LADCO uses machine learning to classify air pollution conditions, identify patterns in pollution events, and to understand the predictors of air pollution in this region.
We use statistical and visualization machine learning tools to understand the causes of air pollution in our region.
Machine learning projects at LADCO are included below.
Classification and Regression Tree (CART) to adjust ozone trends for meteorology
LADCO has applied a Classification and Regression Tree (CART) analysis to ground-level ozone (O3) and meteorology data to control for the impacts of weather and to discern the impact of emissions changes on O3 pollution. We used CART to determine the meteorological conditions most commonly associated with high-O3 days in O3 nonattainment areas in the LADCO region. Any remaining trend in the meteorologically adjusted O3 is assumed to be the result of non-meteorological factors, such as reductions in emissions of O3 precursors.
- LADCO’s 2025 report: Classification and Regression Tree (CART) Analysis for LADCO Ozone Nonattainment Areas: 2001-2023
- Examines trends in data from 2001 to 2023
- LADCO’s 2021 report: Classification and Regression Tree (CART) Analysis for LADCO Ozone Nonattainment Areas
- Examines trends in data from 2005 to 2020
Self Organizing Maps (SOM)
Generalized Additive Models (GAM) to adjust ozone trends for meteorology
In the summer of 2022, LADCO hired a summer intern, Hantao Wang, to apply a generalized additive model (GAM) to adjust ozone trends for meteorology and explore the impact of meteorology on ozone formation at two sites in the region: Sheboygan Kohler-Andrae and the SWFP site in Chicago. He employed quantile regression to examine trends in meteorologically adjusted ozone at various ozone concentrations from 2000 to 2019.
In 2020, LADCO and Wisconsin DNR contracted with Charles L. Blanchard to evaluate the sensitivity of ozone formation around Lake Michigan to changes in emissions of different ozone precursors. As part of this contract, Blanchard applied a GAM to model ozone at 20 sites around Lake Michigan over the years 2000 to 2019, with a particular focus on 2015 to 2019.
