PCF Sensitivity, Uncertainty & Hotspot Analysis Calculator | ISO 14067 & GHG Protocol Product
Interrogate a completed product carbon footprint — rank contribution hotspots, run one-at-a-time sensitivity, and propagate ecoinvent-pedigree uncertainty through Monte Carlo to a confidence interval, all in CO2e against your functional unit.
What this calculator does — and what it deliberately does not. This is a diagnostic layer that sits on top of a finished product carbon footprint (PCF). It does not look up emission factors and does not compute a footprint from activity data — you supply the per-process contributions from a PCF you have already built (in the ISO 14067 cradle-to-gate PCF calculator, an LCA tool, or a spreadsheet). The engine then answers three separate questions about that footprint: where the emissions concentrate (hotspot analysis), which assumptions move the result (sensitivity analysis), and how confident you can be in the headline number (uncertainty analysis).
Entry model. You build the PCF line by line — one row per process or life-cycle stage — entering a label, a stage, a contribution in your chosen unit (g / kg / t CO2e), and a data-quality rating. Negative contributions are allowed for recycling credits, Module D, and biogenic removals. Choosing Custom data quality opens an advanced drawer per line: either the five-indicator ecoinvent pedigree matrix or an explicit distribution (lognormal GSD, normal ±%, triangular, or uniform).
Hotspot analysis ranks each line by its share of the gross footprint and reports a running cumulative (Pareto) percentage, so you can state that the top N contributions account for X% of the total. Sensitivity analysis swings each input across its own confidence interval and presents the result as a tornado centred on the best estimate. Uncertainty analysis converts each line’s pedigree score into a lognormal geometric standard deviation, then propagates all lines through a Monte Carlo simulation to produce a confidence interval, a coefficient of variation, and a per-line ranking of contribution to output variance.
Uncertainty method. Pedigree scoring uses the ecoinvent matrix (Frischknecht et al. 2005; Weidema & Wesnæs 1996): five indicators each map a 1–5 score to an uncertainty factor, a basic uncertainty factor of 1.05 is applied to every line, and the line’s geometric standard deviation is the exponential of the root-sum-of-squares of the log factors. Monte Carlo confidence bounds use z-factors of 1.645 (90%), 1.959964 (95%), and 2.575829 (99%). The simulation is reproducible for a fixed seed and run count.
This is the ecoinvent pedigree matrix, not the PEF/EF DQR. The two are different systems and are not interchangeable — see the dedicated section below. Scope: the calculator analyses the PCF you give it; it does not reach back into the underlying LCA model, so the quality of its diagnosis is bounded by the quality of the contributions and pedigree scores you enter.
Set any line’s data quality to Custom… to model it here — either the 5-indicator ecoinvent pedigree matrix (data quality → geometric standard deviation) or an explicit distribution. Lines left on a High / Medium / Low tier use the tier’s default spread.
Enter your footprint contributions above to analyse
Results appear instantly. The total with its confidence interval, the hotspot (Pareto) ranking, a sensitivity tornado, the “where to improve data” variance drivers, a Monte Carlo output distribution and the full audit trail appear after calculation.
Results are an analysis of a Product Carbon Footprint you supply — a hotspot (contribution) ranking (ISO 14044:2006 §4.4.5), a sensitivity assessment, and an uncertainty quantification (ISO 14067:2018 §6.4.4; GHG Protocol Product Life Cycle Standard) — not an emission-factor calculation. The tool does not look up or apply emission factors; every contribution, its sign, and its data-quality rating are your own inputs. Uncertainty is propagated by both IPCC 2006 approaches (analytical error propagation under GUM / JCGM 100:2008, and a seeded Monte Carlo simulation); where they diverge, the Monte Carlo interval governs because it captures the skew of lognormal terms. Per-line spread comes from a data-quality tier or, in Custom mode, from the ecoinvent pedigree matrix (Frischknecht et al. 2005; Weidema & Wesnæs 1996) or an explicit distribution; the pedigree uncertainty factors are published constants, not MasterBrain data. The combined interval, the tornado, and the contribution-to-variance ranking are only as sound as the per-line judgments behind them — refine the most material and most uncertain lines with primary data, and complete a critical review under ISO 14071 before using this PCF in an EPD, a PACT / Catena-X exchange, or a public comparative assertion.
