Spatio-temporal dynamics and explainable machine learning classification of microplastics in the eastern tropical Pacific: Evidence from a decade-long (2008-2018) particle-level dataset.

Microplastic pollution is an increasing threat to marine ecosystems worldwide. Here we present a detailed particle-level analysis of a unique long-term dataset (Bermúdez, 2025) containing 16,179 individually measured microplastic particles from 141 plankton-net samples collected over 11 years (2008-2018) at four stations approximately 20 km offshore of mainland Ecuador in the Eastern Tropical Pacific. Using explainable machine learning, we show that microplastic type (Fiber vs. Film) can be predicted from morphological features with substantial skill (Matthews correlation coefficient 0.63, balanced accuracy 0.88, and average precision of 0.81 on the minority Film class against a prevalence baseline of 0.16). SHAP interpretability analysis revealed aspect ratio and particle size as the dominant predictors. Linear trend estimation with paired bootstrap confidence intervals detected no statistically significant change in concentration at any of the four stations; every interval spans zero, and the two southern stations with the largest positive point estimates (Pto López, +0.030; Salinas, +0.026 #/L yr-1) are not distinguishable from no change. Gaussian Process regression was used for visualisation only, not for significance testing. Unsupervised clustering further distinguished two morphologically distinct particle groups differing in size and elongation: small films and large fibres. This structure is driven by the morphological type label included among the clustering features. With that label removed the silhouette criterion selects seven clusters and agreement with the two-group partition falls to an adjusted Rand index of 0.07, so we report the clustering as a negative result and make no source attribution. This study demonstrates the value of combining long-term monitoring data with machine learning approaches to extract additional information from archived monitoring material. The results strengthen the evidence base for targeted mitigation measures in tropical coastal regions and highlight the importance of sustained monitoring programmes.