1 Introduction

Fire-prone regions are of significant concern. Portugal, located in the western Iberia region, is one of the Mediterranean regions with the highest wildfire incidence (Amraoui et al. 2015). This alarming trend has led to extensive property damage and loss of life (San-Miguel-Ayanz et al. 2021; Santos et al. 2023a, b). From 2009 to 2023, an average of 96,000 hectares was burned annually (EU JRC, 2023), making Portugal the European country with the largest burned area (EU JRC, 2023). Although 24% of the territory has burned at least once since the early 1980s (dos Santos et al. 2023a, b), time-series data of the total annual burned area in Portugal show two main characteristics: significant year-to-year variability and a downward trend in the total number of fires since 2000 (San-Miguel-Ayanz et al. 2021; dos Santos et al. 2023a, b), despite extensive fires (Tedim et al. 2018; Cardil et al. 2019; Rodrigues et al. 2020). "This high fire activity and behaviour are related to a variety of factors: (1) lack of land management (Beighley and Hyde, 2018; Cardil et al. 2019; Moreira et al. 2020); (2) depopulation of rural areas (Vieira et al. 2020; Calheiros et al. 2022); (3) build-up of fuels in both amount and extent (e.g., Fernandes et al. 2019; Sá et al. 2022); (4) negligent fire ignitions (Meira Castro et al. 2020); and (5) an increase in hot and dry weather events prone to fire activity (Calheiros et al. 2022; Santos et al. 2023a, b). Although all these factors contribute to the outcome of high fire activity, large wildfires tend to occur under severe weather conditions (Pereira et al. 2005; Abatzoglou et al. 2019; Jolly et al. 2015; Turco et al. 2019; Calheiros et al. 2022). More specifically, Portugal has faced extreme wildfire events in recent years, particularly in 2003, 2005, and 2017, which are connected to extreme atmospheric conditions (Pereira et al. 2005; Tedim et al. 2018; Ribeiro et al. 2020).

Fire weather refers to atmospheric conditions that influence the likelihood, behaviour, and intensity of wildfires. Wind speed and direction, temperature, humidity, and precipitation have significant impacts on the behaviour and spread of wildfires (Abatzoglou and Kolde 2013; Amraoui et al. 2015; Jolly et al. 2015; Cardil et al. 2019). The most severe fire conditions arise from the convergence of various weather extremes, such as the simultaneous occurrence of high temperatures, low humidity, and strong winds (Gouveia et al. 2016; Rodrigues et al. 2020). Portugal has a Mediterranean climate characterised by mild to cold wet winters and windy, dry, hot summers, creating conditions conducive to ignition, rapid spread, and development of extreme wildfires (Trigo et al. 2016; Ruffault et al. 2017; Tedim et al. 2018; Royé et al. 2020; Santos et al. 2023a, b). Global temperatures are now one degree higher than those in the pre-industrial era (Intergovernmental Panel on Climate Change, IPCC AR6). Worldwide emissions from greenhouse gases are still growing, corresponding to an increase in temperature that is likely to reach three degrees by the end of this century (IPCC AR6). Climate research suggests that, in Portugal, the "new normal" will be characterised by high temperatures and drier conditions (Carvalho et al. 2021a), with increased susceptibility to aridity (Carvalho et al. 2021b) and a decline in annual rainfall (e.g. Pereira et al. 2020). The former is the perfect setup for more frequent and severe wildfires.

A common strategy used to examine weather conditions and the potential for wildfires is the application of fire-weather indices. Fire weather indices are quantitative measures based on central atmospheric variables (Haines 1988; Srock et al. 2018), or a combination of atmospheric variables and environmental conditions. Typical indices include the Keetch-Byram Drought Index (KBDI, Keetch et al. 1968), McArthur Forest Fire Danger Index (FFDI) (McArthur 1967), and Canadian Fire Weather Index (FWI, van Wagner 1987), which integrate factors such as temperature, humidity, wind speed, and precipitation, helping to estimate the dryness of vegetation and the overall fire risk in a given area (van Wagner 1987). The FWI system is widely used in fire management to assess and communicate the level of fire risk, helping authorities and firefighting agencies take preventive measures and allocate resources effectively (Field et al. 2015; Jolly et al. 2015; Karali et al. 2023). Since 2015, the European Forest Fires Information System (EFFIS), the agency responsible for safeguarding forests against fires in the EU and neighbouring countries (EU JRC 2021), has incorporated FWI into its operations. At the local level, the Portuguese National Service IPMA uses FWI to assess fire danger.

In recent research, FWI has been used as a proxy to identify regions within the Iberian Peninsula that exhibit homogeneous fire weather patterns (Rodrigues et al. 2020), to show a link between fire incidence and past burned areas (Alves et al. 2021), and to establish daily fire weather thresholds linked to the progression towards larger fires within specific regions of Portugal (Fernandes 2019). Recently, several studies have investigated the impacts of climate change on fire danger and activity in southern Europe. These regional studies used FWI projections to rate fire danger based on weather conditions simulated under different climate scenarios (Dupuy et al. 2020 and references within). Most of these studies calculated the FWI or its components using data from a regional climate model (RCM), driven by a global climate model (GCM) or one GCM, from early climate simulation initiatives, such as PRUDENCE and ENSEMBLES (Moriondo et al. 2006; Carvalho et al. 2011; Amatulli et al. 2013; Herrera et al. 2013; Pereira et al. 2013; Bedia et al. 2014; Karali et al. 2014). Similar research has been conducted using regional simulations from EURO-CORDEX models (Calheiros et al. 2021; Aparício et al. 2022; Bento et al. 2023) or GCMs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) initiative (Abatzoglou et al. 2019).

