Revisiting black carbon emission estimates in the Third Pole: a hierarchical Bayesian synthesis inversion perspective

Reliable estimates of black carbon (BC) emissions over the Third Pole, also referred to as High Mountain Asia (HMA), remain challenging due to sparse observations, complex terrain, and uncertainties in emission inventories. We present a hierarchical Bayesian synthesis inversion framework that integrates surface BC observations from 91 sites with MATCHA, a 12 km regional chemical reanalysis based on WRF-Chem spanning 2003–2019 with tagged anthropogenic BC tracers from 10 Asian regions. We perform inversions using both daily and monthly observations to optimize regional BC emissions and quantify errors in both priors and observations. Monthly inversions provide the best performance, resolving 8–9 source tags and reducing biases at pollution hotspots and high-altitude sites. Posterior emissions show up to a fourfold underestimation for Tibetan Plateau and Bangladesh, moderate adjustments for China and India (±20%), and underestimated biomass-burning contributions over Southeast Asia. Sensitivity analysis reveals that site density and strategic location, particularly remote, high-altitude stations, strongly influence inversion performance, more so than observation duration. Transport errors dominate model–observation mismatches, particularly in vertical mixing and long-range horizontal advection across complex terrain. A machine learning (ML) surrogate of the model sensitivity matrix confirms underrepresentation of source-receptor pathways in WRF-Chem, especially over complex terrain. Our results underscore the importance of improving emission inventories in underrepresented regions, enhancing vertical transport and deposition processes in models, and expanding observational networks. The integrated Bayesian-ML approach provides a robust framework for refining BC emissions that can contribute to more accurate climate impact assessments over HMA.