Reviews and syntheses: Spatiotemporal Dynamics, Drivers, and Uncertainties of Global Wildfire Carbon Dioxide Emissions

Liang, Yu; Ma, Tianxiao; Liu, Bo; Liu, Zhihua; He, Hong S.; Yang, Jian; Pan, Yude; Yue, Chao; Wang, Xianli; Wu, Mia; Xu, Wenru; Zhu, Jiaojun

doi:10.5194/egusphere-2026-2906

Preprints

https://doi.org/10.5194/egusphere-2026-2906

Preprints

04 Jun 2026

| 04 Jun 2026

Reviews and syntheses: Spatiotemporal Dynamics, Drivers, and Uncertainties of Global Wildfire Carbon Dioxide Emissions

Yu Liang, Tianxiao Ma, Bo Liu, Zhihua Liu, Hong S. He, Jian Yang, Yude Pan, Chao Yue, Xianli Wang, Mia Wu, Wenru Xu, and Jiaojun Zhu

Abstract. Wildfire carbon dioxide emissions (WCEs) are increasingly recognized as a major and highly uncertain component of the global biogeochemical carbon cycle, reflecting limitations in constraining their historical evolution, future trajectories, and controlling mechanisms. This review synthesizes evidence from satellite-based emission products, fire-enabled model reconstructions and scenario projections, and paleoenvironmental archives (charcoal and ice-core black carbon) to evaluate global WCE dynamics from 1700 to 2100. Historical reconstructions indicate relatively stable global WCEs during 1700–1850, whereas pronounced divergence emerges during 1851–2000 due to differing representations of land-use change and industrialization. Contemporary satellite observations show declining WCEs over 2001–2020 (8.72 ± 0.67 Pg CO₂ yr⁻¹), with 83 % originating from tropical ecosystems because of large burned area and high combustion efficiency. Importantly, these fluxes represent a critical shift in the net carbon balance of tropical and boreal biomes. Multi-model projections suggest increases through the 2040s, followed by growing divergence under alternative socioeconomic pathways. Across datasets and models, annual WCE estimates differ by up to 40 %, driven by uncertainties in fire detection, combustion completeness, emission factors, and fire–climate–human interactions. Despite regional heterogeneity, climate change emerges as the primary regulator of interannual variability in WCEs. Approximately 43 % of global vegetated areas have experienced increasing extreme wildfire seasons, particularly in northern high-latitude forests where recurrent burning amplifies carbon losses and weakens ecosystem carbon sinks. We conclude by identifying priorities for reducing uncertainty through tighter integration of multi-source observations with fire-enabled process models, improved representation of coupled fire–climate–carbon feedback and post-fire recovery, and the application of artificial intelligence to better constrain WCE spatiotemporal variability and enhance predictive capability under increasingly nonlinear fire–climate interactions.

Received: 20 May 2026 – Discussion started: 04 Jun 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Yu Liang, Tianxiao Ma, Bo Liu, Zhihua Liu, Hong S. He, Jian Yang, Yude Pan, Chao Yue, Xianli Wang, Mia Wu, Wenru Xu, and Jiaojun Zhu

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-2906', Anonymous Referee #1, 18 Jun 2026
Review of “Spatiotemporal Dynamics, Drivers, and Uncertainties of Global Wildfire Carbon Dioxide Emissions” by Liang and co-authors. This paper synthesizes evidence on global wildfire carbon dioxide emissions (WCEs) from 1700 to 2100 by integrating multiple sources of analyses, including satellite-based fire emission products, fire-enabled dynamic vegetation and Earth system models, and paleoenvironmental proxies such as charcoal and ice-core black carbon records. Using these complementary datasets, the authors evaluate historical trends, contemporary spatial patterns, future projections, key drivers, associated uncertainties, and they come up with ways to improve this
Review papers are very helpful but I felt the authors are not fully up to speed with the research field they reviewed (mostly concerning the description of the contemporary satellite-based era), and therefore may cause more confusion than clarity. That is a major concern which is not one that can be easily addressed. Another major concern -which can be addressed- is that the paper is outdated. Simple example, since Roteta et al. (2019, https://doi.org/10.1016/j.rse.2018.12.011) we know the coarse resolution burned area datasets -and their derived emissions- are biased low, but in this review paper those coarse resolution datasets are still described and averaged as the best available knowledge.
Please find below a list of examples of verbose, outdated, and occasionally unfocused writing. Simply addressing these would not make this paper publishable in my opinion (also because this is just a subset of the issues). Converting this paper to a publishable version requires a major overhaul, providing much more depth, including analysing the literature that provides independent information on fire patterns compared to the models (e.g., fire proxies), but most of all becoming familiar with the most recent literature. I apologize for being so negative as I do realize the authors have done their best.
L23: “8.72 ± 0.67 Pg CO₂ yr⁻¹” does not refer to the trend but is the average I assume? That makes it a somewhat confusing sentence. Also, it may be good to have fewer significant digits for numbers so uncertain. And it is not clear where the 0.67 refers to. Guess spread between inventories? Good to clarify

