Air quality monitoring in Mendoza, Argentina: machine learning approaches for PM10 prediction.

In this study, different statistical methodologies were combined to assess the relationship between PM₁₀ concentrations and meteorological variables (temperature, relative humidity, wind direction and speed, and atmospheric pressure) and their associations with other pollutants (CO, NO₂, NO, and O₃) recorded during the period 2021-2024 at Mendoza City, Argentina. The results indicate that increased humidity and temperature might reduce PM₁₀ levels by enhancing particle dispersion and deposition. Positive correlations between PM₁₀, NO, and NO₂ suggest a shared origin, likely from vehicle emissions. To further analyze PM₁₀ behavior, prediction models were developed to categorize PM₁₀ levels as "good" (≤ 45 μg/m³) or "bad" (>45 μg/m³) based on a air quality guidelines from WHO. The performance of the random forest (RF) and logistic regression (LR) algorithms were evaluated and compared. Additionally, the influence of atmospheric variables and pollutant concentrations was also assessed to determine their impact on PM₁₀ predictions. RF model demonstrated the highest predictive performance for PM₁₀ level. Results indicate that NOx (NO₂ and NO) significantly contribute to PM₁₀ formation, likely due to shared anthropogenic sources. Temperature, humidity, and wind speed also impact PM₁₀ predictions, though to a lesser extent than pollutant concentrations. The inclusion of these variables highlights the role in the dispersion and transformation of air pollutants. Implementing such models could provide policymakers with real-time data to enhance pollution control and public health protection.