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

Heavy metal contamination in riverine and coastal sediments is a persistent environmental challenge. This study assesses eight trace metals (Cd, Cr, Cu, Fe, Mn, Ni, Pb, Zn) in surface sediments collected annually from 2000 to 2025 at four monitoring sites in the Tangier-Tetouan coastal corridor, Morocco: Oued Moghogha, Oued Negro, Oued Souani, and Rejet Fnideq. The analytical framework combines descriptive statistics, non-parametric trend analysis (Mann-Kendall, Theil-Sen), inter-metal correlation, principal component analysis (PCA), hierarchical cluster analysis (HCA), and multi-model forecasting (naive drift, linear trend, ARIMA, random forest, XGBoost). The four sites maintained distinct contamination signatures: Oued Moghogha enriched in Cr, Mn, and Ni; Oued Negro characterized by Cu; Oued Souani dominated by Pb; and Rejet Fnideq enriched in Cd and Zn. A regime shift around 2016-2017 affected Pb and Zn most strongly, while Cr and Mn declined consistently. Multivariate analyses revealed three metal associations (Cr-Mn-Ni, Cd-Zn, Pb-dominated) and annual observations clustered by sector. Forecasting for 2026-2030 suggests that the post-2017 regime will persist, with continued decline or stabilization for Cr, Cu, Mn, and Ni, and sustained elevated concentrations for Pb and Zn. Naive drift performed best for most metals, while random forest was preferred for Mn; Fe forecasts showed poor reliability. These findings demonstrate the value of long-term multi-site monitoring and combined statistical and forecasting approaches for understanding sediment contamination dynamics.