In this study, we address the effects of future weather conditions on the fire ignition and propagation risk in Portugal. To this end, a high-resolution dynamical downscaling of the most recent future climate projections from one of the CMIP6 earth system models was carried out, to a local-scale resolution of 6 km for three emission scenarios, SSP27, SSP245, and SSP585, considering two future periods, 2046–2065 and 2081–2100. To the best of our knowledge, no CMIP6 has been dynamically downscaled to a resolution of less than 10 km available for public use. This study had the following goals.

  1. (1)

    To further our understanding of how climate change affects wildfire danger in a fire-susceptible region (Portugal).

  2. (2)

    presents a detailed map of the FWI spatial distributions for future climate scenarios for Portugal.

The remainder of this paper is organised as follows. Data and methods are discussed in the next section. Section 3 presents the FWI series and its components derived from future and historical simulations of CMIP6 and ERA5 along with the meteorological variables used in the FWI calculation. Additionally, it discusses the assessment results of fire danger conditions in comparison with historical fire incident records. Section 4 discusses the individual performance of meteorological variables and conducts a qualitative assessment of fire danger projections. Finally, in Section 5, the main conclusions and suggestions for future research are discussed.

2 Methods

In this study, we analysed future changes in fire danger in one fire-prone region using the Fire Weather Index as an indicator of potential fire-weather conditions. To this end, we used data from the ECMWF ERA5 reanalysis and the CMIP6 ESM projections.

2.1 Study area

Portugal, situated on the western edge of the Iberian Peninsula, has diverse climate and topography (Fig. 1). The study area spans several mountain ranges characterised by rugged slopes and deep valleys in the northern part of the country to great plains in the south.

Fig. 1
figure 1

Study area: Western Iberia (Portugal,red)

Full size image

Portugal has a Mediterranean climate, characterised by hot and dry summers and mild-to-wet winters. Due to their diverse topography, coastal areas have milder temperatures influenced by the Atlantic Ocean, whereas inland regions tend to have more extreme temperatures.

Portugal's precipitation patterns show considerable irregularity on both temporal and spatial scales (Pereira et al. 2020). This irregularity can be attributed to large-scale circulation features, with disturbances from the polar front prevailing in the winter months and the Azores High moving poleward in the summer. Additionally, a seasonal asymmetry divides the northern region from the south, with the latter being drier, resulting in significant spatial variability in precipitation. These patterns have been documented in previous studies, including that by Corte-Real et al. (1998), Trigo and DaCamara (2000), Goodess and Jones (2002), Santos et al. (2005) and Costa and Soares (2009). This large-scale circulation also explains the predominance of northern winds during the summer. The varied landscapes combined with dry, hot, and windy climates set the conditions for wildfires to occur.

2.2 Data

2.2.1 Reanalysis model data

This study used the fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis ERA5 to validate the model results. The ERA5 reanalysis data are a cutting-edge global atmospheric reanalysis dataset designed and developed to study climate and provide gridded value data for model validation Hersbach et al. 2020). This dataset is very useful for overcoming weather station dependency. ERA5 covers the entire globe, with a spatial resolution of approximately 0.25° (approximately 31 km) horizontally and 137 vertical levels. It provides data on various atmospheric variables, including temperature, humidity, wind speed and direction, and precipitation on a regular grid. These variables are available at various temporal resolutions, ranging from hourly to monthly means, from 1979 onwards. ERA5 is considered one of the most accurate and reliable reanalysis datasets available owing to advancements in data assimilation techniques, improved model physics, and the incorporation of a vast amount of observational data. However, one should be aware of the potential biases and uncertainties associated with reanalysis datasets, particularly in regions with sparse observational coverage or complex terrains (Hersbach et al. 2020).

2.2.2 Regional climate model data

ESMs provide valuable information about overall trends and large-scale patterns of climate change on a global scale. However, their coarse spatial resolution (~ 100 km) limits their ability to capture the fine-scale features of climate variability and local effects. Dynamic downscaling is a process in climate modelling that allows for a more detailed representation of regional climate features by using a regional climate model (RCM) to simulate the climate at a higher resolution. This process involves running the RCM nested within the ESM grid, meaning that the output of the global model serves as a boundary condition for the RCM. This allows the RCM to simulate finer-scale processes while still being influenced by large-scale climate patterns provided by large-scale models.

