B5 · Publication Volume 10

Age Uncertainty and Geological Interpretation

analytical age, geological events and discordance

Learning goals

After this lesson, you should be able to distinguish an analysis, a calculated date, an age population and a geological event; identify random, correlated, calibration and geological uncertainty; evaluate whether a weighted mean is justified; and interpret discordance or excess scatter without automatic rejection. You should be able to write an age statement that names material, domain, method, uncertainty coverage and event meaning.

The lesson treats uncertainty as information about the evidence chain, not a decorative plus-minus value. A smaller analytical uncertainty improves discrimination only when sampling, correction and event assignment are valid. Geological interpretation can remain ambiguous even when a ratio is measured precisely.

From measured domain to event claim

Hierarchy from measured isotope ratios to analytical dates, domain populations and geological event interpretation
Hierarchy from measured isotope ratios to analytical dates, domain populations and geological event interpretation

An analysis is a measured solution, spot, increment or domain. Data reduction produces isotope ratios and a date under constants and corrections. Several dates may define a statistically and geologically coherent population. An age is the time assigned to a geological boundary such as crystallisation, growth, cooling, deposition or resetting. An event may span time and contain several dated phases.

These levels must not be collapsed. Ten spot dates from one overgrowth are not ten independent geological events. Conversely, ten spots from cores, rims and fractures are not one population merely because they belong to the same mineral species. Textures, chemistry and relative-age relations define plausible groups before a mean is interpreted.

A complete statement resembles: “Domains identified petrographically as phase-X rims yield a calculated population date under the stated system and corrections; this is interpreted as the timing of rim growth because the rims overgrow the earlier fabric and are cut by the later vein.” Every link can be reviewed separately.

Uncertainty layers and covariance

Within-analysis uncertainty includes signal counting, baseline, fractionation and correction propagation. Between-analysis random uncertainty includes repeatability and local heterogeneity. Calibration uncertainty arises from tracers, monitors, reference values and instrument bias. Decay-constant uncertainty affects conversion from ratio to time. Geological uncertainty includes mixed domains, initial components, loss, gain, inheritance and event duration.

Some terms are independent; others are shared. A calibration value can shift every date in a session together, so it should not be reduced by averaging more analyses. Common-component correction can correlate two ratios within a spot. Ignoring covariance changes regression slope and uncertainty. A reported internal uncertainty may be appropriate for comparing analyses reduced together but inadequate for comparison with an independently calibrated timescale.

State coverage: one standard uncertainty, two standard deviations, expanded uncertainty or a confidence interval under a named model. State included components. Rounding should follow the uncertainty; an age of 100.1234 ± 1.7 Ma advertises meaningless digits.

Uncertainty in event assignment is not always expressible as a symmetric interval. Two alternative events may be separated by hundreds of millions of years even though each date has a small analytical error. Report model alternatives explicitly.

Weighted means, regression and populations

For independent dates t_i with standard uncertainties \sigma_i, an inverse-variance weighted mean is

$\bar t=\frac{\sum t_i/\sigma_i^2}{\sum1/\sigma_i^2}.$

Its formal uncertainty assumes the dates estimate one common value, uncertainties are correctly specified and independence or covariance is handled. The mean square of weighted deviations can test excess scatter, but a statistic does not decide geological grouping. Overdispersion may be real age variation, mixed domains, unmodelled correlation or underestimated error.

Regression is required when age information lies in a slope or intercept. Both axes may have uncertainty and correlation. Ordinary least squares that treats one ratio as exact is often inappropriate. Inspect residuals, leverage and uncertainty ellipses. A high correlation coefficient does not prove an isochron; mixing can create a line.

Mixture models can explore populations, but the number of components and priors influence results. Use textures and chemistry to constrain components. A histogram bin or probability peak is not a geological event until it maps to domains and sequence.

Discordance, inheritance and resetting

Discordance contains direction and geometry. Daughter loss, parent gain or loss, inheritance, mixed ablation depths, common-component correction and analytical interference predict different movements in isotope-ratio space. Evaluate these mechanisms before assigning a generic “disturbed” label.

Inherited cores can be scientifically important source evidence. A younger rim may date crystallisation while an older core constrains provenance. Partial resetting can create age gradients correlated with grain size, cracks or alteration. Complete resetting may yield a coherent younger population. Each is a different geological outcome.

Rejection rules should be defined before selecting a preferred result where possible and must be recorded with rejected data. Criteria may include documented contamination, failed quality control, mixed interaction volume or a predeclared discordance threshold. A point should not be rejected solely because it worsens the mean. Conversely, retaining every analysis in one mean is not transparency when the domains are demonstrably different.

Data presentation and reporting design

Display individual analyses with uncertainty and domain metadata. Appropriate plots may include concordance relations, isochrons, age-versus-position profiles, ranked dates, probability or density views and petrographic maps. No one plot is sufficient. Avoid graphics that hide rejected points or truncate axes to exaggerate agreement.

Tables should include sample and domain identifier, material, measured or corrected ratios, date, uncertainty coverage, correlation terms where required, common-component or disequilibrium treatment, quality-control status and rejection reason. Preserve machine-readable full precision separately from rounded publication values.

The narrative should separate three paragraphs: analytical result and data quality; population or model result and statistical fit; geological interpretation and alternatives. Comparisons with external timescales or other systems must include relevant systematic uncertainty and constant conventions.

Worked synthetic example

Three invented mineral domains yield dates of 100.0\pm1.0, 101.0\pm1.0 and 109.0\pm1.0 Ma, where each uncertainty is one standard deviation and treated as independent for illustration. Equal uncertainties give an all-data weighted mean of 103.33 Ma. The weighted sum of squared residuals is

$(-3.33)^2+(-2.33)^2+(5.67)^2=48.67.$

With two degrees of freedom, the mean square of weighted deviations is about 24.3, far above the single-population expectation under this simple model. Quoting 103.33\pm0.58 Ma as one precise age would be misleading.

The first two dates have mean 100.5 Ma and a mean square of weighted deviations of 0.5. That may support one population if the domains are petrographically equivalent. The 109 Ma date should not be discarded automatically. If it is an inherited core, it records an older component; if it is a rim with failed correction, the analytical model needs revision; if all three are texturally identical, geological scatter or underestimated uncertainty remains.

Now add a shared calibration uncertainty of 0.8%. It shifts comparisons with an independent timescale but does not shrink by averaging the two coherent spots. Report internal and full comparison uncertainty separately rather than combining them without explanation.

Review workflow and completion exercise

  1. Define the geological event and domain criteria before statistical pooling.
  2. Audit raw ratios, corrections, constants, reference materials and covariance.
  3. Plot all analyses with textures and quality-control status.
  4. Test concordance, regression or population fit appropriate to the system.
  5. Investigate excess scatter through analytical and geological hypotheses.
  6. Apply documented rejection rules and retain rejected records.
  7. Calculate internal and systematic uncertainty for the intended comparison.
  8. Write date, population and event statements separately.
  9. Identify the observation that would change the preferred interpretation.

For completion, review a synthetic set of twelve dated domains containing one inherited population, one growth population, two altered spots and a shared calibration term. Produce a domain map, a transparent inclusion table, appropriate regression or population statistics and two competing event models. Do not force one mean when the evidence supports several histories.

Review questions:

  • When is a weighted mean physically meaningful?
  • Which uncertainty components remain correlated across analyses?
  • What evidence distinguishes inheritance from analytical contamination?
  • Why can a precise date have a weak event interpretation?
  • How should rejected analyses remain visible to a reviewer?

Sources and further reading