Analysis of the Sentinel-5P NO2 Data Download Process during 2023 in Chile Joaquı́n Fernández Verdugo June 8, 2025 1 Contents 1 Download and Conversion Methodology 1.1 Initial Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Main Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 3 2 Analysis of Obtained Data 2.1 Technical Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Method Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Detected Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Latitudinal Data Loss . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Spatial Variability Induced by GeoJSON Sources . . . . . . . . . 2.3.3 Source Comparison: Sentinel-5P vs. SINCA . . . . . . . . . . . . 4 4 4 5 5 8 13 3 Technical Recommendations 15 4 Conclusions 15 Abstract This work describes an automated system for downloading and processing satellite data of nitrogen dioxide (NO2 ) obtained from the TROPOMI instrument on the Sentinel-5P satellite, using Python scripts. The extracted data correspond to vertical column densities of NO2 , expressed in mol/m2 , allowing for a continuous spatial representation of the total pollutant load in the atmosphere. Unlike ground stations, which provide surface concentrations (ppb or µg/m3 ), this metric offers a comprehensive view of the atmospheric behavior of the gas. The methodology, operational advantages, and limitations of the system are documented, along with comparisons with ground sources and recommendations for future improvements. 2 1 Download and Conversion Methodology The following procedure was developed to extract, convert, and store satellite nitrogen dioxide (NO2 ) data from the Sentinel Hub platform, with the goal of obtaining estimated concentrations in units comparable to those reported by ground monitoring stations (ppb). 1.1 Initial Configuration • Authentication with Sentinel Hub using personal credentials (Client ID and Client Secret). • Selection of the SENTINEL5P collection corresponding to the TROPOMI instrument. • Definition of a custom evalscript to extract the tropospheric NO2 band and filter observations with minimum quality (qa value ≥ 0.75). 1.2 Main Workflow The complete process can be divided into seven functional stages: 1. Initialization: Input parameters are defined (download dates, label language, path to GeoJSON file). 2. Reading area of interest (AOI): Coordinates of the region of interest are extracted from the GeoJSON file and transformed to the appropriate reference system. 3. Generating request to Sentinel Hub: A SentinelHubRequest object is constructed including temporal, spatial, and processing metadata (resolution, resampling method, etc.). 4. Data download: The query is executed and files in GeoTIFF format are obtained with NO2 column values (in mol/m2 ). 5. Optional conversion to volumetric concentration (µg/m3 ): Assuming a homogeneous tropospheric column height of 10 km, the formula is applied: Cµg/m3 = Ccol · M · 106 H where Ccol is the concentration in mol/m2 , H = 10 000 m and M = 46.01 g/mol. 6. Optional conversion to parts per billion (ppb): Estimation is performed with: Cppb = 24.45 · Cµg/m3 M where 24.45 is the molar volume of an ideal gas at standard conditions (STP). 7. Obtaining statistical values of interest: Indicators such as mean, maximum, and standard deviation are automatically calculated for each region and month. 8. Geographic visualization of obtained concentration: The system generates regional maps showing the spatial distribution of NO2 clearly and georeferenced. 3 9. Result storage: Converted data is stored, and the request object is returned for traceability and eventual auditing. The general process, from reading input parameters to the final unit conversion, is represented in the following diagram: Figure 1: Complete workflow of the NO2 data download, conversion, and storage system from Sentinel Hub. Includes reading the area of interest, query execution, physicochemical processing, and generation of outputs in standardized atmospheric units. Optionally, a spatial interpolation module is incorporated using geostatistical methods (e.g., Kriging), to obtain a continuous representation from the discrete values provided by the sensor. 2 Analysis of Obtained Data 2.1 Technical Specifications Parameter Value Units Spatial resolution Temporal resolution Coverage Output format mol/m2 7 × 3.5 km Daily Global GeoTIFF Table 1: NO2 data specifications 2.2 Method Advantages The implemented system has characteristics that make it especially suitable for systematic and reproducible processing of large-scale satellite NO2 data. Its main advantages are listed below: • Full automation: Allows massive downloads without manual intervention, covering multiple defined regions and periods. Facilitates periodic database updates. 