The core problem the researchers set out to solve is deceptively simple to state and notoriously difficult to solve. Regulatory networks measure outdoor pollution at fixed stations, and increasingly sophisticated satellite products and land-use regression models can map street-level concentrations at fine spatial resolution. But people in temperate European cities spend the overwhelming majority of their time indoors, where concentrations of nitrogen dioxide are governed by a second set of processes entirely: infiltration of outdoor air through windows, cracks, and ventilation systems; indoor combustion sources such as gas stoves and boilers; and the building characteristics that determine how quickly pollutants accumulate or disperse. An exposure estimate that ignores these indoor dynamics can be badly biased, and the bias is rarely random. It tends to track income, housing age, building density, and access to ventilation, meaning that the people most poorly represented by outdoor monitors are often those whose true exposure is most underestimated.

The study’s answer to this challenge is a two-stage architecture that separates the problem into an outdoor prediction stage and an indoor transfer stage. In the first stage, the researchers estimate outdoor nitrogen dioxide concentrations at high spatial resolution across Barcelona, drawing on the established toolkit of land-use regression and related spatial models that relate measured concentrations to traffic intensity, road network characteristics, land cover, population density, and meteorology. This stage produces a continuous urban surface of ambient pollution, effectively filling in the gaps between monitoring stations so that every building in the city can be assigned a plausible outdoor concentration. The approach reflects two decades of methodological development in exposure science, refined in recent years by machine learning techniques that can capture nonlinear relationships between urban form and pollution gradients.

The second stage is where the study makes its distinctive contribution. Rather than assuming that indoor concentrations simply mirror outdoor levels, the model estimates how outdoor pollution is translated into indoor air for individual dwellings. This translation depends on the infiltration factor, the fraction of outdoor particles or gases that penetrate and persist indoors, which varies systematically with building type, construction era, window behavior, and the presence of indoor sources. Gas cooking is a particularly important modifier for nitrogen dioxide, because a gas flame releases the pollutant directly into the kitchen air. By combining predicted outdoor concentrations with information on building characteristics and household features, the second stage produces estimates of the concentrations people actually experience inside their homes, the locations where exposure is typically longest and most sustained.

Barcelona is an ideal proving ground for this kind of model. The city’s compact Eixample district, with its characteristic chamfered blocks and enclosed interior courtyards, creates extraordinarily sharp pollution gradients: a dwelling on a wide traffic artery can face dramatically different ambient conditions from one a few tens of meters away on an inner courtyard. At the same time, Barcelona’s housing stock is dominated by apartment buildings of varying ages and construction quality, with a substantial share of households relying on gas appliances for cooking. This combination of steep spatial variability and heterogeneous building stock means that outdoor-only exposure estimates are likely to misclassify large numbers of residents, and it gives the two-stage model a demanding test case in which its added realism can matter most.

The practical payoff of the approach is a city-wide picture of indoor exposure that no measurement campaign could realistically deliver. Monitoring indoor air directly requires recruiting households, installing instruments, and sustaining them over weeks or months, which limits studies to samples of dozens or a few hundred homes. Those measurements remain indispensable for calibrating and validating models, but they cannot by themselves reveal how exposure is distributed across an entire population. The two-stage framework bridges that gap: a limited set of indoor observations anchors the model, and the model then extends those observations to every address in the city, generating exposure estimates that can be linked to health records, school locations, or demographic data. This capacity to produce individual-level or small-area exposure estimates at scale is precisely what modern environmental epidemiology requires, particularly for studying outcomes such as childhood asthma, where the indoor environment is believed to play a decisive role.

The findings carry implications that extend well beyond academic modeling. Nitrogen dioxide is a respiratory irritant with well-documented associations with asthma exacerbations, reduced lung function growth in children, and cardiovascular effects, and the World Health Organization has repeatedly tightened its air quality guidelines for the pollutant. If a meaningful fraction of exposure occurs indoors, then policies that focus exclusively on tailpipe emissions and traffic restriction, while essential, will not fully protect public health. The modeling framework makes it possible to ask targeted questions: which neighborhoods combine high outdoor pollution with poor building envelopes and prevalent gas cooking; how much exposure reduction would follow from electrifying household cooking versus tightening vehicle standards; and whether interventions such as improved ventilation or filtration deliver the benefits their proponents claim. Each of these questions becomes answerable once indoor exposure can be predicted systematically rather than measured only sporadically.

The study also speaks to a broader methodological shift in exposure science, one in which hybrid models that fuse measurements, spatial statistics, and increasingly machine learning are replacing both pure monitoring and purely statistical surrogates. The two-stage design has a particular virtue: interpretability. Because outdoor prediction and indoor transfer are modeled separately, researchers can diagnose which stage contributes most to uncertainty, and policymakers can see transparently how a change in traffic emissions or in housing characteristics propagates through to human exposure. This modularity also makes the framework portable. A city with a different climate, building stock, or pollution profile can retain the architecture while re-estimating the stage-specific parameters from local data, a flexibility that matters as exposure scientists attempt to generalize findings from well-studied European cities to rapidly urbanizing regions where monitoring infrastructure is thin.

Limitations remain, and the authors are candid about them. Indoor models are only as good as the household-level information feeding them, and data on cooking fuel, ventilation behavior, and window-opening habits are difficult to obtain at population scale. Seasonal variation adds another layer of complexity, since infiltration and ventilation patterns shift markedly between Barcelona’s mild winters and hot summers. Uncertainty in the second stage is therefore typically larger than in the first, and the resulting exposure estimates are best understood as probabilistic characterizations rather than precise measurements of any single dwelling’s air. Nonetheless, the study demonstrates that even with these constraints, two-stage modeling yields exposure surfaces that are demonstrably more faithful to the environments people inhabit than outdoor concentrations alone.

For residents of Barcelona and cities like it, the research reframes a familiar anxiety in sharper terms. The pollution that matters most to long-term health is not only the visible haze over a traffic-choked avenue but the quieter accumulation inside apartments, kitchens, and bedrooms, shaped by the building itself and the appliances within it. By giving researchers and policymakers a rigorous way to estimate that hidden half of the exposure equation, the two-stage approach moves the field closer to interventions that meet people where they actually live. As cities worldwide grapple with tightening air quality targets and aging housing stocks, models of this kind are likely to become standard instruments of environmental health policy, translating sparse measurements into the dense, actionable picture that protecting public health demands.

Subject of Research: Two-stage modeling of indoor nitrogen dioxide exposure in Barcelona residences

Article Title: A two-stage modeling approach to estimate indoor NO2 exposure: a Barcelona case study

Article References: A two-stage modeling approach to estimate indoor NO2 exposure: a Barcelona case study. (n.d.). https://doi.org/10.1038/s41370-026-00969-1

Image Credits: AI Generated

DOI: 10.1038/s41370-026-00969-1

Keywords: nitrogen dioxide, indoor air pollution, exposure modeling, Barcelona, land-use regression, infiltration factor, gas cooking, environmental epidemiology, air quality, respiratory health, two-stage model, urban pollution