A product carbon footprint is a single number standing on hundreds of assumptions. Two analysts handed the same bill of materials can return PCFs that differ by a factor of two — not because either made an arithmetic error, but because they chose different allocation methods, different system boundaries, different background datasets, and different biogenic-carbon conventions. The number on the label tells you nothing about how firmly it is held.
Sensitivity, uncertainty, and hotspot analysis are how you find out whether a PCF will survive scrutiny — which line dominates the total, which assumption would flip the conclusion if challenged, and how wide the confidence interval really is. ISO 14067 and the GHG Protocol Product Standard both require this diagnostic layer; this calculator runs all three analyses on a footprint you have already built.
PCF hotspot analysis ranks each process by its share of the total footprint; sensitivity analysis swings each assumption to see which moves the result most; uncertainty analysis propagates per-input data quality through Monte Carlo to a confidence interval. They answer three different questions — the largest contributor is rarely the largest source of uncertainty, and the two rankings should be read separately.
What PCF Sensitivity, Uncertainty & Hotspot Analysis Actually Is
A product carbon footprint quantifies the greenhouse gas emissions associated with a product across a defined system boundary, expressed per functional unit. Building one is a forward calculation: activity data multiplied by emission factors, summed across processes. Sensitivity, uncertainty, and hotspot analysis run in the opposite direction — they take the finished footprint and ask how robust it is. The three are routinely conflated, but they answer distinct questions and produce distinct outputs.
Three diagnostics, three different questions
Hotspot analysis
Where does the footprint concentrate? Ranks each process by its share of the gross total and reports the cumulative Pareto curve. Answers: which lines are worth managing.
Sensitivity analysis
Which assumptions move the result? Swings each input across its plausible range and measures the effect on the total. Answers: which choices you must defend.
Uncertainty analysis
How confident is the number? Propagates per-input data quality through Monte Carlo to a confidence interval. Answers: how wide the error bars are, and where to collect better data.
The practical payoff is that these three rarely point at the same line. The biggest contributor may be well-characterised and low-uncertainty; a smaller contributor based on a proxy dataset may dominate the confidence interval. Treating “hotspot” as a single idea hides exactly the distinction a practitioner needs.
Where it sits in the ISO 14067 / GHG Protocol Product workflow
Both governing standards place this work in the interpretation phase, after the inventory is complete. Under ISO 14040/14044, the interpretation phase explicitly requires a completeness check, a sensitivity check, and a consistency check; ISO 14067 inherits that structure and adds carbon-footprint-specific requirements. The GHG Protocol Product Standard devotes a full chapter to uncertainty and instructs reporters to perform and disclose it. In every case the diagnostic comes last — you cannot sensitivity-test a footprint you have not yet built, which is why this calculator takes contributions as input rather than computing them.
Included vs out of scope
| This calculator does | This calculator does not |
|---|---|
| Rank contribution hotspots and the Pareto cumulative | Look up emission factors or compute a footprint from activity data |
| Run one-at-a-time sensitivity and present a tornado | Reach back into the underlying LCA model or background database |
| Convert ecoinvent pedigree scores to a lognormal GSD per line | Apply the PEF/EF DQR four-dimension quality score (different system) |
| Propagate uncertainty through Monte Carlo to a confidence interval | Decide your allocation method or system boundary for you |
| Rank each line’s contribution to output variance | Substitute for primary data collection on high-variance lines |
| Accept negative contributions (credits, Module D, biogenic removal) | Validate that your boundary or functional unit is correctly defined |
Hotspot Analysis — Contribution and Pareto Ranking
Hotspot analysis is the most familiar of the three and the easiest to read: it answers “where do my emissions come from” by expressing each process as a percentage of the gross footprint. The standards call this contribution analysis, and it is the foundation for any reduction strategy — you cannot prioritise what you have not ranked.
Contribution by life-cycle stage
The calculator groups your line entries by life-cycle stage and shows each stage’s value alongside its share of the total. The illustrative distribution below uses the worked example further down the page — a five-line cradle-to-grave footprint of 7.0 kg CO2e per unit, before the recycling credit is shown separately.
Shares here are expressed against the gross footprint (the sum of positive contributions, 7.5 kg before the credit). The Module D recycling credit of −0.5 kg is shown as a separate line and is not netted into the percentage base — consistent with how credits and removals are kept distinct from gross emissions in product accounting. The net result of 7.0 kg CO2e per unit is the gross less the credit, reported as two figures rather than one.
