A methodology to select particle morpho-chemical characteristics to use in source apportionment of particulate matter from livestock houses

Intensive poultry and pig houses are major point sources of particulate matter (PM) emissions. The knowledge on the contribution of individual sources to PM in different fractions is essential to improve PM reduction from livestock houses. We developed a methodology to investigate which input data (particle chemical, morphological or combined characteristics) were best to distinguish amongst specific sources of airborne PM in livestock houses. We used a validation procedure with classification rules based on decision trees and analyzed misclassification errors. The PM from two livestock species (poultry and pigs), and in two different fractions (fine and coarse) was studied. Results showed the selection of the best input data varied with the sources, which depend on livestock species. Using only particle chemical characteristics resulted in higher overall classification accuracies (62–68%) than using only morphological characteristics (40–64%) in poultry and pigs. Particle morphological characteristics can add value when sources show distinctive and well defined morphologies or differ in size. Using combined chemical and morphological resulted in the highest overall classification accuracies (average of 69% of particles correctly assigned to their source) and lowest misclassification errors. This study provides a methodological approach to assess input data and identifies the most effective characteristics to apportion PM in livestock houses. These data are promising to determine the contribution of different sources to PM in livestock houses and give insight in under- and overestimation errors in the source apportionment.

► An approach to assess particle characteristics in the air of livestock houses.
► Accuracy using particle chemistry (62–68%) is higher than with morphology (40–64%).
► Particle morphology adds value when sources differ in shape and size.
► Combined morpho-chemical data results in the highest accuracy (69%).
► Insight in under- and overestimation errors in the source apportionment is given.