A health-and-environment comparison of drinks usually collapses into a single ranking, and yet that ranking rests on two hidden decisions: what goes into the denominator, and how heavily each of the four environmental components is weighted. This page is a byproduct of the drink–health–footprint research thread — an interactive chart on which the same 100-beverage dataset rearranges itself as you move the weights and switch the denominator.
Ranking beverages looks like a property of the drinks, and it is largely a property of the measurement frame. Measured per litre, wine sits at the top of the carbon-footprint list; per serving — because you don't guzzle it — it slides to the bottom: the same fact reverses when the denominator changes. The environmental side compounds this, because water, CO₂, soil humus and biodiversity arrive in different units and can only be combined once you choose how much each one counts. Whatever default you pick is therefore an editorial claim, not a neutral measurement.
The two measured axes rest on established datasets — carbon on Poore & Nemecek's 2018 global life-cycle synthesis, water on the Water Footprint Network figures of Mekonnen & Hoekstra. The two remaining axes, soil humus and biodiversity, are modelled, because clean per-beverage numbers for them do not exist. The map below wears that difference openly through the transparency of each point.
The vertical axis is health, the horizontal axis is the environmental footprint, and both are weighted blends you control. On the left, the four footprint sliders mix water, CO₂, humus and biodiversity; behind the advanced panel, four more sliders split health into bodily, metabolic, mental and nutrition, so the vertical axis becomes a choice too rather than a fixed verdict. The top-left corner is the sweet spot: healthy and light. Move the sliders, try the presets, and flip the denominator between serving and litre to watch the ranking rearrange.
How much should each factor count? The horizontal axis is their weighted mix.
Per serving or per litre? This switch reverses the ranking — that is the lesson.
The health axis is not one number either. Split it into four parts and weight them yourself. Nutrition subtracts, because it is a benefit rather than a harm.
Lean on the mental axis and the whole map fades — that axis carries the weakest individual-level evidence.
On the Interference stage the map stops being a private tool and becomes a shared instrument. A researcher — a life-cycle analyst or a nutrition scientist — sits beside an artist, and the projected chart responds live while the room decides, together, which weighting feels honest. When someone pushes "water only" and the dairy and wine points slide to the far right, the move is visible to everyone at once, and the argument that follows is the point of the evening rather than a footnote to it. Péter Csermely, as intellectual curator of the series, frames the underlying question that the sliders only make tangible: whether a single number can ever be fair to a thing that serves so many different purposes.
Carbon and water values are drawn from peer-reviewed literature; the humus (soil organic matter) and biodiversity axes are modelled 0–100 scores derived from roughly ten production-system archetypes — cane and beet sugar, cereal, grape, orchard, paddy rice, pasture and confinement dairy, legume, tree-nut, and shade-versus-sun coffee. Health is expressed as burden per standard serving and stays fixed under the denominator switch. Point opacity encodes relative data confidence, with the least certain points shown at maximum transparency. These are illustrative estimates assembled for public understanding, not an audited life-cycle assessment.
Common backbone. Both scores follow one logic: impact per litre = land occupation × per-area impact factor, then rescaled 0–100 across the whole dataset. Land occupation (O, m²·yr per litre) comes from the product-level land-use data in Poore & Nemecek (2018). The final values are ordinal 0–100 scores calibrated to the literature below, not a live per-supply-chain LCA computation, which is why the more heavily modelled points are shown fainter.
Humus (soil organic matter), H. H = rescale( O × f_LU × f_MG × f_I ). The land-use factor f_LU is the relative soil-organic-carbon (SOC) stock of that system versus native vegetation, calibrated from Guo & Gifford (2002) — forest→cropland loses roughly 42% of SOC, forest→pasture gains about 8% — and the IPCC Tier-1 SOC stock-change factors. The management factor f_MG (tillage, residue removal, fallow) and input factor f_I (cover crops, rotation, manure) follow the same IPCC scheme and meta-analyses such as Poeplau & Don (2015) and Conant et al. (2017). The global magnitude is framed by Sanderman et al. (2017): human land use has run up a soil-carbon debt on the order of 116–133 Pg C.
| Archetype | H band |
|---|---|
| Cane / beet sugar (annual, erosive) | 65–70 |
| Grape · rice | ~50 |
| Cereal · tree-nut · dairy | 35–45 |
| Orchard · shade coffee | ~35 |
| Legume (soy, pea) | ~25 |
Biodiversity, B. B = rescale( O × CF_bio × V_region ). CF_bio is the land-use characterization factor as Potentially Disappeared Fraction of species (PDF) per m², from the countryside species-area relationship (cSAR) of Chaudhary et al. (2015), which spans 804 ecoregions, five taxa and six land-use types, refined by management intensity in Chaudhary & Brooks (2018) and consolidated in LC-IMPACT — the method the UNEP Life Cycle Initiative currently recommends for land-use ecosystem-quality damage; Newbold et al. (2015, PREDICTS) corroborates the direction. V_region is an ecoregional vulnerability multiplier built from endemic richness and IUCN threat level, which is why the same crop scores worse in a biodiverse, water-scarce region (e.g. Californian almonds) than in a species-poorer temperate one.
| Archetype | B band |
|---|---|
| Cocoa (tropical, deforestation) | 65–68 |
| Tree-nut in arid biodiverse region | 55–60 |
| Cane sugar · dairy · sun coffee | 45–55 |
| Grape · cereal · soy | 35–45 |
| Oat · pea (temperate rotation) | 28–30 |
Normalisation and uncertainty. Both H and B are min–max rescaled to 0–100 so they sit alongside the water and CO₂ axes. Each point's uncertainty reflects how well the feedstock and sourcing region are characterised and how heterogeneous the archetype is; that uncertainty drives the opacity, and the mental health weight feeds into it as well.