The Pareto cumulative and the materiality threshold
Ranking contributions from largest to smallest and accumulating their shares produces the Pareto curve. It is the basis for the familiar finding that a small number of processes drives most of a footprint. In the worked example, primary materials and manufacturing energy together account for 80% of the gross footprint — the two lines a reduction programme would address first.
The practical use is setting a materiality threshold for data-collection effort. A common convention is to target primary data for every process above a chosen cumulative-share cutoff and accept secondary or proxy data below it. The Pareto cumulative tells you exactly where that cutoff falls. One caution: a materiality threshold set on contribution share alone can miss a small-but-uncertain line — which is why the variance ranking, covered below, is read alongside the contribution ranking rather than instead of it.
Sensitivity Analysis — Which Assumptions Move the Result
Sensitivity analysis tests how the footprint responds when an input changes. Where hotspot analysis ranks contributions as they stand, sensitivity analysis asks the counterfactual: if this number were different, how different would the total be? It is how you identify the assumptions that carry the conclusion — the ones an auditor or a critic will press on.
One-at-a-time swings and the tornado
The standard technique is one-at-a-time (OAT) sensitivity: hold every input at its best estimate, swing one across its plausible range, record the effect on the total, then restore it and move to the next. Plotting each swing as a horizontal bar centred on the best estimate produces a tornado chart — widest bars at the top, narrowing downward, the shape that gives the chart its name. The calculator swings each input across its own confidence interval, so the bar widths reflect both how large a line is and how uncertain it is.
The ordering above is illustrative of the tornado’s shape rather than exact swing magnitudes — read the precise bar widths off the live tool, which centres each bar on the best estimate and labels it with the swung range. The point the chart makes is structural: the lines at the top are where a challenged assumption would change your reported number most, and therefore where your methodology justification needs to be strongest.
Allocation and boundary as the dominant levers
In most product footprints the largest sensitivities are not the activity-data values but the methodological choices that sit underneath them. Allocation method — how shared emissions are split between a product and its co-products — can shift a PCF substantially, which is why it warrants its own treatment in the PCF allocation methods calculator. System boundary (cradle-to-gate versus cradle-to-grave) and the biogenic-carbon convention are the other two high-leverage choices. These are scenario sensitivities rather than numeric swings: you re-run the footprint under each defensible choice and report the spread. The PCF frameworks comparison calculator is the tool for testing how a footprint moves between ISO 14067, GHG Protocol Product, and PEF conventions.
Uncertainty Analysis — Quantifying Confidence
Uncertainty analysis attaches error bars to the footprint. Sensitivity tells you which inputs matter; uncertainty tells you how much the inputs you have could be wrong, and what that does to the headline number. The two are complementary — a high-sensitivity input with low uncertainty is well-managed, while a high-sensitivity input with high uncertainty is the one to worry about.
The ecoinvent pedigree matrix
The calculator scores data quality with the ecoinvent pedigree matrix, the most widely used semi-quantitative scheme in LCA. Each input is rated 1 (best) to 5 (worst) on five independent indicators, and each rating maps to an uncertainty factor. The factors below are the engine’s cited constants, taken from the ecoinvent matrix (Frischknecht et al. 2005; Weidema & Wesnæs 1996) and verified against the Brightway reference implementation.
| Score | Reliability | Completeness | Temporal correlation | Geographical correlation | Further tech. correlation |
|---|---|---|---|---|---|
| 1 (best) | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
| 2 | 1.05 | 1.03 | 1.03 | 1.01 | 1.00 |
| 3 | 1.10 | 1.10 | 1.10 | 1.02 | 1.20 |
| 4 | 1.20 | 1.20 | 1.20 | 1.02 | 1.50 |
| 5 (worst) | 1.50 | 1.50 | 1.50 | 1.10 | 2.00 |
The five indicators capture different ways a dataset can be a poor match for the process it represents: how the data was acquired (reliability), whether it covers the full process (completeness), how old it is (temporal), whether it is from the right region (geographical), and whether it is from the right technology (further technological correlation). A dataset can be perfectly reliable yet geographically wrong — the indicators are scored independently for exactly that reason.
