Predicting burn probability: Dimensionality reduction strategies enable accurate and computationally efficient metamodeling.

Predicting the probability that a given location will be burnt by a wildfire is an important part of understanding the risk that wildfires pose and how our management actions (e.g., prescribed burning) can reduce this risk. Existing methods to quantify this burn probability involve simulating the spread of many thousands of individual wildfires, making them highly computationally expensive. To reduce this expense, we propose strategies that enable the development of computationally efficient machine learning assisted metamodels for estimating burn probability, which are demonstrated for a case study in South Australia. Artificial neural networks are used as the metamodel to emulate the outputs of a landscape fire simulation model. Development of the metamodel is facilitated by reducing the input and output dimensionality of the simulation model by a factor of 10,000-1,000,000, while still being able to predict burn probabilities with high accuracy (approximately ± 7.4% error, on average) and only requiring 0.6% of the computational time compared with an approach using landscape fire simulation models. This opens the door to obtaining many thousands of spatially distributed estimates of burn probability, as is required when optimising fuel treatment strategies.

Short- and long-term wildfire threat when adapting infrastructure for wildlife conservation in the boreal forest.

Managers designing infrastructure in fire-prone wildland areas require assessments of wildfire threat to quantify uncertainty due to future vegetation and climatic conditions. In this study, we combine wildfire simulation and forest landscape composition modeling to identify areas that would be highly susceptible to wildfire around a proposed conservation corridor in Québec, Canada. In this measure, managers have proposed raising the conductors of a new 735-kV hydroelectric powerline above the forest canopy within a wildlife connectivity corridor to mitigate the impacts to threatened boreal woodland caribou (Rangifer tarandus). Retention of coniferous vegetation, however, can increase the likelihood of an intense wildfire damaging powerline infrastructure. To assess the likelihood of high-intensity wildfires for the next 100 years, we evaluated three time periods (2020, 2070, 2120), three climate scenarios (observed, RCP 4.5, RCP 8.5), and four vegetation projections (static, no harvest, extensive harvesting, harvesting excluded in protected areas). Under present-day conditions, we found a lower probability of high-intensity wildfire within the corridor than in other parts of the study area, due to the protective influence of a nearby, poorly regenerated burned area. Wildfire probability will increase into the future, with strong, weather-induced inflation in the number of annual ignitions and wildfire spread potential. However, a conversion to less-flammable vegetation triggered by interactions between climate change and disturbance may attenuate this trend. By addressing the range of uncertainty of future conditions, we present a robust strategy to assist in decision-making about long-term risk management for both the proposed conservation measure and the powerline.

Modeling the effects of different fuel treatment mosaics on wildfire spread and behavior in a Mediterranean agro-pastoral area.

Wildfire spread and behavior can be limited by fuel treatments, even if their effects can vary according to a number of factors including type, intensity, extension, and spatial arrangement. In this work, we simulated the response of key wildfire exposure metrics to variations in the percentage of treated area, treatment unit size, and spatial arrangement of fuel treatments under different wind intensities. The study was carried out in a fire-prone 625 km2 agro-pastoral area mostly covered by herbaceous fuels, and located in Northern Sardinia, Italy. We constrained the selection of fuel treatment units to areas covered by specific herbaceous land use classes and low terrain slope (<10%). We treated 2%, 5% and 8% of the landscape area, and identified priority sites to locate the fuel treatment units for all treatment alternatives. The fuel treatment alternatives were designed create diverse mosaics of disconnected treatment units with different sizes (0.5-10 ha, LOW strategy; 10-25 ha, MED strategy; 25-50 ha, LAR strategy); in addition, treatment units in a 100-m buffer around the road network (ROAD strategy) were tested. We assessed pre- and post-treatment wildfire behavior by the Minimum Travel Time (MTT) fire spread algorithm. The simulations replicated a set of southwestern wind speed scenarios (16, 24 and 32 km h-1) and the driest fuel moisture conditions observed in the study area. Our results showed that fuel treatments implemented near the existing road network were significantly more efficient than the other alternatives, and this difference was amplified at the highest wind speed. Moreover, the largest treatment unit sizes were the most effective in containing wildfire growth. As expected, increasing the percentage of the landscape treated and reducing wind speed lowered fire exposure profiles for all fuel treatment alternatives, and this was observed at both the landscape scale and for highly valued resources. The methodology presented in this study can support the design and optimization of fuel management programs and policies in agro-pastoral areas of the Mediterranean Basin and herbaceous type landscapes elsewhere, where recurrent grassland fires pose a threat to rural communities, farms and infrastructures.

Assessing Climate Change Impacts on Wildfire Exposure in Mediterranean Areas.

