Spatio-temporal data on the air pollutant nitrogen dioxide derived from Sentinel satellite for France.

Monitoring of air pollution is an important task in public health. Availability of data is often hindered by the paucity of the ground monitoring station network. We present here a new spatio-temporal dataset collected and processed from the Sentinel 5P remote sensing platform. As an example application, we applied the full workflow to process measurements of nitrogen dioxide (NO2) collected over the territory of mainland France from May 2018 to June 2019. The data stack generated is daily measurements at a 4 × 7 km spatial resolution. The supplementary Python code package used to collect and process the data is made publicly available. The dataset provided in this article is of value for policy-makers and health assessment.

Data description

Remote sensing data for air quality monitoring is important for health research [1]. The advantage of remotely sensed air pollution data includes, but is not limited to, large coverage at a useful spatial and temporal resolution. Sentinel 5P is a rather new remote sensing data source but requires downloading and computationally intensive processing that is often a barrier to public use. For illustrative purposes, we focus on generation of a NO2 spatio-temporal measurements across mainland France. The shared workflow repository, however, contains other air pollutants including ozone (O3), sulfur dioxide (SO2), carbon monoxide (CO) (for details, see the GitHub page [2]). The spatial resolution of the measurements (3.5 × 7 km2 for all trace gases, except for CO and Methane (CH4) that is 7 × 7 km2) allows observations and mapping of air pollution at a finer scale (e.g. at the scale of an administrative area) (see Table 1 and Fig. 5).

Table 1

Average tropospheric vertical column of NO2 by administrative area.

Administrative area	Average tropospheric vertical column of NO2 ( × 1019 molec/m2)
Corse	1.782
Île-de-France	3.923
Hauts-de-France	3.345
Nouvelle Aquitaine	1.670
Normandie	2.445
Pays de la Loire	1.985
Centre-Val de Loire	2.208
Grand Est	2.913
Provences-Alpes-Côtes d’Azur	2.170
Bretagne	1.922
Bourgogne-France-Comté	2.174
Occitanie	1.659
Auvergne-Rhône-Alpes	1.951

Note: Row data is publicly accessible on Mendeley repository [4].

Fig. 5

Annual average of NO2 measured between May 2018 and June 2019 by administrative area.

The data consist of a netCDF file containing Sentinel 5P measurements between May 2018 and June 2019, with multiple attributes (e.g., latitude/longitude, WGS84 projection, and date of measurement), allowing both spatial and temporal observations of air pollutants. We cleaned the data using a quality flag parameter (noted by ‘qa_value’ varying between 0 (no data) and 1 (full quality data). We used ‘qa_value’ above 0.5 provided by the Copernicus Sentinel 5 Precursor Tropospheric Monitoring Instrument (S5p/TROPOMI [3]) to filter cloud cover and so to ensure high quality data. The dense measurements performed every 24 hours allow accurate annual averaging at fine-grained spatial resolution as shown in Fig. 1.

Fig. 1

Annual average of NO2 concentrations in France between May 2018 and June 2019.

The data were also grouped by date allowing temporal assessment of pollutants density and quantitative observations such as monthly pollutants distribution, both spatially and temporally (Fig. 2). Data allow also analysis for weekday variations (Fig. 3).

Fig. 2

Distribution of NO2 measurements collected over France between May 2018 and June 2019.

Fig. 3

Distribution NO2 measurements collected over France by weekday and weekend.

Fig. 4 illustrates that the seasons fall and winter (October to February) face higher NO2 pollutants in the Northern and Eastern part of France. This spatial contrast reduces to a point where only city urban areas show strong pollution (e.g., June 2019 with Paris).

Fig. 4

Spatial distribution of NO2 measurements across France by month.

Experimental design, materials, and methods

Satellite measurements

Launched in October 2017 by the European Space Agency (ESA), Copernicus Sentinel 5P [5] monitors the density of several atmospheric gases, aerosols, and cloud distributions affecting air quality and climate. The measurements are made by the state of the art instrument called TROPOspheric Monitoring Instrument (TROPOMI). The TROPOMI is a multispectral imaging spectrometer that detects solar radiation reflected or scattered back to space from Earth's atmosphere and surface. As the spectral fingerprint of each target atmospheric trace gas is known, its concentration can be calculated through the identification of the unique fingerprints of these constituents in different part of the electromagnetic spectrum. Sentinel 5P is able to achieve global coverage every 24 hours, giving access to dense measurements over the entire globe. TROPOMI has more spectral bands than its predecessors: ultraviolet and visible (270–500 nm), near-infrared (675-77 nm), and shortwave infrared (2305–2385 nm). This allows TROPOMI to measure a wider range of atmospheric trace gases such as nitrogen dioxide (NO2), ozone (O3), sulphur dioxide (SO2), methane (CH4), and carbon monoxide (CO). In addition, it observes clouds and aerosols-related parameters, which can be fed into the retrieval algorithms of trace gases [3]. The list of standard S5P/TROPOMI L2 products is given in Table 2.