From pedigree to geometric standard deviation
The five indicator factors combine with a basic uncertainty factor of 1.05 — applied to every line regardless of pedigree — to produce a lognormal geometric standard deviation (GSD). The aggregation takes the root-sum-of-squares of the natural logs of all six factors, then exponentiates:
σ = √( Σ [ln(Uindicator)]² + [ln(1.05)]² ) · GSD = exp(σ)
Each line’s GSD describes a lognormal distribution around its contribution. A line scored all-1s carries only the basic uncertainty (GSD ≈ 1.05); a line scored all-5s carries a much wider spread. The 95% range of a lognormal input runs from value ÷ GSD² to value × GSD².
Lognormal is the default distribution for LCA inventory data because emission factors are strictly positive and right-skewed — a factor can be several times too high but cannot fall below zero. For lines where you know the distribution explicitly, the advanced drawer accepts a direct lognormal GSD, a normal ±%, a triangular (min-mode-max), or a uniform (min-max) instead of pedigree scoring.
Monte Carlo versus analytical propagation
The calculator reports both propagation approaches, mirroring the two methods the GHG Protocol Product Standard describes. The analytical approach (sometimes called Approach 1) combines the per-line uncertainties using error-propagation formulae — fast, closed-form, and exact for simple sums, but it assumes the inputs are independent and the output is near-normal. The Monte Carlo approach (Approach 2) samples every input from its own distribution thousands of times and builds the output distribution empirically — slower, but it captures skew, correlation, and the asymmetry that analytical formulae miss. The calculator surfaces a divergence note in percentage points when the two disagree, which is itself diagnostic: a large divergence signals that the footprint is skewed enough that the analytical interval understates the true spread.
Reading the output distribution
The Monte Carlo run produces a full distribution, summarised in a statistics grid. The live calculator renders the histogram with percentile markers; the table below is the stat grid that accompanies it. Figures shown are from the worked example at 95% confidence and 10,000 runs — read the exact rendered digits off the live tool, since Monte Carlo values are reproducible bit-for-bit only at a fixed seed and run count.
| Statistic | Meaning | Worked example |
|---|---|---|
| Best estimate | Deterministic total from the entered contributions | 7.0 kg CO₂e/unit |
| Mean (MC) | Average across all Monte Carlo runs | [verify against live tool] |
| Median (MC) | 50th percentile of the output distribution | [verify against live tool] |
| Std deviation | Absolute spread of the output | [verify against live tool] |
| Coefficient of variation | Std deviation as a % of the mean — the comparable spread metric | [verify against live tool] |
| 95% confidence interval | 2.5th to 97.5th percentile of the distribution | [verify against live tool] |
| Skewness | Asymmetry of the distribution (positive = longer upper tail) | [verify against live tool] |
| Analytical ± | Closed-form interval, for comparison with the MC interval | [verify against live tool] |
Because the inputs are lognormal, the output distribution is right-skewed: the mean sits above the median, the upper tail runs longer than the lower, and the confidence interval is asymmetric about the best estimate. This is the expected and correct shape — a symmetric ± interval reported on a skewed PCF understates the upside risk. The coefficient of variation is the figure to quote when comparing the robustness of two footprints, since it normalises the spread against the mean.
The Two Hotspot Rankings Are Not the Same
This is the distinction that separates a competent uncertainty analysis from a superficial one, and the calculator surfaces it deliberately. There are two ways to rank the lines in a footprint, and they answer different questions.
Contribution ranking
Ranks lines by share of the total footprint. Tells you where the emissions are — the target for reduction effort.
Variance ranking
Ranks lines by contribution to output variance. Tells you where the uncertainty is — the target for better data.
A line can rank high on one and low on the other. A large, well-measured contributor — primary materials backed by a supplier-specific EPD — may dominate the footprint while contributing almost nothing to its uncertainty. A small contributor based on a geographically and temporally mismatched proxy dataset may barely register in the contribution ranking yet drive the confidence interval. Reduction effort follows the contribution ranking; data-collection effort follows the variance ranking.
The single most useful output of an uncertainty analysis is the answer to “where do I collect primary data first?” That answer is the variance ranking — not the contribution ranking. Spending a data-collection budget on the largest contributor when a smaller, proxy-based line drives the variance leaves the confidence interval exactly as wide as it was. The calculator reports both rankings precisely so the two budgets — emissions reduction and data improvement — are allocated against the right list.
Pedigree Is Not PEF DQR
Two data-quality systems are common in product footprinting, and they are frequently — and incorrectly — treated as the same thing. This calculator uses one of them; knowing which, and why it is not the other, matters for how you report.
