Wildfire Burnt Area and Associated Greenhouse Gas Emissions under Future Climate Change Scenarios in the Mediterranean: Developing a Robust Estimation Approach

Abstract

Wildfires burn annually over 400,000 ha in Mediterranean countries. By the end of the 21st century, wildfire Burnt Area (BA) and associated Green House Gas (GHG) emissions may double to triple due to climate change. Regional projections of future BA are urgently required to update wildfire policies. We present a robust methodology for estimating regional wildfire BA and GHG emissions under future climate change scenarios in the Mediterranean. The Fire Weather Index, selected drought indices, and meteorological variables were correlated against BA/GHG emissions data to create area-specific statistical projection models. State-of-the-art regional climate models (horizontal resolution: 12 km), developed within the EURO-CORDEX initiative, simulated data under three climate change scenarios (RCP2.6, RCP4.5, and RCP8.5) up to 2070. These data drove the statistical models to estimate future wildfire BA and GHG emissions in three pilot areas in Greece, Montenegro, and France. Wildfire BA is projected to increase by 20% to 130% up to 2070, depending on the study area and climate scenario. The future expansion of fire-prone areas into the north Mediterranean and mountain environments is particularly alarming, given the large biomass present here. Fire-smart landscape management may, however, greatly reduce the projected future wildfire BA and GHG increases.

1. Introduction

The Mediterranean region is highly susceptible to wildfires, a situation exacerbated by climate change [1,2]. The frequency and magnitude of large wildfires will likely increase by 14% by the end of the century (2071–2100) under the RCP4.5 climate scenario and by 30% under the RCP8.5 throughout the Mediterranean Basin [3]. Dupuy et al. [1] reviewed 23 studies on future wildfire danger and activity in Southern Europe. They found a consensus that wildfire danger and burnt areas will increase, necessitating more robust fire management strategies. Studies emphasize the need for a paradigm change in wildfire management in Mediterranean-type regions, including the need for integrated management approaches, as traditional suppression methods may not suffice under future climate conditions [4,5,6].

Recent research has focused on understanding how future climate scenarios might influence wildfire behavior, particularly in terms of burnt areas. Statistical approaches to estimate Burnt Area from climate variables, mainly using Multiple Linear Regressions (MLR) to model burned areas as a function of weather parameters, are complementary to the process-based land-surface models [1,7,8,9,10]. Regression models require no local high-resolution datasets and may be applied over large areas, making them useful regional assessment tools.

Mean annual Burnt Area (BA) and Green House Gas (GHG) emissions by wildfires in the Mediterranean are projected to double or triple by 2070 under future climate change scenarios [7,11,12,13], assuming non-significant changes in external factors (such for example, ignition type and rate, land-use, fire suppression, and human activities). Emissions of GHG by wildfires are strongly related to area-specific BA vegetation characteristics. There are large regional differences in future projections across the Mediterranean that may relate to regional variability. However, studies use different datasets and techniques to estimate future BA and GHG emissions. This makes comparative analyses between studies across the Mediterranean not feasible.

This paper addresses this issue by providing a statistical method for estimating wildfire BA and associated GHG emissions, which aims to be applicable throughout southern Europe, using a novel regression model that is driven by selected climatic and fire-danger indices. This regression model needs to be calibrated at each target site with observational data and is, therefore, area-specific. It assumes implicitly that all other future variables are stable: no changes are assumed in land-use and -cover, fire-suppression methods, and fire-ignition causes, as well as climate-fire and fire-vegetation feedback. This approach is common for statistical projection studies.

There is an urgent need to implement alternative fire management approaches throughout the Mediterranean as existing suppression methods are likely not sufficient under future climate change [4,5,6]. To support those efforts, this paper provides an estimation of the effect of fire-smart landscape management on the reduction of future wildfire BA in the study areas, using a GIS-based approach described in the methodology section. Fire-smart landscape management is “an integrated approach primarily based on fuel treatments through which the socio-economic impacts of fire are minimized while its ecological benefits are maximized” [14]. Long-term observational studies that detail the effects of fire-smart landscape management are not available. However, modeling studies suggest that fire-smart management of around 3% of the total study region may lead to a 10–20% reduction in annual averaged BA [15,16,17].

The methodological approach outlined here permits up-scaling of the study results throughout the Mediterranean. It will enable local stakeholders to formulate regional climate change mitigation and adaptation plans and help them access new funding instruments. The study supports the European Green Deal and regional climate-resilient landscape planning policies and strategies, providing quantitative information to the estimation of the Forest Reference Level (FRL) for the “LULUCF (Land Use, Land Use Change and Forestry)” EU Regulation 2018/841.