L25: “high combustion efficiency”. In the fire literature combustion efficiency usually refers to the fraction of total carbon emitted as CO2, but I doubt that is what you mean here?

L25: “Importantly, these fluxes represent a critical shift in the net carbon balance of tropical and boreal biome”. One can argue that fires in tropical forests change the carbon balance, but fires in savannas and the Boreal zone have been there for a long time and there needs to be some discussion about background versus elevated fire levels.

L30: “Despite regional heterogeneity, climate change emerges as the primary regulator of interannual variability in WCE”. In the main text you rightfully state that interannual variability is related to ENSO, but that is not climate change.

L43 “dominant source of the atmospheric CO₂ “ -> dominant source of the rise in atmospheric CO₂

L44: “Meanwhile, satellite observations from this period suggest that wildfires annually burned 3.8–4.8% of global vegetated land and wildfire carbon dioxide emissions (WCEs) emitted approximately 20% of CO₂ released from fossil fuel combustion (Jones et al., 2022; Van Der Werf et al., 2017; Zheng et al., 2021)”. Chen et al. (2023, https://doi.org/10.5194/essd-15-5227-2023) reports 5.9%. To me it looks the authors are not aware of the latest literature and developments which is crucial when writing a review. In addition, comparing WCE with fossil fuel emissions or with annual CO2 increases is comparing apples and oranges; the vast majority of fire emissions are balanced by regrowth and not a net source of CO2. The way it is framed now (“WCEs accounted for 35–52% of the net annual atmospheric CO₂ increase“) is simply wrong. What are the insights from recent studies using higher resolution burned area (for example Sentinel-2 or Landsat)? How do the estimated emissions compare to atmospheric inversion studies? Why is the recent version of GFED (GFED5) so much higher, etc. That is the kind of information that would be good to focus a review paper on and would push the field forwards. The same counts for a proper discussion and evaluation of fire models.

L94: “The ensemble mean of four fire emission datasets (Table 1), including the Global Fire Emissions Dataset with small fires (GFEDv4s), the Global Fire Assimilation System Dataset (GFASv1.2), the Quick Fire Emissions Dataset (QFEDv2.6r1), and the Fire Energetics and Emissions Research Dataset (FEER), indicates that the global burned area averaged 440 ± 50 Mha yr-1 from 2001 to 2020”. Then Table 1 shows that only one of those databases is based on burned area, can one call this an ensemble estimate then? And how should I interpret the ± 50 Mha? Please be crystal clear with your writing and take the reader by the arm so (s)he does not have to guess what you mean.

L103: “Tropical fires, primarily within 20°S-15° N, account for approximately 83% of global WCEs (Fig. 1), largely due to tropical dry forests in Amazonia.“ In most datasets African fire emissions exceed those from South America, and tropical dry forests may be a relatively small fraction of South American forests.

L117 “–32.92 Tg CO₂ yr⁻¹, significant” So the other ones are not significant? I guess so, but please be clear in how you write things down

L119 “seven major positive hotspots” Not clear what you are after and how these are defined

L137-147 is very confusing and to me it is not clear what you want to say here, also because there is a downward trend in burned area in Africa

L167 “Although global emissions remained relatively stable prior to 1850 (±0.1% yr⁻¹), reflecting predominantly natural climate variability”. Please make sure readers understand that the emissions being stable is a model output. In a review I expect also to get some background on why climate anomaloies like the Medieval Warm Period do not show up, and why some reconstructions based on proxies give different results. For the post 1850 period you do give some insight in potential dynamics and deficiences later on which is great, but there (L177-178) you also draw very firm conclusions. Please elaborate which DGVM’s have these deficiencies, have they been improved over time, etc.

L182 “Between 2001 and 2020, the four widely used datasets (GFED, QFED, CAMS, and FEER) reveal broadly comparable global averages” This is not fully surprising as CAMS is tuned to GFED3, and QFED also underwent tuning. Therefore the range is also something completely different than the uncertainty. A discussion about this (and include GFED5) would be useful for the readers.

