Using TROPOMI-Based Estimation of Daily Ozone Ground Levels to Assess the Impact of COVID-19 on Ozone Concentrations in China Open Access

Xu, Muwu (Spring 2021)

Permanent URL:



Using TROPOMI-Based Estimation of Daily Ozone Ground Levels to Assess the Impact of COVID-19 on Ozone Concentrations in China


By Muwu Xu

Background: While China’s strict quarantine policy during the COVID-19 pandemic reduced the formation of ozone precursors from transportation and industrial sectors, ground observations have reported an increase in ozone levels in many Chinese cities. However, few studies have been able to evaluate the ozone level change in the entire country.


Methods and Materials: We developed a machine learning model using nadir ground level ozone column from the TROPOspheric Monitoring Instrument (TROPOMI), ozone profiles from the Ozone Monitoring Instrument (OMI), metrological parameters, and land use data to estimate full-coverage daily ground ozone concentrations across China at 0.05° spatial resolution.


Results: We built two separate models for the pandemic year (11/2019 – 04/2020) and the reference year (11/2018 – 04/2019), respectively. There were 209, 654 daily measurements from a total of 1, 500 AQS monitor during the study period. The out of bag R2 was 86.7% in the reference year model and 90.06% in the pandemic year model. During the phase of lockdown in Covid-19 (Jan/23/2020-Feb/13/2020) , defined as high level quarantine phase, a significant increase of concentration of ozone took place comparing to the concentration in pre-lockdown episode (Jan/1/2020to Jan/22/2020) in China (95% CI: 9.80 μg/m3, 9.88 μg/m3; p < 0.0001).

Conclusion: Our study demonstrates the possibility and utilization of TROPOMI product for modelling Ozone at a fine spatial and temporal resolution, which will allow us for construction of long-term daily Ozone measurements at 5km2 spatial resolution and support further epidemiological and environmental studies about ground Ozone.

Table of Contents

Table of Contents

1. Introduction:

2.Data and methods:

2.1 Study Design

2.2 Datasets:

3. Results and discussion:

3.1 Descriptive statistics of the training dataset:

3.2 Random forest model performance and cross-validation:

3.3 Importance ranking of model predictors

3.4 Spatial and temporal trends of O3 predictions during COVID:


Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.


Table 1.

Table 2.


Figure S1:

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
Subfield / Discipline
  • English
Research Field
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files