Local climate zones as predictors of NO2 in Barcelona: Towards healthier and more sustainable cities

• LCZ-based modelling captures spatial patterns of NO 2 concentrations in the Barcelona metropolitan area. • A dominant morphological gradient explains 58% of NO 2 spatial variability. • An extended model including road and building density improves performance (R 2 up to 0.72) • Road density emerges as a key driver, highlighting the role of traffic-related emissions. • LCZs provide an integrative framework, while process-based variables capture local variability and hotspots. Although urban form shapes the neighbourhood-scale distribution of air pollutants, standard land-use datasets rarely capture this influence, and the mechanisms linking urban morphology to NO 2 variability remain insufficiently understood. This study evaluates the Local Climate Zone (LCZ) framework as a morphology-based approach to explain spatial variability in annual mean NO 2 concentrations across the Metropolitan Area of Barcelona (Spain) in 2023. Twenty-two monitoring stations were buffered (500 m radius) and intersected with a high-resolution (100 m) LCZ map. Within each buffer, the relative cover of sixteen LCZ classes was summarized and analysed using principal component analysis (PCA), and the resulting components were regressed against measured NO 2 . The first principal component represents a dominant urban morphological gradient contrasting densely built environments (e.g., LCZ 2 – Compact Mid-rise; LCZ E – Bare Rock/Paved) with open and vegetated areas (e.g., LCZ 6 – Open Low-rise; LCZ A – Dense Trees; LCZ C – Scrubland). This gradient explains 58% of the spatial variance in NO 2 (β = –2.62 µg m⁻ 3 , p < 0.001), highlighting the central role of urban form in shaping pollutant distribution. To further investigate the drivers of NO 2 variability, an extended model incorporating process-based variables (building density, road density, and distance to coastline) was developed. This approach improved the explained variance to 72% and enhanced predictive performance (LOOCV R 2 = 0.58; RMSE = 4.65 µg m⁻ 3 ; MAE = 3.84 µg m⁻ 3 ), highlighting the importance of explicitly accounting for emission-related and structural factors. Road density emerged as the most influential predictor, reinforcing the role of traffic-related emissions in shaping local NO 2 patterns. Application of the calibrated model to the LCZ composition produced a continuous NO 2 surface (1–57 µg m⁻ 3 at 100 m resolution). Most urban land (∼56%) falls within the 20–30 µg m⁻ 3 range, while higher concentrations (>40 µg m⁻ 3 ) are associated with compact urban fabrics and traffic-intensive areas. While both models reproduce the general spatial structure of NO 2 , the extended model captures greater spatial variability and a higher representation of local high-concentration areas, whereas the LCZ-based model produces smoother gradients. These findings should be interpreted as spatial associations rather than direct causal effects, as urban morphology co-occurs with other factors such as traffic intensity, transport infrastructure, land-use functions, and long-term urban development patterns. These results demonstrate that LCZ composition can be reduced to a physically interpretable morphological gradient that explains a substantial fraction of NO 2 spatial variability. At the same time, LCZs provide an integrative description of urban form that complements more detailed process-based variables. This approach provides a robust and transferable proxy for exposure assessment and planning-oriented analysis, while supporting the prioritization of actionable measures such as traffic reduction, low-emission zones, and targeted street-level interventions.