This calculator uses the ecoinvent pedigree matrix: five indicators, each scored 1–5, mapped to uncertainty factors and aggregated into a lognormal GSD that feeds Monte Carlo. It does not use the PEF/EF Data Quality Rating, which averages four dimensions into a single 1–5 quality score with named bands (excellent, very good, good, fair, poor). The pedigree matrix produces a distribution; the PEF DQR produces a score. They are different methods with different outputs and are not interconvertible — a pedigree GSD is not a DQR, and a DQR cannot be dropped into the Monte Carlo engine.
The distinction is not pedantry. The ecoinvent pedigree matrix exists to quantify uncertainty for propagation — its whole purpose is to produce the spread that Monte Carlo samples. The PEF DQR exists to gate data acceptability against a quality benchmark — its purpose is to decide whether a dataset is good enough to use under the Product Environmental Footprint method, not to propagate error. If you are reporting under PEF you will compute a DQR for compliance; if you are propagating uncertainty you will use pedigree GSDs. A report can legitimately contain both, computed separately, but the numbers are never substituted for one another.
Standards Basis — ISO 14067, GHG Protocol Product, ISO 14044
Sensitivity and uncertainty analysis are not optional add-ons in product footprinting — the governing standards require them, with differences in emphasis worth understanding. The background uncertainty data and pedigree provenance trace to the ecoinvent LCI database, the source of the matrix factors used here.
What each framework requires for sensitivity and uncertainty
| Framework | Sensitivity analysis | Uncertainty analysis | Data quality |
|---|---|---|---|
| ISO 14040/14044 | Required in interpretation — explicit sensitivity check on significant inputs and methodological choices | Required where relevant; consistency and completeness checks mandated | Data quality requirements specified (time, geography, technology coverage) |
| ISO 14067 | Inherits ISO 14044 interpretation; sensitivity of significant assumptions assessed and reported | Uncertainty addressed and documented; methods not prescribed | Data quality assessed per ISO 14044; pedigree-style approaches accepted |
| GHG Protocol Product | Sensitivity analysis recommended to identify influential assumptions | Dedicated uncertainty chapter; both analytical and Monte Carlo approaches described and disclosure expected | Data quality indicators defined; qualitative and quantitative assessment |
| PEF / EF method | Sensitivity of choices assessed | Uncertainty acknowledged; not the primary quality gate | Four-dimension DQR averaged to a 1–5 score with named bands (distinct system — see above) |
The common thread is that every framework treats the diagnostic layer as part of a credible footprint, not a separate exercise. The GHG Protocol Product Standard is the most explicit on uncertainty propagation, describing both the analytical and Monte Carlo approaches this calculator reports. ISO 14067 leans on the ISO 14044 interpretation phase for its structure. PEF is the outlier on data quality, using its own DQR rather than a pedigree-to-distribution approach.
Worked Example — Full Diagnostic on One PCF
This example runs the three analyses on a single five-line cradle-to-grave footprint. It mirrors the state the calculator boots with on load, so you can reproduce every figure in the tool above. All contributions are in kg CO2e against a functional unit of one product unit.
Settings: unit kg CO2e · functional unit “1 unit” · boundary cradle-to-grave (A–C with Module D shown separately) · confidence 95% · Monte Carlo 10,000 runs · seed 12345 · inputs independent · GWP basis AR6-100.
| Line | Stage | Contribution (kg CO₂e) | Share of gross | Data quality |
|---|---|---|---|---|
| Primary materials | A1 Raw materials | 4.2 | 56.0% | Medium |
| Manufacturing energy | A3 Manufacturing | 1.8 | 24.0% | High |
| End-of-life | C End-of-life | 0.9 | 12.0% | Low |
| Inbound transport | A2 Inbound transport | 0.6 | 8.0% | Low |
| Recycling credit | D Beyond boundary | −0.5 | separate line | Medium |
Hotspot. Gross footprint = 4.2 + 1.8 + 0.9 + 0.6 = 7.5 kg CO2e. Primary materials at 56% and manufacturing energy at 24% together reach 80% of the gross — the two-line Pareto target. The Module D recycling credit of −0.5 kg is reported separately, giving a net result of 7.0 kg CO2e per unit (gross 7.5 less the 0.5 credit), stated as gross-plus-credit rather than a single netted number.
