NOx and CO Concentrations Associated with Meteorological Variables in the Cities of Brasília and Goiânia

Abstract

Air pollution is a major consequence of anthropogenic activities, affecting air quality and urban thermal comfort. Data from the TROPOMI sensor aboard the Sentinel-5P satellite and meteorological data from INMET were used to assess air quality in Brasília and Goiânia—two capitals in the Brazilian Cerrado—between 2019 and 2024. Ridge and Lasso regression models were compared based on predictive performance (MSE and R²) and variable selection capability, incorporating 1- and 2-day time lags. Goiânia exhibited higher average CO concentrations than Brasília during the study period. Temperature and relative humidity were the variables that contributed most to explaining pollutant variance. Increased atmospheric pressure was associated with reduced concentrations, whereas wind speed and precipitation showed coefficients of lower magnitude. The Ridge model demonstrated slightly better predictive performance than the Lasso model (with R² values 0.01 to 0.06 higher, depending on the pollutant and city), while the Lasso model proved more efficient at variable selection by setting the coefficients of less significant variables to zero