4 • Built-in quality validation: Integrates a quality filter (minQa) that excludes low-reliability observations, reducing the influence of clouds, atmospheric noise, or other artifacts. • Integrated post-processing: Includes an optional spatial interpolation stage (Kriging) that converts sparse data into continuous, comparable, and mappable surfaces. • Structured file organization: Storage follows a hierarchy by year, region, and month, facilitating traceability, exploration, and auditing of results. • GIS interoperability: Outputs are generated in standard formats (GeoTIFF, CSV), compatible with tools like QGIS, ArcGIS, and Python libraries, easing integration with other spatial analyses. • Obtaining statistical values of interest: Metrics such as mean, maximum, and standard deviation are automatically calculated for each region and period. • Geographic visualization of obtained concentration: The system generates regional NO2 distribution maps, ready for visual or communication analyses. 2.3 Detected Limitations 2.3.1 Latitudinal Data Loss Throughout this study, monthly 2023 nitrogen dioxide (NO2 ) data were downloaded for Chile’s 16 regions (2015 administrative division). Although this division is prior, it fully covers the national territory, so it does not limit the validity of the analysis. Number XV I II III IV V XIII VI VII VIII IX XIV X XI XII Region Arica and Parinacota Region Tarapacá Region Antofagasta Region Atacama Region Coquimbo Region Valparaı́so Region Santiago Metropolitan Region Libertador General Bernardo O’Higgins Region Maule Region Biobı́o Region La Araucanı́a Region Los Rı́os Region Los Lagos Region Aysén del General Carlos Ibáñez del Campo Region Magallanes and Chilean Antarctica Region Table 2: Chile’s regional division in 2015 The first diagnosis, shown in Figure 2 and Figure 3, revealed the regional sets most affected by losses. Figure 2 shows the monthly average of obtainable valid data under the established download conditions. It shows that the upper and lower limits of this average 5 were Magallanes and Chilean Antarctica with 17853.6 and Arica and Parinacota with 1241.9 data respectively. On the other hand, Los Rı́os and Arica and Parinacota were the regions most affected by percentage loss compared to the Metropolitan Region, with up to 25% fewer valid pixels than the capital. On the other side, northern regions like Tarapacá and central regions like Valparaı́so exceeded the metropolitan reference. Figure 2: Regional average of valid data for 2023 Figure 3: Percentage loss of valid data in regions relative to the Metropolitan Region A finer inspection of valid pixel density by latitude confirmed a systematic pattern: the correlation coefficient between latitude and density is ρ = −0.82, implying that 6 at higher southern latitudes, satellite coverage degrades. In numbers, southern regions (Aysén, Magallanes) concentrate up to 85% less data than northern regions; the most extreme case was Magallanes, with only 5 valid pixels in June 2023. This variability in the amount of validated data becomes noticeable when studying its behavior throughout the year. The following figure shows the monthly regional count with a diagram for each region. Figure 4: Count of validated samples by month in 2023 for each Region of Chile When analyzing each case individually, an increasing trend of sample numbers towards the end of the year is noticeable; however, a deeper look reveals significant differences in these quantities among the regions studied. For example, Magallanes and Chilean Antarctica shows a volume of 30667 data points in December, corresponding to its fourth highest value recorded that year. On the other hand, Coquimbo peaks in that month but with 4627 data, nearly 85% less. This can be seen in depth from the following images: 7 Figure 5: Dispersion of validated data (Qa = 0.75) for Coquimbo and Magallanes and Chilean Antarctica regions in 2023 The images show the spatial distribution of raw data collected for December, represented in a color scale according to the measured value (in mol/m²). Although the Magallanes and Chilean Antarctica region has a total of 30,667 observations, visual coverage is interrupted by large areas with no apparent information. In contrast, the Coquimbo region presents a visually more continuous and complete map, despite having 85% fewer data than the former. This contrast is particularly striking, as it suggests that not only the total amount of data determines the spatial quality or completeness of the visualization, but also factors such as the distribution and spatial density of valid pixels. Additionally, geographical differences between both regions, with Magallanes located at much more southern latitudes and with a more extensive territory in longitude, could influence effective satellite coverage due to factors such as viewing angle, higher cloud or ice presence, and the satellite’s orbital geometry. This disparity is explained by the observation geometry of the Sentinel-5 Precursor satellite: its orbit is near-polar and sun-synchronous, inclined 98.7° relative to the equator, so that zenith angles increase at latitudinal extremes and reduce effective resolution according to: RESeffective = RESnominal cos(θ) (1) These findings confirm that the coverage of the TROPOMI instrument in Chile has a significant latitudinal bias, particularly in the far south, which must be critically considered in national-scale atmospheric assessments. 