This paper provides projections of wildfire BA and associated GHG emissions under different scenarios of (i) future climate change and (ii) fire-smart landscape interventions for different Mediterranean target landscapes. The three target landscapes (Figure 1) are located in southeast France (the Luberon-Lure National Regional Park), southern Greece (W Crete, Chania Prefecture), and south-east Montenegro (focusing on the Prokletije/Komovi region).

2. Materials and Methods

This paper presents a robust methodology for estimating regional wildfire BA under future climate change scenarios in the Mediterranean using publicly available data. The following approach was taken. Firstly, a statistical model was developed and calibrated for each study area, associating BA with selected climatic variables and indices derived from gridded observational meteorological data. Annual GHG emission data from past wildfires were downloaded from GWIS (https://gwis.jrc.ec.europa.eu/, accessed on 10 May 2024) and related to BA in the target areas. Secondly, future climate projection data were obtained from state-of-the-art regional climate models (horizontal resolution of 12 km, developed within the EURO-CORDEX initiative) to feed the calibrated statistical projection models for calculation of wildfire BA and GHG emissions under three future climate change scenarios (RCP2.6, RCP4.5 and RCP8.5) up to 2070. Finally, numerical correction factors were applied to future wildfire simulations to derive the potential reduction in fire BA and GHG emissions under fire-smart landscape management. Figure 2 summarizes our procedure in a flowchart.

2.1. Data Sources

Meteorological data were obtained from the state-of-the-art ERA5-Land reanalysis dataset [18] at a horizontal resolution of 9 km. Two meteorologically based indices were also used. Firstly, we used the Fire Weather Index (FWI), used worldwide to estimate fire danger, which consists of different components that account for the effects of fuel moisture and wind on fire behavior and spread [19]. The FWI has been adopted at the EU level (since 2007) by the European Forest Fire Information System (EFFIS) of the Copernicus Emergency Management Service (CEMS) to assess fire danger levels in a harmonized way. Secondly, we used the Standardized Precipitation Evaporation Index (hereafter: SPEI), a multi-scalar drought index based on climatic data [20]. The index needs monthly temperature and precipitation data for its calculation and is standardized against the long-term climatic data of an area.

Low-resolution BA data (recording areas of >300 × 10³ m²) for Montenegro were derived from the Global Wildfire Information System (GWIS: https://gwis.jrc.ec.europa.eu/, accessed on 10 May 2024) for the years spanning 2002–2019. High-resolution BA data (recording areas of >1 × 10³ m²) for Chania Province and the Luberon-Lure National Park were provided by the corresponding regional fire services, covering 2000–2021. GHG emission data from past wildfires at the respective study sites were downloaded from GWIS (https://gwis.jrc.ec.europa.eu/, accessed on 10 May 2024).

2.2. Statistical Model

To build the statistical models that link regional wildfire BA to the current climate for each of the target areas, we tested diverse meteorological variables that studies have related to BA throughout the Mediterranean [1,3,4,7,8,9,12]. These variables include FWI, maximum, minimum, and mean temperatures (TX, TN, and TG, respectively), precipitation (RR), as well as the 3-month and 6-month Standardized Precipitation Index as proposed by Hargreaves (SPEI3H and SPEI6H, respectively). In the latter index, the potential evapotranspiration (PET) is estimated using the Hargreaves equation, which considers the maximum and minimum daily temperature data and the latitude of the location [21].

GAMs (Generalized Additive Models [22]) were used to model the response variable (total BA) from the candidate independent variables. Both the statistical model optimization and the variable selection were performed using leave-one-out cross-validation and tested against the observational data. The final models for each one of the areas are in the form of log₁₀ BA = s(FWI) + s(SPEI6H) + s(TX). All of the post-processing, the statistical modeling, and the following analysis of the data were performed using the R project for statistical computing, R version 4.4.0 [23].

Published studies indicate that our optimum variables are strongly correlated to BA over the fire season (JJAS) in the Mediterranean area [11,12,13,17]. Climate variables of the months (JFMAM and OND) outside the fire season show no significant correlation with BA in any of the study areas. To calculate the total annual BA, we assumed that the percentage (%) of BA over the fire season vs. the other months was stable for each of the target areas.

