Table. Impact of Covariates on Prediction Performance

District: Kunming
Covariate Configurations
Single Univariate
+Weather Single + Meteorological factors
+Air Single + Air quality
+NWP Single + NWP model
+Weather+Air Single + Weather + Air quality
+Weather+NWP Single + Weather + NWP model
+Weather+Air+NWP All covariates combined
Red Bold Best performance
Blue Underline Second best
Temperature (T) and Visibility (VIS) Predictions
Model Configuration T VIS
Basic Extreme Events Basic Extreme Events
MAE Corr (24th) DTW HRQE LRQE HWRQE LWRQE HCSRQE LCSRQE MAE Corr (24th) DTW (×107) HRQE LRQE (×107) HWRQE LWRQE (×107) HCSRQE LCSRQE
GraphWaveNet Single
+Weather
+NWP - - - - - - - - - - -
+Weather+Air
+Weather+NWP - - - - - - - - - - -
+Weather+Air+NWP
DGCRN Single
+Weather
+Weather+Air
+Weather+Air+NWP
PM2.5 Predictions
Model Configuration PM2.5
Basic Extreme Events
MAE Corr (24th) DTW HRQE LRQE HWRQE LWRQE
GraphWaveNet Single
+Weather+Air
+Air+NWP
+Weather+Air+NWP
Notes
• MAE and all RQE metrics: Lower is better
• Corr: Higher is better
• Best/Second best are evaluated across all configurations within each metric
• VIS metrics LRQE and LWRQE are scaled by 107
• "-" indicates data not available for this configuration