The Indo-Gangetic Plain, one of the most densely populated areas on earth, suffers from a combination of mobile source emissions, industrial point sources, agricultural biomass burning and other sources that lead to extreme levels of PM_2.5 concentrations. We analyze multiyear time series of hourly concentrations made by reference monitors in five major cities around the bay as well as measurements from low cost sensors at three locations spanning the meridional extent of the Maldives. We use a Generalized Additive Model (GAM) to estimate the impacts of long-range transport and biomass burning as well as local wind effects and vertical mixing. Trajectory Cluster Concentration Impacts (TCCI) were calculated based on hourly back-trajectories for all the sites. We estimate the impact of biomass burning by combining the back-trajectories with the Fire INventory from NCAR (FINN). The strength of the GAM simulations is that they calculate smooth splines that directly relate input variables to PM_2.5 concentrations. In this way, they are a type of Interpretable Machine Learning (IML) / eXplainable Artificial Intelligence (XAI) that provide quantitative information on the sources of PM_2.5. The method demonstrates that there is a region-wide plume of air pollution that is diluted as it moves south. We estimate that long-range transport accounts for 32 μg/m³ of PM_2.5 at Dhaka which corresponds to one third of the long-term average PM_2.5 loading. In the Maldives, long-range transport accounts for around 3 μg/m³ which also corresponds to one third of PM_2.5 concentrations. Biomass burning contributes 16 μg/m³ in Dhaka and Kolkata and around 3 μg/m³ in the Maldives. Overall, the results show that improving air quality in the region will require a concerted effort to reduce long-range transport and biomass burning, but also that local reduction