The forcing model can be chosen from a multitude of EMSs that are part of the Sixth Phase of the Coupled Model Intercomparison Project (CMIP6; Eyring et al. 2016). The CMIP6 model simulations were designed to provide global climate model simulations of past, current, and future climates under different radiative forcing scenarios at a horizontal resolution of 1.25°latitude and longitude. The scenario narrative, part of the Shared Socioeconomic Pathways (SSPs), is a future world in which socioeconomic development follows different pathways of greenhouse gas emissions and consequent radiative forcing. SSP2-4.5 is a moderate emissions scenario, with CO2 peaking by 2040 and declining thereafter, resulting in a radiative forcing peak of 4.5 W/m2 by 2100. This would result in a rise in the global average temperature of approximately 2.5 °C compared to pre-industrial times and 1.5 °C compared to the current era. The SSP3 represents a scenario with high mitigation and adaptation challenges. In this scenario, CO2 emissions will continue to rise throughout the century, resulting in a radiative forcing peak of 7.0 W/m2 by 2100. This translates to a global mean temperature increase of approximately 3.7 °C compared to pre-industrial levels. SSP5-8.5 is a business-as-usual scenario, with CO2 emissions steadily rising until 2080, before a slight decline, leading to a 5 °C increase compared to pre-industrial levels by 2100. This scenario resulted in a radiative forcing peak of 8.5 W/m2 by 2100. More details and a complete description and comparison of the different CMIP6 future climate scenarios can be found in O’Neill et al. (2016) and Riahi et al. (2017). In this study, RCM results were obtained using the Weather Research and Forecast model (WRF; Skamarock et al. 2019) configured for climate simulations (Marta-Almeida et al. 2016) and driven by the CMIP6 Max Planck Institute (MPI, Mueller et al. 2018; Mauritsen et al. 2019) ESM for three emission scenarios, SSP2-4.5, SSP3-7.0, and SSP5-8.5.

A high-resolution 6-km grid over Portugal was obtained by configuring the WRF model with 2-nested domains and 27 vertical levels unevenly spaced and various choices of physics schemes. The larger domain corresponding to the MPI ~ 100-km resolution grid is large enough to encompass a vast Atlantic area west of Peninsula, and thus captures the principal mesoscale systems and large-scale flow affecting the precipitation and wind in the North Atlantic patterns. Following Marta-Almeida et al. (2016), Table 1 lists the parameters used in the numerical experiments.

Table 1 Parameterisation options for WRF model
Full size table

The WRF model was run over three 20-year periods: 1995–2014 (historical), 2046–2065 (medium-range future), and 2081–2100 (long-range future). These are the reference periods for the past-present and future periods defined by IPCC (IPPC AR6).

The following every six hours meteorological variables from both datasets (model and reanalysis) are required to calculate the fire weather index: 2-m temperature, surface pressure, specific humidity, precipitation (daily accumulated), 10-m wind speed. The relative humidity was derived from the specific humidity using the relation described by Bolton (1980).

2.3 Fire Weather Index (FWI) calculations

The Fire Weather Index (FWI) is a tool used by national meteorological institutions and fire management agencies to assess potential fire danger in a given area (Wagner 1987). It combines various weather and environmental factors into six components to quantify the level of fire danger (Fig. 2).

Fig. 2
figure 2

Fire Weather Index flow chart, structure, and FWI’s components

Full size image

The FWI consists of several components, each representing various fire weather conditions.

  1. 1.

    Fine Fuel Moisture Code (FFMC): reflects the moisture content of fine fuels, such as grass, leaves, and twigs. It is influenced by the temperature, relative humidity, wind speed, and precipitation.

  2. 2.

    Duff Moisture Code (DMC): Measures the moisture content of decomposed organic material beneath the forest floor. It is affected by temperature, humidity, and precipitation.

  3. 3.

    Drought Code (DC): Indicates the moisture content of deep organic layers in the soil. This integrates the effects of temperature and precipitation over longer periods.

  4. 4.

    Initial Spread Index (ISI): The rate of fire spread immediately after ignition is estimated. It is influenced by the wind speed and FFMC.

  5. 5.

    Buildup Index (BUI): represents the overall fuel moisture content and the amount of fuel available for combustion. It is calculated from the DMC and DC.

Fire Weather Index (FWI) combines FFMC, ISI, and BUI to provide an overall assessment of fire danger. FWI is typically calculated using empirical equations based on meteorological data, such as temperature, relative humidity, wind speed, and precipitation. These equations consider the moisture content of different fuel types and the influence of weather conditions on fire behaviour. Dowdy et al. (2009) provides a detailed description and application of these equations. The FWI is expressed as a numerical value, with higher values indicating a higher fire danger. Different thresholds can be used to classify fire danger levels, ranging from low to extreme. Fire management agencies use these classifications to identify fire prevention measures, such as implementing fire bans, increasing surveillance, or deploying firefighting resources.

It offers a simple way to quantify meteorological conditions influencing fire ignition and spread; thus, it is widely used for understanding the impact of climate variability on fire weather relevant to both observational and model data. In addition, the three fuel moisture codes display the moisture in the three soil layers critical for fire ignition, fire spread, and fire suppression (Wotton 2009). Because it accounts for environmental variables, it is useful for detecting and attributing human influences on fire risk, although with a limited perspective, as it excludes vegetation/fuel availability, ignition sources, and management strategies (Wotton 2009).