We used simulation modeling to assess potential climate change impacts on wildfire exposure in Italy and Corsica (France). Weather data were obtained from a regional climate model for the period 1981-2070 using the IPCC A1B emissions scenario. Wildfire simulations were performed with the minimum travel time fire spread algorithm using predicted fuel moisture, wind speed, and wind direction to simulate expected changes in weather for three climatic periods (1981-2010, 2011-2040, and 2041-2070). Overall, the wildfire simulations showed very slight changes in flame length, while other outputs such as burn probability and fire size increased significantly in the second future period (2041-2070), especially in the southern portion of the study area. The projected changes fuel moisture could result in a lengthening of the fire season for the entire study area. This work represents the first application in Europe of a methodology based on high resolution (250 m) landscape wildfire modeling to assess potential impacts of climate changes on wildfire exposure at a national scale. The findings can provide information and support in wildfire management planning and fire risk mitigation activities.

Assessing wildfire exposure in the Wildland-Urban Interface area of the mountains of central Argentina.

Wildfires are a major threat to people and property in Wildland Urban Interface (WUI) communities worldwide, but while the patterns of the WUI in North America, Europe and Oceania have been studied before, this is not the case in Latin America. Our goals were to a) map WUI areas in central Argentina, and b) assess wildfire exposure for WUI communities in relation to historic fires, with special emphasis on large fires and estimated burn probability based on an empirical model. We mapped the WUI in the mountains of central Argentina (810,000 ha), after digitizing the location of 276,700 buildings and deriving vegetation maps from satellite imagery. The areas where houses and wildland vegetation intermingle were classified as Intermix WUI (housing density > 6.17 hu/km2 and wildland vegetation cover > 50%), and the areas where wildland vegetation abuts settlements were classified as Interface WUI (housing density > 6.17 hu/km2, wildland vegetation cover < 50%, but within 600 m of a vegetated patch larger than 5 km2). We generated burn probability maps based on historical fire data from 1999 to 2011; as well as from an empirical model of fire frequency. WUI areas occupied 15% of our study area and contained 144,000 buildings (52%). Most WUI area was Intermix WUI, but most WUI buildings were in the Interface WUI. Our findings suggest that central Argentina has a WUI fire problem. WUI areas included most of the buildings exposed to wildfires and most of the buildings located in areas of higher burn probability. Our findings can help focus fire management activities in areas of higher risk, and ultimately provide support for landscape management and planning aimed at reducing wildfire risk in WUI communities.

Incorporating Resource Protection Constraints in an Analysis of Landscape Fuel-Treatment Effectiveness in the Northern Sierra Nevada, CA, USA.

Finding novel ways to plan and implement landscape-level forest treatments that protect sensitive wildlife and other key ecosystem components, while also reducing the risk of large-scale, high-severity fires, can prove to be difficult. We examined alternative approaches to landscape-scale fuel-treatment design for the same landscape. These approaches included two different treatment scenarios generated from an optimization algorithm that reduces modeled fire spread across the landscape, one with resource-protection constrains and one without the same. We also included a treatment scenario that was the actual fuel-treatment network implemented, as well as a no-treatment scenario. For all the four scenarios, we modeled hazardous fire potential based on conditional burn probabilities, and projected fire emissions. Results demonstrate that in all the three active treatment scenarios, hazardous fire potential, fire area, and emissions were reduced by approximately 50 % relative to the untreated condition. Results depict that incorporation of constraints is more effective at reducing modeled fire outputs, possibly due to the greater aggregation of treatments, creating greater continuity of fuel-treatment blocks across the landscape. The implementation of fuel-treatment networks using different planning techniques that incorporate real-world constraints can reduce the risk of large problematic fires, allow for landscape-level heterogeneity that can provide necessary ecosystem services, create mixed forest stand structures on a landscape, and promote resilience in the uncertain future of climate change.

Wildfire exposure and fuel management on western US national forests.

Substantial investments in fuel management activities on national forests in the western US are part of a national strategy to reduce human and ecological losses from catastrophic wildfire and create fire resilient landscapes. Prioritizing these investments within and among national forests remains a challenge, partly because a comprehensive assessment that establishes the current wildfire risk and exposure does not exist, making it difficult to identify national priorities and target specific areas for fuel management. To gain a broader understanding of wildfire exposure in the national forest system, we analyzed an array of simulated and empirical data on wildfire activity and fuel treatment investments on the 82 western US national forests. We first summarized recent fire data to examine variation among the Forests in ignition frequency and burned area in relation to investments in fuel reduction treatments. We then used simulation modeling to analyze fine-scale spatial variation in burn probability and intensity. We also estimated the probability of a mega-fire event on each of the Forests, and the transmission of fires ignited on national forests to the surrounding urban interface. The analysis showed a good correspondence between recent area burned and predictions from the simulation models. The modeling also illustrated the magnitude of the variation in both burn probability and intensity among and within Forests. Simulated burn probabilities in most instances were lower than historical, reflecting fire exclusion on many national forests. Simulated wildfire transmission from national forests to the urban interface was highly variable among the Forests. We discuss how the results of the study can be used to prioritize investments in hazardous fuel reduction within a comprehensive multi-scale risk management framework.

