Environmental Engineering Reference
In-Depth Information
4.6 Clustering Analysis of Air Quality Model
Performance
Christopher Fung
1
, Paul Yau
1
, Connie Lam
2
, Philip Yu
2
, and Linda Yu
1
1
Hong Kong Environmental Protection Department, 33/F Revenue Tower, 5 Gloucester
Road, Hong Kong Special Administrative Region, China
2
Department of Statistics and Actuarial Science, University of Hong Kong, Pokfulum, Hong
Kong
Abstract
A model performance evaluation approach based on matching clusters
was developed to provide a first indication of an air quality modeling system's
performance. Observed weather, modeled weather, observe air quality, modeled
air quality parameters are clustered and the optimum number of clusters chosen.
These sets of clusters are then compared against each other to determine the
performance of the individual modules and also how well the modeling system
can reproduce the relationship that exists between meteorology and air quality
measurements. An index is used to quantify the matches between the different
clusters. Change in the values of this index resulting from changes in the system
(formulation, input data …etc.) can be interpreted as real improvements or
deteriorations of the modeling system's performance.
Keywords
Air quality modeling system, clustering, performance, evaluation
1. General Framework
The performance of a photochemical air quality modelling system depends on the
accuracy of each module - emissions, meteorology and transport and chemistry.
Yet, the amount of outputs of these modules can be so voluminous that it is often
difficult to determine from them whether the system has indeed 'improved' after
some measures have been taken to that effect.
Figure 1
gives a schematisation of an approach based on clustering to overcome
the difficulty with overwhelming data. Treating emissions as the quasi-steady
unknown, the approach recognises that there are recurrent weather patterns which
give 'clusters' in the weather parameters. These 'weather clusters' also affect air
pollution in distinctive ways through three dimensional transport, availablity of
sun light, etc. giving air quality clusters. On the other hand, air quality responds to
factors other than meteorology, e.g. spatial and temporal distribution of emissions.
By first clustering observed weather parameters (Box 1) and air quality parameters
