Satellite Greenness Reveals When City Trees Really Do Clean the Air

Planting trees has become one of the most politically attractive weapons against urban air pollution, yet scientists have struggled for decades to answer a deceptively simple question: does greener vegetation actually mean cleaner air? A new seven-year study from two Brazilian metropolitan regions offers one of the most detailed answers to date, and its verdict is both encouraging and sobering. Higher vegetation greenness was generally associated with lower concentrations of three major pollutants, but the strength of that link depended heavily on where researchers looked, how far they measured from their sensors, and what time of year it was.

The research, published in the journal Clean Technologies and Environmental Policy, focused on two adjacent but contrasting metropolitan areas in São Paulo State: Campinas and Piracicaba. Campinas is a densely urbanized hub of industry, technology and logistics, where vehicle and factory emissions dominate the air quality picture. Piracicaba, by contrast, is shaped by agro-industry, particularly sugarcane cultivation and processing, with cropland covering much of its landscape. Both regions sit within the Atlantic Forest biome, one of the world’s great biodiversity hotspots, now heavily fragmented by urban expansion and agriculture. The researchers, led by Ana Laura Fragoso Favoreti of the Universidade Estadual de Campinas, reasoned that these two very different landscapes would provide a natural experiment in how land use modulates the relationship between greenness and pollution.

The team assembled an unusually rich dataset. Monthly concentrations of coarse particulate matter (PM10), fine particulate matter (PM2.5) and nitrogen dioxide (NO2) came from the automated monitoring stations operated by CETESB, the São Paulo State environmental agency, covering January 2019 through December 2025. Hourly measurements passed through strict quality control: physically implausible values were discarded, and daily and monthly averages were only accepted when at least 75 percent of the expected observations were valid. Vegetation greenness was measured using the Normalized Difference Vegetation Index, or NDVI, computed from cloud-masked Sentinel-2 satellite imagery at ten-meter resolution within Google Earth Engine. Crucially, the team did not rely on a single greenness figure per station; instead, they calculated average NDVI within concentric buffers of 100, 250, 500 and 1000 meters around each monitoring station, explicitly testing whether the vegetation-pollution relationship changes with spatial scale.

The headline finding is a consistent negative correlation between greenness and pollution. Spearman rank correlations between monthly NDVI and pollutant concentrations were predominantly negative across both regions, ranging from −0.24 to −0.83 for PM10, −0.17 to −0.82 for PM2.5 and −0.31 to −0.75 for NO2. In plain terms, months and places with lusher vegetation tended to record dirtier air less often. Some of the strongest relationships emerged at stations embedded in agro-industrial landscapes: Rio Claro and Santa Gertrudes showed correlations approaching −0.83 for PM10 and −0.82 for PM2.5, while the central Campinas station, hemmed in by dense traffic, produced the weakest and least consistent signals.

But raw correlations in environmental data can mislead, because both vegetation and pollution follow strong seasonal cycles. The study’s statistical core therefore consisted of multiple linear regression models in which pollutant concentrations were log-transformed and adjusted for temperature, relative humidity, wind speed, and categorical fixed effects for month and year, with autocorrelation-robust standard errors to guard against the serial dependence that plagues monthly time series. These models explained between 75 and 96 percent of the variance in concentrations, and within them, the vegetation signal sharpened in one region and faded in the other. In Piracicaba, NDVI retained statistically significant negative associations across most buffer scales for all three pollutants, with standardized coefficients exceeding −1.0 at the 100-meter buffer. In Campinas, by contrast, nearly all adjusted associations were statistically indistinguishable from zero, a result the authors attribute to smaller sample sizes, reduced statistical power, and strong collinearity between NDVI and meteorological predictors driven by shared seasonality.

The spatial scale of the analysis proved decisive. Vegetation effects in Piracicaba were strongest at the smallest buffers, within 100 to 500 meters of the sensors, and weakened or vanished at 1000 meters, suggesting that nearby canopy exerts a genuinely local influence on measured air quality. One striking exception hinted at a cautionary tale: at the 1000-meter buffer for PM10 in Piracicaba, the coefficient reversed sign, becoming strongly positive. The likely explanation is that broad-scale greenness in an agricultural landscape can flag croplands, unpaved surfaces and fire-prone areas that actually generate coarse particles. In other words, at larger scales, the satellite index may be measuring the geography of emissions rather than the geography of pollution removal.

Seasonality dominated the temporal record. Pollutant concentrations rose sharply during the dry season from May to August, when reduced rainfall suppresses wet deposition and weak winds limit dispersion. The seasonal contrast was especially dramatic for PM10 in Piracicaba, where dry-season median concentrations of 53.59 micrograms per cubic meter were 2.5 times the wet-season value. Peak particulate concentrations arrived in September in both regions, likely reflecting long-range transport of biomass burning aerosols from central Brazil, where fire activity intensifies through August and September. Vegetation greenness moved in exact anti-phase, peaking in the January-to-March wet season and bottoming out in September precisely when particulate pollution peaked. The authors are candid that this synchronized seasonality means part of the observed correlation may reflect shared seasonal dynamics rather than a direct cleansing effect of leaves.

Regional comparisons added further nuance. PM10 and NO2 concentrations were significantly higher in Piracicaba than in Campinas, consistent with the influence of agro-industrial sources, soil resuspension and combustion, while PM2.5 showed no significant difference between regions, hinting that fine particles are more regionally mixed and shaped by secondary formation and long-distance transport. Inter-annual variability told its own story: 2024 stood out as an anomalously polluted year in both regions, coinciding with Brazil’s exceptional drought and intensified fire activity, while Campinas recorded a conspicuous drop in NO2 in 2020 during COVID-19 lockdowns, before concentrations rebounded to their highest level in 2023.

What should planners take away? The authors emphasize that NDVI is an indirect structural proxy for vegetation vigor, not a direct measure of pollutant filtration. It cannot capture canopy height, leaf area, species composition or street geometry, all of which determine whether vegetation intercepts particles or, in dense street canyons, traps them by blocking wind. The study’s observational design likewise rules out causal claims: the results describe statistical associations between surrounding greenness and pollutant concentrations, adjusted for weather and time, not proof that trees remove the pollution. Still, the message for green infrastructure policy is clear. Greening works best as one component of integrated air quality management, deployed with attention to local emission sources, urban morphology and spatial scale, rather than as a standalone cure. In rapidly urbanizing, agro-industrial landscapes like Campinas and Piracicaba, the authors conclude, pairing vegetation planning with air quality monitoring and land use management will deliver far more environmental benefit than simply increasing greenness alone.

Subject of Research: The spatiotemporal relationship between urban vegetation greenness measured by NDVI and air pollutant concentrations in two São Paulo metropolitan regions from 2019 to 2025.

Article Title: Urban green spaces and air quality in the State of São Paulo: a spatiotemporal analysis of the metropolitan regions of Campinas and Piracicaba

Article References: Favoreti, A. L. F., Rodrigues, B. N., Emiliano, W. M., Canteras, F. B., & Molina Junior, V. E. (2026). Urban green spaces and air quality in the State of São Paulo: a spatiotemporal analysis of the metropolitan regions of Campinas and Piracicaba. Clean Technologies and Environmental Policy, 28(10), Article 255. https://doi.org/10.1007/s10098-026-03604-7

Image Credits: AI Generated

DOI: 10.1007/s10098-026-03604-7

Keywords: urban green spaces, air quality, NDVI, particulate matter, nitrogen dioxide, green infrastructure, remote sensing, Sentinel-2, Google Earth Engine, São Paulo, Campinas, Piracicaba