The Gold Standard of Volcanology | Comparing Event Rates When Magnitude Thresholds Change

An event catalogue is a constructed record of identified events. Its entries depend on observations, processing, association, location and the rules used to include a result. A physical event and a database row are related, but they are not interchangeable definitions.

For this lesson, one row represents one distinct event in a fixed fictional region. Duplicate detections from several instruments have already been reconciled. The magnitude convention and event classification are held comparable between the windows unless a chapter explicitly changes them.

Those assumptions prevent several accidental changes of denominator. Counting station detections would give a different unit from counting distinct earthquakes. Including a larger region would change the eligible population. Including a new event class could change the total without changing the rate of the original class.

Write a short inclusion sentence before calculating: distinct events of this stated kind, in this region, during this window, above this magnitude threshold. If the sentence cannot be completed from the supplied record, identify the missing field. Accurate division cannot repair an undefined event population.

Contents · Next chapter

A count names how many qualifying events appear in a window. An observed rate divides that count by the relevant duration. Two windows of unequal length need that time basis before their totals can be compared as rates.

If eight events appear in two hours and twelve in six hours, the observed rates are four and two events per hour. The second count is larger while its rate is smaller. The statements do not conflict because their denominators differ.

For the main fictional ledger, observation is continuous throughout each stated window. That is a stronger assumption than simply having timestamps at the start and end. Later we will examine incomplete coverage.

An observed rate remains a description of a selected record. It is not automatically a long-term mean, a future rate or a parameter known without sampling uncertainty. Keep the word observed when the result comes directly from a finite count. That small word helps prevent a descriptive calculation from becoming a stronger scientific assertion than the data support.

Contents · Previous chapter · Next chapter

Window A lasts eight hours. The supplied catalogue is complete under the exercise’s ideal rule for events of magnitude two and above. Window B lasts four hours and is complete for magnitude one and above.

Within A, forty events have magnitude at least two, and four of those have magnitude at least three. Within B, eighty events have magnitude at least one, eight of those have magnitude at least two, and two of those have magnitude at least three.

The counts are nested. The four A events above three are already included in the forty above two. Adding forty and four would count some events twice.

The unresolved cell is deliberately not zero. It says that the exercise does not establish the complete count above one for A. Preserving that distinction is essential to every later comparison.

Contents · Previous chapter · Next chapter

If someone compares each window’s broadest complete count, A has forty and B has eighty. The count doubles. Dividing by duration gives five events per hour for A above two and twenty per hour for B above one.

It would be arithmetically correct to say that twenty is four times five. It would be scientifically incomplete to present that as a fourfold increase in the same event rate. The threshold changed at the same time as the window.

The comparison combines a real numerical operation with a changed question. In A the question concerns events above two. In B it includes smaller events that were excluded from A’s complete population.

Ask a learner to write both rates with their threshold attached before drawing an arrow between them. The labels make the mismatch visible. A graph caption saying only events per hour can conceal the changed selection, even when every plotted value was calculated correctly. The remedy is to preserve the population definition rather than simply replace the count graph with a rate graph.

Contents · Previous chapter · Next chapter

Magnitude two is within the supplied complete range for both windows. A therefore has forty divided by eight, or five events per hour, above that threshold. B has eight divided by four, or two per hour.

The B-to-A observed rate ratio is 2/5 = 0.4. On this defined comparison, the observed rate is sixty percent lower in B. This reverses the impression from the mismatched broadest counts.

The conclusion remains narrow: the supplied complete record above two has a lower observed rate in the second window. It does not establish that all seismic processes weakened, that gas behaviour changed, or that an eruption became less likely.

A common threshold removes one particular comparability problem. It does not remove every possible problem involving event classification, magnitude calibration, region, location or short-window variation. State which assumptions were supplied, then state the result conditional on them. The result becomes more useful when its exact scope is visible, not when it is inflated into a general verdict on the volcano.

Contents · Previous chapter · Next chapter

At magnitude three and above, A has four events in eight hours and B has two in four hours. Both observed rates are 0.5 events per hour. The rate ratio is one.