The Fire Weather Index (FWI) was computed in three stages. First, moisture codes (FFMC, DMC, and DC) are calculated based on previous day values initiated with prescribed values for DC, DMC, and FFMC (Wagner 1987). In the second stage, the Build-up Index (BUI) and Initial Spread Index (ISI) are derived from these moisture codes. Finally, in the third stage, the FWI is calculated from the BUI and ISI, showing the fire danger and potential fire intensity. The first version of FWI used midday temperature, precipitation, wind speed, and relative humidity observations (Wagner 1987). However, model data from reanalysis or the CMIP initiative can also be used. CMIP6 data are available at daily resolution, in these cases daily maximum temperature, total 24-h precipitation, minimum or mean relative humidity, and mean wind speed (Gallo et al. 2023) can replace the original reference time.

2.4 Methodology

Our analysis focused on 1995–2014 (historical), 2046–2065 (medium-range future), and 2081–2100 (long-range future) from April to October for the three CMIP6 future climate scenarios scenarios. The months chosen aim to emphasize the fire season, which starts in late April and continues for about 20 weeks, with the highest risk period spanning from 1 July to 30 September (Artigo 2.º da Lei n.º 76/2017). We began by validating FWI by comparing the values obtained by the model for the historical period with those of the ERA5 reanalysis. Validation was performed based on the statistical parameters that reflected the means, extremes, and variability. These parameters included mean bias, mean absolute error, and properties related to probability distributions and extremes. The Perkins Skill Score was used to test the degree of overlap between the two distributions (Perkins et al. 2007). This metric quantifies the overlap between two probability density functions (PDFs) by calculating the cumulative minimum value for each bin. A perfect match between a model simulation and the observed conditions results in a skill score of one. The score is defined,

$${S}_{score}= \sum_{1}^{N}minimum({Z}_{0}{,Z}_{m})$$
(1)

where N is the number of bins and Z0 and Zm are the frequencies from observations and from the model in the bin, respectively. This metric has been successfully applied in studies related to climate change to account for changes in probability distributions (Deng et al. 2021; Ridder et al. 2022). Unlike traditional tests that focus on medians, the Perkins score is simple and robust to apply (Perkins et al. 2007; Divine et al. 2018). In the Wilcoxon-Mann–Whitney U test, the null hypothesis is not solely focused on medians; it also considers the differences in scale parameters and symmetry. Therefore, if the test rejects the null hypothesis, it does not necessarily mean that the rejection is solely due to a shift in medians. Simulations demonstrate that samples with equal medians but differing scales and shapes can cause the Wilcoxon-Mann–Whitney U test to fail completely (Divine et al. 2018).

Next, the number of days with an FWI within a certain range was calculated and categorised into danger classes. The European Forest Fire Information System (EFFIS) defines fixed FWI thresholds (Vitolo et al. 2019) for Europe. These limits do not fit perfectly with Portugal, which is a region with a particular combination of atmospheric conditions and vegetation. Therefore, we decided to follow the classes of danger according to the values established by Fernandes (2019), which were tailored and refined to suit the fireline intensity conditions of the region's predominant vegetation, namely maritime pine (Pinus pinaster Ait.). In this classification, values above 38.2 are categorized as very high danger. Subsequently, the upper class was divided into three classes: the extreme danger class and the exceptional danger class (IPMA 2023). The classes and thresholds are presented in Table 2.

Table 2 Classification of Fire Weather Index (FWI) categories based on daily thresholds (Fernandes 2019) and corresponding types of fires that can be ignited by such fire weather conditions (IPMA 2023)
Full size table

3 Results

3.1 Validation of FWI and its components

The results are shown for the different components of FWI (see Section 2.3 for details) for two reasons. First, the different components of the FWI can be considered indices on their own (Amiro et al. 2004), and since FWI aggregates several components if one of them fails, it could compromise the final index.

In this study, we compared the FWI values of WRF-ERA5 and WRF-MPI. Our goal is to assess how well climate simulations align with reanalysis data during the historical period of 1995 to 2014. For this analysis, the daily FWI values were calculated from the daily meteorological variables at 1200 UTC and averaged for the fire season (April to October).

Figure 3 (left panel) shows the quantile–quantile (QQ) plots and probability distributions (PDF, right panel) for FWI and its components. The QQ plot is a probability scatterplot that helps to visually compare how well the two distributions agree with each other in terms of quantiles. Each point is a coordinate pair that represents a quantile from one dataset and the corresponding quantile from another dataset. If both distributions agree, then the points on the QQ plot form a straight line. In our case, the components of the FWI and the index PDFs derived from the model skewed differently with the corresponding ERA5 variables, with an average degree of overlap (Sscore) of 0.90 (Fig. 3).

Fig. 3
figure 3figure 3

Quantile–quantile plots for FWI and its components (left side) and probability distribution functions (PDF, right side). The labels in the quantile–quantile plots represent the percentiles. The Perkins Skill Score (Sscore) is shown in the PDF. The components of the FWI shown from the first row are FFMC (a), DMC (b), DC (c), ISI (d), BUI (e), and in the last row, FWI (f)

