Selection of a representative set of chemical accidents from a complex data matrix for the development of...

Chemical accidents often lead to negative consequences for the environment. Preparedness and proper actions are, therefore, essential components in order to minimise environmental effects. To assist and facilitate this work, a proposed planning tool, the environment-accident index (EAI), was formulated by Scott [J. Hazard. Mater. 61 (1998) 305]. As a result of a first validation of the index, based on 21 chemical accidents, the database was complemented with 42 additional accidents covering a broader spectrum of chemicals. The additional accidents were collected by means of an inquiry and their environmental consequences are, so far, unknown. The collected data had an overrepresentation of accidents involving petroleum products (69%). Because of the overrepresentation of this group of chemicals in the material, the data was skewed with respect to chemical properties. Since the model should be valid for a variety of chemical accidents, a method was needed which enabled a proper and unbiased selection of a representative subset of accidents to be used in development and validation of the model. For this purpose, the possibility to use multivariate data analysis in combination with statistical design was investigated. The result showed the feasibility of this method in the selection of a representative subset from a complex and skewed large dataset. Within the new dataset, 53% were accidents involving petroleum products and 47% involved other chemicals. The selected accidents will be used in further work to evaluate the environmental consequences, for model development and model validation.