| 30 Jul 2025
Towards better black carbon emission estimates in Europe: assimilating observations with a Bayesian inversion framework
Abstract. There are large discrepancies between different black carbon (BC) bottom-up emission estimates. This affects our understanding of how BC affects population health and climate. Previous studies have presented top-down estimates in various domains but estimates in Europe for more than a few months are lacking. With the Lund University Modular Inversion Algorithm (LUMIA) and the FLEXible PARTicle dispersion model (FLEXPART), we aim to calculate top-down estimates of anthropogenic BC emissions by assimilating surface observations from 24 background sites in Europe (15° W–35° E, 33–73° N) during 2021. The results show that the bottom-up BC inventory generally underestimates emissions in the domain. Annually, the resulting top-down emissions are 411 ± 10 Gg of BC, 18 % higher than the bottom-up estimate of 349 ± 30 Gg used as a prior. The largest increases in emissions occur in Eastern parts of Europe during spring and summer, while emissions in Poland and Italy are reduced during winter and autumn. The overall posterior emissions are most sensitive to changes in the observational network on the periphery, resulting in a reduced standard deviation compared to the prior emissions in the central domain, where the network density is high. A cross-validation scheme, where one site at the time is removed for validation, show that the posterior for a majority of sites fit independent observations better than the prior emissions.
This study addresses significant discrepancies in European black carbon (BC) emission inventories by employing the LUMIA inversion algorithm and FLEXPART dispersion model to assimilate observational data from 24 background sites during 2021, thereby providing the top-down estimates of anthropogenic BC emissions at an annual scale. The study’s innovative high-resolution inversion of European BC emissions is commendable, but flaws in background concentration treatment undermine its core conclusions. The claim of “18% underestimated European BC emissions” lacks credibility unless critical methodological gaps are resolved.
Line 109: The settings for wet deposition efficiency (rainout/cloud scavenging) significantly impact BC lifetime (4-10 days), yet no tests evaluate how different parameters (e.g., snow scavenging efficiency) affect posterior emissions.
The FLEXible PARTicle (FLEXPART) dispersion model has too many assumptions. How did the authors address this issue?
Lines 128-130: The manuscript claims that “total emissions are irrelevant for background calculations; only the ratio between intra- and extra-domain emissions matters.” While theoretically valid under proportional scaling assumptions (e.g., doubling both intra- and extra-domain emissions preserves their relative contribution ratio to background concentration ybg), this contains serious flaws in practical application. The authors tested the impacts of prior uncertainties and observational errors but did not test the sensitivity of ybg to extra-domain emission errors. The manuscript simultaneously states “all emission sources are assumed to have identical mass absorption cross-sections (MAC)” (Line 169), yet MAC values for extra-domain BC (e.g., dust-influenced regions) versus intra-domain BC (e.g., European vehicular emissions) differ significantly, further invalidating the proportionality assumption. Line 405 indicates: “At most, 52% of the average concentration is attributed to ybg at Jungfraujoch (jfj) and Pic du Midi (pmi).” The poor performance of these sites in cross-validation (Figure 7) directly reflects the unreliability of background concentration calculations. If ybg contains biases, the system will misattribute external errors to European emissions, rendering the results invalid. The reported “18% underestimation of European BC emissions” may partly compensate for extra-domain emission/transport errors rather than reflect actual missing sources. How did the authors address this issue?
Line 184: Correction factors (1.76) for the AE33 aethalometer at five stations relied on manual verification, with methods for other stations unspecified. This risks introducing systematic biases into observational data.
Line 200: The manuscript only selected afternoon data (low-altitude sites) or nighttime data (high-altitude sites), discarding >50% of valid observations despite intending to reduce boundary-layer modeling errors. The authors did not verify whether this operation introduces selection bias.
From Section 3.2, it can be seen that the measurement of BC utilized multiple instruments and assumed that the MAC values from different sources would not change. However, the AAE of BC is not exactly equal to 1. How did the author consider the impact of these errors on the posterior results?
Line 261: Remove the redundant parenthesis after “dashed and dotted lines”.
Line 365: Significant increases in Eastern European emissions during spring/summer (Figure 5) potentially originate from agricultural burning (discussed in Sect. 6), yet no comparison was made with satellite fire detection data or existing fire emission inventories (e.g., GFED) for verification. How did the authors address this issue?