This does not invalidate the comparison above two. The two calculations concern different nested populations. Events between two and three contribute to one result and not the other.

For A, thirty-six events lie in the interval from two inclusive to three exclusive. Their observed rate is 4.5 per hour. For B, six lie in that interval, giving 1.5 per hour. Adding those bin rates to the respective above-three rates recovers five and two.

The exercise shows why there is no single threshold-free observed activity ratio hidden inside this table. If a question concerns larger qualifying events, the above-three calculation is relevant. If it concerns all events above two, use that population. Choose the question first rather than selecting the threshold after seeing which result looks most dramatic. Preserve both valid calculations when explaining why their conclusions differ.

Contents · Previous chapter · Next chapter

An event-rate comparison is one part of a volcanic monitoring account. Its interpretation can require event locations, types, magnitude information, deformation, gas observations and knowledge of the volcanic system.

The existing Volcanology root introduces the relationship between several lines of evidence. Here the narrower task is to establish what a count comparison actually says before those wider observations are considered.

Nothing in the fictional ledger supplies an eruption probability or a decision rule. Neither the decrease above two nor the unchanged rate above three provides a public-safety instruction. The example is deliberately constructed to teach comparability, not to imitate an observatory bulletin.

For a practical explanation, use two sentences. First describe the selected numerical change. Then state that its volcanic significance is unresolved by the table alone. This retains the useful result without forcing it to carry a conclusion for which the lesson has supplied no evidence. Scientific restraint can coexist with precise arithmetic and an inquisitive next question.

Contents · Previous chapter · Next chapter

USGS research on volcanic earthquake catalogue enhancement describes how detection and relocation methods can add events and refine their locations. This supports an important distinction: a revised catalogue can contain more information about past activity without the past physical events changing.

That source does not establish a detection probability for our fictional windows. It motivates checking how a record was produced. Its reported research examples should not be substituted for the invented counts here.

For a classroom evidence record, preserve the catalogue version and processing description alongside the query. If a later version adds smaller events, a comparison should say whether both windows were reprocessed consistently.

Deleting the earlier version would make the history harder to audit. Keeping both versions with a clear explanation lets another reader distinguish a change in observation from a change in the underlying period being described. That distinction also applies to revised rainfall datasets and improved image classifications, but the event definitions and scientific interpretation remain specific to their subjects.

Contents · Previous chapter · Next chapter

Now leave the finite ledger and consider a separate original model. Let λ₀ be the expected event rate above a reference magnitude m₀. For m ≥ m₀, stipulate λ(m) = λ₀ × 10^[−b(m − m₀)], where b is positive.

This exponential threshold relation has the familiar Gutenberg–Richter form. Here it is an assumed teaching model, not a fitted claim that one real volcano follows it across every magnitude. λ(m) is an expected rate, not an exact count that every short window must contain.

Increasing the threshold by Δ reduces the model rate by a factor 10^(−bΔ). The exponent is dimensionless because b is expressed per unit of the stated magnitude convention.

This model lets us separate a population-rate parameter from the effect of a threshold. It cannot justify extrapolation below a supported completeness range merely because an equation can be evaluated there. Keep the hypothetical parameter calculation separate from an empirical claim about undetected earthquakes. The model is useful precisely because its assumptions are explicit enough to change and test.

Contents · Previous chapter · Next chapter

Set m₀ = 1 and b = 1 in both model windows. In Model A, let λ₀ = 100 events per hour and the complete threshold be two. The expected detected rate is then ten per hour.

In Model B, let λ₀ fall to sixty per hour while the complete threshold drops to one. Its expected detected rate is sixty per hour. The expected catalogue rate rises sixfold even though the reference-population rate falls to sixty percent of its earlier value.

The example changes two things deliberately. Lowering the threshold increases the included fraction, while changing λ₀ alters the reference population. Neither effect should be hidden inside a single word such as increased activity.

At a common threshold of two, the model rates are ten and six per hour, recovering the 0.6 ratio because the slopes are equal. This is a model-level result distinct from the earlier finite ledger. Do not combine the fictional observed counts from one example with expected parameters from the other as though they were one dataset.

