Ambient and indoor air pollution is a major cause of premature mortality, and has been associated with more than three million preventative deaths per year worldwide. Most of these health impacts are from the effects from fine particulate matter. It is suspected that PM2.5 health effects vary by composition, which depends on the mixture of pollutants emitted by sources. This has led to efforts to estimate relationships between sources of PM2.5 and health effects. The health effects of PM2.5 may be preferentially dependent on specific species; however, recent work has suggested that health impacts may actually be caused by the net effect of the mixture of pollutants which make up PM2.5. Recently, there have been efforts to use source impacts from source apportionment (SA) studies as a proxy for these multipollutant effects. Source impacts can be quantified using both receptor and chemical transport models (RMs and CTMs), and have both advantages and limitations for their use in health studies. In this work, a technique is developed that reconciles differences between source apportionment (SA) models by ensemble-averaging source impacts results from several SA models. This method uses a two-step process to calculate the ensemble average. An initial ensemble average is used calculate new estimates of uncertainties for the individual SA methods that are used in the ensemble. Next, an updated ensemble average is calculated using the SA method uncertainties as weights. Finally, uncertainties of the ensemble average are calculated using propagation of errors that includes covariance terms. The ensemble technique is extended to include a Bayesian formulation of weights used in ensemble-averaging source impacts. In a Bayesian approach, probabilistic distributions of the parameters of interest are estimated using prior distributions, along with information from observed data. Ensemble averaging results in updated estimates of source impacts with lower uncertainties than individual SA methods. Overall uncertainties for ensemble-averaged source impacts were ~45 - 74%. The Bayesian approach also captures the expected seasonal variation of biomass burning and secondary impacts. Sensitivity analysis found that using non-informative prior weighting performed better than using weighting based on method-derived uncertainties. The Bayesian-based source impacts for biomass burning correlate better with observed levoglucosan (R2=0.66) and water soluble potassium (R2=0.63) than source impacts estimated using more traditional methods, and more closely agreed with observed total mass. Power spectra of the time series of biomass burning source impacts suggest that profiles/factors associated with this source have the greatest variability across methods and locations. A secondary focus of this work is to examine the impacts of biomass burning. First a field campaign was undertaken to measure emissions from prescribed fires. An emissions factor of 14±17 g PM2.5/kg fuel burned was determined. Water soluble organic carbon (WSOC) was highly correlated with potassium (K) (R2=.93) and levoglucosan (R2=0.98). Results using a biomass burning source profile derived from this work further indicate that source apportionment is sensitive to levels of potassium in biomass burning source profiles, underscoring the importance of quantifying local biomass burning source profiles. Second, the sensitivity of ambient PM2.5 to various fire and meteorological parameters in was examined using the method of principle components regression (PCR) to estimate sensitivity of PM2.5 to fire data and, observed and forecast meteorological parameters. PM2.5 showed significant sensitivity to PB, with a unit-based sensitivity of 3.2±1 µg m-3 PM2.5 per 1000 acres burned. PM2.5 had a negative sensitivity to dispersive parameters such as wind speed.