Satellite-based burned area reconstructions in land surface models: impacts on the land carbon cycle

Stripp, Tim; Guo, Zhixuan; Friedlingstein, Pierre; O’Sullivan, Michael; Sitch, Stephen; Bastos, Ana; Ciais, Philippe; Goncalves de Souza, Jefferson; Ito, Akihiko; Li, Wei; Ma, Lei

doi:10.5194/egusphere-2026-3584

Preprints

https://doi.org/10.5194/egusphere-2026-3584

Preprints

30 Jun 2026

| 30 Jun 2026

Satellite-based burned area reconstructions in land surface models: impacts on the land carbon cycle

Tim Stripp, Zhixuan Guo, Pierre Friedlingstein, Michael O’Sullivan, Stephen Sitch, Ana Bastos, Philippe Ciais, Jefferson Goncalves de Souza, Akihiko Ito, Wei Li, and Lei Ma

Abstract. Fire is a key Earth system process influencing vegetation distribution and the land carbon cycle, yet its representation within dynamic global vegetation (DGVM) models is generally poor, adding to the high level of inter-model uncertainty around simulated burned area (BA), fire emissions, and hence land carbon sink estimates. Here, we address this uncertainty by applying a satellite-based long-term reconstruction of burned area from 1901 to 2020 to five DGVM models, in order to assess the impacts of diagnostic BA on global carbon cycle estimates.

By diagnosing BA, we reduced the inter-model standard deviation in global fire emissions by 59 % compared to the prognostic BA simulations, and significantly increased estimate agreement with observations spatially and regionally. We also moderately improved net land carbon sink estimates, reducing bias by 0.3 PgC year^-1 against top-down constraints on average over 2001–2020, however the inter-model spread persisted. Additionally, regional simulated fire combustion rate (fire carbon emissions per unit BA) magnitudes and trends did not align with observations under diagnostic BA, influencing simulated land sink trends. This indicates poor representation of fire dynamics, specifically around the representation of fuel loads and combustion completeness, fire-induced mortality, fire intensity, and vegetation recovery processes.

We also explore land sink uncertainty associated with various diagnostic BA reconstructions using the JULES DGVM. By performing two JULES simulations using two reconstructed BA datasets derived from distinct satellite products (FireCCI5.1 and GFED5), we found a substantial difference in the land carbon sink (0.8 PgC year^-1) between the two simulations over 2001–2020. This shows that the underlying uncertainty in BA mapping and reconstructed BA trends can add a large source of uncertainty to modern day land sink estimates. Using a counterfactual simulation, we show that this uncertainty likely stems from differences in reconstructed pre-satellite era burned area dynamics leaving legacy impacts on both fire emission and ecosystem productivity estimates.

Our results show that although diagnostic burned area can greatly improve fire emission realism and constrain DGVMs, an enhanced representation of post-fire ecosystem impacts, fire intensity and fuel loads is needed in order to capture regional impacts of fire on the carbon cycle.

Received: 18 Jun 2026 – Discussion started: 30 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

Preprint (PDF, 4883 KB)

Supplement (5961 KB)

Download & links

Tim Stripp, Zhixuan Guo, Pierre Friedlingstein, Michael O’Sullivan, Stephen Sitch, Ana Bastos, Philippe Ciais, Jefferson Goncalves de Souza, Akihiko Ito, Wei Li, and Lei Ma

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-3584', Anonymous Referee #1, 08 Sep 2026
Overall assessment
This manuscript applies long-term satellite-based burned-area reconstructions to five DGVMs to assess how diagnostic burned area affects fire emissions and the terrestrial carbon sink, with additional JULES experiments examining the legacy effects of historical burned-area uncertainty. The results show that prescribing BA reduces fire-emission spread and bias, while substantial disagreement remains in combustion rate and land-sink responses. While the study is relevant and the combustion-rate and historical-legacy analyses are potentially valuable, I have several major concerns regarding the framing and novelty of the main conclusions, the independence of the emission evaluation, and the strength of the process attribution and generalization of the JULES results.

Major comments:
The manuscript places substantial emphasis on the reduction in fire-emission bias and inter-model spread after prescribing satellite-constrained BA. This is useful but also expected, since BA is a direct driver of fire emissions and a major source of disagreement among prognostic fire models. In addition, the GFED emission benchmarks are not fully independent of the prescribed BA, so part of the apparent improvement in “realism” is imposed by construction. In my view, the manuscript contains two more scientifically important results that deserve much greater emphasis. First, once BA is controlled, substantial disagreement remains in fire emissions per unit BA, revealing structural uncertainty in fuel, combustion, mortality, and related processes. Second, the JULES experiments show that historical BA differences can propagate into modern NBP through NEP, not only direct fire emissions. I recommend reorganizing the Results and Discussion around this hierarchy, with the BA to emission improvement presented more concisely as an expected baseline and the other two having more emphasis. This framing would make the manuscript's genuine scientific contribution much clearer and more compelling.