2.3.2 Spatial Variability Induced by GeoJSON Sources During satellite data processing for the Aysén region in May 2023, the impact of using different vector files (GeoJSON) as Area of Interest (AOI) delimiters was evaluated. 8 Although both sources officially represent regional boundaries, the obtained results reveal relevant spatial inconsistencies in the analysis of NO2 derived from TROPOMI. Figure 6: NO2 values for Aysén in May 2023 using the first GeoJSON source 9 Figure 7: NO2 values for Aysén in May 2023 using the second GeoJSON source It can be observed that although both figures exhibit comparable spatial distributions concerning magnitude and concentration, the regional delineation notably displays displacements and gaps. Close examination reveals that specific borders contain either missing or overrepresented pixels, especially on the eastern boundary. This indicates a lack of complete coincidence in the projection or geographic definition between the two sources. This difference is also reflected in the statistical metrics of the collected data, as seen in the following distributions: 10 Figure 8: Frequency distributions of NO2 values for Aysén in May 2023 with two GeoJSON sources Both histograms approximate a normal distribution; however, small variations in skewness and dispersion are observed. To quantitatively characterize these differences, the files response statistics jlhonora/geo.csv and response statistics caracena/chile-geojson. were used, whose values are summarized below: Statistic Mean Standard deviation Maximum Minimum N° of Data jlhonora/geo −2.26 × 10−6 6.56 × 10−6 1.9940513 × 10−5 −2.9138968 × 10−5 6053 caracena/chile-geojson −2.24 × 10−6 6.55 × 10−6 1.9940513 × 10−5 −2.9138968 × 10−5 6011 Table 3: Comparative statistics of NO2 for Aysén in May 2023 by GeoJSON source Despite small absolute differences in statistical values, a systematic variation is evident, originating exclusively from the geometric definition of the region of interest. This effect proves particularly sensitive in areas with low valid pixel density or complex geometry, as occurs in southern zones. The Aysén region, located at a high latitude, shows greater sensitivity to variations in spatial definition due to the viewing geometry of the TROPOMI sensor. Variability in borders generates changes in the total number of pixels selected for analysis and can introduce unintentional biases in longitudinal or comparative studies. To visually highlight the geometric differences between both vector sources used, the boundaries of the Aysén region according to definitions available in the repositories jlhonora/geo and caracena/chile-geojson are presented below: 11 Figure 9: Regional polygons from jlhonora/geo Figure 10: Regional polygons from caracena/chile-geojson Superficially, no major differences are apparent between both images. However, this superficial similarity contrasts with the measurable effects observed in previous analyses. Therefore, it is concluded that data derived from remote sensors can be susceptible to noticeable variations even when geometric differences in the area of interest definition seem minimal. This finding highlights the need to adopt a single standardized vector source, with validated projection and explicit compatibility with the reference system used in satellite products. The lack of standardization between AOI files—even for the same administrative unit—can lead to systematic spatial selection errors and thus compromise the comparability and reproducibility of atmospheric analyses. 12 2.3.3 Source Comparison: Sentinel-5P vs. SINCA The Sistema de Información Nacional de Calidad de Aire (SINCA) or National Air Quality Information System in english is an official platform of Chile’s Ministry of the Environment that compiles, validates, and publishes atmospheric pollutant measurements from a network of ground stations distributed throughout the country. These stations operate in real time and report data with high temporal resolution (hourly or daily), under international quality and calibration standards. In particular, the Parque O’Higgins station in Santiago represents one of the most used reference points for evaluating urban NO2 concentration. During 2023, SINCA reported a monthly average NO2 concentration at this station with an annual mean of 22.65 ppb, a monthly maximum of 35.47 ppb, and a minimum of 12.64 ppb (Figure 12). In contrast, the satellite estimate processed from products of the TROPOMI instrument on board the Sentinel-5P satellite provides a considerably lower monthly average for the Metropolitan Region, with a maximum value in May of only 0.76 ppb and an annual mean near 0.23 ppb (Figure 11). Figure 11: Monthly average NO2 concentration in 2023 according to Sentinel-5P for the Metropolitan Region 13 NO2 - registro diario - promedio mensual x1:Parque O’Higgins, 0003, RAT[M], Value Tipo deGrafico: Grafico de Barras Serie Tiempos seleccionada: x2:Parque O’Higgins, 0003, VAL[M], Value 01/01/2023 00 - 31/12/2023 00 