1. Introduction
Mediterranean areas are experiencing rapid expansion of communities into rural areas with multiple adverse impacts in terms of invasive species, deforestation, and human activities that increase fire ignitions [
1]. The expanding intermix of flammable vegetation and low density development along with a warming climate and more ignitions [
2,
3] has created a wicked wildfire problem due to the significant economic burden of wildfire management, incomplete knowledge, the number of people and opinions involved, and the interconnected nature of these problems with other governance challenges [
4,
5,
6]. Risk governance systems have failed to keep pace with the change in these anthropogenic fire regimes and new policies [
7,
8,
9,
10,
11,
12,
13] proposed to address the wildfire problem have yet to be implemented. These policies aim to improve landscape fuels management but stop short of identifying specific spatial strategies and their outcome in terms of reducing future losses [
3]. The core issue is identifying the scale of risk to ensure that fire management activities consider the source and key values exposed to large wildfire events. Overlaying the spatial context of risk are the multiple and complementary strategies that address the pinch points from ignition to containment. Increasingly, literature from North America [
14,
15,
16,
17,
18,
19] and Southern Europe [
20,
21,
22] has shown that investing more in fuel management and pre-fire planning is critical to mitigate upcoming challenges that emerge from climate change and land abandonment effects on fuel patterns.
In Greece, different stakeholders see the wildfire issue as a symptom of a higher order problem but tend to focus on specific solutions rather than a broad-spectrum strategy. For example, some point to poor forest and fuels management, while others argue that wildfire problems could be solved if there were only more resources for fire suppression. Greece implements minimal levels of fuel treatments in terms of extent, size, and intensity, and the government funds a broad mix of fire prevention with local governments with a 2020 budget of 17 million euros, with an additional 2 million euros for fire protection in the Forest Service [
23,
24]. However, only a fraction of this funding is applied to treat fuels in forested areas. Rarely do forest management agencies combine forest and fuel treatments on management units at a scale comparable to the burned area of predicted fire events. Thus, risk management is entirely dependent on fire suppression to protect people, private property, and other values-at-risk. Managing fuels in Greece has multiple challenges such as rapid re-growth of fuels with vegetation that exceeds the flammability of what was removed, and restrictions that exclude fuel management with prescribed fire. The bulk of the fuel treatments are fuel breaks around communities and are located without strategic assessment of the likely wildfire paths from wildlands [
25]. Around the fuel breaks, the vegetation has the potential to burn in crown fires that generate spot fires kilometers ahead of the fire perimeter [
26]. By contrast, many studies have shown that landscape scale programs that are strategically allocated are effective at changing wildfire behavior before fires arrive at community boundaries [
22,
27,
28,
29,
30,
31,
32]. Much of the evidence is experimental, obtained through use of scenario planning tools to evaluate alternative management strategies and estimation of potential trade-offs [
28,
33,
34,
35,
36,
37]. For instance, previous studies have shown that reduction in large fire growth is obtainable through the collective effect of many treatment units occurring on the landscape with specific patterns and densities [
27,
38,
39]. Random treatment patterns are inefficient in changing large fire growth rates compared with strategic designs, because they permit fire to easily move laterally around treatments, unless large portions of the landscape are treated [
40]. Several alternative fuel treatment methods exist that can be applied as prescriptions in pre-selected units on the landscape, including silvicultural tending with mechanical means (e.g., thinning, pruning), prescribed burning, and grazing for reducing surface and ladder fuels.
In this study, we first used wildfire simulation methods to assess the relative contribution of different land cover types to the overall fire problem on a large and diverse fire prone landscape on the island of Lesvos, Greece. The goal was to show the connectivity in terms of fire spread and exposure among the major land types in a typical, highly fragmented Greek landscape. We then used wildfire and fuel treatment simulation methods to develop an optimized fuel treatment program in the extensive, fire prone conifer forests. The objective of the fuel treatments design was to effect broad landscape fire risk reduction in terms of fire severity and control by suppression forces, rather than fuel breaks focused in a specific location. Fuel treatments consisted of the removal of small trees and saplings by thinning and cleaning the understory to reduce fuels. We assumed these treatments would be achieved through grazing or mechanical means since prescribed fire is not legal in Greece. Forest canopy fuels would be treated with silvicultural improvement methods including thinning and pruning to increase tree canopy base height and reduce ladder fuels. We assess how the varied land uses and associated vegetation interact and spread fire and examine how investments in forest and fuel management can bring about landscape improvements in fire resiliency in the pine forests. We discuss how the current practice of relying on fuel breaks will not significantly alter the growing wildfire risk problem without coupling with more extensive and intensive forest and fuel management.
