The research team’s central insight is that a forested landscape can be treated as a network. Each patch of land is a node, connected to its neighbors by edges that carry information about fuel load and the likelihood that fire will spread from one patch to the next. When a fire ignites, its potential path across the landscape can be traced through this network, and any given ignition under a given set of weather conditions produces what the authors call a fire propagation graph, a map of which patches the fire could plausibly reach. By generating a large set of such scenarios, each representing a different combination of ignition location, weather, and fire behavior, the researchers built a probabilistic picture of how fire moves through the landscape. Fuel breaks then enter the picture as modifications to the network: treating a patch reduces the fuel along the edges connected to it, and the optimization problem becomes one of choosing which patches to treat so that the expected burned area across all scenarios is minimized.

What distinguishes this framework from earlier fuel break planning models is its explicit treatment of failure. A fuel break is not a guarantee. Under extreme fire weather, wind-driven embers can leap across a cleared corridor, or the fire may find a path around the break’s flanks. The model captures this uncertainty by assigning widths to candidate fuel breaks and using those widths to track the probability that a break will fail to contain a spreading fire. A wider break is more reliable but more expensive and more disruptive to the landscape, so the optimization must balance the cost and footprint of each break against its expected contribution to hazard reduction. This scenario-based approach, formulated as a mixed-integer programming problem, allows the model to search the enormous space of possible fuel break configurations and identify the allocation that delivers the greatest reduction in expected burn area for a given level of investment.

The second, and arguably more novel, dimension of the work is conservation. Fragmenting a forest with fuel breaks does not only affect fire; it also alters the structure of habitat and the connectivity of the landscape for wildlife. Species that depend on large, contiguous tracts of old forest can be harmed when corridors of movement are interrupted. The researchers focused on woodland caribou, Rangifer tarandus caribou, a population listed as endangered in Canada whose ranges overlap some of the most fire-prone landscapes in western North America. Caribou are particularly sensitive to landscape disturbance because they rely on extensive, connected habitat to move between seasonal ranges and to avoid predators. A fuel break network that efficiently stops fires but severs caribou migration routes could solve one problem while deepening another, so the model was designed to protect a target proportion of known migration routes while pursuing its fire hazard objectives.

Formally, the planning problem is expressed as a mixed-integer optimization in which binary decision variables indicate whether each candidate fuel break segment is built, and continuous or integer variables capture the widths of those segments. The objective function minimizes the expected area burned across the full set of fire spread scenarios, subject to constraints that ensure a specified share of caribou migration routes remains intact. Because each scenario is a separate propagation graph, the model can evaluate how a given fuel break configuration performs across the whole distribution of possible fires rather than against a single average fire. The authors also explored an alternative formulation that minimized the expected area of the largest fires, the extreme events that cause the most damage, alongside the fuel connectivity of the landscape. For any given hazard reduction goal, the optimal solutions are those that impose the least disruptive impact on the wildlife movement network, making the trade-off between fire protection and conservation explicit and quantifiable rather than left to negotiation after the fact.

To demonstrate the framework, the team applied it to the Red Rock-Prairie Creek area of Alberta, Canada, a fire-prone landscape in the foothills of the Rocky Mountains that also supports woodland caribou populations. This region exemplifies the tension at the heart of the study: it is exactly the kind of place where aggressive fuel management could reduce catastrophic fire risk, and exactly the kind of place where such management could fragment critical habitat. The researchers assembled data on fuel loads, predicted fire behavior, and documented caribou movement routes, then ran the optimization to identify the configuration of fuel breaks that best reduced wildfire spread potential while respecting the conservation constraints. The results showed that the approach could meaningfully lower the potential for fire to spread across the area, at the cost of moderate impacts on known caribou migration routes, a trade-off that the model makes transparent and that planners can adjust by changing the target proportion of protected routes.

The technical machinery behind the model deserves attention because it reflects broader trends in how complex environmental problems are being solved. Mixed-integer programming is a mature operations research technique, but applying it to landscape-scale fire planning requires encoding fire spread, a spatially explicit and stochastic process, into a form that an optimization solver can handle. The scenario-based graph representation achieves this by precomputing the propagation possibilities and letting the solver reason over them. The use of fuel break width as a proxy for failure probability is an elegant simplification: rather than modeling fire physics in full detail within the optimization, the model captures the essential relationship that wider breaks are less likely to be breached, and lets the optimizer decide how much reliability to buy with each segment. This keeps the problem computationally tractable at landscape scale while still acknowledging that fuel breaks are probabilistic tools, not walls.