L195: The use of SSP5-85 requires a bit of discussion, it is not just a “high emission scenario”. Best to avoid using it or provide a very strong disclaimer (see Van Vuuren et al., 2026, https://doi.org/10.5194/gmd-19-2627-2026)

Figure 4; trends for Canada will change dramatically when adding 2023-2025. This also highlights how sensitive the trends are to individual years

L245 “and fire emission inventories”. So uncertainties in fire emissions inventories are uncertain because of assumption in fire emissions inventories?

L245 combustion completeness refers to more than burning depth

Table 2 would benefit from possible a general part (factors that are everywhere the same) and a region specific part. Also, a figure may be better here.

L322: “Specifically, projections suggest a 29% global increase in area burned due to climate change”. Which projections? How do they fit in the earlier story wrt future fire?

L344: In a paper about WCE I would expect this equation to be mentioned much earlier, and then a systematic discussion about the various parameters

L367: So the warming will be around 4 degree (50% / 12%) over this century? Or does that depend on the climate scenario? This could be a misunderstanding from my part, apologies, but it would help to write this down in a different way.

L355: 7% in all types of forests?

L408: “This operational definition identifies wildfire seasons characterized by exceptionally large, fast-spreading, and intense wildfires that deviate markedly from the historical range of variability in burned area, fire severity, and associated carbon emissions. “ How do we know this?

L456: “far exceeding typical annual Canadian emissions“. From fires or from fossil fuel burning?

L467: Greenpeace Internaional (2021) is not scientific literature, please be careful with using grey literature

L480: “The fires burned an estimated 24 to 33 million hectares and emitted about 0.72 Pg CO₂ (Binskin et al., 2020; Van Der Velde et al., 2021), accounting for approximately one quarter of global WCEs in 2019.” How should I reconcile your text that 0.72 Pg C is a quarter of global WCE (which would thus be 4 x 0.72 = 2.88 Pg CO2) with your earlier number of 8.7 Pg CO2?

L491: “representing an 83 percent increase compared with 2018” Yes, but 2018 was a very low fire year. In total, 2019 was not exceptional in the Amazon and according to most of the datasets used even below average

Figure 6b is mentioned in the caption but I do not see it

Finally, I was wondering why the paper is about carbon dioxide and not simply carbon? There is no discussion about CO2 emission factors and all the text refers to the carbon dynamics of fires.
Citation: https://doi.org/10.5194/egusphere-2026-2906-RC1
- AC1: 'Reply on RC1', Yu Liang, 08 Aug 2026
  
  The comments and responses were uploaded in the form of a supplement.
  
  Citation: https://doi.org/10.5194/egusphere-2026-2906-AC1
RC2:
'Comment on egusphere-2026-2906', Anonymous Referee #2, 19 Aug 2026
In their manuscript ‘Reviews and syntheses: Spatiotemporal Dynamics, Drivers, and Uncertainties of Global Wildfire Carbon Dioxide Emissions’, Liang et al. integrate multiple datasets, including remotely-sensed and offline and coupled model simulations, to understand different aspects of global wildfire carbon dioxide emissions, including spatial patterns, trends in present and future projections, drivers and associated uncertainties, and finally make suggestions on future research directions.
While I think idea of the paper is interesting and timely, I have many concerns. Given I recommend a rejection based on this, I will only provide my main criticisms:
In general, I would describe this manuscript as a literature review rather than a synthesis. I appreciate that this is a massive undertaking, however, I find the different areas the authors cover are descriptive and disconnected: large parts of text for example quantify extensively uncertainties the authors find in custom analyses of different data sources, which I think during the description could have used more direct context of existing literature given this is a review, and then in different parts in similar detail drivers are described and also potential sources of uncertainty. What I’m missing for this to be a useful synthesis however is linking those findings: Based on literature, can the authors come to any conclusions on what driver drives uncertainty where, to which degree, and what would be pathways forward to constrain this uncertainty?

My impression is that in parts the authors need to be more careful about the presentation of data sources and what they mean / what the most up-to-date sources are:

The abstract boldly states ‘paleoenvironmental archives (charcoal and ice-core black carbon)’ are evaluated in the manuscript. Doing a quick search of ‘paleo’ in the manuscript renders 6 results, all of which mention other studies have used those types of data without mentioning any actual conclusions, and therefore don’t offer valuable insights on what can be gained from those paleoenvironmental archives.