The combustion-rate analysis is one of the strongest parts of the manuscript because, after BA is constrained, substantial inter-model disagreement remains in fire emissions per unit burned area, including large regional biases and failure to reproduce the observed positive trend. However, the discussion sometimes moves too quickly from this diagnosis to specific causal explanations involving fuel loads, combustion completeness, fire intensity, mortality, and SOC combustion. These are plausible mechanisms, but the current analysis does not isolate their individual contributions. For example, ORCHIDEE explicitly represents fire intensity but still produces a Boreal Asia trend substantially weaker than GFED, suggesting that the bias cannot be attributed simply to the absence of fire-intensity representation. I therefore recommend tempering the causal language and presenting these processes as plausible sources of disagreement unless additional process-level or sensitivity analyses are provided.

The 0.8 PgC yr⁻¹ difference in modern NBP between the two JULES diagnostic BA simulations is important, but it is a single-model result. Its magnitude depends on JULES-specific representations of post-fire mortality, decomposition, and carbon-pool dynamics, which the manuscript itself acknowledges may exaggerate the NEP response. The result supports a potentially important historical-BA legacy effect in JULES, but not a 0.8 PgC yr⁻¹ uncertainty across DGVMs. This model dependence should be made much more explicit.

The manuscript interprets the 0.8 PgC yr⁻¹ difference between the DSCCI and DSGFED JULES simulations primarily as a legacy effect of differing pre-satellite BA dynamics. Historical BA dynamics clearly matter, as demonstrated by the counterfactual experiment. However, this experiment shows that one historical BA trajectory can produce a legacy effect; it does not demonstrate that differences in historical temporal trends explain the full DSCCI–DSGFED NBP difference. The two reconstructions also differ in mean BA, spatial distribution, and treatment of land-use-associated fires. All of these differences could contribute to the NBP divergence. The attribution should therefore be softened, or additional sensitivity experiments should separate temporal-history effects from differences in BA magnitude and spatial pattern.

Minor comments:
Lines 239: “Diagnosing BA reduced the multi-model mean land sink by 0.3±0.4 PgC year⁻¹ over 2001–2020 compared to the prognostic simulations, despite the concurrent reduction in fire emissions.” The reason for this seemingly counterintuitive response is not clearly explained. Since lower fire emissions alone should increase NBP, the reduced land sink must be driven by changes in NEP that more than offset this effect. Please clarify and, if possible, quantify the contributions from NPP and HR.

Line 270 “This NBP estimate in the simulation with GFED5 diagnostic BA fell below all GCB inversion and land-use dataset change combinations (minimum of 1.2 PgC year−1).” Please also report the mean of the GCB inversion estimates, not only the minimum. From Fig. 6, the PS simulation appears to align more closely with the inversion mean than both diagnostic simulations, which should also be discussed.

Line 292: “The magnitude of year-to-year differences in NBP between the counterfactual and original simulations were aligned much more closely with year to year differences in forest BA rather than total BA” This wording is unclear because NBP and BA have different units. Please clarify whether “aligned” refers to temporal correlation or another quantitative relationship, perhaps referring directly to the R² analysis reported immediately afterward.

Line 298: “Therefore, the counterfactual NBP was 2.7 PgC year⁻¹ over this period which is similar to the PS average of 2.6 PgC year⁻¹ and lies well within 1 S.D of the mean of the GCB inversions and close to the GCB residual sink.” The implication of this comparison is unclear. Is the agreement with the GCB residual intended to suggest that the counterfactual represents a more realistic historical BA trajectory? If not, please clarify what conclusion should be drawn from the closer agreement.

Line 350: “However, vegetation killed by fire does not form a component of fire emissions in ORCHIDEE, so variations in fire-induced mortality do not impact fire emission estimates.” I do not think vegetation killed by fire should necessarily be considered part of direct fire emissions. GFED also primarily represents carbon emitted during combustion rather than subsequent mortality losses.

Lines 361: "Although uncertainties exist around the partitioning of burned area across different land cover classes … suggesting the partitioning of burned area across different land cover classes is less uncertain in these regions" is difficult to follow. Please rephrase to clarify why dominance of woody land cover in boreal regions reduces uncertainty in the partitioning of BA among land-cover classes.

