AI turns 26 sediment samples into a map of an estuary’s entire plastic legacy

Every year, staggering quantities of plastic debris fragment into microscopic particles that drift through rivers, accumulate in lakes and ultimately settle into the sediments of coastal waters. These aquatic sediments are widely regarded as the primary long-term sinks for microplastic pollution, preserving a physical record of humanity’s plastic age layer by layer, much like tree rings record past climates. Yet despite their importance, scientists have struggled to answer a deceptively simple question: how much microplastic is actually stored in the sediments of an entire estuary, bay or coastal region? Direct measurements are slow, expensive and laborious, so most studies rely on a handful of samples that cannot possibly represent the chaotic patchwork of an estuarine environment. A new open-source tool called NIXVEGS, developed by a German research team and described in the journal Microplastics and Nanoplastics, promises to change that calculus dramatically by letting a modest set of measurements paint a region-wide picture.

The study, led by Kristina Enders and Matthias Labrenz of the Leibniz Institute for Baltic Sea Research Warnemünde, together with colleagues from the Leibniz Institute of Polymer Research in Dresden, Christian-Albrechts-University of Kiel and the Federal Waterways Engineering and Research Institute in Hamburg, focused on the Schlei, a narrow estuary on the Baltic Sea coast of Northern Germany. The Schlei is a hydrodynamically complex environment where currents, wind mixing and variable salinity conspire to deposit particles in unpredictable patterns. Existing models of microplastic distribution, the researchers note, have typically been designed for ocean basins or simple hydrodynamic systems and simply lack the flexibility to cope with the fine-grained spatial heterogeneity that estuaries present. That gap is precisely what NIXVEGS, short for a machine-learning pipeline for predicting microplastic concentrations from sediment proxies, was built to fill.

At its heart, NIXVEGS is an empirical modeling pipeline that adapts well-established machine learning components to the peculiar constraints of environmental microplastic science. The central difficulty is data scarcity: microplastic analysis is so demanding that even thorough studies often yield only a few dozen observations, far fewer than conventional data-hungry algorithms expect. Rather than pretending this problem away, the team built their pipeline around it. NIXVEGS employs nested cross-validation, a rigorous scheme in which the data are repeatedly split so that model selection and performance assessment happen on strictly separate subsets, ensuring that reported accuracy reflects genuine generalization rather than memorization. On top of that, the tool uses ensemble modeling, combining multiple predictive models so that their individual weaknesses partially cancel out, a strategy that stabilizes predictions when training data are sparse and noisy.

What makes the approach particularly clever is its choice of predictor variables. Instead of requiring new microplastic measurements, NIXVEGS leans on granulometric proxies, which are the routinely measured physical characteristics of sediments such as grain size distributions. The underlying logic is that microplastic particles behave physically like sediment grains of comparable size and density, so the same hydrodynamic forces that sort sand from silt and clay also sort plastic fragments and fibers across the seafloor. By training the model to learn the statistical relationship between sediment texture and measured microplastic concentrations, and by incorporating measures of spatio-temporal connectivity between sampling locations, the pipeline can then predict concentrations at locations where only standard geological data exist. This transforms cheap, widely available sedimentological information into a window on invisible plastic contamination.

The payoff in the Schlei estuary was substantial. From an original set of just 26 sediment samples in which microplastic concentrations had been directly analyzed, the model’s predictions extended spatial data coverage by a factor of 7.6, effectively interpolating the plastic landscape across the estuary far beyond the reach of laboratory budgets. Each prediction was validated against data the model had never seen during training, which is exactly the kind of honest performance testing that environmental modeling often lacks. The result is a continuous regional map of microplastic concentrations in the Schlei’s sediments, revealing the plastic legacy of the estuary at a resolution that direct sampling alone could never achieve economically.

With that map in hand, the team took a further ambitious step: converting particle counts into an actual mass budget for the region. Using a three-dimensional particle shape-to-mass conversion, which accounts for the geometry of individual particles identified in the samples, they estimated that the sediments of the Schlei hold roughly 20 trillion microplastic particles in the 50 to 5000 micrometer size range, weighing approximately 14.3 tons. That figure represents the first region-wide sedimentary inventory for this estuary and offers a concrete sense of scale for what has until now been an abstract pollution problem. Fourteen tons of microscopic plastic resting quietly on and within the seafloor of a single estuary illustrates why sediments matter so much in the global plastic budget.

The technical significance of the work extends well beyond the Baltic coast. Many of the ideas embedded in NIXVEGS, including nested cross-validation and ensemble approaches, are standard fare in machine learning but have been underused in environmental microplastic research, where methodological rigor is difficult when datasets are small. By explicitly engineering the pipeline for sparse data and by documenting validation on unseen observations, the authors have set a template for how predictive modeling should be done in this field. Equally important, NIXVEGS is open source, meaning that research groups anywhere can inspect, adapt and apply the same workflow to their own water bodies without paying for proprietary software or reinventing the statistical machinery from scratch.

The practical implications for management are equally compelling. Estuarine and coastal sediments serve as long-term archives of pollution evolution, so reliable regional inventories allow authorities to identify contamination hotspots, prioritize dredging or remediation efforts, and track how plastic loads respond to policy interventions such as bans on single-use items. Because the model relies on proxies that sedimentologists and hydrographic offices already collect, extending the approach to other estuaries requires far less investment than building new monitoring programs around direct microplastic measurement. The researchers suggest that NIXVEGS can assist regional plastic management while also contributing data needed to constrain the global microplastic budget, a quantity that remains stubbornly uncertain because so few regions have been inventoried at all.

There are, of course, inherent limitations to any empirical approach. NIXVEGS learns relationships from the data it is given, so its predictions inherit the uncertainties of the original measurements, the analytical methods used to count and identify particles, and the representativeness of the sampling design. Machine learning can interpolate skillfully within the range of observed conditions, but extrapolating to estuaries with very different hydrodynamics or pollution histories would require retraining with local data. The tool is best understood not as a replacement for direct measurement but as a force multiplier, one that squeezes maximum spatial insight from every expensive laboratory analysis performed.

Even so, the study marks a meaningful moment in the effort to account for humanity’s plastic legacy. A field that has long been limited to scattered point measurements now has a demonstrated pathway toward continuous regional maps, built from data that already exist in geological archives and monitoring databases. If similar approaches are rolled out across the world’s estuaries and coastal waters, the murky picture of where microplastics have accumulated could sharpen into something that policymakers, modelers and the public can actually see and act upon. The plastic hidden in the seafloor is no longer entirely invisible; with the right algorithms, twenty-six samples can speak for an entire estuary.

Subject of Research: Machine learning-based prediction of microplastic distributions and inventories in estuarine sediments using sediment proxies

Article Title: Mapping the plastic legacy with NIXVEGS: machine-learning enabled microplastics prediction in sediments

Article References: Enders, K., Lenz, R., Fischer, F., Schwarzer, K., Seiß, G., Fischer, D., & Labrenz, M. (2026). Mapping the plastic legacy with NIXVEGS: machine-learning enabled microplastics prediction in sediments. Microplastics and Nanoplastics. https://doi.org/10.1186/s43591-026-00236-y

Image Credits: AI Generated

DOI: 10.1186/s43591-026-00236-y

Keywords: microplastics, sediments, machine learning, NIXVEGS, estuary, Schlei, Baltic Sea, cross-validation, ensemble modeling, grain size proxies, pollution mapping, environmental monitoring