Scholarlink

<p>Forecasting the spatio-temporal evolution of atmospheric pollutants is a critical challenge for air quality management and public health. Traditional approaches based on Gaussian, Lagrangian, and Eulerian frameworks rely on advection–diffusion equations and meteorological forcing, while recent advances integrate data assimilation and artificial intelligence to improve prediction accuracy. In this work, we propose a novel methodology that combines physics-based modelling and data-driven techniques within a unified framework. The governing equation is decomposed into three contributions: advection, diffusion, and external influences (gap). Advection is addressed using a semi-Lagrangian scheme, diffusion is identified through Dynamic Mode Decomposition (DMD), and the gap is estimated via multilinear regression applied to Proper Orthogonal Decomposition (POD) coordinates of external variables such as temperature, pressure, and water vapour. The approach is validated on a domain centred around Singapore using Copernicus data with a spatial resolution of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(0.75^\circ \)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0</mn> <mo>.</mo> <msup> <mn>75</mn> <mo>∘</mo> </msup> </mrow> </math></EquationSource> </InlineEquation> and a temporal resolution of 3&#xa0;h over the entire year 2024. Despite its apparent complexity, the three-pass learning process achieves predictions in less than one second per iteration, enabling near real-time forecasting. Results demonstrate a monotonic decrease in error norms across successive passes, confirming the robustness of the method. This hybrid strategy opens promising perspectives for operational air quality forecasting systems capable of handling multiple chemical constituents under complex meteorological conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid physics: data framework for real-time forecasting of atmospheric pollutants

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月3日

查看全文
  • Amine Ammar,
  • Francisco Chinesta

摘要

Forecasting the spatio-temporal evolution of atmospheric pollutants is a critical challenge for air quality management and public health. Traditional approaches based on Gaussian, Lagrangian, and Eulerian frameworks rely on advection–diffusion equations and meteorological forcing, while recent advances integrate data assimilation and artificial intelligence to improve prediction accuracy. In this work, we propose a novel methodology that combines physics-based modelling and data-driven techniques within a unified framework. The governing equation is decomposed into three contributions: advection, diffusion, and external influences (gap). Advection is addressed using a semi-Lagrangian scheme, diffusion is identified through Dynamic Mode Decomposition (DMD), and the gap is estimated via multilinear regression applied to Proper Orthogonal Decomposition (POD) coordinates of external variables such as temperature, pressure, and water vapour. The approach is validated on a domain centred around Singapore using Copernicus data with a spatial resolution of \(0.75^\circ \) 0 . 75 ∘ and a temporal resolution of 3 h over the entire year 2024. Despite its apparent complexity, the three-pass learning process achieves predictions in less than one second per iteration, enabling near real-time forecasting. Results demonstrate a monotonic decrease in error norms across successive passes, confirming the robustness of the method. This hybrid strategy opens promising perspectives for operational air quality forecasting systems capable of handling multiple chemical constituents under complex meteorological conditions.