Physics-Based Spatial Oversampling of TROPOMI NO2 Observations to US Neighborhoods Reveals the Disparities of Air Pollution

Abstract

Satellite observations provide continuous and global coverage observations of air pollutants, widely used to inform health impacts and air pollution disparities. Linking satellite retrievals with socioeconomic or health data involves matching the irregularly shaped satellite observations with administrative units. Here, we develop a physics-based approach to spatially oversample nitrogen dioxide (NO₂) retrievals from TROPOspheric Monitoring Instrument (TROPOMI) directly to United States (US) neighborhoods (i.e., block groups). The physics-based oversampling approach considers each satellite pixel as a sensitivity distribution, meaning that satellite instruments are more sensitive to the neighborhoods at the center than at the edge of the observations. We show that directly oversampling satellite observations to administrative shapes is a more accurate and computationally efficient approach than the commonly used gridding approaches, and it is advantageous for shorter temporal windows. Combining the newly developed NO₂ data set with demographic data, we find widespread racial/ethnic and income-related NO₂ disparities across the US. NO₂ disparities are even more pronounced during the most polluted days, suggesting greater acute health effects for overburdened communities. We expect that the resolution-adaptive, neighborhood-level, and GIS-compatible NO₂ data set would lower barriers of the public to access and interpret satellite observations, facilitating the actionable applications of satellite observations.

Keywords: NO2; TROPOMI; air pollution disparity; environmental justice; remote sensing; spatial oversampling.

Figure 1

(a) Illustration of the 2‐D Gaussian spatial response function of a single satellite observation overlaid with the administrative boundaries of 10 block groups in the satellite field of view. The response function is projected to the Cartesian space and normalized by the sum. (b) Comparison of the weight assigned to each block group (labeled in a) based on overlap areas (AWO‐BG) or the Gaussian weight (PGO‐BG). The area is normalized by the area of the satellite pixel polygon.

Figure 2

Map of 5‐year average (2019–2023) NO₂ at block group level over contiguous United States with zoom‐in maps of NO₂ for 10 CBSAs with the highest level of NO₂. The block‐group NO₂ is calculated using PGO‐BG approach. Block groups where NO₂ exceeds top 10% national level (>5.5 × 10¹⁵ molecules/cm²) are aggregated and outlined in black. The numbers in the zoomed‐in maps represent the mean NO₂ column density (in unit of 10¹⁵ molecules/cm²), with the values in brackets indicating the minimum and maximum within each core‐based statistical areas.

Figure 3

Maps of block‐group level TROPOspheric Monitoring Instrument NO₂ for New York City based on 5‐year average (upper panel) and top 5% polluted days (bottom panel) using three oversampling methods: (a) Physics‐based Gaussian oversampling to block groups (PGO‐BG); (b) area weighted average oversampling to block groups (AWO‐BG); (c) area weighted average method oversampling to a regular grid of 0.01° × 0.01° (AWO‐Grid).

Figure 4

Maps of NO₂ disparities related to (a) race/ethnicity and (b) income over all CBSAs of contiguous United States derived from block‐group TROPOspheric Monitoring Instrument NO₂ oversampled with PGO‐BG approach. The racial/ethnic disparity is defined as the relative difference in population‐weighted average NO₂ between white non‐Hispanic and minority groups (including Hispanic, non‐Hispanic Black, Asian, and native American), and the income‐related disparity is defined as the relative difference in population‐weighted average NO₂ between the groups with ratio of income to poverty level below 1.24 (hereafter low‐income groups) and the groups with the ratio above 1.5 (hereafter high‐income groups). The left panel shows the top 20 CBSAs with the largest NO₂ disparity, with numbers on the map indicating their rank. The results using the other two oversampling approaches can be found in Supporting Information S1.

Figure 5

(a) Difference in the racial/ethnic NO₂ disparities between the top 5% polluted days and the 5‐year mean. (b) Comparison of the racial/ethnic NO₂ disparities on the top 5% polluted days versus the 5‐year mean NO₂ in selected CBSAs. NO₂ data are oversampled with the PGO‐BG approach. The top 5% days are selected based on the daily mean TROPOMI NO₂ at each core‐based statistical areas. The CBSAs are ranked by the 5‐year mean NO₂ level with Los Angeles being the highest. The results using the other two oversampling approaches can be found in Supporting Information S1.

Figure 6

Normalized spatial gradient of NO₂ in selected CBSAs: (a) New York; (b) Miami; (c) St. Louis; (d) Los Angeles and (e) Denver based on 5‐year average (top panel) versus top 5% polluted days (middle panel), calculated as the block‐group NO₂ divided by the maximum NO₂, using the PGO‐BG approach to oversample NO₂. The bottom panel shows the percentage of white non‐Hispanic population. The extent of each map is defined based on the full extent of the corresponding core‐based statistical areas, which include the city labeled and the surrounding areas.