Sensitivity. Swinging each line across its confidence interval, primary materials produces the widest tornado bar — it is both the largest contributor and carries a Medium data-quality rating, so its plausible range is wide in absolute terms. Manufacturing energy, despite being the second-largest contributor, swings less because its High data quality gives it a tight range.
Uncertainty. Here the rankings diverge. The two Low-quality lines — end-of-life and inbound transport — punch above their contribution weight in the variance ranking, because a Low rating maps to a wider GSD. End-of-life is only 12% of the footprint but, with Low data quality, contributes disproportionately to output variance. This is the worked example’s central lesson: the line to address for reduction is primary materials; the line to address for better data is end-of-life.
Read the precise Monte Carlo confidence interval, coefficient of variation, and per-line variance shares off the live tool — they are reproducible bit-for-bit at seed 12345 and 10,000 runs. The qualitative result holds regardless of the exact digits: largest contributor and largest uncertainty driver are different lines, and a serious analysis acts on both.
Common Errors in PCF Uncertainty Reporting
Uncertainty and sensitivity analysis is where product footprints most often fall short of their governing standards — not through arithmetic mistakes but through method errors that pass unnoticed until a reviewer or a critic finds them. The eight below are the recurring ones.
- Reporting a symmetric ± interval on a lognormal footprint. Inventory data is right-skewed, so the true confidence interval is asymmetric about the best estimate. A symmetric ± understates the upper tail — the direction that matters most for risk.
- Confusing the contribution ranking with the variance ranking. Collecting primary data for the largest contributor when a smaller, proxy-based line drives the variance leaves the confidence interval unchanged. The two rankings target two different budgets.
- Treating the ecoinvent pedigree matrix and the PEF DQR as interchangeable. One produces a distribution for propagation; the other produces a quality score for a gate. A pedigree GSD is not a DQR and cannot be substituted for one.
- Netting Module D and biogenic credits into the gross before analysis. Credits and removals are reported on a separate line; folding them into the gross before ranking distorts both the contribution shares and the variance attribution.
- Omitting the sensitivity of methodological choices. Allocation method and system boundary often move a PCF more than any activity-data value. A sensitivity analysis that swings only the numbers and leaves the method choices fixed misses the dominant levers.
- Running Monte Carlo without disclosing seed and run count. Results are reproducible only at a fixed seed and run count. A reported confidence interval without those parameters cannot be reproduced or verified.
- Quoting absolute standard deviation instead of coefficient of variation. Absolute spread is not comparable across footprints of different magnitudes. The coefficient of variation normalises spread against the mean and is the comparable robustness metric.
- Ignoring the analytical-versus-Monte-Carlo divergence. A large gap between the two approaches is diagnostic — it signals the footprint is skewed enough that the closed-form interval understates the spread. Reporting only the analytical interval in that case is misleading.
Data Sources, Provenance & Method Constants
This is a factor-less calculator: it reads no emission-factor data and computes no footprint. Its only constants are the uncertainty-method parameters, all of which are cited engine constants rather than looked-up values.
- Pedigree matrix factors — the ecoinvent pedigree matrix (Frischknecht et al. 2005; Weidema & Wesnæs 1996), verified against the Brightway reference implementation. Five indicators, each mapping a 1–5 score to an uncertainty factor, as tabulated above.
- Basic uncertainty factor — 1.05, applied to every line in addition to the pedigree factors, per the ecoinvent method.
- GSD aggregation — σ = √(Σ[ln U]² + [ln 1.05]²); GSD = exp(σ). Lognormal, the default for strictly-positive inventory data.
- Confidence z-factors — 1.645 (90%), 1.959964 (95%), 2.575829 (99%).
- Monte Carlo — configurable run count (1k / 10k / 100k) and random seed (default 12345); results reproducible bit-for-bit at a fixed seed and run count.
Standards governing the analyses are ISO 14067, ISO 14040/14044, and the GHG Protocol Product Standard. The PEF DQR bands referenced for contrast are part of the European Product Environmental Footprint method and are not used by this calculator.
Frequently Asked Questions
They answer three different questions about the same footprint. Hotspot analysis ranks each process by its share of the total, telling you where emissions concentrate. Sensitivity analysis swings each assumption across its range to see which moves the result most, telling you which choices you must defend. Uncertainty analysis propagates per-input data quality through Monte Carlo to a confidence interval, telling you how wide the error bars are. The largest contributor is often not the largest source of uncertainty, which is why the three are run and reported separately.