3. Results
Simulations revealed that most fires burned in olive plantations (42.7%), followed by dense conifer forests (19.7%), agricultural lands (9.9%), and grasslands (8.1%) (
Figure 4A and
Table 3). The land cover types that received the most fire from self-burning ignitions were olives, dense conifers, agricultural lands, and grasslands (
Figure 4B and
Table 3). Sparse conifers received 4.6% of all fires, most of which came from dense conifers (25.5%), olives (25.1%), self-burning (11.6%), grasslands (11.2%), chaparral (10.2%), shrub/grass (7.8%), and agricultural lands (6.2%) (
Figure 4C and
Figure 5A). Dense conifers burned mostly from self-burning fires (33.8%), followed by olives (31.8%), grasslands (9.1%), and chaparral (6.2%) (
Figure 4C). Young conifers received 3.1% of all fires, with 30.8% self-burning, followed by dense conifers (17.4%), olives (16%), chaparral (13.8%), and shrub/grass (9.9%). In
Figure 4A and
Figure 5B, we see that most outgoing fires came from olives (27.2%), affecting dense conifers, chaparral, grasslands, and agricultural lands. Dense conifers sent 18% of area burned, mostly to olives, grasslands, and chaparral. Chaparral and grasslands sent 12% of total outgoing area burned, followed by agricultural lands. Overall, the amount of area burned generated by ignitions on each land cover type was non-proportional to the cover type’s total area, in particular for dense conifers that produced less fire compared to their proportional area cover, i.e., 35% of the study area but produced 19.7% of total fire activity, olives (27.5% cover/42.7% of total fire activity), and grasslands (5.3% cover/8.1% of total fire activity).
The proposed combined surface and canopy fuel treatment units inside conifer forests, as described in
Table 2 and derived from TOM simulations, spanned 7600 ha, with 6000 ha located inside dense, 770 ha inside sparse, and 500 ha inside young conifer forests (
Figure 6A). This can be translated as 35% of all conifer covered lands and 16% of the total simulation landscape (i.e., 46,800 ha). These treatment areas included 75 polygons with an area >10 ha and 64 polygons between 10 and 100 ha. Five treatment units were between 100 and 250 ha, four between 350 and 650 ha, and two between 800 and 900 ha. Two-thirds of all treatments were located on ten units, each with area greater than 200 ha. At the north and eastern parts of the Vouleri-Koukos forest, we found the largest and most continuous treatments sites. The Olympus-Ampeliko forest also had large candidate areas for fuel treatments. Smaller treatment units were established inside the Axladeri and Vrisa-Vatera forests, while the conifer forest east of the chestnut forests in the Megalochori forest had also great potential for successful fuel treatment application.
The percentage of burned area of each land cover type from the total area burned from each large (>50 ha) simulated fire was estimated using the outputs of the two stochastic MTT simulations (pre- and post-fuel treatments modeling), and was compared using boxplots (
Figure A5 in
Appendix B). The number of large simulated fires were 8470 for the baseline conditions and 5510 after fuel treatment modeling, with a notable reduction in the percent area burnt by large fires for dense conifers.
