Air quality surveillance through time-series forecasting: a comparative SARIMA analysis of NO2 and SO2 in Baltimore and New York.

Urban air pollutants exhibit distinct temporal dynamics that reflect differences in emission sources, atmospheric processes, and regulatory influences. This study introduces forecast tractability - the extent to which a pollutant's temporal behavior is captured by standard linear seasonal time-series models - as a diagnostic for distinguishing pollutants by source structure. Monthly nitrogen dioxide (NO₂) and sulfur dioxide (SO₂) concentrations in Baltimore and New York (2018-2024) were analyzed under a unified seasonal SARIMA framework, with 24-month forecasts validated against January-September 2025 observations. NO₂ showed a strong, consistent seasonal pattern in both cities, with recurring winter peaks and summer troughs. SO₂ exhibited moderate seasonality with residual variability not fully captured by linear seasonal specifications; residual dependence was particularly evident for Baltimore SO₂, and no structural regime changes were detected for either pollutant. Findings demonstrate that forecast tractability differs systematically: NO₂, dominated by continuous mobile sources, fits well under standard SARIMA specifications, whereas SO₂, particularly in cities with discrete legacy industrial sources, shows wider forecast intervals and a different residual structure under the fixed specification. These differences inform pollutant-specific monitoring and modeling: predictable pollutants suit forecast-driven regulatory planning, while less predictable ones warrant denser sampling or nonlinear approaches.