Contents · Previous chapter · Next chapter

Suppose the model rate above magnitude two is twelve per hour, and the reference threshold is one. Then twelve = λ₀ × 10^(−b).

There are many compatible parameter pairs. If b = 1, λ₀ = 120. If b = log₁₀4, approximately 0.60206, λ₀ = 48. If b = log₁₀2, approximately 0.30103, λ₀ = 24.

Each pair produces the same rate above two while making a different statement about the reference population above one. A single threshold-rate observation cannot separate the overall rate scale from the magnitude-distribution slope in this family.

An analyst who fixes b to one has supplied an assumption. They have not discovered it from the one observed rate. That may be a useful conditional scenario, but it needs to remain labelled.

For independent checking, substitute each pair into the original relation. The equal outputs demonstrate non-identification constructively. It is stronger evidence of the information gap than merely saying that more data would be helpful, because it exhibits exactly what the supplied observation cannot distinguish.

Contents · Previous chapter · Next chapter

Supply two exact expected rates within the complete range: λ(2) = 12 per hour and λ(3) = 3 per hour. Dividing eliminates λ₀, giving λ(2)/λ(3) = 10ᵇ = 4.

Thus b = log₁₀4, approximately 0.60206. With reference magnitude one, λ₀ = 12 × 10ᵇ = 48 per hour. Two supplied model rates have now separated the slope and reference scale.

The word exact matters. If the inputs were estimated from finite counts, the resulting parameters would inherit sampling variation and other uncertainties. This arithmetic exercise does not provide a statistical estimator, confidence interval or professional completeness method.

The thresholds must also refer to compatible magnitude definitions and the same population. Dividing a rate from one region by a rate from another does not eliminate a shared scale that was never shared.

Ask a learner to identify which assumption makes the ratio cancellation possible. They should point to one model population and one common λ₀, rather than merely remember that logarithms are used somewhere in earthquake questions.

Contents · Previous chapter · Next chapter

The reconstructed model predicts the expected rate at magnitude 2.5. Halfway between two and three, its logarithmic rate lies halfway between log twelve and log three.

The resulting rate is √(12 × 3) = 6 per hour. It is the geometric mean, not the arithmetic mean of 7.5. Equal magnitude increments correspond to equal multiplicative rate factors in this model.

This prediction can be held out from the parameter reconstruction. If a separately supplied exact model rate at 2.5 were nine, the three values would not all lie on the assumed single exponential threshold curve.

The discrepancy would not tell us which physical explanation is correct. It would tell us that the exact model and supplied values are inconsistent. With finite observations, uncertainty would need consideration before making the corresponding empirical judgement.

For practice, write the predicted six before revealing the third value. Then distinguish the two thresholds used to fit from the one used to check. Reusing a fitted input as confirmation is a weaker test than asking the model to explain something it did not already consume.

Contents · Previous chapter · Next chapter

In another model comparison, keep A at λ₀ = 100 and b = 1, with m₀ = 1. Let B have λ₀ = 60 but b = 0.5.

Above the reference magnitude one, B’s expected rate remains sixty percent of A’s. Above magnitude two, however, A has ten per hour while B has 60/√10, approximately 18.974 per hour. B now has a larger rate for that higher-threshold population.

The common-threshold comparison is still well defined. What changes is the interpretation: equal slopes were needed to equate its ratio with the reference-population ratio. When slopes differ, the event-size composition matters.

It would be incorrect to say that a common threshold has failed. It answered the above-two question correctly. It was never guaranteed to answer every threshold question with the same ratio.

This example is a useful guard against overgeneralising a successful repair. Standardising one selection rule improves comparability for that rule; it does not make the resulting summary independent of population structure. Preserve the threshold in the final sentence even after doing the careful calculation.

Contents · Previous chapter · Next chapter

Suppose λ(2) = 12 is retained, but the only supplied slope information is 0.5 ≤ b ≤ 1. With m₀ = 1, the reference rate is λ₀ = 12 × 10ᵇ.