GHG emissions data from past wildfires, processed using the FAOSTAT methodology [24], were downloaded from GWIS (https://gwis.jrc.ec.europa.eu/, accessed on 10 May 2024) for all study areas. The relationship between the annual BA of the respective target study areas and the annual GHC emissions from regional wildfires was assessed using the Reduced Major Axis regression [25]. This area-specific relationship was subsequently used to derive future estimations of GHG emissions based on total wildfire BA per study area.

2.3. Climate Projections

To estimate BA under future climate change conditions, daily data from Regional Climate Models were used to drive the statistical models for each of the study areas. The models used in this study (

Table 1. Climate models used in this paper.

EURO-CORDEX RCM/GCM PAIRS (Horizontal Resolution~12 km)
Institute	RCM	GCM
Swedish Meteorological and Hydrological Institute (SMHI)	RCA4	ICHEC-EC-EARTH
Swedish Meteorological and Hydrological Institute (SMHI)	RCA4	MPI-M-MPI-ESM-LR
Swedish Meteorological and Hydrological Institute (SMHI)	RCA4	MOHC-HadGEM2-es

) are available at the C3S Climate Data Store (CDS), at a horizontal resolution of 12 km developed within the EURO-CORDEX initiative. These data fed the site-specific statistical projection models to estimate future BA and GHG emissions from wildfires under future climate change up to 2070. The reference period was set at 1971–2000, while three climate change scenarios were used, based on Representative Concentration Pathways (RCPs) 2.6 (ambitious mitigation policies; +1.5 °C), 4.5 (moderately ambitious mitigation policies; +2.5 °C), and 8.5 (business as usual/no mitigation; +4.5 °C).

2.4. Numerical Correction Factors for BA Simulations

Martinoli [26] devised a GIS-based modeling approach using publicly available databases to estimate the reduction in wildfire BA under Fire-Smart Landscape (hereafter: FSL) management. This approach can be applied to any landscape in Europe (excluding Russia, Belarus and Moldova) and to the wider Mediterranean ecoregion (North Africa and Middle East included, excluding Egypt). Part of this methodology is adapted from Carmo et al. [27] and Sequeira et al. [28]. The GIS-derived numerical correction factors for the study areas [26] were applied to the future BA simulations to derive the potential reduction in fire BA and GHG emissions under FSL management.

3. Results

3.1. Baseline Data from the Study Areas

Up to 0.08% of the total surface area (2308 km²) of the Luberon-Lure area is burned annually on average, compared to 0.63% of the surface area (2376 km²) of Chania Province and 1.23% of the total surface area (13,812 km²) of Montenegro. Over 90% of the total annual wildfire BA is lost over the fire season (JJAS). Specifically, the percentage BA over the fire season (JJAS) versus the other months (JFMAM and OND) is 96% vs. 4% for Chania Province, 93% vs. 7% for Montenegro, and 91% vs. 9% for the Luberon-Lure National Park.

Average annual emissions from wildfires amount to 0.4 Gg per km² of BA for Chania Province (annual fire emissions: 6 Gg). It amounts to 0.5 Gg per km² of BA for the Luberon-Lure Biosphere Park (annual fire emissions: 1 Gg). The emission per km² of BA is similar between these areas, indicating a comparable Mediterranean land cover. Average annual emissions from wildfires in Montenegro amount to 0.68 Gg per km² of BA (annual fire emissions: 116 Gg). The very high emissions per km² of wildfire BA in Montenegro are related to the large amount of mature forest burning.

3.2. Future Changes in Wildfire BA and GHG Emissions

3.2.1. Chania Province (Crete, Greece)

Future fire danger (based on the FWI) is set to increase; the multi-monthly average FWI over the fire season (JJAS) is above 35 (high fire danger). The SPEI-6 drought index is becoming progressively more negative in the future, while the average maximum monthly temperature will increase significantly over the future fire season.

Figure 3 shows that BA will increase by 15–20% in the near future (2011–2040) and by 18–25% in the more distant future (2041–2070). As values of BA in the fire season (JJAS: 96.1%) and outside the season (JFMAM and OND: 3.9%) are assumed to be stable, the percentage increases stated here are assumed to be valid on an annual basis. Figure 4 shows that total emissions will increase by about 17% under all scenarios, both in the near future (2011–2040) and in the more distant future (2041–2070).

3.2.2. Montenegro

Future fire danger (based on the FWI) is set to increase over the fire season (JJAS) in Montenegro. The SPEI-6 drought index is becoming progressively more negative in the future, while the maximum average monthly temperature will increase significantly over the future fire season. There is a large variability of these variables, especially of future fire danger and Tmax, in the target area.

