| 11 Jun 2026
Divergent Trends of Black and Brown Carbon Driven by Anthropogenic Emissions and Open Biomass Burning Across Asia
Abstract. Black carbon (BC) and brown carbon (BrC) are the dominant light-absorbing carbonaceous aerosols (LACs) and contribute substantially to regional climate warming. Across Asia, heterogeneous changes in anthropogenic and open biomass burning (OBB) emissions driven by clean air policies and climate variability are reshaping LAC composition, yet how these changes alter the relative abundance and radiative roles of BC and BrC remains poorly understood. Here, using the chemical transport model GEOS-Chem combined with machine-learning attribution, we quantify the responses of BC and BrC to concurrent emission changes across East, South, and Southeast Asia during 2012–2019. We identify pronounced spatial heterogeneity in BrC/BC mass ratio trends. East Asia exhibits a significant upward trend, driven by policy-induced suppression of BC-rich anthropogenic sources alongside enhanced BrC-rich OBB from a seasonal shift in crop residue burning under open-fire regulations. In contrast, South and Southeast Asia show declining ratios attributable to residential energy transitions and wildfire variability, respectively. These asymmetric responses propagate into direct radiative forcing (DRF), with the DRFBrC/DRFBC ratio increasing from 34.1% to 41.3% in East Asia while declining elsewhere. Notably, in East and Southeast Asia, OBB amounts to up to half of anthropogenically driven radiative forcing changes. Our results demonstrate that emission mitigation can redistribute rather than proportionally reduce LAC warming, highlighting that future climate benefits will critically depend on the concurrent management of open biomass burning.
Overall, the methodology is scientifically sound and employs a modern modelling framework. The combination of GEOS-Chem simulations, radiative transfer calculations, and machine-learning-based attribution provides a comprehensive approach for investigating regional changes in BC and BrC. Nevertheless, several aspects of the methodology rely on strong assumptions that require further justification or sensitivity analysis.
The parameterization of BrC is one such example. The estimation of primary BrC emissions depends on several assumptions regarding emission factors, mass absorption coefficients (MAC), and combustion characteristics. In particular, the adopted MAC values are subject to considerable variability in the literature and are known to depend strongly on aerosol source type, atmospheric processing, and measurement location. Given the sensitivity of the simulated BrC concentrations and radiative forcing to these parameters, the uncertainty associated with these assumptions should be discussed more explicitly and, if possible, quantified through sensitivity analyses.
The trend analysis also raises some concerns. The study period spans only approximately seven years, which is relatively short for deriving robust long-term trends, particularly in regions where large interannual variability associated with meteorology and biomass burning is expected. The robustness of the reported trends is therefore uncertain. Furthermore, although STL decomposition is an effective technique for separating seasonal and long-term variability and is useful for visualization, regression performed on the STL-derived trend component can underestimate the true uncertainty because the smoothing procedure reduces short-term variability and introduces serial dependence. A more rigorous statistical assessment of the trends, including confidence intervals and methods that explicitly account for autocorrelation, would strengthen the analysis.
The implementation of the Random Forest–SHAP framework is technically well executed; however, there are important methodological limitations that should be acknowledged. The Random Forest models are trained using GEOS-Chem simulation outputs as the target variables rather than independent observations. Consequently, the SHAP analysis explains the relationships learned from the GEOS-Chem simulations, rather than directly identifying the physical drivers of atmospheric variability. The distinction between explaining the behaviour of the model and demonstrating causal relationships in the real atmosphere should be made clearer throughout the manuscript.
The model evaluation is convincing for East Asia, where multiple long-term BC and OC observational datasets are available and the simulations reproduce both concentrations and temporal variability reasonably well. In contrast, the evaluation is considerably weaker for South Asia and Southeast Asia, where the analysis relies primarily on short-term campaign observations for concentration evaluation and does not validate the simulated long-term regional trends. Since the principal conclusions of the manuscript concern regional trend attribution, the authors should either incorporate long-term observational datasets from South and Southeast Asia or explicitly acknowledge that the regional trends and their attribution in these regions are largely model-derived and therefore subject to greater uncertainty.