Table 1 Summary of different models in weather normalization of PM_2.5.

CTM

Emission inventories.

Meteorology fields.

Initial conditions.

Boundary conditions.

4-D simulation results.

Physical-chemical mechanisms encoded.

Heavy computation resources.

Slow in computation.

Accuracy highly depends on the accuracies of inputs.

Computer cluster needed.

^a

Large spatial scale simulation.

Relative shorter study period depends on available resources.

MLR

Air pollutant observation.

Meteorology variables.

Fast in computation.

Fewer computation resources.

Address assumptions required for statistical models.

Lack of physic-chemical mechanisms.

\({{\rm{PM}}}_{2.5}^{{\rm{EMI}}}\) (13%)

\({{\rm{PM}}}_{2.5}^{{\rm{MET}}}\) (66%)

Site (city) or grid scales.

Relative longer study period depends on observation data.

KZ-MLR

Air pollutant observation.

Meteorology variables.

Fast in computation.

Fewer computation resources.

Address assumptions required for statistical models.

Lack of physical-chemical mechanisms.

\({{\rm{PM}}}_{2.5}^{{\rm{EMI}}}\) (23%)

\({{\rm{PM}}}_{2.5}^{{\rm{MET}}}\) (79%)

Site (city) or grid scales.

Relative longer study period depends on observation data.

RF

Air pollutant observation.

Meteorology variables.

Fast in computation.

Medium computation resources.

Lack of physical-chemical mechanisms.

Performances depend on hyperparameters

\({{\rm{PM}}}_{2.5}^{{\rm{EMI}}}\) (3%)

\({{\rm{PM}}}_{2.5}^{{\rm{MET}}}\) (29%)

Site (city) or grid scales.

Relative longer study period depends on observation data.

XGB

Air pollutant observation.

Meteorology variables.

Fast in computation.

Medium computation resources.

Lack of physical-chemical mechanisms.

Performances depend on hyperparameters

\({{\rm{PM}}}_{2.5}^{{\rm{EMI}}}\) (3%)

\({{\rm{PM}}}_{2.5}^{{\rm{MET}}}\) (28%)

Site (city) or grid scales.

Relative longer study period depends on observation data.