Seeking pilot sites, engineering partners and regulatory collaborators across Europe.

Aftermine
Develop a pilot

instrument · digital twin

Auguste Victoria Colliery, for thirty years after closure

A coupled model of gas, water and abatement at a single named mine in the Ruhr — closed 2015, an estimated 28 million cubic metres of methane a year, flooded status unknown. Every figure carries a confidence interval, an evidence tier and an estimate class. Several of them argue against the case for building this.

Screening and design-basis framing. Not a design, and not a basis for financing a plant.

What is Auguste Victoria about this, and what is not

This is a twin of a specific mine, not a generic one. It is also not a measurement of that mine, and the difference matters more here than it would almost anywhere else.

Taken from Auguste Victoria

  • Closure year — 2015, so the decline curve starts eleven years before the project does.
  • Emission rate — 28 Mm³/yr, the figure carried for AV in the field-study dataset.
  • Workings geometry — a shallow horizon from surface to 300 m and a deep one from 300 m to 1,200 m.
  • Flooded status — unknown, which is the condition the whole venture exists to address.

Not measured at Auguste Victoria

  • The emission rate is MODELLED, not metered. No abandoned mine in the EU inventory has a measured emission rate — that gap is what Regulation (EU) 2024/1787 exists to close.
  • The horizon depths and void areas are assumed. They are stated site inputs at the highest tier, which means 'exact because we chose it', not 'checked against a survey'.
  • The water chemistry — iron, manganese, sulphate — is a REGIONAL prior from the literature, not an analysis of AV's discharge. Two collieries in the same seams can differ by an order of magnitude in iron.
  • The water level is modelled as a free surface rising to a decant. Under the Ruhr's Grubenwasserkonzept it is a regulated setpoint held by pumping, so this trajectory is a stop-pumping counterfactual rather than a forecast.

Which is why the hydrogeology reviewer signed this as a screening model for a general north-west European coalfield and explicitly NOT for a Ruhr release. The mine is named because the geometry and the gas rate are its own. It is not named because anybody has measured it.

What this model says against its own project

A results page showing only the tonnage would be a fair reading of the numbers and a dishonest reading of the model. These four are what the work actually surfaced.

−13,695 t CO₂e

versus a 99 % compliant flare, ETS credits disallowed

Against a compliant flare, the sign flips

Once venting is prohibited, the legal counterfactual is not doing nothing — it is the cheapest compliant option, an enclosed flare at 99 % destruction. Our engine destroys about 98 %. Measured against that baseline the destruction claim is NEGATIVE, and what survives is grid and heat displacement. German and Polish generation sits inside the EU Emissions Trading System, where displacing a capped emission frees an allowance rather than reducing emissions. Disallow those credits, as a verifier would, and the project is a net disbenefit versus a flare.

0.20 – 0.75

prior range; no incumbent can supply the measurement

Capture, not destruction, is the whole question

Gas that never enters the collection system is not abated by anything downstream of it, however good the equipment. Capture efficiency is simultaneously the largest term in the balance, the least well known, unmeasurable by any wellhead instrument, and the one thing that cannot be improved by buying better plant. Every honest thing this model surfaced points at it.

0.0 %

median share of plant-years inside the window

The biofilter's window is essentially never reached

The methanotrophic reactor needs a stream between roughly 0.02 and 1.8 vol %. Across the whole sampled space the drained wellhead stream sits in that band in a median of zero years out of thirty. The Bio layer contributes nothing to any figure on this page. It earns its place by saying what a working biofilter would have to be — and by ruling out the drained stream as the place to put one.

7.9×

undersized for the flow it is asked to treat

And the biofilter as drawn is an order of magnitude too small

The shipped vessel gives 38 seconds of gas residence time against a field-envelope minimum of 300. The geometry was borrowed from air-pollution-control practice, where 15–60 seconds is normal — but methane is thousands of times less soluble than the contaminants that practice was built on, so the transfer does not hold. Reaching a useful removal needs roughly nineteen times the bed volume, and even then removal is capped by channelling rather than by biology.

Results

Thirty-year assessment. P5 and P95 are the 5th and 95th percentiles of the Monte Carlo, not a best and worst case anyone chose.