Full size image

The FFMC measures the moisture content in shaded litter fuels under a forest canopy (Figs. 3a-b). It estimates how easily these fine fuels can catch fire and how flammable they are, with values ranging from 0 to 101. Higher FFMC values imply drier conditions, whereas lower values indicate moister conditions (Beck and Armitage 2004). One feature stands out: WRF-MPI and WRF-ERA showed a closely aligned pattern in a single frequency mode under dry fuel conditions (Fig. 3a). The Duff Moisture Code (DMC) depicted in Figs. 3b1 and 3b2 indicates the moisture content present in the surface layer, which plays a crucial role in preventing the ignition of deeper layers of fuel. The index varies from lower values for humid conditions to higher open-ended values representing dry conditions. The DMC distribution (Fig. 3b) showed a rapid increase in frequency, followed by a flat and large variability among days with higher values. The results indicate that WRF-MPI exhibits greater variability in the extremes compared to the reference dataset, as evidenced by the higher quantiles (over the 75th percentile). The DC (drought code) represents the drying conditions deep into the soil (Fig. 3c) and characterizes the balance between precipitation and evaporation (Miller 2020). The QQ plot highlights a markedly heavy-tailed distribution in the higher percentiles for WRf-MPI values exceeding 600 (> 75th percentile) as depicted in Fig. 3c. The DC PDF is a multimodal distribution with two local maxima corresponding to wet and extremely dry conditions (Fig. 3c2). Typically, values vary between < 334.1 and > 750 according to the Copernicus Emergency Management Service (2019). Other studies have reported a long-term response of approximately 52 days to weather conditions, which can vary from 15 (initial value) to infinity (Vitolo et al. 2019). Figure 3d–e show the distributions of the components that represent fire behaviour, namely ISI and BUI. The ISI integrates the moisture content of surface fuels and wind speed to estimate the potential rate of fire spread. Figure 3d1 shows that extreme values are more likely to occur in model simulations than in ERA reanalysis, suggesting a heavy-tailed distribution with more outliers and greater variability in extreme values (Fig. 3d1). The ISI PDF shows a significant rise, with two local maxima peaking at approximately 1 and 8 value.

The Buildup Index (BUI) integrates the current DMC and DC to determine the total amount of fuel available for combustion by the fire front (Fig. 3e1-e2). The probability distribution for the BUI skews towards the right-hand side and is characterised by a concentration of values in the lower range, which typically suggests wet conditions. A second frequency mode can be found towards the right tail, followed by a stationary frequency in the extreme dry classes. Figure 3f1-f2 shows the FWI distribution over the study domain. The FWI integrates both the BUI and ISI and serves as a key indicator of extreme fire behaviour during dry conditions. The distribution of FWI shows a departure from observations in the upper percentiles (Fig. 3f), which is identical to the ISI. This suggests that model simulations are more prone to extreme values compared to ERA reanalysis data, with an increased occurrence of outliers and greater dispersion in the dataset, particularly at the tails (Fig. 3f2). Over the study domain, the FWI fuel moisture components ranged from to 0–100 (FFMC), 0–700 (DMC), 0–700 (DC) and to 0–140 for FWI.

Figure 4 shows the climatological daily median and climatological daily differences for the median and 90th percentile derived from ERA5-driven downscallings and WRF-MPI for FWI and its components with the respective skill magnitude given by the Perkins score. The mean values for the moisture-related components, FFMC, DMC, and DC, showed a consistent spatial pattern (Fig. 4a1-a3). They exhibited lower values in the northern region and higher values in the southern region, which aligns with the climatic patterns of the domain. The same north–south gradient is also present in the ISI and BUI indices and, hence, in FWI (Fig. 4a4-a6). The north and west coast regions have the lowest fire danger, in contrast to the southern and inland regions. The median differences in the spatial patterns between the WRF-ERA and WRF-MPI simulations are shown in Fig. 4b1-6. The FFMC component exhibited the lowest differences (approximately 7%), showing a high level of agreement between the ERA and MPI datasets. For the remaining components, the largest disparities were observed in the northeastern region, ranging from 20% (ISI; Fig. 4b6) to 70% (BUI; Fig. 4b5). There were also a few isolated instances in which the differences were greater. The same interpretation is valid for the spatial distribution of the 90th percentile and respective differences (Fig. 4c and d). On average, during the period from April to October, much of the country has a mean FWI above the 20 threshold (Table 2), which corresponds to a class of high fire danger, reaching the very extreme to exceptional classes, FWI > 50, in most of the territory, as can be seen in the distribution of the 90th percentile (Fig. 4c1-c6, The spatial analysis of Perkins Skill scores demonstrates a marked alignment between the climatological WRF-ERA and WRF-MPI datasets within the country, suggesting a significant level of similarity. Overlaps are predominantly around 90%, except for the northeast region, where overlaps are notably lower, hovering around 80% (Fig. 5).

Fig. 4
figure 4figure 4

Spatial distribution of the differences in the daily mean and in the daily 90th percentile between WRF-MPI and ERA5 based on the historical period (1995–2014). The maps show the averages over April to October. Each row, numbered from 1 to 6, represents a component of the FWI: from top to bottom, FFMC (1), DMC (2), DC (3), ISI (4), BUI (5), and FWI (6), respectively. The columns, identified by the letters (a) to (d), represent the median (a), differences in medians (b), the mean field at the 90th percentile (c), and differences in the 90th percentile fields (d), respectively. Please note that the colour bar varies from plot to plot

Full size image
Fig. 5
figure 5

Spatial distribution of the Perkins Skill Score (SScore) for different components of the Fire Weather Index (FWI) over Portugal, comparing the WRF-MPI and WRF-ERA models. The indices include (a) the Fine Fuel Moisture Code (FFMC), (b) Duff Moisture Code (DMC), (c) Drought Code (DC), (d) Initial Spread Index (ISI), (e) Build-Up Index (BUI), and (f) Fire Weather Index (FWI). Higher SScore values indicate better agreement between the two models. Percentage values near 100% indicate a strong degree of similarity