Because that expression increases with b, the conditional range is 12√10 ≤ λ₀ ≤ 120, or approximately 37.947 to 120 per hour. Selecting b = 0.75 gives one scenario inside the interval, not a uniquely identified answer.

This is an assumption-based range, not a statistical confidence interval. Its endpoints are only as justified as the supplied slope bounds and exponential model. It says nothing about an actual volcano unless those inputs have appropriate evidence.

The calculation also shows why extrapolation can enlarge uncertainty. The greater the magnitude distance from the observed threshold, the more strongly the output depends on b. If a reader needs a rate near the observed threshold, report that supported quantity directly rather than reaching unnecessarily for a much less certain reference-population estimate. Choosing a useful target can matter as much as choosing a formula.

Contents · Previous chapter · Next chapter

The earlier examples supplied a hard complete threshold. Real detectability need not behave as an exact step. To study the distinction, introduce a separate simplified model in which every eligible event is detected independently with the same probability p.

If the underlying expected eligible rate is λ, the expected observed rate is r = pλ. Independence between events is stronger than needed for this expectation, but is included here to make the thought experiment clear. The common detection probability within the eligible population is essential to the simple product interpretation.

A rate of twelve observed events per hour is compatible with λ = 12 and p = 1, or λ = 20 and p = 0.6. Without information about p, the observed rate does not identify λ.

Do not interpret p as a value inferred merely because the catalogue has a stated magnitude threshold. A threshold and a calibrated detection model are different pieces of information. The exercise isolates one uncertainty so that the missing assumption becomes visible; it does not prescribe how observatories should estimate detection probability.

Contents · Previous chapter · Next chapter

Let the exact expected observed rates be rA = 12 and rB = 15 per hour for the same eligible population definition. Supply detection ranges pA ∈ [0.6,1] and pB ∈ [0.5,1].

Then λA = 12/pA lies in [12,20], while λB = 15/pB lies in [15,30]. With independently feasible detection choices, λB/λA lies in [15/20,30/12] = [0.75,2.5].

The observed ratio is 1.25, but the underlying model ratio could represent a decrease or an increase. Both endpoints can be constructed: pA = 0.6 with pB = 1 gives 0.75; pA = 1 with pB = 0.5 gives 2.5.

The interval would need revision if the two detection probabilities were constrained together. For example, requiring pA = pB would make detection cancel and recover 1.25. That extra relationship is evidence or an assumption, not an arithmetic convenience. This is why uncertainty records should preserve dependencies between inputs rather than list separate ranges and silently choose any combination.

Contents · Previous chapter · Next chapter

A common magnitude threshold is not sufficient if the geographical population changes. Suppose two fictional subregions each have stable rates above that threshold: two events per hour in the north and eight in the south.

A catalogue covering only the north records two per hour under complete detection. A later catalogue covering both records ten. The fivefold total difference comes from a changed spatial boundary, even though neither local rate changed.

Dividing by mapped area would answer another question and would require meaningful area definitions. It would not automatically restore the originally intended population comparison. Equal areas can contain different source environments and event distributions.

For a fair desk exercise, keep the same supplied boundary or report the subregions separately. Then explain what the expanded coverage adds. The map, event query and time window belong together in the evidence record.

This example deliberately changes only coverage. It does not claim that actual seismicity is uniform within a region or that a per-area statistic is always appropriate. The lesson is to identify which population the denominator represents before interpreting a change in its total.

Contents · Previous chapter · Next chapter

Suppose a four-hour window contains only three hours of valid observation. Six qualifying events are recorded during those three hours. The rate during observed time is two per hour; dividing by four gives 1.5 per calendar hour, a different description of the available record.

Neither number establishes what happened during the missing hour. If a stationary-rate model is explicitly supplied, one may calculate an expected full-window count from the observed-time rate. Without that assumption, the gap could coincide with a different period of activity.

Keep the valid intervals, not only their total duration. Three hours distributed across the day can provide different temporal coverage from one continuous three-hour block. Whether that difference matters depends on the scientific question and the process.

The safe reporting distinction is simple: observed-time count and rate are established; the missing-period contribution remains unresolved unless additional evidence or a declared model supplies it. Entering zero for the gap would convert lack of observation into an unsupported absence of events.