Yes. This is a diagnostic layer, not a footprinting tool — it takes per-process contributions as input and does not look up emission factors or compute a footprint from activity data. Build the footprint first in the ISO 14067 cradle-to-gate PCF calculator, an LCA tool, or a spreadsheet, then enter the per-line contributions here to run the sensitivity, uncertainty, and hotspot analyses on it.
The ecoinvent pedigree matrix: five indicators (reliability, completeness, temporal, geographical, and further technological correlation), each scored 1–5 and mapped to an uncertainty factor, aggregated with a basic uncertainty factor of 1.05 into a lognormal geometric standard deviation. It does not use the PEF/EF Data Quality Rating, which averages four dimensions into a single 1–5 quality score. The two are different systems — pedigree produces a distribution for Monte Carlo propagation; the PEF DQR produces a quality score for a compliance gate — and they are not interconvertible.
Because contribution and uncertainty are independent properties of a line. A large contributor backed by a supplier-specific dataset can be both dominant and well-characterised, contributing little to the confidence interval. A small contributor based on a geographically or temporally mismatched proxy can barely register in the contribution ranking yet drive the output variance. The calculator reports a contribution ranking and a separate variance ranking for exactly this reason: reduction effort follows the first, data-collection effort follows the second.
The analytical approach combines per-line uncertainties with closed-form error-propagation formulae — fast and exact for simple sums, but it assumes independent inputs and a near-normal output. Monte Carlo samples every input from its own distribution thousands of times and builds the output distribution empirically, capturing skew and asymmetry the analytical method misses. The GHG Protocol Product Standard describes both. The calculator reports each and flags the divergence between them; a large divergence indicates the footprint is skewed enough that the analytical interval understates the true spread.
Because inventory data is modelled as lognormal — emission factors are strictly positive and right-skewed, so a value can be several times too high but cannot fall below zero. Propagating lognormal inputs produces a right-skewed output: the mean sits above the median and the upper tail runs longer than the lower. The resulting confidence interval is therefore wider on the upside than the downside. Reporting a symmetric ± interval on such a footprint understates the upper-tail risk, which is usually the direction that matters.
Disclose the random seed and the run count alongside the confidence interval. Monte Carlo output is reproducible bit-for-bit only at a fixed seed and run count; the calculator defaults to seed 12345 and offers 1,000, 10,000, or 100,000 runs. A reported interval without those parameters cannot be reproduced or verified by a reviewer. Quote the coefficient of variation rather than the absolute standard deviation when comparing footprints of different magnitudes, since it normalises spread against the mean.
No. Those are methodological choices you make before building the footprint, and they are often the largest sensitivities in a PCF. This tool analyses the footprint you give it under the choices you have already made. To test how a footprint moves between allocation methods, use the PCF allocation methods calculator; to test how it moves between ISO 14067, GHG Protocol Product, and PEF conventions, use the PCF frameworks comparison calculator.
Methodology Notes and Limitations
Diagnostic only — quality is bounded by your inputs. The calculator analyses the contributions and pedigree scores you enter; it does not reach into the underlying LCA model. A confident-looking confidence interval built on optimistic pedigree scores is still optimistic. The diagnosis is only as good as the honesty of the data-quality ratings.
Pedigree matrix, not PEF DQR. Uncertainty is quantified with the ecoinvent pedigree matrix producing lognormal GSDs. The PEF/EF four-dimension DQR is a separate system and is not computed here; if you report under PEF you compute the DQR separately for compliance.
Lognormal default. Pedigree-scored lines are treated as lognormal, the standard assumption for strictly-positive inventory data. Where a different distribution is known, the advanced drawer accepts explicit normal, triangular, or uniform parameters per line.
Independence assumption. The default analysis treats inputs as independent. Real inventories often have correlated inputs (a shared electricity factor across several processes); the analysis settings offer a fully-correlated mode, and the analytical-versus-Monte-Carlo divergence note helps flag when independence is doing more work than it should.
Monte Carlo reproducibility. Results are exact and reproducible at a fixed seed and run count, and shift slightly at higher run counts as the sampling converges. Published figures should be read off the live tool at the seed and run count being reported.
Build the footprint first, then test how firmly it holds.