Full size image

3.2 Climate Change Signal

This section begins with an analysis of projected changes in temperature and precipitation. Figure 6 depicts the differences in the median between the future climate and present climatology for the total precipitation summed from April to October for the 20-year period. Purple shading denotes greater future precipitation, whereas brown shading indicates lower future precipitation. The map illustrates the spatial configuration of precipitation in Portugal, showing a notable disparity between the northern and southern regions. The northern area experiences a considerable amount of precipitation, measuring over 700 mm, whereas the inland regions of the north and south receive less than 600 mm of rainfall. In all cases and over different time periods, rainfall for the west and north coasts is projected to decrease, with the most optimistic scenarios (SSP2-4.5 and SSP3-7.0) potentially resulting in a decrease of up to 20%. In the SSP5-8.5, the increase in extent and magnitude is clear, approximately 7% to 20%, for the mid-century period, and less evident towards the end of the century. Scenarios SSP2-4.5 and SSP5-8.5 both predicted a reduction of almost 40% in rainfall for the northern regions and a positive change, that is, more precipitation in the southern region. The most significant change was a reduction of up to 20% in precipitation in the northern region, which is among the wettest European regions. These results suggest an increase in spring precipitation which can favour the growth of vegetation and, consequently, the amount of available fuel.

Fig. 6
figure 6

Projected percentage differences in median rainfall calculated for the fire season (April to March) between the future and the present (first column) for each of the three scenarios and two time periods: SSP2.4–5 (top row), SSP3.7–0 (middle row), SSP5.8–5 (bottom row) and 2046–2065 (middle column) and 2081–2100 (right column)

Full size image

Figure 7 is identical to Fig. 6, except for the maximum temperature. In all cases, the maps show an increase in the maximum temperature. The largest increase of 20% is projected at the end of the twenty-first century in the SSP5-8.5 and SSP2-4.5 scenarios. In these cases, the greatest increase coincided with the areas with the highest rainfall and wettest climate. The weakest warmth could reach 5% by mid-century for SSP2-4.5 and SSP5-8.5 in areas now experiencing a mild summer.

Fig. 7
figure 7

Similar to Fig. 6, but for maximum temperature

Full size image

The average FWI distribution over the fire season (April to October) exhibited a clear spatial pattern (Fig. 8). The southern and northern interior regions are anticipated to have the highest FWI values, surpassing the 38 threshold. Conversely, the northern littoral regions were less prone to high FWI values. Positive differences are expected in the future, especially in the northern region. Regardless of the scenario and time horizon, this pattern remains consistent in terms of intensity, extent, and form, suggesting an intensification of fire danger conditions in the future. Although the average FWI increased from south to north, the same was not true for the number of days above a certain threshold (Fig. 9). Figure 9 displays the number of days with high fire danger (see Table 2 for thresholds). The spatial pattern of days with high fire danger was relatively homogeneous, showing minor differences from the present. The climate change signal indicated a reduction in these days between 7 to 20%. Figure 10, 11, 12, 13 show the days from very high danger to exceptional danger, respectively. Here, the spatial distributions of the number of days with very high to exceptional fire danger were practically the same throughout mainland Portugal and across scenarios (Fig. 10, 11, 12, 13). In all scenarios, the number of fire danger days is expected to increase. As the number of days increased, the disparities increased. For exceptional days (Fig. 13), we found differences of up to 100%, suggesting that this category is not yet seen in the current climate but will in the future. The most substantial increases occurred in the northern and northeastern regions, with an average increase of approximately 50% compared with the current climate (1995–2014). When examining the differences between scenarios, we noticed that the intermediate scenario SSP3-7.0 yielded the largest differences in the medium- and long-term future climate. During the period 2046–2065, the SSP5-8.5 scenario is comparable to the SSP2-4.5 scenario, which may be related to the design of radiative forcing for each scenario. While there is uncertainty associated with the scenarios and the magnitude of climate change, the signal of an increase in the number of days with high fire potential is unequivocal.

Fig. 8
figure 8

Comparison of FWI projections for three future climate scenarios, SSP2.4–5 (top row), SSP3.7–0 (middle row), and SSP5.8–5 (bottom row), with historical climate (1995–2014) and for two climate horizons, middle-term climate (2046–2065) and far-term climate (2081–2100). Differences in percentage

Full size image
Fig. 9
figure 9

Projections for the number of days with high fire danger for three future climate scenarios, SSP2.4–5 (top row), SSP3.7–0 (middle row), and SSP5.8–5 (bottom row), compared with the historical climate (1995–2014) and for two climate horizons, middle term climate (2046–2065) and far-term climate (2081–2100). Differences in percentage

Full size image
Fig. 10
figure 10

Similar to Fig. 9, but for very high days

Full size image
Fig. 11
figure 11

Similar to Fig. 9, but for extreme days

Full size image
Fig. 12
figure 12

Similar to Fig. 9, but for very extreme days

Full size image
Fig. 13
figure 13

Similar to Fig. 9, but for exceptional days

Full size image

4 Discussion and conclusions

Climate transition zones such as Portugal are significantly exposed to climatic variations (Mohammed et. al., 2018), particularly when susceptible to severe and extensive wildfires (Amraoui et al. 2015). Recent climate projections point to hotter, drier, and windier climate (Pereira et al. 2020; Carvalho et al. 2021a; 2021b). These alterations collectively foster an environment that is conducive to the ignition and propagation of forest fires. This study investigated future fire-weather conditions using the Fire Weather Index and explored the potential for dynamical downscaling in Portugal using three CMIP6 future climate scenarios. Considering that most burned areas during a fire season are typically attributed to the growth of fires on days with extreme fire weather conditions (Jain et al. 2020), the potential for fire spread on such days was also investigated.

