Skip to main content
. 2022 Aug 2;9:466. doi: 10.1038/s41597-022-01488-y

Fig. 4.

Fig. 4

Smog potential raster predictions. A GBM machine learning model was used to predict mean station PM2.5 under stable winter conditions (A). Errors from this model were spatially interpolated using kriging (C). Red areas are where the model underpredicts smog potential while blue areas indicate overprediction. The GBM fit and kriged errors were combined to make the final smog potential raster (B). Panel D shows the kriging prediction variance.