Preprints
https://doi.org/10.5194/gmd-2020-433
https://doi.org/10.5194/gmd-2020-433

Submitted as: model description paper 23 Apr 2021

Submitted as: model description paper | 23 Apr 2021

Review status: this preprint is currently under review for the journal GMD.

Air Control Toolbox (ACT_v1.0): a machine learning flexible surrogate model to explore mitigation scenarios in air quality forecasts

Augustin Colette1, Laurence Rouïl1, Frédérik Meleux1, Vincent Lemaire1,a, and Blandine Raux1 Augustin Colette et al.
  • 1Institut National de l’Environnement Industriel et des Risques (INERIS), Parc Alata, BP2, 60550 Verneuil-enHalatte, France
  • anow at: Amplisim, 96b Boulevard Raspail, 75006 Paris, France

Abstract. We introduce the first toolbox that allows exploring the benefit of air pollution mitigation scenarios in the every-day air quality forecasts through a web interface. Chemistry-transport models (CTMs) are required to forecast air pollution episodes and assess the benefit that shall be expected from mitigation strategies. However, their complexity prohibits offering a high level of flexibility. The Air Control Toolbox relies on machine learning methods to cope with this limitation. It consists of a surrogate model trained on a limited set of sensitivity scenarios to allow exploring any combination of mitigation measures. As such we take the best of the physical and chemical complexity of CTMs, operated on high performance computers for the everyday forecast, but we approximate a simplified response function that can be operated through a website to emulate the main sensitivities of the atmospheric system for a given day and location.

The numerical experimental plan to design the structure of the surrogate model is detailed by increasing level of complexity. The selected structure of the surrogate is a quadrivariate polynomial of first order for residential heating emissions, and second order for agriculture, industry and traffic emissions with three interaction terms. It is fitted to 12 sensitivity CTM simulations, at each grid point and every day for PM10, PM2.5, O3 (both as daily mean and daily maximum) and NO2. The validation study demonstrates that we can keep relative errors below 2 % at 95 % of the grid points and days for all pollutants. Various applications of the toolbox are presented for air quality episode analysis, source apportionment, and chemical regime analysis.

Augustin Colette et al.

Status: open (until 18 Jun 2021)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Augustin Colette et al.

Augustin Colette et al.

Viewed

Total article views: 186 (including HTML, PDF, and XML)
HTML PDF XML Total Supplement BibTeX EndNote
149 34 3 186 11 1 3
  • HTML: 149
  • PDF: 34
  • XML: 3
  • Total: 186
  • Supplement: 11
  • BibTeX: 1
  • EndNote: 3
Views and downloads (calculated since 23 Apr 2021)
Cumulative views and downloads (calculated since 23 Apr 2021)

Viewed (geographical distribution)

Total article views: 174 (including HTML, PDF, and XML) Thereof 174 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 14 May 2021
Download
Short summary
We introduce the first toolbox that allows exploring the benefit of air pollution mitigation scenarios in the every-day air quality forecasts through a web interface. The toolbox relies on the joint use of Chemistry-transport models (CTMs) and machine learning approaches.