Under the assumption that we treated the entire landscape for the stochastic simulations with the MTT algorithm (
Figure 6A), the decrease in burn probability (BP) was greater in the southern section of the study area and in parts of the Olympus-Ampeliko and Axladeri forests (
Figure 7A), while conditional flame length (CFL) reduction was moderate for most of the dense conifer forests and higher in parts of the Olympus-Ampeliko and Vrisa-Vatera forests (
Figure 7B). We noticed that conifer fuel treatments influenced the value of burn probability on different land cover types (
Figure 7A), but regarding CFL, almost all pixels with a reduction greater than 1 m were inside conifer forests (
Figure 7B). The greatest number of times a pixel could be burned on the base condition simulations was by 395 (out of the 10,000 fires simulated), and decreased to 319 on the ideal conditions simulation (burn probabilities were 0.0395 and 0.0319, respectively). Simulations after applying fuel treatments revealed that sparse conifer forests experienced on average 60 fewer fires and CFL was reduced by 1.17 m; dense conifer forests experienced 50 fewer fires and a CFL reduction of 1.47 m; and young conifer forests experienced 20 fewer fires and a CFL reduction of 0.09 m. We found that more than a third (7800 ha) of conifer forests had a minor reduction in CFL (<1 m), while 9500 ha had a moderate reduction (>1 m up to 2 m), 2300 ha moderate-high reduction (>2 m up to 3 m), and 1400 ha high to very high reduction (>3 m). About 14,500 ha of conifer forests were projected to show a moderate or low decrease in burn probability, 3500 ha a moderate-high decrease, and 3000 ha a high to very high reduction. For the treatment sites selected by TOM, the reduction in both BP and CFL was high to very high for 20% of their area (approximately 1500 ha), while approximately 50% for both metrics showed a moderate reduction.
4. Discussion
Our results showed the how diverse land cover types and land use practices contribute to fire spread on a typical Greek landscape, and how a fuel management program targeting the conifer forests can reduce area burned and fire intensity. One key finding is that olive plantations were predicted to be a substantial contributor to fire exchange among the land cover types even though they cover 22% less area compared to the conifer forests. Olive plantations interspersed with abandoned agricultural lands and grasslands create a fuel mosaic with high rates of overall fire spread. All the major land cover types showed high fire connectivity as measured by fire transmission metrics, with more than half of the outgoing area burned originating from olive plantations and dense conifers. Olives received substantial fire originating on conifer forests, since formerly cultivated olive groves have been invaded by conifers, and on highly productive lands, farmers expanded their cultivations through conifer deforestation (
Figure A6A in
Appendix C). The result of these land transitions is an increase in shared boundary, and high interdependence in terms of fire spread.
When fuel treatments were located with the TOM optimization process, most of the treatments were placed near boundaries of conifer forests with olives, underscoring the importance of this intermixed land type in the management of fire. Most of the conifer forest was of low priority for receiving fuel treatments because past management practices (agroforestry, silvo-pastoralism, grazing, and resin collection) have collectively kept fuel loads lower relative to the olive–pine intermix. Fuel treatments were also not allocated by the TOM process in areas with a high historical fire density. Overall, the results of the fuel treatment simulation showed that treatments can reduce burn probability and fire intensity on both land cover types. Most treatment units were between 200 and 900 ha, creating a network of adjacent units.
The network diagram of fire exchange among the land uses (
Figure 4C) supports the need for a comprehensive fuel treatment program that extends beyond confer forests to consider all land cover types and their respective contribution to large fires. Limited budgets need to be allocated to efficiently address fire risk, considering the spatial pattern of ignitions, fuel loading, weather patterns, values-at-risk, and suppression strategies on the different land cover types [
39,
68]. An effective fire risk management strategy should consider the role of direct fire suppression and indirect fuel management on fire size distributions, and understand how fire regimes will be naturally affected by climatic changes through changes in fire weather conditions or changes in dominant forest cover types [
69].
In Lesvos, and Greece in general, fire-resilient landscapes have historically been created and maintained by land use practices associated with agroforestry and silvo-pastoralism. These practices were prevalent until the early 1970s and included abundant and frequent low-intensity fire. This is a useful reference condition similar to that used in western US conifer forests to describe resilient forests maintained by natural fires [
70]. However, after thousands of years of human activities in Greece, natural and anthropogenic influences on fire regimes are inseparable, except for small enclaves in northern Greece where virgin forests are unaffected by humans with a high-severity stand-replacing fire regime (200–400 years). By contrast, the contemporary fire regime is characterized by <50 years of high-severity stand-replacing human-ignited fires burning in abandoned former agricultural areas with live fuel accumulation and spreading into unmanaged forested areas with dead fuel accumulation. This fire regime resulted from the rural exodus in the 1970s, and aggressive fire suppression policies that created positive feedbacks on the fire regime over time.
