A random forest partition model for predicting NO₂ concentrations from traffic flow and meteorological conditions

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• High atmospheric nitrogen dioxide (NO2) concentrations in urban areas negatively impact resident health.
• Existing models may not fully capture the complex factors influencing NO2 levels across different concentration ranges.

Purpose of the Study:

• To develop and validate a novel partitioned model for atmospheric NO2 concentration forecasting in urban environments.
• To improve the accuracy of NO2 pollution modeling by distinguishing between low and high concentration scenarios.

Main Methods:

• Utilized a data mining technique, specifically random forests, to build a partitioned model (M) with separate components for lower (ML) and upper (MU) concentration values.
• Applied an iterative method to determine the optimal boundary concentration (y˜) minimizing mean absolute deviation error.
• Validated the model using NO2 concentration data from Wrocław, Poland (2015-2017).

Main Results:

• The partitioned model achieved a higher R² value (0.82) compared to a classical random forest model (0.60).
• In the lower concentration model (ML), traffic flow was the primary influencer, followed by meteorological factors (wind direction and speed).
• In the higher concentration model (MU), traffic flow remained dominant, but temperature and relative humidity showed significantly increased influence.

Conclusions:

• The partitioned modeling approach is justifiable and effective for predicting urban NO2 concentrations.
• Understanding variable importance differences between low and high pollution scenarios is crucial for targeted air quality interventions.