Line 367 “This further emphasizes the need for models to produce estimates of fire intensity and relate these to emissions in order to better replicate observed magnitudes and trends in fire combustion rates.” Does ORCHIDEE, which explicitly calculates fire intensity, actually reproduce the Boreal Asia trend better than the other models? In Lines 224–226, ORCHIDEE is grouped with VISIT-UT and OCN as producing significant but substantially weaker trends than GFED. If including fire intensity does not clearly improve the trend in ORCHIDEE, the argument that missing fire intensity is a primary explanation for the model bias should be qualified.
Citation: https://doi.org/10.5194/egusphere-2026-3584-RC1
- AC1: 'Reply on RC1', Tim Stripp, 29 Sep 2026
  
  Thank you for providing these constructive comments.
  In terms of the major comments, we understand your concerns regarding the framing of the fire-emission improvements. We will revise the framing to place less emphasis on these emission improvements, and a greater emphasis on the uncertainty in fire emissions per unit burned area and the influence of historical burned area dynamics on contemporary NBP. Also, we will soften the language regarding the potential causes of uncertainty in fire emissions per unit BA, and highlight the fact that the 0.8 PgC per year difference in NBP between the two diagnostic BA JULES simulations may be JULES specific. Finally, we will further consider the most appropriate way to assess the influence of pre-satellite burned-area dynamics on this difference in NBP.
  We also appreciate the minor comments and will address these in the revised manuscript.
  
  Citation: https://doi.org/10.5194/egusphere-2026-3584-AC1
RC2:
'Comment on egusphere-2026-3584', Anonymous Referee #2, 09 Oct 2026
Major Comments
In the Results, section 4, the spatial analysis is gradually reduced as the paper progresses, from a grid-cell-scale analysis to a more regional-scale aggregation, and finally to being almost absent in the later sections. It would be helpful if the authors could include a short paragraph about this change in spatial scale. If the reason the coarser resolution is used is due to the availability of the data to resolve features at smaller scales, comparability of models, the concept of observational uncertainty, or to facilitate interpretability then this should be stated. This might help the reader understand whether the reduction in scale is due to the nature of the data or some other factor.

Figure 2 provides important comparison of the global fire-emission estimates, yet it is not explicitly cited or analyzed in the main text. The authors could consider to cite it in the Results section and to provide a brief analysis of its main implications. In particular, it would be important to analyze the differences across the prognostic and diagnostic simulations, the reduction of the fire-emission estimate using the diagnostic burned area, and the large difference between the GFED4.1s and GFED5 reference estimates, with implications for assessing how well the simulations compare to observational constraints.

Minor Comments
In the Abstract, the manuscript uses the phrase “dynamic global vegetation (DGVM) models.” The authors may consider moving the abbreviation to the end of the full term, i.e., “dynamic global vegetation models (DGVMs),” which would also be consistent with the terminology used throughout the rest of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2026-3584-RC2

Tim Stripp, Zhixuan Guo, Pierre Friedlingstein, Michael O’Sullivan, Stephen Sitch, Ana Bastos, Philippe Ciais, Jefferson Goncalves de Souza, Akihiko Ito, Wei Li, and Lei Ma

Supplement

https://doi.org/10.5194/egusphere-2026-3584-supplement

Tim Stripp, Zhixuan Guo, Pierre Friedlingstein, Michael O’Sullivan, Stephen Sitch, Ana Bastos, Philippe Ciais, Jefferson Goncalves de Souza, Akihiko Ito, Wei Li, and Lei Ma

Viewed

Total article views: 413 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
265	101	47	413	50	38	46

HTML: 265
PDF: 101
XML: 47
Total: 413
Supplement: 50
BibTeX: 38
EndNote: 46

Views and downloads (calculated since 30 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	53	32	9	94
Aug 2026	39	15	7	61
Sep 2026	149	45	25	219
Oct 2026	24	9	6	39

Cumulative views and downloads (calculated since 30 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	53	32	9	94
Aug 2026	39	15	7	61
Sep 2026	149	45	25	219
Oct 2026	24	9	6	39

Viewed (geographical distribution)

Total article views: 368 (including HTML, PDF, and XML) Thereof 368 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 09 Oct 2026

Short summary

Fire is a key process within the Earth system, yet it is often poorly represented in land carbon cycle models used to estimate how much carbon land ecosystems absorb from the atmosphere. We improved these models by applying long-term burned area data reconstructed from satellite observations. This reduced biases and uncertainty in fire carbon emission estimates and improved estimates of land carbon uptake, implying this method could enhance global and regional carbon budget estimates.


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