Increasing fuel loadings from land abandonment and afforestation need to be addressed with expanded use of agroforestry and silvo-pastoralism [
71,
72] in the areas surrounding conifer forests. Agroforestry reduces understory vegetation while also providing revenue from the sale of biomass as food or fuel [
73]. Silvo-pastoralism can be used to target specific fuels, since grasses and herbaceous vegetation are preferred by cows, horses, and pigs, while goats have a preference for feeding on woody shrubs and young trees [
71]. Grazing could also be intensified and reinforced with subsidies to peri-urban livestock farms in those areas we showed through simulated treatments can reduce fire spread rate [
74]. One strategy to expand these practices is through implementation of the European Union’s Common Agricultural Policy (CAP), which can be leveraged to reduce fire hazard by: (1) promoting the reintroduction of livestock grazing in areas prone to abandonment [
74], (2) creation of agricultural low-hazard belts around urban areas and values-at-risk, (3) the regulation of burning by shepherds and fire use, and (4) directing management to high fire risk areas, giving preference to agroforestry [
75]. The result of these efforts could be heterogeneous agroforestry mosaics that allocate crops in an aggregated pattern (10 km yr
−1 for Lesvos Island), thus providing more opportunities to suppress fires. CAP could also be leveraged to promote regional plans that sustain rural activities in remote areas to ensure the maintenance of croplands, orchards, or pastures over time [
76].
In terms of community wildfire protection, it is useful to note that there are distinct land use patterns around developed areas and resulting fire exposure and mitigation strategies. For instance, the three major activities around the communities of Lesvos Island are olive cultivation, animal production, and tourism, where each has a typical fuel complex (
Figure A6A,B) and optimal strategy for fuel management. Communities surrounded by conifer forests (
Figure A6B) are a high priority for large landscape fuel management programs as demonstrated in this study. Communities in
Figure A6C,D are surrounded by olive or other cultivations and orchards, in which case fire risk reduction is dependent on many small landowners managing fuels inside the plantations. The importance of olive plantations in the exchange of fire among land uses was a key finding in the study, where one third of all transmitted fire originated from these areas. Fuel management should be prioritized and spatially coordinated among landowners to create fuel break systems that can facilitate fire suppression as part of community protection programs.
The dense conifer forests were also a significant source of fire transmission to other land uses (one fifth of the total) and creating fire resilient forests will require significant expansion of traditional fuel management practices. The Greek Forest Service has the authority to license individuals from forested communities to extract forest biomass for household or commercial reasons, a cost-effective and socially accepted alternative to fuel treatments performed by contractors. Biomass extraction for bioenergy is one solution for Mediterranean policy makers to consider, but these programs need to be scaled over large areas to substantially reduce area burned during extreme weather events [
77], and treatments need to include unmanaged or abandoned lands and private forests, and targeted to source areas that cause high exposure with high hazard. Currently, industrial timber production in high productivity timber stands (>100 m
3 ha
−1 and DBH ≥ 5 cm) is applied on 400,000 ha across all Greece, while in less productive stocks (<100 m
3 ha
−1), it is applied in more than one million ha [
72]. The annual timber production is 1.1 million m
3, 30% of which comes from private forests; one third is produced from conifer species, with most of the production (~65%) used as fuelwood. Collaborative actions between private contractors and the Greek Forest Service to perform silvicultural thinning could reduce wildfire activity and improve the ecology and health of future forests (e.g., stewardship contracting; see [
78]). For the protection of private conifer forests, compulsory fuel removal on surrounding fire prone farmlands could help reduce the risk of transmitted wildfire from those lands.
The selection of fuel management projects presents a challenge to forest management agencies to reduce fuels at landscape scales while addressing, among others, the presence of human infrastructure and settlements, the smoke effects on human health and recreation, the protection of highly valued resources, such as wildland habitat and drinking water quality, and constraints such as timber targets set in forest management plans. Policies, constraints, and regulations that restrict treatment location, type, and total area treated can significantly degrade the performance of these strategies [
56]. Using modeling approaches as described in this work can provide forest managers with a number of different potential treatment polygons that, in turn, using on-ground knowledge and spatial data, can be selected to design a strategy that will minimize negative effects while achieving fire risk reduction targets. Newer methods such as scenario planning and trade-off analyses can help predict optimal treatment locations to apply to each strategy [
79,
80].
