Data Quality & Uncertainty

The first thing to internalise is that there is no single true carbon factor for a material - only a distribution of values that depend on how, where and when it was made. Take "concrete": its carbon per cubic metre depends on the cement content, the type of cement, how much of the clinker was replaced with fly ash or slag, the aggregates, the water content and the manufacturing energy - so two concretes both honestly called the same grade can differ substantially in carbon. "Steel" varies even more starkly, because steel made from iron ore in a coal-fired blast furnace carries roughly several times the carbon of steel recycled from scrap in an electric arc furnace running on a clean grid. The material name tells you little; the production route tells you almost everything.

On top of production route sit three amplifiers of variation. Region is the largest: because making materials is energy-intensive, the carbon reflects the local energy grid and process mix, so the same product made in a country with a coal-heavy grid can carry markedly more carbon than one made where the grid is cleaner - a first-order issue for India, whose grid and cement and steel routes differ from the European averages that populate many databases. Time matters because industry is (slowly) decarbonising and because measurement methods improve, so a factor's vintage affects its value. And method and boundary matter, because whether a factor counts only the factory stages or also transport, or how it allocates the credit for recycled content, changes the number even for the identical physical product.

The consequence is that any carbon factor you use is a representative estimate with a spread around it, not a constant. This is not a flaw to be embarrassed about; it is the nature of the thing, shared with most environmental data. But it means the honest question is never "what is the carbon factor for concrete?" - it is "which concrete, made where, by what route, measured how, and how sure are we?" Everything about data quality flows from taking that variability seriously rather than collapsing it into a single confident digit.

Given that variation, the central choice in data quality is between generic and specific factors, and knowing when each is appropriate. A generic factor is an industry or database average for a material category - "concrete, C30", "structural steel", "float glass" - drawn from sources like the ICE database or national datasets. It represents a typical product, is easy to get, and is perfectly suited to early estimates and to materials that do not dominate your total. Its weakness is precisely that it is an average: it cannot capture that you specified a low-clinker mix or high-recycled-content steel, so if your design choice is exactly to use a better-than-average product, a generic factor will hide the very saving you made.

A product-specific factor comes from an Environmental Product Declaration - a verified, standard-based document stating a particular manufacturer's product carbon. Where you can get one for a material that matters, it is far more accurate and lets you claim a real, specified improvement rather than an assumed average. Its costs are that EPDs are not available for every product (and are especially patchy for Indian-made materials), that reading and comparing them correctly is a skill in itself, covered in Module 2.3, and that an EPD is still bounded by its own scope and vintage. Between the two extremes sits a ladder - global average, national average, manufacturer average, product-specific - and moving up it improves quality at the cost of effort and availability.

The discipline, then, is to spend your data-quality effort where it changes the answer. Use good product-specific data for the few lines that dominate the total and for any material whose low-carbon credentials you intend to claim; accept generic averages for the long tail where precision would not move the result. Record, for every factor, which rung of the ladder it came from and its source, region and date - because a total built from a mix of verified EPDs and rough proxies is only as honest as your willingness to say which lines are which. A specified saving backed by a generic average is not yet a saving you can honestly claim.

Once you accept that every factor carries a spread, the logical form of a carbon result changes: it stops being a point and becomes a range. A building is not "412.7 kgCO2e/m2"; it is "roughly 350 to 520, best estimate around 430", where the width of that range reflects how much of the total rests on shaky data. Reporting the range is not hedging - it is the accurate statement, and it is far more useful, because it tells the reader how much weight the number can bear. A narrow range says the big lines are backed by good data and the answer is firm; a wide one says treat this as directional. Collapsing that to a single decimal throws away the most important information you have.

The tool for understanding your range is sensitivity analysis, and it is simpler than it sounds. Take the two or three lines that dominate the total, swing each one across a plausible low-to-high span of its factor, and see how much the total moves. This does two things. It tells you how uncertain the answer really is - if a plausible swing in the concrete factor moves the whole building by fifteen per cent, your total is exactly that soft. And it tells you where better data would pay - the lines that most move the total are the ones worth chasing an EPD for, while a line that barely moves it can stay a rough proxy. Sensitivity turns a vague sense of doubt into a targeted plan for firming the number up.

This reframes the whole exercise around the big, uncertain items rather than false completeness. It is far better to know your total to plus-or-minus twenty per cent and to know exactly which three factors drive that spread than to quote a single confident number whose fragility is invisible. It also disciplines comparisons: if two design options differ by five per cent but each carries a twenty per cent uncertainty, the honest conclusion is that they are indistinguishable on carbon and you should decide on other grounds - a call you can only make if you carry the ranges rather than the points. Uncertainty, made explicit, is not weakness; it is what lets you use numbers responsibly.

All of this converges on a single professional habit: matching your claim to your certainty, and never exceeding it. False precision - quoting a figure to several digits when the underlying data supports maybe two - is the quiet form of dishonesty in carbon work, because it dresses an estimate as a measurement and invites a confidence nobody has earned. The louder form is over-claiming a reduction: asserting a building or product is some exact percentage better when that gap sits well inside the combined uncertainty, or claiming a saving computed from a generic average that could never have detected it. Both mislead, and both are avoidable simply by being honest about what the data can and cannot say.

The honest practice has a few concrete moves. Report results as ranges with a stated best estimate, not as false points. Always state the boundary and the key assumptions, so a reader knows what the number covers. Label the data quality - which lines are product-specific, which are generic proxies - so the mix is visible. When comparing options, ask whether the difference exceeds the uncertainty before you call one better. And resist the pressure, which is real, to produce a single clean, impressive figure for a brochure or a rating when the data only justifies a caveated range. Saying "about this, give or take, and here is what would firm it up" is not weaker than a confident decimal; it is more credible, and it is true.

This is where data quality rejoins the course's spine. Carbon claims must be earned by measurement, and a claim is only as good as the data beneath it, so being honest about data quality and uncertainty is the same discipline as being honest about greenwash - practised at the level of a single number. It is also what protects you: a caveated, well-documented range stands up to scrutiny, while a confident figure built on hidden proxies collapses the moment someone asks how you know. And the binding, reported claim - a declared, certified figure a project stands behind - still belongs to a qualified assessor working to the standards with documented, verified data and a stated uncertainty, not to a tidy number pulled from a tool. Humility about the data is not a lack of rigour; it is what rigour actually looks like.