THE USE OF BOOTSTRAPPING TO ESTIMATE CONDITIONAL-PROBABILITY FIELDS FOR SOURCE LOCATIONS OF AIRBORNE POLLUTANTS

A receptor model has been developed in which meteorological informatio n in the form of air parcel back trajectories are combined with on the atmospheric constituent concentration data to produce conditional pro bability fields pointing to areas that are likely to have made signifi cant contributions to samples with higher than average concentrations. This approach, potential source contribution function (PSCF) analysis , has proven quite successful in producing maps that have a good corre spondence with areas of known high emissions on a variety of spatial s cales from large urban scale problems in the air basin that includes L os Angeles, CA to regional transport of pollutants to southern Ontario to semi-global scale transport to several sites in the high Arctic. H owever, there are cells having a limited numbers of endpoints because trajectories to that region have low probabilities and there is estima te of the uncertainties in the PSCF values. Thus, we have examined the use of bootstrapping to provide better estimates of the probability v alues and their uncertainties. This approach has been tested on data f rom several locations at differing levels of geographical scale for va rying numbers of trajectories selected and trials made. The results of the studies for data from the high Arctic at Ny Alesund on Spitsberge n (78 degrees 55' N, 11 degrees 57' E, 5 m above mean sea level) are p resented. The results of these studies for the transport of pollutants to the Arctic basin suggest that in many cases the bootstrapped PSCF maps are clearer and more easily interpreted in terms of known sources .