0
Letter ? AI-assigned paper type based on the abstract. Classification may not be perfect — flag errors using the feedback button. Sign in to save

Scaling laboratory results with machine learning is no silver bullet to strengthen global (micro)plastic mitigation policy

Journal of Chest Surgery 2025 2 citations ? Citation count from OpenAlex, updated daily. May differ slightly from the publisher's own count.
Elke Brandes, Peter Fiener, Arthur Geßler

Summary

This commentary challenged a study estimating 4.11–13.52% global crop yield losses from microplastic and nanoplastic (MP/NP) pollution, arguing that the machine learning model was trained predominantly on seedlings in controlled hydroponics at high concentrations rather than field-grown mature crops. The authors identified three fundamental gaps: methodological bias in model training data, neglect of natural MP/NP concentrations in aged soil, and failure to account for complex crop yield drivers, questioning the validity of upscaling laboratory results to global hunger and sustainability policy.

Study Type Review

Zhu et al. ( 1 ) estimated "annual global losses of 4.11 to 13.52% [] for main crops" due to microplastic (MP) and nanoplastic (NP) pollution and concluded that "These findings underscore the urgency of integrating plastic mitigation into global hunger and sustainability initiatives."These alarming results triggered substantial press echo.However, we challenge the legitimacy of such statements, given the fundamental gaps in i) process understanding of MP/NP interaction with plants in natural systems and ii) data limitations regarding global MP/NP contamination.Drawing on our expertise, we highlight three main points concerning the impact of MP/ NP on terrestrial photosynthesis ( Fig. 1 ). Methodological Bias in Model Training and UpscalingMost studies included in the meta-analysis investigated physiological mechanisms under controlled conditions, typically growing seedlings in hydroponic setups exposed to high concentrations of pristine MP/NP ( Fig. 2 ).Only 11% of data points stemmed from soil-grown maize, rice, or wheat cultivated to maturity.This raises concerns about training a machine learning-based effect size model and applying it to real-world conditions.Such data neglect natural concentrations of aged NP/MP in soils and the complexity of crop yield drivers, including water and nutrient availability, plant-microbial interactions, and crop genetics.

Share this paper