Soil toxic metals — agricultural threshold, worst metal

Source: Hou et al. (2025), Science — "Global soil pollution by toxic metals threatens agriculture and human health" Year: ~2020 Category: Physical risks · Enabling services · Soil condition Coverage: Global Format: Raster grid (~2 km) Used in risk analysis: No (contextual layer; not used in materiality scoring)

What it shows

The probability, between 0 and 1, that at least one of seven toxic metals in the soil exceeds its agricultural threshold: arsenic, cadmium, cobalt, chromium, copper, nickel and lead.

The value in a cell is the maximum over the seven metals, which is what "worst metal" means in the title. It answers "is any of these metals likely to be over the line here", not "how much metal is in the soil" and not "which metal it is". It is the agricultural twin of Soil toxic metals — human-health threshold, worst metal: same metals, same construction, a different set of thresholds, so the two surfaces do not have to agree on the same ground.

How it is built

Hou et al. (2025) train a machine-learning model on 796,084 sampling points compiled from 1,493 regional studies, then predict over roughly 2 million pixels worldwide. The prediction is a probability of exceedance rather than a concentration, which is how the uncertainty of an interpolated soil measurement is carried into the map instead of being hidden behind a single number.

Darwin imports the published surface on a 2 km grid and computes the per-cell maximum across the seven metals. No reclassification, no rescaling and no Darwin threshold are applied on top.

Consistency check at import. On the imported raster, 19.4 % of agricultural cells sit above 0.5, against the "14 to 17 % of croplands" the article reports. The order of magnitude matches; the gap is not reconciled, and it is the reason a cell should be read as a likelihood rather than as a verdict.

How to read it

Read the value as a probability of exceedance in [0, 1], on a sequential scale: 0 is the bottom of the scale, not a neutral midpoint between two directions. Higher means it is more likely that some metal in that soil is above the agricultural threshold, which is the reading that matters for cropland, grazing and sourcing questions.

Two cautions travel with it. It is a screening surface at 2 km, interpolated from point samples, so a cell states a regional likelihood and never substitutes for a site soil test. And it is the worst metal, so a high value does not say which metal, at what concentration, or what a grower would have to do about it.

Class thresholds

The layer publishes no risk classes and no proximity rule. The colour ramp is continuous from 0 to 1, with stops labelled at 0, 0.25, 0.5, 0.75 and 1, and nothing in the platform raises a flag from it.

The only threshold in the underlying science is the agricultural threshold of each metal, which comes from Hou et al. (2025) and their regulatory references, not from Darwin.

Why it is contextual

It is a present state with no trajectory. A Nature Stress Test shock is a difference between a reference and a horizon; with a single point in time there is nothing to subtract, and no shock can be derived without inventing an assumption about how soil metal loads evolve. The Stress Test reads its Soils component from the soil-organic-carbon shock layers, which publish a baseline and scenario horizons; this layer publishes neither.

It therefore sits in the catalogue beside the pesticide layers, as a map to read rather than an input to a computation. Nothing in the engine samples it.

Source

Hou, D. et al. (2025). Global soil pollution by toxic metals threatens agriculture and human health, Science. Data: Dryad.

Comparison with the WWF Risk Filter Suite

There is no equivalent indicator in the WWF Risk Filter Suite. Its soil indicator, S2_1 Soil Condition, is built on soil organic carbon and is answered in Darwin by Soil condition (Global Soil Organic Carbon); soil metal contamination is a separate quantity that WWF does not carry. This layer is a Darwin extension.

Legend

Symbolised field: Exceedance probability, continuous ramp from 0 to 1.

Generated from darwin/layers/layer-soilmetal-at-max.toml (develop).