The implications extend well beyond Alberta. Fire-prone forest landscapes across North America, and indeed around the world, face the same dual pressure: intensifying fire regimes driven by climate change and fuel accumulation on one hand, and biodiversity conservation obligations on the other. Planning frameworks that treat these as separate problems tend to produce solutions that optimize one at the expense of the other, or that require cumbersome manual reconciliation. By embedding both objectives in a single optimization, the new approach gives managers a defensible, quantitative way to explore the trade-off space. They can ask, for example, how much additional fire hazard reduction is achievable if the protected share of caribou routes is relaxed by ten percent, or how much conservation can be gained for a small sacrifice in expected burn area. The methodology is timely, the authors note, for developing strategies that reduce wildfire hazard while achieving conservation management goals in other regions with similar conflicts.

There are, of course, practical considerations in translating such a model into on-the-ground decisions. The quality of the optimization depends on the fidelity of the fire spread scenarios, the accuracy of fuel maps, and the completeness of wildlife movement data, all of which carry uncertainty. Real-world implementation also involves budgets, jurisdictional boundaries, and social factors that a mathematical model does not fully capture. Yet the framework’s value lies precisely in its ability to structure the conversation: instead of debating fuel breaks in the abstract, planners can examine a set of optimal or near-optimal configurations, each with quantified consequences for both fire risk and caribou connectivity, and choose among them with a clear view of the costs and benefits. As wildfires grow larger and more severe, tools that make these trade-offs explicit, rather than implicit, may become as essential to landscape management as the fuel breaks themselves.

Subject of Research: Optimization of fuel break placement to reduce wildfire hazard while conserving woodland caribou migration routes

Article Title: A network optimization framework for fuel break planning in wildfire-prone landscapes: Accounting for fuel break failures and wildlife conservation goals

Article References: Yemshanov, D., Liu, N., Neilson, E. W., Thompson, D., Koch, F. H., & Gritter, K. (2026). A network optimization framework for fuel break planning in wildfire-prone landscapes: Accounting for fuel break failures and wildlife conservation goals. PLOS Complex Systems, 3(8), e0000117. https://doi.org/10.1371/journal.pcsy.0000117

Image Credits: AI Generated

DOI: 10.1371/journal.pcsy.0000117

Keywords: wildfire, fuel breaks, network optimization, mixed-integer programming, woodland caribou, habitat connectivity, fire spread modeling, landscape planning, conservation, Alberta, fire hazard, wildlife migration

The research, led by Michele Salis and Liliana Del Giudice of Italy’s National Research Council Institute of BioEconomy, together with colleagues from Spain, Iran and other institutions, focused on two time windows: a historical baseline from 1985 to 2014 and a near-future period from 2035 to 2064. The team drew on three state-of-the-art climate models from the Coupled Model Intercomparison Project Phase 6, known as CMIP6, namely ACCESS-ESM1-5, EC-EARTH3 and MPI-ESM1-2-HR, each run under three Shared Socioeconomic Pathways spanning low to very high greenhouse gas emissions. What sets this study apart is the resolution: the raw climate model output, which arrives at a coarse scale of tens of kilometers, was statistically downscaled to a grid of just one square kilometer using a two-step technique combining parametric quantile mapping with geostatistical regression, before feeding into fire danger calculations and fire spread modeling.

To quantify fire danger, the researchers applied the Canadian Forest Fire Weather Index System, the same harmonized method used across Europe by the European Forest Fire Information System since 2007. The FWI integrates daily maximum temperature, minimum relative humidity, wind speed and cumulative precipitation into a single rating of potential fire intensity, and it depends on three fuel moisture codes that track the dryness of fine fuels, duff and deep organic layers. The team also calculated the Daily Severity Rating, a metric that amplifies the influence of extreme FWI values and estimates how difficult a fire burning that day would be to suppress. Both indices were computed for every pixel of the island on every day of the two study periods, for each climate model and each emissions scenario.

The signal that emerged was consistent across all three models, though its magnitude varied. Under the historical baseline, days with very high FWI values were rare, concentrated almost entirely in the May-to-October fire season, when high and very high danger conditions already accounted for roughly 43 to 46 percent of days. By the 2035-2064 period, the picture shifts dramatically. Under the highest emissions scenario, SSP5-8.5, the EC-EARTH3 model projected that very high FWI days would more than double during the fire season, climbing from about 4.6 percent to 10.8 percent of days, while the ACCESS model showed a rise from 4.0 to 10.1 percent. Even the more conservative MPI model projected a near doubling, from 4.5 to 8.2 percent. The Daily Severity Rating told a parallel story, with very high severity days under EC-EARTH3 rising from 8.4 percent to 17.1 percent of fire-season days under the worst-case pathway.