In this study, we used the Fire Weather Index computed from WRF simulations driven by a CMIP6 ESM (and ERA5 reanalysis for validation), with a resolution of 6 km. Subsequently, we identified the number of days with an elevated fire danger potential within the study region. The validation process was carried out by comparing the WRF-CMIP6 ESM and WRF-ERA5 downscalled data, with the aim of examining the agreement between the two datasets.

During the summer months, the persistence of intense surface heating over the Iberian Peninsula led to a shallow convection low-pressure system. This near-stationary thermal low exhibits daily cyclical behaviour with maximum intensity in the afternoon (Mohammed et. al., 2018) with strong wind gradients from coastal areas to inland areas (Portela and Castro 1996) and no association with the occurrence of precipitation (Trigo et al. 2002). The strong late afternoon winds help the rapid spread and transportation of moist Atlantic air, subsequently resulting in enhanced humidity levels during the night. This weather phenomenon imprints a diurnal variation in air temperature and moisture that can affect the fuel content and, hence, the FWI values. On dry days, fuels reached a minimum moisture content in the late afternoon, increasing during the night with the advection of humid air, starting to dry out in the early morning (Beck and Armitage 2004). This can explain the multimodal frequency in the probability distributions found in the components related to the fuel moisture content, namely FFMC, DMC, and DC (Fig. 3). Our results differ from the climatological distributions presented by others (Amiro et al. 2004, and references within); namely, our distributions presented a long right tail, suggesting that there are outliers or extreme values at the lower end of the scale. These outliers may represent unusually dry conditions or specific situations in which DMC and DC are exceptionally high. There are several possible explanations for this finding. The first reason is the weather conditions. The lack of precipitation and high evaporation rates make Portugal prone to persistent drought (Sousa et al. 2011; Silva et al. 2022). During the 20-year period of 1995–2014 there are three major drought episodes both in intensity and extent, with the 2004/06 drought being the most extensive (100% of the territory affected) and the most intense, taking into account the consecutive months of severe and extreme drought (https://www.ipma.pt/, last accessed 8 April 2024). Second, our climatology is based on model data, and other studies have used meteorological data from past fire episodes (Amiro et al. 2004). Third, the regions are not the same, and the distinctive character of our climatic situation is more favourable to episodes of prolonged drought (Pereira et al. 2020; Carvalho et al. 2021a; 2021b). Finally, studies on these indices (DC and DMC) and their components in the western Iberian region are lacking. One approach is to compare DC with other indices. Drought codes typically refer to numerical or categorical representations of drought conditions that can be derived from indices such as the Palmer Drought Severity Index (PDSI). However, Pereira et al. (2007) mentioned the modification of the PDSI for Southern Portugal, suggesting that the modified PDSI responds better to observed drought conditions in the region, particularly in relation to different soil water-holding capacities (Pereira et al. 2007). This implies that the drought codes, if derived from the modified PDSI, would likely vary more accurately with the actual drought conditions in Portugal than the original PDSI. All these points highlight the need for a more in-depth study of this issue.

Our understanding of future changes in time and space of FWI has been greatly boosted by using numerical models. it is necessary to run computer simulations using regional climate models, typically at scales of around a kilometre, forced by global models. This process of downscaling produces outputs that deviate slightly from observations and may compromise the model skill. In this study, dynamical downscaling using MPI as a global forcing model and WRF as a local model proved skilful. The WRF-MPI was able to reproduce the climatological representation and distribution of FWI derived from WRF-ERA5, as well as its different components, which provide a solid confidence in this work findings.

The Fire Weather Index (FWI) is crucial for understanding fire danger and how it might change in the future. The FWI value progressively rises from the northern regions towards the southern parts of the Portuguese territory, adhering to the climatic patterns that prevail in the country. Specifically, it increases from areas with higher precipitation levels, which in turn results in higher fuel moisture content, to those with lower precipitation amounts and dry northern prevailing winds that are common during the summer season (Portela and Castro 1996; Trigo et al. 2002). The spatial distribution of the FWI reflects this gradual change in climatic conditions. From the FWI spatial distribution, we divided the Portuguese territory into two distinct regions: a northern coastal region with FWI values below the 35 threshold and a southern inland region with FWI values above the 35 threshold. Other studies have identified regions with identical fire incidences with a similar north–south pattern (Sousa et al. 2015; Trigo et al. 2016; Calheiros et al. 2021). They demonstrated a relationship between the northern region and the high-burnt area, with high vegetation dynamics and high annual mean precipitation (Trigo et al. 2016). A local study in Italy and Greece found that the FWI values were linked to the rate of fire frequency. Low values of FWI (< 15) are associated with a low fire frequency rate, whereas values above 45 thresholds are linked to a higher fire frequency (Moriondo et al. 2006). Overall, these findings were consistent with our findings. A northern region with high FWI values, more fire events, and more burnt areas reflects vegetation cover dynamics and weather conditions.

