EXPLORING ERROR TYPES AND PERFORMANCE OF AN AIR QUALITY MODEL THROUGH CLUSTERING ANALYSIS

PINEDA ROJAS A.L.; KROPFF E.

Congreso; 20th conference on "Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes"; 2021

In this work, a simple approach to identify input data conditions that are associated with different model performance levels is presented. Using four years nitrogen dioxide (NO2) hourly concentrations measured at three air quality sites in the city of Buenos Aires and DAUMOD-GRS model results, a clustering analysis is applied over three performance metrics (FB, NMSE and R) to group days according to their levels of model performance. Four clusters are found to better describe such differences at the three sites. At the urban background site, wind speed and air temperature present the largest statistical differences between model performance clusters. In turn, at the residential industrial site, clusters show clear significant differences in most meteorological variables which suggest a potential role from the emissions coming from the power plants that are located on the coast. Overall, a better understanding of the DAUMOD-GRS model performance and how it changes with different conditions is obtained.