AI Models Track 25 Years of Toxic Metals in Moroccan River Sediments

Along the Mediterranean coast of northern Morocco, where the cities of Tangier and Tetouan drain their wastewater into a handful of small river mouths, researchers have now assembled one of the most complete long-term portraits of sediment contamination anywhere in North Africa. A new study published in Environmental Monitoring and Assessment traces the concentrations of eight trace metals—cadmium, chromium, copper, iron, manganese, nickel, lead, and zinc—in surface sediments collected annually from 2000 to 2025 at four monitoring sites in the Tangier–Tetouan coastal corridor. By combining classic environmental statistics with modern machine learning forecasts, the work offers something rare in contamination research: not just a record of what has happened, but a quantified projection of what is likely to happen next.

The four sites examined in the study are far more than arbitrary sampling points. Each receives effluent from a distinct urban and industrial catchment, and each has developed its own chemical fingerprint over the quarter-century of observation. Oued Moghogha has been consistently enriched in chromium, manganese, and nickel, a pattern the researchers link to industrial discharges typical of metal-processing and tanning activities in its watershed. Oued Negro stands out for copper, while Oued Souani is dominated by lead. The Rejet Fnideq site, near the coastal town of Fnideq, shows a signature enriched in cadmium and zinc, metals commonly associated with urban runoff, galvanized surfaces, and mixed domestic and commercial waste streams. The persistence of these distinct signatures across twenty-five years of sampling demonstrates how tightly sediment chemistry mirrors the character of the land use feeding each outfall.

Methodologically, the team layered several complementary analytical techniques. Descriptive statistics established baseline concentrations and variability for each metal at each site. Non-parametric trend analysis using the Mann–Kendall test, paired with Theil–Sen slope estimation, allowed the researchers to detect monotonic increases or decreases in contamination over time without requiring the data to follow a normal distribution—a crucial consideration for environmental monitoring data, which are frequently skewed and censored. Inter-metal correlation analysis, principal component analysis, and hierarchical cluster analysis then served as multivariate lenses, grouping metals that behave together and clustering years and sites that share similar contamination profiles. Finally, five forecasting models—a naive drift method, a linear trend model, the autoregressive integrated moving average approach known as ARIMA, and two machine learning algorithms, random forest and XGBoost—were trained on the historical records to project concentrations through 2030.

One of the most striking findings is the identification of a regime shift around 2016–2017. Before this transition, several metals, most notably lead and zinc, followed one trajectory; after it, they settled into a distinctly different pattern. A regime shift of this kind typically signals a structural change in the sources or transport of contaminants—perhaps altered wastewater treatment, changes in industrial activity, or shifts in sediment transport driven by rainfall and coastal dynamics. In the Tangier–Tetouan corridor, the shift affected lead and zinc most strongly, while chromium and manganese told the opposite story: they declined consistently across the monitoring period, suggesting sustained reductions in the inputs driving their enrichment.

The multivariate analyses sharpened the picture considerably. Principal component analysis revealed three coherent metal associations—chromium–manganese–nickel, cadmium–zinc, and a lead-dominated grouping—that map cleanly onto the site-specific contamination signatures. These associations are more than statistical convenience. Metals that co-vary in sediments often share common sources or similar geochemical behavior, so identifying a chromium–manganese–nickel cluster at one site and a cadmium–zinc cluster at another provides circumstantial evidence about where the contamination originates and how it might respond to changes in land use. Hierarchical cluster analysis of the annual observations similarly grouped years by sector rather than scattering them randomly, confirming that each monitoring location has maintained its own contamination identity throughout the entire record.

The forecasting results carry a clear practical message. Projections for 2026 to 2030 suggest that the post-2017 regime will persist: chromium, copper, manganese, and nickel are expected to continue declining or to stabilize, while lead and zinc are forecast to remain at sustained elevated concentrations at the affected sites. For environmental managers in the Tangier–Tetouan region, this asymmetry matters. It implies that whatever processes reduced the first group of metals are working, while the sources behind lead and zinc contamination either remain active or reflect legacy pollution locked in the sediment that will not dissipate on its own. Forecasting, in this sense, converts a monitoring archive into an early-warning tool.