FireLossRate: An R package to estimate the loss rate of residential structures affected by wildfires at the Wildland Urban Interface.

To inform proactive management actions supporting community resilience to wildfires, we developed a new software package called FireLossRate. This package in R helps the user to compute wildfire impacts on residential structures at the Wildland Urban Interface (WUI). The package integrates spatial information about exposed structures, empirical equations that estimate the loss rate of structures affected by wildfires as a function of fireline intensity and distance from fire edge with fire growth modeling outputs from fire simulation software and burn probability models. FireLossRate helps to quantify and produce spatially explicit data on structural exposure and loss for single and multiple fires. The package automates post hoc analyses on simulations that include single or multiple wildfires and enables result mapping when combined with other packages available in R. In this paper, we describe the functionality of the FireLossRate package and introduce users to the interpretation of impact indicators of wildfires at the WUI. FireLossRate is available for download at https://github.com/LFCFireLab/FireLossRate.•FireLossRate allows the computation of wildfire impacts indicators on residential structures at the Wildland Urban Interface in support of community fire risk management.

Spatial planning of fire-agency stations as a function of wildfire likelihood in Thasos, Greece.

Even though wildfires constitute a natural phenomenon, they may have severe implications with respect to the socioeconomic structure of the affected population and the ecological wealth of a territory, especially when they burn under high intensities. Timing of the initial attack is thus crucial to fire control in areas that fires are considered to be under high threat of burning. The aim of this paper is to investigate the combined use of simulation modeling and spatial optimization to assess the pre-positioning of fire-management resources on a small Greek island, Thasos, based on the current and desired fire agency capabilities, maximization of environmental protection, and rationalization of financial resources. The estimation of burn probability (BP) depicted specific areas of high fire hazard in the southern, central, and western part of the island, where essential preventive measures should be undertaken. Based on this result, BP was then used as a primary input for the assessment of optimal locations of fire operation agencies in order to achieve the maximal coverage under certain (already available) and minimum number of fire-fighting vehicles in different time windows. The results generated three differentiated optimal location schemes [8 available vehicles within either 10 (immediate response time) or 31 min (average response time) with the current fire resources; 19 and 2 required vehicles within 10 and 31 min, respectively, based on a minimum number of fire resources]. This type of information enables us to propose a relocation of the current fire agency in a southern town of the island. The flexibility and interaction of the models provide a framework for appropriate decision making under a set of political and financial constraints.

Wildfire risk for main vegetation units in a biodiversity hotspot: modeling approach in New Caledonia, South Pacific.

Wildfire has been recognized as one of the most ubiquitous disturbance agents to impact on natural environments. In this study, our main objective was to propose a modeling approach to investigate the potential impact of wildfire on biodiversity. The method is illustrated with an application example in New Caledonia where conservation and sustainable biodiversity management represent an important challenge. Firstly, a biodiversity loss index, including the diversity and the vulnerability indexes, was calculated for every vegetation unit in New Caledonia and mapped according to its distribution over the New Caledonian mainland. Then, based on spatially explicit fire behavior simulations (using the FLAMMAP software) and fire ignition probabilities, two original fire risk assessment approaches were proposed: a one-off event model and a multi-event burn probability model. The spatial distribution of fire risk across New Caledonia was similar for both indices with very small localized spots having high risk. The patterns relating to highest risk are all located around the remaining sclerophyll forest fragments and are representing 0.012% of the mainland surface. A small part of maquis and areas adjacent to dense humid forest on ultramafic substrates should also be monitored. Vegetation interfaces between secondary and primary units displayed high risk and should represent priority zones for fire effects mitigation. Low fire ignition probability in anthropogenic-free areas decreases drastically the risk. A one-off event associated risk allowed localizing of the most likely ignition areas with potential for extensive damage. Emergency actions could aim limiting specific fire spread known to have high impact or consist of on targeting high risk areas to limit one-off fire ignitions. Spatially explicit information on burning probability is necessary for setting strategic fire and fuel management planning. Both risk indices provide clues to preserve New Caledonia hot spot of biodiversity facing wildfires.