top of page

Fine Particulate Matter (PM2.5)

We developed machine learning ensemble models (random forest, light gradient boosting, and neural network) to predict monthly PM2.5 concentrations in South Korea.

Model Information
We constructed three machine learning-based models, namely random forest, light gradient boosting, and neural network, to predict monthly PM2.5 averages using a 1 km × 1 km grid during 2002–2020. A total of 112 predictor variables collected from the GEE, SEDAC, regional socioeconomic database, and others were used as input variables, and monthly-average PM2.5 concentrations were predicted as the outcome values. Please refer to the following figures and tables for more specific information.

Table 1. Test results by test R2 and rooted mean squared error (RMSE) for each study year, month, and area.


Figure 1. Annual Mean PM2.5 Concentrations (㎍/㎥)


Figure 2. Monthly Mean PM2.5 Concentrations (㎍/㎥)


Figure 3. Comparisons of the predicted values with the monitored values

Figure 4. Time Series Mean PM2.5 Concentrations (㎍/㎥)

bottom of page