OutputP5MedianP95UnitEstimate class
Methane destroyed, lifetime136884958089839t CH₄Wider than Class 5
Methane emitted, lifetime89793224795393664t CH₄Wider than Class 5
Net warming avoided40694714675362842897t CO₂eWider than Class 5
Lifetime site abatement12.4 %21.2 %31.7 %%Class 5
Peak annual emission123491738624840t CH₄/yrClass 5†
Years to decant, or horizon✳5.82230yrWider than Class 5
Heat delivered, lifetime079668758380MWhWider than Class 5
Peak thermal power0334511667kWWider than Class 5
Mean-flow wetland area03.738.5haWider than Class 5
Mean-flow lime demand01541643t CaO/yrWider than Class 5
Years with unverified engine slip62130yr of 30Wider than Class 5
Years in the biofilter's window0.0 %0.0 %3.3 %%Wider than Class 5
  • Wider than Class 5 — The spread is wider than AACE's loosest estimate class, so no class applies. Read the interval; do not read a class.
  • † Set by the evidence tier, not by the spread — the interval shown is NARROWER than the real uncertainty.
  • ✳ Right-censored: 35.7 % of draws never reach the decant within thirty years and report the horizon instead. Read the median, never the P95 alone.

What actually drives the answer

Marginal Spearman rank correlation. These are not variance shares: the parameters are correlated with each other, so they do not partition the spread and are not claimed to.

  1. recharge_m3PerYear−0.66
  2. deepStorage_m2+0.52
  3. maxCh4Volfrac+0.49
  4. gwp+0.23
  5. startDepth_m+0.16

The water parameters dominate the methane answer. Recharge and storage area outrank every abatement parameter — the coupling made visible, and invisible to any model that treats gas and water as separate projects.

74 of 87 parameters sit below |ρ| 0.02. For most of them that is a REACHABILITY result rather than an unimportance one: all four flare parameters are inert because the flare route is never selected. Publishing the full list would invite the conclusion that they were tested and found unimportant, when most were never reached.

Move the assumptions

The same coupled model, run live at one parameter vector. These are the three terms the review identified as decision-changing; everything else sits at its prior median.

Share of the mine's methane that reaches the plant. The dominant term, and the one no wellhead instrument can measure.

Water entering the workings each year. Drives how fast the mine floods, which decides both sealing and displacement.

How hard the plant draws. Too little leaves gas in the ground; too much pulls air in behind it and the stream goes lean.

…

This runs the deterministic central case, not the Monte Carlo. It shows how the answer responds, not how uncertain it is — the table above is the uncertainty.

Four independent expert reviews

The model was reviewed adversarially across four domains, with a second opinion in each. Verdicts as the reviewers wrote them.

Bioprocess

Yes for screening, and yes for the shape. The magnitude may be published provided it is stated that the model is anchored to a single 2014 laboratory column and has never been compared to a field installation. Not yet a basis for design.

Hydrogeology

Yes for screening. No for a Ruhr or Upper Silesia release. Reviewed as a screening model for a general north-west European coalfield; not suitable for the Ruhr or Upper Silesia until a managed-level regime and chloride are added.

Uncertainty quantification

The mathematics was already sound. What was wrong was the layer that turns a distribution into a claim. I regard the resulting figures as fit for screening and design-basis framing, and not for design — which is what the document already claims for them.

Utility MRV and verification

Sign the numbers for verification: yes, unqualified. Fund a measurement campaign: yes. The accounting is better than most of what gets third-party verified in this sector, and the reason is that it reports the things that count against it.

The single most useful thing the review produced: a model can be numerically flawless and still tell an asset owner something false. Every defect found across four streams sat not in the mathematics but in the layer between a correct computation and what a reader is told — the part that looks like presentation, and is almost never reviewed.

Read the full methodology and review record

How to check this

Every number below comes from one deterministic run. The same seed and sample size reproduce the table exactly.

Monte Carlo
n = 6000
Seed
aftermine-twin-v1
Declared correlations applied
10
Parameters sampled
87 (18 of which verified against a cited source)

Reproduce the table

node -e 'var S=require("./public/seam-lib/twin-scenario.js"),P=require("./public/seam-lib/twin-priors.js");console.log(S.run(P.flatten(), {n: 6000, seed: "aftermine-twin-v1"}).results)'