Characterisation of errors in an urban scale atmospheric dispersion model through clustering of performance metrics

PINEDA ROJAS, ANDREA L.; BORGE, RAFAEL; KROPFF, EMILIO

Air Quality, Atmosphere and Health

Springer Science and Business Media B.V.

Año: 2022

1873-9318

This work introduces a methodology to perform a comprehensive analysis of the performance of atmospheric dispersion models, using the urban scale DAUMOD-GRS model as a testing ground. The estimation of NO2 concentration of this model at an urban background site in the city of Buenos Aires is compared with long-term observed concentration series. Days are classified through clustering analysis according to their model performance level on a three-dimensional metric space (fractional bias, normalised mean square error and correlation coefficient), and four clusters are obtained. An assessment of whether or not classes occur under different typical meteorological conditions is then performed. The largest statistical differences amongst clusters occur for wind speed and air temperature. The method is also used to assess the performance of a modified version of DAUMOD-GRS. The modification leads to a general improvement, mostly due to greater accuracy during night-time, which improves the ability of the model to estimate the NO2 peak concentration values, occurring at early morning and late evening hours. This is an example of how clustering techniques can be used to identify specific conditions under which a model underperforms, understand its causes and test potential model improvements.