Underlying these danger trends is a straightforward climatic mechanism. All models project warmer and drier conditions for Sardinia, with maximum temperature increases reaching up to about 3.3 degrees Celsius annually and 3.9 degrees during the fire season under high emissions. Precipitation declines across the board, particularly in western and central sectors of the island, and the number of rainy days drops markedly, especially during the hottest months, pointing to longer and more frequent dry spells that desiccate fuels. Intriguingly, the analysis of wind conditions on extreme danger days showed that the projected increase in very high fire weather comes mostly from more days with low and moderate winds rather than stronger winds, suggesting that rising temperatures and reduced rainfall, not windier conditions, will be the dominant drivers of intensifying fire risk.

But the study went a crucial step beyond danger indices. Weather alone does not make a wildfire; fires need ignitions, flammable fuels and terrain that allows spread. To capture those interactions, the team built a probabilistic fire simulation framework using the Minimum Travel Time algorithm, which computes the fastest possible paths of fire growth across a landscape and calculates spread rates with the Rothermel surface fire model, while crown fire initiation follows the Van Wagner approach. The researchers ran tens of thousands of simulated fire footprints, 50,000 under baseline conditions alone, at a fine 150-meter resolution, randomly drawing weather streams, wind directions and fire durations from frequency distributions derived for three weather zones covering northern, central and southern Sardinia. Ignition points were placed according to probability grids built from two decades of historical fire records, and fuel moisture inputs were anchored in field measurements collected on the island, including live fuel moisture values between 65 and 100 percent depending on vegetation type.

The hazard outputs are striking. Under historical conditions, the models simulated an average annual burned area of roughly 11,700 to 15,400 hectares, consistent with the observed mean of about 14,700 hectares per year, or 0.65 percent of the island’s surface. In the future simulations, that figure nearly doubles or more. Under EC-EARTH3 and SSP5-8.5, the average annual area burned jumps to approximately 33,590 hectares, a 250 percent increase over the baseline, while ACCESS projects more than 30,000 hectares per year under the two higher-emission scenarios. Crown fire area, a measure of the most intense and dangerous fire behavior in forests, rises from between roughly 730 and 1,060 hectares per year historically to as much as about 2,020 hectares per year under the worst-case combination of model and scenario.

Perhaps most important is where these changes unfold. Historically, burn probability hotspots have been concentrated in southern Sardinia, with annual burn probabilities exceeding 4 hectares per square kilometer in localized zones. The future projections show those hotspots expanding in both magnitude and extent, pushing into western and interior regions of the island. Under SSP5-8.5, much of the island shifts toward higher burn probability classes, with localized increases above 10 hectares per square kilometer in the main southern hotspot zones. Crown fire probability rises in parallel, with the most pronounced gains in southern and western Sardinia, where some areas reach values exceeding 8.5 hectares per square kilometer. This westward and inland expansion matters because those regions, including areas hit by major fires such as the Montiferru blaze of 2021, are already among the most fire-affected parts of the island, and inland zones lack the moderating influence of the sea that cushions coastal areas.

One of the study’s most thought-provoking findings is a caveat about what fire danger indices can and cannot tell us. Despite the systematic rise in FWI values, the simulated average flame length and average fire size did not increase consistently from the historical to the future period, partly because the extra extreme-danger days occurred disproportionately under low and moderate winds. The authors stress that extreme fire weather is necessary but not sufficient for large wildfires: a blaze must also start, escape initial suppression and find continuous fuels to spread across. Simply equating projected FWI maps with future fire risk, as some earlier studies have done, therefore overstates the certainty of the picture. Probabilistic spread modeling, which integrates weather, ignitions, fuels and landscape connectivity, provides a far more realistic estimate of where fire exposure will actually grow.

The researchers acknowledge limitations, notably that fuel layers and ignition patterns were held constant to isolate the climate signal, an assumption that could either understate or overstate future hazard depending on how land abandonment, fuel accumulation and fire activity themselves evolve. Still, the message for the Mediterranean is unambiguous. The authors argue that suppression-oriented strategies alone cannot cope with the projected intensification, and they call for proactive, integrated risk management: landscape-scale fuel treatments, grazing, prescribed burning, fire-resilient land-use mosaics and spatial planning that accounts for expanding fire-prone zones. As record seasons sweep southern Europe, with more than one million hectares burned across the EU in 2025 alone, Sardinia’s high-resolution glimpse of the future offers both a warning and a practical roadmap for landscapes on the front line of a warming world.

Subject of Research: Projected effects of climate change on wildfire danger and hazard in Sardinia, Italy

Article Title: Climate change will reshape future wildfire danger and hazard in Sardinia, Italy

Article References: Climate change will reshape future wildfire danger and hazard in Sardinia, Italy. (n.d.). https://doi.org/10.1007/s11069-026-08320-5

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08320-5

Keywords: wildfire, climate change, Sardinia, Mediterranean Basin, Fire Weather Index, burn probability, fire spread modeling, CMIP6, SSP scenarios, crown fire, Daily Severity Rating, fire management