Acumulacion hasa Valores Mensuales x3:Parque O’Higgins, 0003, LIN[M], Value 40 36 32 28 24 20 16 12 8 4 0 19/01/2023 16/03/2023 Registros validados, Mensual media (ppb) m=22.65 s=7.772 min=12.64 max=35.47 n=12 11/05/2023 06/07/2023 31/08/2023 26/10/2023 21/12/2023 Registros preliminares, Mensual media (ppb) m=--- s=-0 min=--- max=--- n=0 Registros no validados, Mensual media (ppb) m=--- s=-0 min=--- max=--- n=0 Figure 12: Monthly average NO2 concentration in 2023 according to SINCA Parque O’Higgins station (RM) The discrepancy between both sources is notable: satellite values represent between 1% and 4% of those reported by the ground station. This difference can be explained by multiple technical factors: • Measurement height: SINCA ground stations measure surface concentrations, while satellites observe total tropospheric NO2 columns, diluting values throughout the atmospheric thickness. • Spatial resolution: Sentinel-5P has a nominal resolution of 3.5 × 5.5 km in nominal mode, limiting its ability to capture intense point sources like dense urban hotspots. • Atmospheric conditions: Cloud cover, surface reflectance, and aerosol presence affect the validation of satellite data, reducing sensitivity in dense urban environments. • Comparison units: Both instruments measure NO2 but under different methodologies; the satellite provides data in molecular columns (mol/m2 ) that must be converted and weighted to compare with parts per billion (ppb) units used by SINCA. To perform this comparison more coherently, additional processing was applied to the raw satellite data. From the column values obtained via the SentinelHub API, concentrations were estimated in units equivalent to SINCA (ppb), assuming a homogeneous tropospheric column 10 km thick. The conversion procedure included two stages: 14 1. Conversion from column (mol/m2 ) to volumetric concentration (µg/m3 ) using the formula: Ccol · M · 106 Cµg/m3 = H where: • Ccol is the tropospheric column • H the estimated height (10 km) • M the molar mass of NO2 (46.01 g/mol). 2. Conversion to ppb using the stoichiometric factor for gases at standard conditions: 24.45 · Cµg/m3 Cppb = M Despite these adjustments, differences between both sources remain evident. This underscores that, although it is possible to approximate measurement units, there are inherent limits to the comparability between remote platforms and ground stations. Consequently, satellite estimates should be interpreted as regional representations of atmospheric load, not as direct substitutes for human exposure measured at surface level. 3 Technical Recommendations Given the exploratory nature of the system and the observed limitations, the following improvements are proposed for future versions: • Incorporate adaptive geometric correction according to extreme viewing angles, especially in southern zones. • Implement latitudinal adjustment factors based on historical series to compensate for systematic coverage losses. • Complement with additional satellite missions (OMI, future post-TROPOMI sensors) in low-coverage regions. • Apply specific spatial reconstruction models for winter months with higher cloud incidence. • Strengthen validation with ground stations in sectors with high discrepancy between sources. 4 Conclusions The analysis shows that the orbital geometry of Sentinel-5P introduces systematic data losses in southern Chile, limiting its applicability for homogeneous national studies. However, the developed system allows automated and efficient extraction in central and northern regions, integrating standard formats and basic validations. As a prototype, there is room for improvement. Nevertheless, its modular and replicable design provides a solid foundation for future iterations incorporating geometric corrections, multi-sensor integration, and more robust validation. Its evolution into a useful tool for national-scale satellite environmental monitoring is projected. 15 References • Copernicus Data Space Ecosystem. OpenEO API Collections. Available at: https: //documentation.dataspace.copernicus.eu/APIs/openEO/Collections.html • Copernicus Sentinel-5P. S5P Processing and Products. Available at: https:// sentiwiki.copernicus.eu/web/s5p-processing • Caracena, M. (2021). Official GeoJSON of Chile by administrative division (2015). GitHub repository: https://github.com/caracena/chile-geojson • Honora, J.L. (2019). Chilean regions GeoJSONs (alternative version). GitHub repository: https://github.com/jlhonora/geo/tree/master • SINCA - National Air Quality Information System. Chilean Ministry of the Environment. Available at: https://sinca.mma.gob.cl/ • Cressie, N. (1993). Statistics for Spatial Data. Wiley-Interscience, New York. • Savenets, M. (2021). Air pollution in Ukraine: the view from Sentinel-5P. Időjárás – Quarterly Journal of the Hungarian Meteorological Service, 125(2), 271–289. 16
0
You can add this document to your study collection(s)
Sign in Available only to authorized usersYou can add this document to your saved list
Sign in Available only to authorized users(For complaints, use another form )