Related Experiment Video
Updated: Aug 6, 2026
09:10
Multimodal Analysis of Microplastics in Drinking Water using a Silicon Nanomembrane Analysis Pipeline
Published on: June 13, 2025
Unraveling Microplastics in Lakes with a Mechanism-Informed, Few-Shot Data-Driven Liquid Neural Networks
Yihan Li1,2, Hua Wang1,2, Yichuan Zeng1,2
1Key Laboratory of Integrated Regulation and Resource Development on Shallow Lake of Ministry of Education, College of Environment, Hohai University, Nanjing210024, China.
Environmental Science & Technology
|July 20, 2026
Summary
This study introduces a novel data-driven framework for simulating microplastic (MP) transport in aquatic environments. The approach enhances prediction accuracy using limited data, offering a transferable solution for environmental pollutant modeling.
Area of Science:
- Environmental Science
- Computational Fluid Dynamics
- Machine Learning
Background:
- Simulating microplastic transport is challenged by physical model errors and sparse observational data for machine learning.
- Existing methods struggle with error accumulation and discontinuous field data in aquatic environments.
Purpose of the Study:
- To develop a mechanism-informed, few-shot data-driven framework for accurate microplastic transport prediction.
- To address the limitations of purely physical models and data-scarce machine learning approaches.
Main Methods:
- Utilized Liquid time-constant networks, a continuous-time neural ordinary differential equation model.
- Integrated the MIKE21 hydrodynamic particle tracking model to generate physics-informed training data.
- Employed a physically constrained closed-form continuous-time (CfC) neural network trained on sparse observations.
Main Results:
- Achieved high R2 values (0.867 dry, 0.895 wet seasons) in Poyang Lake simulations.
- Demonstrated superior performance over benchmarks like Random Forest.
- Identified key drivers: flow velocity, water depth, and suspended sediment concentration via SHAP analysis.
Conclusions:
- The proposed framework effectively simulates microplastic transport in data-sparse lacustrine environments.
- This paradigm offers a transferable solution for environmental pollutant prediction and simulation.
- Continuous-time neural networks show promise for environmental modeling with limited data.