Contents · Previous chapter · Next chapter

A threshold comparison also depends on how magnitudes are represented. Suppose the rule includes events whose underlying supplied magnitude is at least two. An event with value 1.96 displayed to one decimal place as 2.0 creates a potential mismatch if a reader applies the threshold to the rounded display.

The correct query follows the catalogue’s documented precision and inclusion convention. A displayed numeral is not automatically the unrounded value used in processing. Preserve the convention rather than inventing a correction event by event.

Revised magnitudes can also move events across a threshold between catalogue versions. The physical event remains, while the current estimated value or classification changes. A reproducible comparison therefore includes the version or retrieval date.

For a classroom task, give both unrounded and displayed columns and ask which one the supplied selection rule uses. The learner should identify the rule before counting. This is another example of meaning preceding calculation: a carefully computed percentage can still compare differently selected populations if the threshold was applied to incompatible representations.

Contents · Previous chapter · Next chapter

Finite counts vary. Equal expected rates do not imply identical counts in every pair of short windows. The first ledger’s observed ratios were exact descriptions of its supplied records, but they were not confidence statements about unknown long-term parameters.

If a scientific analysis uses a stochastic count model, it needs assumptions appropriate to the events, possible clustering, changing conditions and observation process. This guide does not assume that volcanic events are independent Poisson arrivals or derive an uncertainty interval from that assumption.

Small counts particularly deserve careful language. Two events versus four can create a large percentage change while leaving substantial uncertainty about a repeatable underlying pattern. More decimal places in the ratio do not remove that uncertainty.

Write the count and duration alongside the ratio. A reader can then see whether the numerical summary comes from a large catalogue or a small handful of events. The observation remains useful without being presented as an exact measurement of a permanent physical rate.

Contents · Previous chapter · Next chapter

A straight line on a logarithmic rate plot can be consistent with the stipulated exponential relation. That consistency does not by itself establish its range of applicability or identify a volcanic mechanism.

Two exact rates determine the two parameters of this teaching family. A third compatible rate is an additional model check. Further checks might change the observation window, spatial subset or processing version while preserving the defined comparison.

Those checks answer whether a representation is consistent with the supplied observations. Explaining why a volcano produced those events requires a separate scientific account. Avoid replacing that account with the fitted line’s slope.

For learners, keep three columns: inputs used to choose the model, observations used to test it, and wider interpretations requiring other evidence. A single number should not quietly occupy all three columns. This record makes it easier to notice circular reasoning, especially when a model-derived value later appears in a table as if it were a new independent measurement.

Contents · Previous chapter · Next chapter

Try this new fictional record. Window C covers ten valid hours and is supplied complete above magnitude 1.5. It contains thirty qualifying events, including six above 2.5. Window D covers five valid hours and is complete above 2.5, with four events above that threshold.

The broadest complete observed rates are three and 0.8 per hour, but their thresholds differ. At the common threshold 2.5, C has 0.6 per hour and D has 0.8. D’s common-threshold observed rate is one-third larger.

Can the complete count above 1.5 in D be recovered from this record? No. The catalogue does not supply it. Multiplying four by a remembered factor of ten would add a slope model that the exercise has not provided.

Now add a separate ideal model statement: the same threshold-rate slope b = log₁₀5 applies in both windows over this range. Under that added assumption, D’s expected above-1.5 rate would be five times its expected above-2.5 rate. Explain why the extra calculation needs the model statement rather than following from completeness alone.

Contents · Previous chapter · Next chapter

The reference m₀ is a parameterisation choice. It should not change the threshold-rate curve when parameters are transformed consistently.

For the ideal curve with λ(2) = 12 and b = log₁₀4, using m₀ = 1 gives λ₀ = 48. Using m₀ = 2 instead gives λ₀ = 12. Both predict λ(3) = 3 and λ(2.5) = 6.

If someone changes the reference magnitude but retains λ₀ = 48, they have changed the model rather than merely relabelled it. The predicted curve shifts because the scale parameter’s meaning changed.