The authors have not included GFED5, despite it being available before submission of the manuscript. Given GFED5 also provides a significantly different estimate in burned area and therefore presumably fire carbon emissions, this would be an crucial addition to this manuscript.

If I understand correctly, the authors rely on the FireMIP tier 1 simulations ranging from 1700-1900 for their analysis of historical reconstructions. Again, I’m missing the acknowledgment of the shortcomings of this part of the simulation. The experiment protocol reads ‘The historic simulations were run from 1700 through 2013. Population and land use were changed annually from the beginning of this simulation, and CO2 values were changed annually from 1751 onwards. However, because the CRU-NCEP and lightning forcing data were not available for 1700–1900, the 1901–1920 forcings were recycled for the first 200 years of the simulation; this allowed natural climate variability to be captured while incorporating only minimal human influence. From 1901 to 2010, time-varying values of all variables were used.‘ (Rabin et al., 2017). The data can of course be used as a first-order estimate of historical fire activity, but I think the manuscript should be much clearer that these are model simulations rather than proxy-constrained reconstructions. In particular, the recycling of 1901-1920 reanalysis and lightning forcing over 1700-1900 limits the realism of temporal variability in this period and should be acknowledged when interpreting these results.

I am also concerned that some of the interpretations suggest that the authors may not have sufficiently distinguished between processes that are physically relevant to wildfire emissions and processes that are explicitly represented in the datasets and models being analysed. For example, they claim i) ‘strengthened fire-suppression policies’ are accounted for in the moderate SSP1-2.6 and SSP2-4.5 scenarios without providing a reference (I looked through the scenario descriptions and couldn’t find any support for this statement), ii) highlight permafrost and peatland feedbacks as a likely source for increasing fire emissions when not many fire-enable models include these processes, iii) mis-identify LPJ-GUESS-SIMFIRE-BLAZE as a model with explicit anthropogenic ignitions although Rabin et al. (2017) explicitly classifies LPJ-GUESS-SIMFIRE-BLAZE as a purely empirical model with no concept of ignition count, iv) implicitly oversimply anthropogenic suppression to reductions in number of ignitions, v) if the future projections are based on concentration-driven CMIP6 ScenarioMIP simulations, these cannot capture the full climate-fire-carbon feedback because carbon-cycle changes do not feed back onto prescribed atmospheric CO₂ concentrations [...]

I like the suggested analysis in Figure 6, however find the conclusions the authors make based on it quite speculative. It might be the way the results are presented, but to me the map looks like there is substantial spatial variability that makes it hard to identify the clear patterns they authors appear to find.

The recommendations for future research directions remain very general. Many of the proposed future directions would apply broadly to almost any Earth-system process. I would expect a synthesis focused specifically on wildfire carbon emissions to identify more concrete, fire-specific research priorities emerging from the uncertainties diagnosed in this study.
Citation: https://doi.org/10.5194/egusphere-2026-2906-RC2

Yu Liang, Tianxiao Ma, Bo Liu, Zhihua Liu, Hong S. He, Jian Yang, Yude Pan, Chao Yue, Xianli Wang, Mia Wu, Wenru Xu, and Jiaojun Zhu

Viewed

Total article views: 309 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
135	161	13	309	7	15

HTML: 135
PDF: 161
XML: 13
Total: 309
BibTeX: 7
EndNote: 15

Views and downloads (calculated since 04 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	38	12	4	54
Jul 2026	60	39	5	104
Aug 2026	37	110	4	151
Sep 2026	0

Cumulative views and downloads (calculated since 04 Jun 2026)

Month	HTML	PDF	XML	Total
Jun 2026	38	12	4	54
Jul 2026	60	39	5	104
Aug 2026	37	110	4	151
Sep 2026	0

Viewed (geographical distribution)

Total article views: 262 (including HTML, PDF, and XML) Thereof 262 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 03 Sep 2026

Short summary

Wildfire CO₂ emissions are a major but uncertain part of the global carbon cycle. We reviewed satellite emission datasets, fire-enabled models, future projections, and paleo-archives to assess global wildfire CO₂ emissions from 1700 to 2100. Annual estimates differ by up to 40 % because of uncertainties in fire detection, fuel consumption, and emission factors. Tropical fires dominate global CO₂ wildfire emissions, while extreme wildfire seasons increasingly shape year-to-year variability.


Total:	0
HTML:	0
PDF:	0
XML:	0