| 18 Aug 2025
Structural uncertainty in the direct human forcing of future global burned area
Abstract. The first fire model intercomparison project (FIREMIP) gave rise to two distinct proposals around how best to improve the fire modules of dynamic global vegetation models. The first proposal was to develop representation of direct human impacts on burned area, particularly managed fire use in agriculture and other land management. The second proposal was to improve representation of the ecological dimensions of fire, including relationships of fuel load, connectivity, dryness and fire. Here, we present future projections from two models that have attempted to advance model representation and understanding of the human (WHAM-INFERNO) and ecological (Haas) dimensions of global fire regimes. The models project radically different future burned area for the same sets of scenario forcings. There is particularly strong disagreement regarding direct human impacts (or “direct human forcing”) of global burned area: differences in model assessment of the impact of direct human forcing is greater between models than between scenarios. We show how such structural uncertainty constrains understanding of climate change adaptation, including its limits and pitfalls. Differences in model outputs are largely traceable back to model assumptions. Hence, we argue that advances made by the two models could be combined in a future fire model that better captures the socio-economic and ecological drivers of burned area. We identify key challenges to the development of such integrated socio-ecological models, highlighting crucial uncertainties around how anthropogenic and biophysical factors interact to produce patterns of fuel fragmentation and hence fire spread. Overall, advancing understanding of the interactions between human and biophysical drivers of fire remains a central challenge in fire science.
This paper compares future burned area projections from two global fire models — WHAM-INFERNO (focused on socio-economic drivers) and the Haas GLM (focused on ecological drivers) — under SSP1-2.6 and SSP3-7.0, finding that the two models disagree radically on future burned area, particularly regarding the role of direct human forcing. The topic is timely and the comparison is a useful contribution. However, the generalizability is limited by reliance on a single ESM and only two SSPs, and the central claim lacks formal statistical support, together with some other minor issues. I recommend moderate revision.
Major Comments
1. Limited scenario and ESM diversity undermines generalizability. The study uses only two SSPs (SSP1-2.6 and SSP3-7.0) forced by a single Earth System Model (UKESM). The authors themselves acknowledge that UKESM projects unusually rapid change in the Amazon (lines 503–505). Using only one ESM makes it impossible to disentangle structural uncertainty in the fire models from forcing uncertainty. At minimum, a middle-of-the-road scenario (SSP2-4.5) and one additional ESM should be included, or the limitation needs much more detailed discussion with clear justification for why this was not done.
2. The "direct human forcing" experiment design conflates multiple effects. The 2020 DHF runs hold cropland constant but allow natural vegetation to change transiently (lines 213–215). This is a reasonable pragmatic choice, but it means the DHF signal includes interactions between fixed human land use and dynamically changing natural vegetation. The paper does not adequately discuss how this confounding affects interpretation. A sensitivity test, or at least a thorough discussion of this issue, is needed.
3. No formal uncertainty quantification or statistical testing of inter-model differences. The paper's central claim — that inter-model disagreement on direct human forcing exceeds inter-scenario disagreement — is supported only by comparing Pearson correlation coefficients (e.g., r = 0.35 between models vs. r = 0.52 within WHAM-INFERNO across SSPs; lines 357–360). No confidence intervals, significance tests, or formal uncertainty decomposition are provided. Given that this is the paper's headline finding, more rigorous statistical support is essential.
Minor Comments
4. Lines 341–348: The IAV comparison finds that WHAM-INFERNO underestimates and the Haas model overestimates IAV relative to GFED5. However, the Haas model's IAV is substantially higher in SSP1 (61.8 Mha) than in SSP3 (40.8 Mha), which is counterintuitive — shouldn't stronger climate forcing produce higher IAV? This deserves explanation.
5. Line 171: Typo — "Mangeon et al., 20216" should be "Mangeon et al., 2016."
6. Lines 475–483: The discussion of anthropogenic fragmentation is framed almost entirely in terms of fire spread mechanics. However, the motivation for better representing fragmentation in fire models should also be grounded in its demonstrated ecological importance — for example, anthropogenic fragmentation significantly impacts biodiversity (https://doi.org/10.1111/brv.12519) and animal behavioral adaptation (https://doi.org/10.1002/ece3.71721).
7. Lines 232–238: The comparison of IAV between model outputs for 2020–2029 and GFED5 observations for 2011–2020 is acknowledged as indirect, but the decade shift also implies different climate baselines. This issue should be stated more explicitly.