Identification and Stability Assessment of Nonlinear Thresholds in Relationships between PM₁₀, NO, and O₃ Concentrations and Meteorological, Anthropogenic, and Topographic Factors

Introduction: Atmospheric pollutant concentrations may exhibit nonlinear relationships with meteorological, topographic, and anthropogenic factors. PM₁₀, NO, and O₃ were selected to examine particulate matter dynamics, combustion-related primary emissions, and secondary photochemical processes, respectively. This study aimed to identify potential thresholds in pollutant-environment relationships and assess their statistical stability.

Methods: Data from 351 air quality monitoring stations across Türkiye were analyzed for a one-year period spanning April 2025 to April 2026. Explanatory variables included annual mean temperature, relative humidity, wind speed, and precipitation; elevation; population within a 5-km-diameter area around each station; and distances to forest and industrial areas. A separate generalized additive model was fitted for each pollutant, with all variables considered jointly. Performance was evaluated using 10-fold cross-validation. Marginal effects were examined through partial dependence functions, thresholds were estimated using segmented regression, and stability was assessed through 100 bootstrap iterations with 95% confidence intervals.

Results: The PM₁₀ model showed the most stable cross-validation performance. PM₁₀ concentrations declined more markedly above wind-speed and precipitation thresholds of 2.78 m/s and 3.18 mm/day. Stronger winds enhance atmospheric mixing and dilution, while precipitation removes particles through wet scavenging. NO concentrations were higher below wind-speed and temperature thresholds of 2.34 m/s and 15.19 °C. Weak winds and reduced convective mixing under cooler conditions restrict dispersion, favoring near-surface accumulation of primary NO emissions. O₃ concentrations increased above a wind-speed threshold of 3.07 m/s. Enhanced regional advection and vertical exchange transport ozone-rich air toward the surface. Bootstrap resampling yielded generally consistent threshold estimates across iterations, with uncertainty quantified using 95% confidence intervals.

Conclusions: The findings suggest that pollutant-environment relationships may change around particular values. Combining generalized additive models, segmented regression, and bootstrap analysis provided an applicable analytical framework for identifying potential thresholds and assessing their stability.