This is a useful independent transfer check. Ask a learner to choose a new reference, transform its rate and reproduce one held-out prediction. They should be able to explain why two different parameter numbers describe the same population curve.

The exercise also protects interpretation: a larger fitted λ₀ in another report might partly reflect a lower reference threshold. Compare definitions before treating parameter names as directly comparable physical quantities.

Contents · Previous chapter · Next chapter

Record the event unit, region, time window, valid coverage, event class, magnitude convention, precision and threshold. Include catalogue version and the source of any completeness or detection assumption.

Preserve nested counts without adding them. If bins are used, define their endpoints so each event belongs to exactly one bin. Report counts and durations alongside rates and ratios.

Separate observed finite quantities from expected model rates. State which parameters were fixed, fitted or bounded, and whether uncertainty choices can vary independently. Save any held-out prediction before incorporating its checking observation.

A practical review asks another reader to reproduce one comparison and identify one conclusion the record does not establish. They should not need to guess the threshold from the article’s title or infer valid coverage from a smooth graph.

This desk-based exercise needs no field monitoring or current volcano access. Its purpose is to make the numerical account inspectable, so that a later scientific interpretation rests on a clearly defined observation rather than a misleading total.

Contents · Previous chapter · Next chapter

For repair, begin with count, duration and inclusion rule. Use two windows and one common threshold. Require a full sentence naming the event population and observed rate.

For stabilisation, retain two common thresholds and reconcile their different results through the intervening magnitude bin. Then change the window duration without changing the included events’ definition.

For extension, use the threshold-rate model, parameter non-identification and uncertain detection examples. Ask which added information separates λ₀ from b or observed rate from underlying rate. Include an inconsistent held-out value so that checking can genuinely challenge the model.

An independent task should leave method selection to the learner. Give a catalogue version change, a missing observation interval or a revised spatial boundary and ask which earlier comparison remains justified. Recognising that the original result no longer answers the changed question is a form of progress, even when the next answer is a carefully described information gap.

Contents · Previous chapter · Next chapter

Does a lower detection threshold mean the volcano produced more events?

It can mean more events enter the record. The effect on the underlying physical activity is a separate question requiring comparable observations and interpretation.

Is the highest common threshold always the best scientific choice?

A threshold must suit the question and available evidence. A stricter threshold can improve a particular completeness comparison while leaving fewer events. This guide supplies no universal selection procedure.

Can a common-threshold ratio describe every magnitude population?

No. Different populations can change differently. The original ledger and changed-slope model demonstrate that explicitly.

Does a fitted b value predict an eruption?

No such inference follows from the exercises. They concern catalogue selection and model rates, not an eruption forecasting system.

Should missing observation time be entered as zero events?

No. Preserve the gap and distinguish observed-time quantities from assumptions about the unobserved period.

Contents · Previous chapter · Next chapter

For the broader scientific account, return to Volcanology and Seismology. The root’s worked unequal-window comparison supplies an introductory route into rates; this article adds threshold-specific populations, constructive parameter ambiguity and conditional detection bounds.

For introductory evidence reading, use the existing volcano-monitoring guide and the aftershock sequence guide. They introduce threshold and catalogue context; this article adds original nested ledgers, ideal parameter recovery, changed-slope comparisons and attainable detection bounds.

The primary contextual source is Tan and colleagues’ Volcanic earthquake catalog enhancement using integrated detection, matched-filtering, and relocation tools, published 9 May 2023 and checked 26 October 2026. It supports the role of detection and catalogue processing. The ledgers and model exercises here are original; no source dataset, operational threshold or reported parameter is reproduced.

Use the Education parent to connect definition, calculation and independent transfer. Continue through How X Works when another subject raises a similar question about what a changing record actually counts.

For primary research on observable magnitudes and completeness-sensitive slope estimation, see Martinsson and Jonsson, A New Model for the Distribution of Observable Earthquake Magnitudes and Applications to b-value Estimation, submitted 15 March 2018 and checked 26 October 2026. The ideal threshold-rate exercise here does not implement that paper’s estimation method.

Contents · Previous chapter · Return to Volcanology