the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
An integrated framework for wildfire risk assessment under climate change: empirical validation and application in Greece
Abstract. Climate change is increasing the frequency and severity of wildfires across the Mediterranean, posing growing risks to ecosystems, infrastructure and human communities. Robust methodologies for assessing wildfire risk under both present and future climate conditions are therefore essential to support climate change adaptation. In this study, we develop and empirically validate an integrated wildfire risk assessment framework for Greece following the Intergovernmental Panel on Climate Change risk concept. Wildfire hazard, exposure and vulnerability are represented through climate, environmental and socioeconomic indicators, while alternative hazard representations and aggregation approaches were systematically evaluated against observed wildfire occurrence to identify the most accurate methodological configuration. The validated framework was then applied to assess wildfire risk for the reference period (1995–2014) and under SSP2-4.5 and SSP5-8.5 scenarios for the near (2041–2060) and distant (2081–2100) future. The results demonstrated that aggregation behaviour plays a critical role in determining the predictive performance of composite wildfire risk indices, with an Ordered Weighted Averaging (OWA) approach providing a statistically significant improvement over conventional linear aggregation. Analysis of component contributions identified hazard as the dominant contributor to wildfire risk across most municipalities, although exposure and vulnerability remained important determinants of the final risk patterns. Future projections indicated a widespread increase in wildfire risk throughout Greece, with the largest increases occurring under the high-end SSP5-8.5 scenario in the late century. The proposed framework provides a transparent and empirically validated basis for wildfire risk assessment that can support climate change adaptation planning at national and regional scales.
- Preprint
(1888 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 19 Oct 2026)
- RC1: 'Comment on egusphere-2026-4656', Anonymous Referee #1, 29 Sep 2026 reply
-
RC2: 'Comment on egusphere-2026-4656', Anonymous Referee #2, 30 Sep 2026
reply
Karali and colleagues present a wildfire risk assessment for Greece that includes projections of risk under different climate change emissions scenarios. Their study helpfully partitions risk into hazard, exposure and vulnerability (the latter partitioned into sensitivity and adaptive capacity). The authors use an interesting expert judgement-driven method to assess the relative importance of different risk components. They also explore the influence of aggregation method of individual risk components on the overall (composite) wildfire risk index. They find that hazard tends to be the most important contributor to risk (of course with lots of nuance around this), and that high emissions scenarios lead to greater increases in risk than low emissions scenarios. They also find that an aggregation method that allows for additional tuning of input weights resulted in a more faithful representation of wildfire risk than using the weights derived from expert elicitation.
Congrats to the authors on this interesting, relevant and well-presented manuscript, which makes a useful contribution to the field of wildfire risk modelling.
My main comments concern the nature of the risk model.
- Why are the terms added, rather than multiplied? What are the implications of this choice?
- Some definitions of risk place fire within the hazard category (acknowledging it can still be modelled in terms of individual drivers, as well as embellished with details about intensity, rate of spread etc). What is the reason for separating out the drivers of fire in the way that the authors have here, with a surface weather index as hazard but things like fuel load, fuel moisture and ignitions as exposure? What are the implications?
- It would be good to see a little discussion about the (sorry to keep saying this word) implications of selected indicators – how might results change with different proxies for these indicators, and with a different (or additional) set of indicators?
- Can we see a little more of the fire data? e.g. a time series and/or map?
- I would like to see an acknowledgement of the relatively ‘flat’ nature of the validation data – which seems to be a binary variable of >=1 fire or not. This misses out on the impacts of fire (essential to risk) and doesn’t distinguish between different types of fire, including extreme ones
- Can you please clarify, did you only evaluate the predictive power of the fire weather metrics (in terms of your binary fire variable) or did you also use the total risk score? Would it be possible to compare the performance of the hazard score with that of the exposure and vulnerability scores?
- I struggled a little to follow the details of the OWA method. If I understand correctly, you are still summing the components to get a risk score – if so, I suggest adjusting Figure 1 to show that addition is still involved. It also seems to be the case that at least some of the OWA weights sum to one (per Table A1 in the appendix) – so is it the weights of individual variables that no longer sum to one? If the form of the equation is still an addition of terms with weights applied (as with framework A), then I think that should be shown (even if it means adding more detail like a layer of local weights, for example)
- For Figure 4, I suggest a harder demarcation between nCCI values, to make it really clear when a particular component accounts for >50% of the total score. Perhaps other categories could be 0-10, 11-30 and 31-50? It might be interesting here to highlight when hazard is *not* the main contributor
- There is a pretty big literature on wildfire risk modelling in Europe, North America and Australia, among other places, but there is virtually no discussion of how these results and methods fit into that context. This is a fairly major omission in my view.
- I really liked the expert elicitation element of the methods, but there was very little discussion of them, which seems like a missed opportunity. Are those results being saved for a separate manuscript?
- BTW The description of the weighting of hazard, vulnerability and exposure possibly overstates their separation – it seems plausible to me that those figures are all within the same margin of error. (around L390).
Citation: https://doi.org/10.5194/egusphere-2026-4656-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 230 | 99 | 45 | 374 | 44 | 41 |
- HTML: 230
- PDF: 99
- XML: 45
- Total: 374
- BibTeX: 44
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The paper addresses a relevant scientific topic and falls within the scope of NHESS. It provides an important step toward improving fire risk assessment in Greece, particularly under climate change scenarios, where fire activity is likely to increase. It is clear that the authors have dedicated considerable effort to provide an extensive analysis. I do, however, have concerns regarding the overall structure of the paper, especially the clarity, transparency and organization of the Methods section.
Please see the PDF attached for further details.