Figure 5 shows a large BA variability under different scenarios in the near future (2011–2040), ranging from −33% to +47%, while increases in the more distant future (2041–2070) range from 36–127%. Percentages of BA occurring in the fire season (JJAS: 93.4%) and outside the season (JFMAM and OND: 6.6%) are assumed to be stable; therefore, the percentage-changes stated here are on an annual basis.

Figure 6 indicates that total emissions related to wildfires show similar percentual changes as the BA, both in the near future (2011–2040) and in the more distant future (2041–2070). Note that the Montenegro BA and emissions data have a coarse resolution (see Section 2.1), and simulated climate data show significant variability, which explains the large variability in the future wildfire projections.

3.2.3. Luberon-Lure

Fire danger (based on the FWI) is set to increase over the future fire season in the Luberon-Lure region, especially under RCP4.5 and RCP8.5 in the more distant future (2041–2070). The SPEI-6 drought index is becoming progressively more negative in the future, while the maximum average monthly temperature will increase significantly over the future fire season, especially under RCP4.5 and RCP8.5 in the more distant future. All variables, and especially of future fire danger, show large variability in this target area.

Figure 7 shows BA increases of 38–40% in the near future (2011–2040) and 32–111% in the more distant future (2041–2070). Percentages of BA occurring in the fire season (JJAS: 91.3%) and outside of it (JFMAM and OND: 8.7%) are assumed to be stable. Therefore, the percentage increases stated here are on an annual basis. Figure 8 indicates that total emissions will increase by about 100% under all scenarios (from 1 Gg to 2 Gg). Note that these are rounded averaged increases.

3.3. Potential Impact of Fire-Smart Landscape Interventions

This paper assessed the potential impact of FSL management on BA, as outlined in Section 2.4. It was assumed that 2% and 5%, respectively, of the total study areas were subjected to FSL interventions. The BA under these management intervention scenarios was compared to the BA under management-as-usual (

Table 2. Numerical estimates of the effectiveness of Fire-Smart Landscape (FSL) intervention scenarios in reducing annual BA for each of the study areas.

Study Area	Reduction in Annual BA, Under:
	2% FSL Interventions	5% FSL Interventions
Chania Province (Crete, Greece)	−3.2%	−7.9%
Luberon-Lure NP (France)	−4.7%	−11.8%
Prokletije/Komovi NP (Montenegro)	−5.6%	−13.9%

). Our estimated impacts of the FSL management intervention scenarios on reducing BA are at the lowest end of published studies [15,16,17].

The numerical estimates detailing the effect of FSL interventions in each of the study areas (Table 2) were applied to future BA simulations. The results (

Table 3. Projected future BA and emissions associated with wildfires in the study areas under (i) a no management-change scenario, (ii) a scenario where 2% of the area experiences FSL intervention measures, and (iii) a scenario where 5% of the area experiences FSL intervention measures.

	BA Change vs. Present		Emissions Change vs. Present
	Near Future (2011–2040)	Distant Future (2041–2070)	Near Future (2011–2040)	Distant Future (2041–2070)
Chania—no change	+15 to +20%	+18 to +25%	+17%	+17%
Chania—2% intervention	+12 to +16%	+15 to +21%	+17% (only RCP8.5)	+17%
Chania—5% intervention	+6 to +10%	+9 to +15%	0	0%/+17% (RCP4.5)
Montenegro—no change	−33% to +47%	+36 to +127%	−33% to +47%	+36 to +127%
Montenegro—2% intervention	−37% to +39%	+29 to +114%	−37% to +40%	+19 to +115%
Montenegro—5% intervention	−42% to +27%	+17 to +95%	−42% to +27%	+18 to +96%
Luberon—no change	+38 to +40%	+32 to +111%	+100%	+100%
Luberon—2% intervention	+29 to +33%	+26 to +101%	+100%	+100%/0% (RCP2.6)
Luberon—5% intervention	+22 to +26%	+19 to +90%	+100% (0% RCP2.6)	+100%/0% (RCP2.6)

) indicate that FSL management reduces the future increases in wildfire BA and emissions. Projections show that future increases in wildfire BA are reduced by 10% (Chania, Greece), 20–30% (Montenegro), and 10–20% (Luberon, France) by 2070 under a 5% FSL intervention scenario. Similarly, this FSL intervention scenario may limit or even halt, future wildfire emission increases in the study areas under most (or all) RCP scenarios.