All fuel management scenarios considered in this study assumed the use of mechanical means to reduce forest fuels and excluded the use of prescribed fire since it is currently illegal in Greece. Legislative reforms to allow the careful application of prescribed fire could substantially accelerate policy to reduce fuel loadings by providing a low-cost method for fuel management. The effectiveness of fire in fuel management is widely supported by recent studies that found higher fuel treatment effectiveness under extreme conditions for treatments included broadcast burning, compared to thinning and/or pruning alone [
32,
81,
82,
83]. The latter treatments can contribute to surface fuel biomass accumulation and more severe wildfire effects. While it is still common and legal for farmers to use fire (pile and burn) in olive groves to remove fuels (
Figure A7), frequent burning is required to maintain low fuels loads under Mediterranean conditions and prevent severe fires [
1,
32,
84,
85]. A recent study in Catalonia suggested that applying prescribed fire treatments (15,000 ha yr
−1 in an area of 32,000 km
2, i.e., 750 ha yr
−1 for Lesvos Island), can greatly contribute to a decrease in high intensity fire and extreme fire events [
86]. In addition, allowing prescribed burning under a controlled forage burning program administered by an authorized agency would reduce the risk from frequent illegal fires set by livestock farmers to increase forage production.
We note several limitations and assumptions to this work. Fire behavior modeling of proposed fuel treatments can overestimate the effectiveness of potential fuels treatment to reduce fire behavior [
87] if fuel models do not accurately reflect the post-treatment conditions. Fire managers are required to evaluate and justify the effectiveness of planned fuel treatments resulting from modeling processes in modifying fire growth, behavior and effects on resources and assets [
88]. Validation of fire behavior modeling outputs was a crucial component of this analysis, and we performed local accuracy assessments to validate that the modeled fire behavior of each fuel model closely resembles the observed one (
Appendix A). Accuracy and results validation for MTT simulations in Greece have been assessed during previous studies [
47,
89,
90] (
Appendix A). Surface fuel conditions and canopy cover have not experienced major disturbances within the study area since the time of field inventories (circa 2009), but a major source of potential simulation error comes from the accuracy of spatial inputs regarding fuel conditions in the canopy layer, especially regarding crown bulk density. This is a result from both mapping error (see [
48]) and forest growth since the time of field inventories. Additionally, TOM assumes that reduction in large fire growth is obtainable through the collective effect of many units occurring on the landscape [
38]. We also assumed that simulated wildfires were larger than the fuel treatment units to allow the analysis to focus on the directions in which fires move rather than their start locations [
53]. We also did not account for the benefits of the proposed treatments on fire suppression, and thus we underestimated the reduction in area burned in the treated landscape. Another assumption is that we estimated the impact of large fires that can escape initial attack by burning under the most frequent, but also average–extreme weather conditions (average worst-case fire potential), thus simulating the proposed treatments to perform under these target conditions. A limitation of TOM is that it requires a single modeled weather scenario; thus, we restricted the modeling on a single fire front from one wind direction and speed. Less frequent weather scenarios and from different fire fronts could potentially highlight different treatment locations, but with lower probability of experiencing a wildfire. To address this, during stochastic MTT simulations, we used thousands of ignitions under different weather scenarios to account for both spatial and weather variability. Although we used three different weather scenarios, they describe only the current dominant weather conditions without considering the uncertainty related to climate warming scenarios and how climate change can exacerbate fire risk in Mediterranean Europe in the coming decades [
91]. This makes even more critical the need for proactive fuel reduction treatments simultaneously combined with other management options (i.e., a more proactive and integrative management).
The spatial optimization of fuel treatments in this study assumes that all treatment units will be implemented at a given instance in time, but in reality, treatments are accomplished on an annual basis and treatment effects in reducing fire behavior diminish with time, approximately one decade after implementation [
29]. Given the historical treatment rate that the Greek forest management agencies can implement annually, to effectively treat 7600 ha (16% of the landscape) as simulated here in the span of a decade, the annual treatment rate must be no lower than 760 ha yr
−1 (1.6% of the landscape), which is actually twice as much as the percentage locally treated with best-case administrative conditions in Greece (e.g., the local Forest Service branch of Kassandra in Chalkidiki, northern Greece, with an area similar to our case study). Finally, we did not exclude high slope (20% of the proposed treatment is located on slopes >20°), rugged and high elevation sites, which can substantially decrease the available land for fuel treatments.