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

Abstract

Aquatic sediments act as major sinks for microplastic (MP) pollution, yet accurately quantifying their MP distributions and inventories on a regional scale remains challenging, especially in hydrodynamically complex estuarine environments. Existing distribution models focus on ocean basins or simple hydrodynamic systems but lack sufficient adaptability to the spatial heterogeneity typical of estuaries. To address this limitation, we developed NIXVEGS, an open-source tool that predicts regional MP distributions from readily obtainable sediment proxies. It is an empirical pipeline that adapts established machine learning components, such as nested cross-validation and ensemble modeling, to the sparse data typical of environmental MP studies. By integrating granulometric proxies and spatio-temporal connectivity, it combines data-driven model selection with performance validation on unseen data, making the best possible use of the available observations. The model predictions increased spatial data coverage 7.6-fold from an original set of 26 analyzed MP sediment concentration samples in the Schlei estuary (Baltic Sea coast, Northern Germany). We demonstrated the application of NIXVEGS and used its predictions to estimate the region’s sedimentary MP inventory at ~ 20 trillion particles (50–5000 μm) or ~ 14.3 tons, based on a 3D particle shape-to-mass conversion. Estuarine sediments are key environments for monitoring MP pollution evolution, where NIXVEGS can assist regional plastic management and contribute to the understanding of a global MP budget. We developed NIXVEGS, a predictive modeling tool leveraging readily obtainable sediment proxies, to significantly enhance the spatial prediction of microplastic concentrations from limited direct measurements.