Equally instructive is the study’s candid assessment of its own models. Counterintuitively, the simplest method—a naive drift forecast that extrapolates recent rates of change—performed best for most of the metals, while random forest was the preferred model for manganese. Iron forecasts proved unreliable across the board, likely because iron behaves largely as a crustal, geogenic element whose concentrations reflect natural sediment mineralogy more than anthropogenic inputs, making its temporal behavior noisy and difficult to predict from short annual series. The lesson echoes a broader finding in the forecasting literature: with limited observations and relatively smooth temporal dynamics, sophisticated algorithms do not automatically outperform simple baselines. Comparing multiple models, as this study did, is essential to avoid placing unwarranted confidence in any single projection.

The broader significance of the work lies in its demonstration that long-term, multi-site monitoring combined with statistical and machine learning tools can disentangle the complex dynamics of coastal pollution. Heavy metals in sediments are persistent by nature: unlike organic pollutants, they do not degrade, and they accumulate in fine-grained particles where benthic organisms can ingest them, entering food webs that extend to fisheries consumed by millions of people along the Mediterranean. Regional reviews of heavy metal pollution in Mediterranean sediments have repeatedly flagged urbanized coastal corridors like Tangier–Tetouan as hotspots, and earlier studies of Tangier Bay and the Tetouan coast documented concerning enrichment decades ago. What the new study adds is temporal resolution—the ability to say not merely that contamination exists, but whether it is improving, worsening, or holding steady, and for which metals and where.

For Morocco, where rapid coastal urbanization, tourism development, and industrial expansion continue to place pressure on the Mediterranean littoral, the findings arrive at a consequential moment. The forecast of persistent lead and zinc elevation suggests that targeted interventions—improved wastewater treatment at the responsible outfalls, control of urban runoff, or remediation of contaminated hotspots—would yield the greatest benefit at the Oued Souani and Rejet Fnideq systems. Meanwhile, the consistent declines in chromium and manganese offer tentative evidence that past management measures or changes in industrial discharge have had measurable effects. Sustaining the annual monitoring program that made these conclusions possible is itself a policy recommendation embedded in the research: without a continuous, standardized record, neither the 2016–2017 regime shift nor the site-specific fingerprints would ever have been detectable, and the machine learning forecasts would have had no foundation on which to stand.

The study also contributes methodologically to a growing global movement that treats environmental monitoring data as a resource for predictive modeling. As climate change alters rainfall patterns and sediment delivery, and as coastal cities across the Mediterranean and beyond continue to grow, the ability to forecast contamination trajectories—even with simple, transparent models—will become an increasingly important component of adaptive environmental management. The Tangier–Tetouan record, spanning a quarter century and four contrasting catchments, stands as a template for how such forecasting can be built from long-term fieldwork, rigorous statistics, and a healthy skepticism about complexity for its own sake.

Subject of Research: Long-term spatiotemporal analysis and machine learning forecasting of heavy metal contamination in Mediterranean riverine sediments in the Tangier–Tetouan region of Morocco

Article Title: Spatiotemporal dynamics and machine learning forecasting of heavy metal contamination in Mediterranean riverine sediments: a case study of the Tangier–Tetouan region, Morocco

Article References: El Hir, S., El Macouti, N. E. H., Maanan, M., Lagrat, I., & Ait Errouhi, A. (2026). Spatiotemporal dynamics and machine learning forecasting of heavy metal contamination in Mediterranean riverine sediments: a case study of the Tangier–Tetouan region, Morocco. Environmental Monitoring and Assessment, 198(10), Article 1065. https://doi.org/10.1007/s10661-026-15913-x

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15913-x

Keywords: heavy metals, sediment contamination, machine learning, forecasting, Mediterranean, Morocco, Tangier, Tetouan, Mann–Kendall trend analysis, random forest, XGBoost, environmental monitoring