4. Discussion

Over the observational period, ~0.63% of the 2376 km² surface area of Chania Province burned annually on average. Emissions from wildfires were at ~0.4 Gg per km² of BA. Future increases in BA of 18–25% (2041–2070) may be significantly reduced (to 9–15%) under FSL interventions. Future wildfire emission increases of about 17% may be largely avoided under FSL intervention scenarios.

On average, ~0.08% of the 2308 km² surface of the Luberon-Lure region burned annually over the observation period. Emissions from wildfires were at ~0.5 Gg per km² of BA. Future BA is projected to increase by 32–111% (2041–2070), but increases are significantly less (at 19–90%) under FSL interventions. Future climate change-related emissions do not significantly decrease under FSL scenarios, except for interventions under climate scenario RCP2.6. However, absolute future emissions are still very low compared to the other study areas, even under a 100% emission increase, as present annual wildfire emissions are low at 1 Gg.

At present, absolute wildfire BA and GHG emissions are six times higher in Chania Province compared to the Luberon-Lure region. These two areas have virtually the same surface area and a similar biomass and climate. The much lower annual wildfire BA in the Luberon-Lure area is likely mainly related to more effective fire management and suppression efforts. This strongly suggests that improving fire management in Chania Province would help to significantly reduce wildfire BA.

About 1.23% of the 13,812 km² surface area of Montenegro burned annually, on average, over the observation period. Emissions from wildfires are high, amounting to 0.68 Gg per km² of BA due to the large amount of forest burning. Funding for fire management is much lower than in France or Greece, while strong relief hampers fire suppression efforts [27]. Future BA is projected to increase by 36–127% (2041–2070). However, under FSL interventions BA increases are significantly less at 17–95%. Future climate change-related emission increases of 36–127% may be reduced to 18–96% under FSL scenarios.

The largest absolute future increases in BA and GHG emissions from wildfires are projected in Montenegro. This is consistent with published studies that project the future expansion of fire-prone areas into the north Mediterranean and into higher-altitude Mediterranean mountain environments [7,12,13] due to changing climatic variables. This is a growing concern as much larger biomass is present at these locations [1,3]. However, in more arid Mediterranean areas the climate-induced wildfire BA increases may be limited due to fuel constraints [13,17]. For example, in Chania, fire danger over the fire season is already very high, with frequent fires. Furthermore, a lower biomass is present as there is less (and open) forest. The impact of climate change on wildfire BA and associated emissions is, therefore, likely more limited in this region.

Integrated landscape management approaches focused on fire prevention may help limit future wildfire BA increases [1,4,5,6]. This study suggests that FSL management may reduce the projected future increases in BA and GHG emissions from wildfires even under our low-reduction management scenarios. For example, future BA increases are reduced by up to 40–50% for Chania, by 25–50% for Montenegro and 20–35% for the Luberon. Future GHG emissions from wildfires are not increasing (Chania), or the increase is reduced by up to 22–50% (Montenegro). In the case of the Luberon-Lure region, the decrease in wildfire GHG emissions is non-significant; however, this is likely an artifact of the very low initial GHG emission values in this study area. Much larger decreases in wildfire BA and GHG emissions are possible under FSL management, according to the literature [15,16,17].

5. Conclusions

The statistical model projections in this study offer a higher wildfire BA resolution than statistical projections from existing Mediterranean-wide studies, e.g., [12,13] that focus on global trends. The paper’s novel approach to estimating future BA and GHG emissions from Mediterranean wildfires is quick and robust while enabling direct comparisons between different regions of the Mediterranean. It is, therefore, a useful rapid assessment tool to estimate the impact of future climate change on regional wildfire BA and associated GHG emissions. Our approach relies on publicly available data and is thus applicable throughout the Mediterranean region. The wildfire projections are detailed enough to help local/regional government and other stakeholders formulate area-specific climate change mitigation- and adaptation plans while helping them access new funding instruments.

Temporal and spatial scales are different from those that are usual in fire landscape management. As wildfires are linked to climate, we need access to long-term average weather characteristics (spanning > 15 years), as well as BA and GHG emission data series of the same temporal length. Furthermore, fire-related BA (and associated GHG emissions) can only be linked to climate variables when looking at large surface areas (upwards of 50 × 50 km). Finally, it should be noted that this assessment is based on historical relationships between wildfire, BA, and climate conditions. These relationships may change in the future, as an extension of the fire season and changes in the vegetation are likely. This study should, therefore, be considered conservative in its estimates of increases in wildfire BA and GHG emissions under future climate change scenarios.