Introducing the Freshwater Ecosystem Condition Index: A 0–100 eDNA Score for UK Rivers and Streams

Freshwater ecosystems are under more pressure than at any point in recorded history. Monitored freshwater species populations have fallen by an average of 85% since 1970. Yet the data infrastructure available to organisations responsible for managing, restoring or reporting on UK rivers and streams has remained largely unchanged for decades: labour-intensive kick-net surveys, chemical spot-checks, and a regulatory monitoring network too sparse and too infrequent to support the scale of investment and disclosure now required.

Launching today, the Freshwater Ecosystem Condition Index (ECI) is the first commercially available freshwater eDNA ecosystem condition tool in the UK — a 0–100 score of river and stream ecological condition derived from just a single litre of water, grounded in the biological signal of the macroinvertebrate community, and calibrated against UK Water Framework Directive (WFD) status data. A full technical white paper detailing the modelling methodology and validation is in preparation and will be published shortly.

Why now — and why NatureMetrics

This is something we have been working towards for several years. Developing a robust, nationally applicable freshwater condition index from eDNA is not straightforward — the biological signal is complex, the regulatory standards demanding, and the modelling requires a training dataset of sufficient scale, consistency and quality to be genuinely reliable.

That last point matters more than it might appear. Most freshwater eDNA research has been built on data collected by multiple groups, using different sampling kits, laboratory protocols, and bioinformatic pipelines. Combining those datasets introduces noise that is difficult to separate from genuine ecological signal. NatureMetrics is in a fundamentally different position. Over the past decade, we have built one of the largest proprietary freshwater eDNA archives in the world — collected using the same kits, the same field protocols, and the same laboratory and sequencing processes throughout. That consistency is the scientific foundation that makes a reliable, nationally applicable model possible.

‍

“I am extremely proud of what the skilled ecologists and data scientists at NatureMetrics have accomplished. We have been working towards this for years, and what makes it possible is something most people overlook: our entire training dataset was collected using the same kits, the same protocols, and the same laboratory processes from the very beginning. That uniformity is a vast difference from the multi-group research datasets most models are built on — and it is what gives us confidence in the score.”

Dr. Bastian Egeter, Ecology Director, NatureMetrics

‍

“To build this index, we stopped trying to force eDNA data into conventional metrics and instead leaned into what the eDNA data was telling us that conventional methods couldn’t. Terrestrial species’ DNA in the water as evidence of agricultural run-off, fine-scale taxonomic resolution to differentiate sensitive groups like Chironomids, and sediment-dwelling meiofauna as indicators. A species community tells you a lot about what habitat is available, how a river/stream is physically structured, and whether pollution-sensitive species are likely absent. Alongside all this new signal, we’re also able to leverage data on conventional indicator species like mayflies, stoneflies, caddisflies (EPT) to ground and validate our model alongside decades of ecological knowledge.

What’s also exciting is that we’ve adopted modelling concepts from the world of language models to sidestep any dependency on reference sequence databases. Put simply, our model doesn’t need to know it’s looking at DNA from a stonefly to know it’s looking at data from a sensitive species. Incomplete reference sequence databases are when working with eDNA data, but novel modelling approaches like this avoid the problem entirely. We’re able to use the taxonomic annotations that are produced to understand how the model thinks and makes decision, but importantly they’re not required as an input to the decision-making process.”

Dr. James Whiting, Quantitative Ecologist, NatureMetrics

‍

What is the Freshwater Ecosystem Condition Index?

The Freshwater ECI converts a single 1-litre water sample into a 0–100 score of river or stream ecological condition. The score is derived from environmental DNA (eDNA) — the genetic material that organisms continuously shed into the water as they live, feed and move. By sequencing and analysing that genetic signal, we can characterise the macroinvertebrate community present across a river reach without anyone entering the water.

