Despite the ubiquitous and persistent presence of microplastic (MP) in marine ecosystems, knowledge of its potential harmful ecological effects is low. In this work, we assessed the risk of floating MP (1 μm-5 mm) to marine ecosystems by comparing ambient concentrations in the global ocean with available ecotoxicity data. The integration of twenty-three species-specific effect threshold concentration data in a species sensitivity distribution yielded a median unacceptable level of 1.21 ∗ 105 MP m-³ (95% CI: 7.99 ∗ 103-1.49 ∗ 106 MP m-³). We found that in 2010 for 0.17% of the surface layer (0-5 m) of the global ocean a threatening risk would occur. By 2050 and 2100, this fraction increases to 0.52% and 1.62%, respectively, according to the worst-case predicted future plastic discharge into the ocean. Our results reveal a spatial and multidecadal variability of MP-related risk at the global ocean surface. For example, we have identified the Mediterranean Sea and the Yellow Sea as hotspots of marine microplastic risks already now and even more pronounced in future decades.

Microplastic pollution is an issue of concern due to the accumulation rates in the marine environment combined with the limited knowledge about their abundance, distribution and associated environmental impacts. However, surveying and monitoring microplastics in the environment can be time consuming and costly. The development of cost- and time-effective methods is imperative to overcome some of the current critical bottlenecks in microplastic detection and identification, and to advance microplastics research. Here, an innovative approach for microplastic analysis is presented that combines the advantages of high-throughput screening with those of automation. The proposed approach used Red Green Blue (RGB) data extracted from photos of Nile red-fluorescently stained microplastics (50-1200 μm) to train and validate a 'Plastic Detection Model' (PDM) and a 'Polymer Identification Model' (PIM). These two supervised machine learning models predicted with high accuracy the plastic or natural origin of particles (95.8%), and the polymer types of the microplastics (88.1%). The applicability of the PDM and the PIM was demonstrated by successfully using the models to detect (92.7%) and identify (80%) plastic particles in spiked environmental samples that underwent laboratorial processing. The classification models represent a semi-automated, high-throughput and reproducible method to characterize microplastics in a straightforward, cost- and time-effective yet reliable way.