Table 2

List of S5P/TROPOMI level 2 data products.

Product	Main Parameter	(Planned) released
UV aerosol index	Aerosol index	Released
Aerosol layer height	Mid-level pressure	Mid-2019
Carbon monoxide (CO)	Total column	Released
Cloud	Fraction, albedo, top pressure	Released
Formaldehyde (HCHO)	Total column	Released
Methane (CH4)	Total column	Released
Nitrogen dioxide (NO2)	Total, tropospheric, stratospheric column	Released
Ozone profiles	Total and tropospheric profiles	Late-2019
Sulfur dioxide (SO2)	Total column	Released
Ozone (O3)	Total column	Released
Tropospheric ozone (O3)	Tropospheric column	In development
Ultraviolet (UV)	Surface irradiance erythemal dose	In development

Overall, the dataset provides:

geolocated total columns of ozone, sulfur dioxide, nitrogen dioxide, carbon monoxide, formaldehyde and methane-geolocated cloud and absorbing aerosol index

other products are under development and will made available at a later date. They include geolocated tropospheric columns of ozone, geolocated vertical profiles of ozone, aerosol layer height, ultraviolet index, etc.

All of the mission's measurements of atmospheric gases and aerosols are ‘column data’, which means they cover the full depth of the atmosphere. For some gases, advanced techniques and algorithms like ozone profile retrieval, the convective cloud differential, and the cloud slicing methods allow to have access to ‘tropospheric column densities’, ‘stratospheric column densities’ and ‘vertical density profiles’. When available, these variables are included in the dataset.

TROPOMI has a very high spatial resolution (3.5 × 7 km2 for all trace gases, except for CO and CH4 that is 7 × 7 km2). Further, TROPOMI has an improved signal-to-noise ratio (2–5%) for measurements under low albedo conditions. Data gaps are documented by Copernicus and can be consulted on the mission page.

Processing workflow

For further analysis of TROPOMI data, we produced an aggregated product on a regular grid with spatial resolution of 3.5 × 7 km (0.01 × 0.01 arc degrees). Every orbit that TROPOMI measures has a different spatial distribution of grid cells, depending on the viewing zenith angle at the moment of the observation. For this reason, we resampled each product for the area of interest on this single grid and binned the dataset by latitude/longitude WGS84 projection. The quality of the individual observations depends on many factors, including cloud cover, surface albedo, presence of snow-ice, saturation, geometry etc. In Sentinel 5P, a layer summarizing the different factors affecting the quality of the measurements is provided. This aggregate measure called 'quality assurance value' (‘qa_value’) can be used to screen poor quality pixel. This ‘qa_value’ is a continuous variable, ranging from 0 (no date) to 1 (all is well). To filter out errors and problematic retrievals we excluded measurements with a 'qa_value' < 0.5 following the Copernicus specifications [3].

Specifications Table

Subject	Environmental science
Specific subject area	Pollution
Type of data	TableImageDatasetPython code
How data were acquired	The data was collected via the Sentinel Hub API.
Data format	Processed raw dataAnalyzedFiltered
Parameters for data collection	The daily air pollution data was collected from May 2018 to June 2019.
Description of data collection	All the data was obtained from the Sentinel 5P satellite using the application programming interface. The collected data are processed by a Python code.
Data source location	France
Data accessibility	Data is supplied on Mendeley (Public repository)Repository name: https://doi.org/10.17632/zxwc456xy9.5

Value of the Data• The collected air pollution data offer a (cost-)effective and scalable source for advancing the monitoring of NO2 at a large scale. • Policy makers can generate yearly, monthly, or even near real-time datasets on air pollution maps to support government efforts in reducing air pollution. • The automated workflow to process Sentinel 5P data is transferable to any other study area on the globe. • Satellite data can be merged with ground-based measurement or survey data.