export reaches ~45.8 m, from a handful of heterogeneous rows at very low slip — the regime the paper excludes as high-uncertainty. The small discrepancy between the two numbers is unresolved, and immaterial to anything shown here. ↩

6. Chiama et al. (2025), which assumes measured vertical separation is "similar enough to the scarp heights to foster these comparisons", citing the FDHI report in support. ↩

7. SURE_EVENT_MAGNITUDES in subprojects/python/src/eps_ground_rupture/config.py, with every value confirmed against the SURE 2.0 data descriptor — Nurminen, F., et al. (2022), "SURE 2.0 — new release of the worldwide database of surface ruptures for fault displacement hazard analyses," Scientific Data 9, DOI 10.1038/s41597-022-01835-z. ↩