Background: Ambient fine particulate matter less than 2.5 µm in aerodynamic diameter (PM2.5) has been linked to various adverse health outcomes. However, the sparsity of air quality monitors greatly restricts the spatio-temporal coverage of PM2.5 measurements, limiting the accuracy of PM2.5 health studies and leading to the continual interest to develop and improve exposure estimates. Methods: We develop a method to combine estimates from two current data integration approaches for PM2.5: satellite-retrieved aerosol optical depth (AOD) and simulations from the Community Multi-scale Air Quality (CMAQ) modeling system. While most previous methods utilize AOD or CMAQ separately, we aim to leverage advantages offered by both methods in terms of resolution and coverage. Borrowing from weather forecasting techniques, we adapt Bayesian ensemble averaging to statistical downscaler models. The cross-validated predictive performance for AOD and CMAQ at a monitor determines the spatially varying averaging weights. Results: A preliminary analysis of 63 monitoring locations in the southeastern United States from 2003 to 2005 indicates good predictive performance of AOD and CMAQ individually with R2 values of 0.762 and 0.765, respectively, when estimating daily PM2.5 levels in cross-validation experiments. However, our ensemble approach shows that the usefulness of AOD and CMAQ differs spatially, as evidenced by estimated averaging weights varying from 0.024 to 0.969 for CMAQ; the ensemble averaging approach also increases the overall R2 to 0.802. Conclusion: The ensemble framework for harnessing predictive powers of multiple models can be used to improve health effect estimation and health impact assessment. The approach is also highly applicable for estimating other environmental risks, particularly in hydrology and climate science, that utilize information from both satellite imagery and numerical model simulation.