The launch model is built on macroinvertebrate eDNA: the genetic signal of invertebrate orders including Ephemeroptera (mayflies), Plecoptera (stoneflies), Trichoptera (caddisflies)and Diptera (true flies), alongside other invertebrate families detected in the water column. This is deliberate. Macroinvertebrates are the backbone of UK freshwater ecological assessment. They are sensitive to a wide range of pressures — nutrient loading, organic pollution, sedimentation, abstraction — and they integrate those pressures over time in a way that chemical sampling cannot. They sit at the heart of WFD ecological status assessment for precisely these reasons, and they are the taxon group for which our training dataset is deepest and most robust.

The score maps directly to WFD ecological status bands, giving teams a biologically grounded, regulatory-aligned output that is immediately interpretable without specialist expertise. Work is already underway to extend the index to additional species groups, including fish, as datasets meet our strict statistical confidence thresholds.

How the Freshwater ECI works

The index rests on three components.

1. Macroinvertebrate detection through eDNA metabarcoding

A field operator filters one litre of water from the bank. No waders, no disturbance to the riverbed, no specialist entomology. The filter is sent to our laboratory, where we extract, amplify and sequence the macroinvertebrate DNA present in the sample. The result is a community profile — a quantified picture of which invertebrate families are present and at what relative abundance across a reach of approximately two kilometres upstream.

2. Machine learning calibrated against WFD status data

The community profile is passed to a neural network model trained on NatureMetrics’ proprietary eDNA database with verified WFD ecological status labels. The model was developed following a rigorous pipeline that forces it to learn generalisable representations of species communities across the water quality gradient. Model training and performance was explicitly evaluated on data from geographic regions that the model had never “seen” during training; this minimises the risk of overfitting.

Nature is complex and there’s no single species community that represents High or Bad water quality across the extent of Britain. Neural networks allow for multiple representations of “good” and “bad” communities to be categorised in a way where the model knows that ahigh quality community from the Welsh lowlands is different from a high quality community in the Scottish highlands. The model framework is ordinal, meaning the model explicitly respects the ecological ordering of the WFD status bands.

The model architecture takes DNA sequences, not taxonomic labels like species names or families, as input. This means the model is able to use 100% of the eDNA data generated, avoiding the usual requirement to assign DNA to species based on taxonomic reference databases. These databases are incomplete and closely-related species can be difficult to differentiate where reference sequences are highly similar. The model has also been developed from the outset to offer explainable outputs, giving ecologically-interpretable insights and avoiding the “black box” issue.

In blind testing on hundreds of held-out samples, 85% of predictions landed within one WFD status band of the true value. Critically, the families the model associates with high and low ecological status align closely with decades of established freshwater bioassessment science: families such as Chloroperlidae (sallflies) and Brachycentridae (humpless casemaker caddisflies) at the clean-water end; Corixidae (water boatmen), Physidae (bladder snails) and Glossiphoniidae(freshwater leeches) at the degraded end. The model was given no access to ecological trait databases or expert sensitivity scores — it learned those associations from the eDNA and status data alone. That alignment gives us strong confidence it is capturing genuine ecological signal rather than statistical artefact. Full validation details will be set out in the forthcoming technical white paper.

3. A standardised, scalable condition score

Results are expressed as a 0–100 score that maps to WFD ecological status bands, making them directly comparable across sites, seasons and years. Scores aggregate from reach to catchment to portfolio level, so condition can be reported at the scale decisions are actually made — whether that is a single abstraction point, a WINEP investigation zone, or a corporate nature disclosure covering hundreds of sites.

Because sampling requires no in-water entry and minimal specialist skill, the method scales in a way that conventional kick-net surveys cannot. Operational field teams, environmental contractors and citizen science partners can all collect samples to the same standard, dramatically reducing the cost per data point while expanding geographic coverage.

Expanding beyond Great Britain

The Freshwater ECI launches with Great Britain coverage, where our training data density is highest and the WFD regulatory framework provides a robust validation standard. But freshwater biodiversity loss is a global challenge, and we are confident in our ability to extend the index to other regions as the necessary reference datasets are established.

If you are working in a region not yet covered and have a need for catchment-scale freshwater condition monitoring, we would very much like to hear from you. The methodological framework is in place; geographic expansion is a question of data, and that is something we can build together with the right partners and clients.
‍

Talk to our team about applying Freshwater ECI to your project: Contact Us

‍