9.1. NO2 Calibration with OLS Regression

Another way of calibration is to use a regression analysis. In vein with the previous section, this section implemented four separate regressions between the modelled results and observations: GAM-observation, UK-observation, GAM Final-observation, UK Final-observation. The regression model was applied to groups at a 10-day time step and employed a Density-based clustering approach to capture different R² by clusters. Unlike the common clustering methods, such as K-means clustering, this method increases the accuracy especially when the scatterplots looks like a hockey stick. The downside, however, that it will only consider points allocated as a cluster, which in other words the outliers are excluded. --> R² might be lower than measurement

Summer

• Regression results for 6 models in Summer.

Models	Aug: Top	Aug: Mid	Aug: Bot	Sep: Top	Sep: Mid	Sep: Bot
GAM-UK	0.979	0.978	0.965	0.981	0.97	0.982
GAM final-UK final	0.995	0.993	0.992	0.992	0.971	0.987
GAM final-UK final2					0.806	0.927
GAM-Obs	0.56	0.58	0.484	0.673	0.713	0.81
UK-Obs	0.568	0.635	0.541	0.681	0.739	0.85
GAM final-Obs	0.679	0.703	0.613	0.676	0.498	0.861
GAM final-Obs2	0.619	0.63	0.249	0.471		0.787
UK final-Obs	0.648	0.753	0.648	0.689	0.812	0.745
UK final-Obs2	0.641	0.657	0.381	0.503	0.707

August

September

Winter

• Regression results for 6 models in Winter

Models	Dec: Top	Dec: Mid	Dec: Bot	Jan: Top	Jan: Mid	Jan: Bot	Feb: Top	Feb: Mid	Feb: Bot
GAM-UK	0.978	0.985	0.986	0.996	0.991	0.995	0.994	0.967	0.985
GAM final-UK final	0.993	0.991	0.977	0.924	0.99	0.997	0.975	0.978	0.842
GAM-Obs	0.58	0.875	0.804	0.847	0.801	0.818	0.835	0.746	0.757
UK-Obs	0.635	0.889	0.824	0.844	0.783	0.81	0.813	0.689	0.754
GAM final-Obs	0.703	0.89	0.618	0.621	0.643	0.762	0.733	0.793	0.732
GAM final-Obs2	0.63	0.908				0.499			0.788
UK final-Obs	0.753	0.903	0.648	0.727	0.609	0.74	0.725	0.821	0.74
UK final-Obs2	0.657	0.913				0.483			0.568

December

January

February