Various studies have assessed how FWI reacts to climate change in different geographical regions. In British Columbia, Canada, there is evidence of a rising trend in the location parameter of the maximum FWI during May, July, and August, especially in areas with elevated maximum FWI values. This suggests a potential increase in the frequency of days with highly elevated or extreme FWI values (Hübnerová et al. 2020). In Greece, projections point to an increase in days with extreme fire danger of as many as 45 days per fire season by the end of the century (Karali et al. 2013). For other latitudes, such as the United Kingdom, a calibration of the FWI system based on percentiles revealed a correlation between the occurrence of wildfires and periods when the FWI component values surpassed the 99th percentile. This suggests that elevated FWI values, particularly those classified as very high or extreme, are linked to increased fire activity (Jong et al. 2016). In the Mediterranean region, there is a general trend of increasing fire danger, which is consistent with longer and more intense summer droughts (Karali et al. 2013; Varela et al. 2019). Our findings are consistent with these trends, suggesting an increase in the number of days from 30 to 40, with high to extreme fire danger. In conclusion, for Portugal, it is reasonable to infer that similar trends may be observed owing to similarities in vegetation and climate. The reviewed literature suggests that there is a general trend towards an increase in the number of days with very high or extreme FWI values in various regions, which is consistent with expectations under climate change scenarios.

However, the low FWI regions (littoral and some areas of the northern region) can be explained by the increased precipitation pattern (Fig. 5) which by itself may reinforce the contradictions identified in previous studies in this region (Santos et al. 2019; Pereira et al. 2020). Different regions show conflicting trends: some may experience more heavy rain in autumn (Santos et al. 2019), whereas others may experience less rain throughout the year but more intense rain in winter and late spring (Pereira et al. 2020). However, studies have consistently indicated a trend towards drier conditions in Portugal, particularly in the northwestern regions, with an overall decrease in mean seasonal precipitation, except in winter (Pereira et al. 2020).

Future projections of temperature and precipitation in Portugal suggest significant changes (Pereira et al. 2020). Studies suggest a substantial increase in temperature across the Iberian Peninsula, with the southern part of Portugal experiencing particularly high temperature rises and more frequent hot days (Carvalho et al. 2021a). In terms of precipitation, a decrease was projected, particularly over northwestern Portugal, with an increase in the number of consecutive dry days (Carvalho et al. 2021a). However, while the overall precipitation is expected to decline, extreme precipitation events may become more intense (Pereira et al. 2020), contributing to a higher proportion of total precipitation, particularly in southern Portugal (Andrade and Contente 2020; Santos et al. 2019).

We conclude that the results for the DC and DMC components, especially DC, reinforce the importance of the role of land surface processes such as soil moisture as a significant factor that impacts future climate projections of temperature and precipitation (Jerez et al. 2012). The SSP5-8.5 scenario, despite being the hottest, produced the fewest days with very high to extreme fire danger values. This may be linked to the thermal low phenomenon, which explains why Portugal's Atlantic coastal zone has the smallest future changes in the number of days with fire weather conditions and low-to-moderate FWI values. The number of days with exceptional fire danger was projected to increase over the study region for all scenarios, with the highest change relative to the historical period found for the inland areas. These results are consistent with those obtained by other studies that projected increases in FWI using various model periods and scenarios (Pereira et al. 2005; Calheiros et al. 2022; Bedia et al. 2014; Moriondo et al. 2006). This may be linked to the increase in warm summer nights in the southern regions, which could be attributed to trends in the Iberian thermal low and North Atlantic Anticyclone (Mohammed et. al., 2018).

This study provides significant contributions to the field of climate science and fire weather research, particularly in the context of Portugal's climate transition zones. By utilizing the Fire Weather Index (FWI) computed from high-resolution WRF simulations driven by a CMIP6 Earth System Model (ESM) and validated with ERA5 reanalysis data, the study offers a detailed analysis of future fire-weather conditions under different climate scenarios. Key contributions may include enhancing understanding of fire weather patterns in Portugal by identifying days with high fire danger and examining regional FWI distribution. The study validates dynamical downscaling with WRF against ERA5 data, showing reliable modeling of fire conditions. It highlights that climate change will likely increase fire risk, especially inland, due to hotter, drier, and windier conditions, and underscores the importance of soil moisture in climate projections. Future research should validate models with more data, monitor fire weather long-term, integrate socioeconomic factors, study extreme events, conduct comparative studies, and inform public policy and awareness on increasing fire risks.

Climate transition zones like Portugal are highly vulnerable to climatic variations, particularly severe wildfires. Future research should further explore the impact of soil moisture, validate models with more data, and integrate socioeconomic factors to improve fire risk predictions and inform public policy. Overall, studies suggest that Portugal will become hotter with less rain and more extreme weather conditions. These changes will deeply impact a country's environment, economy, and society. Adaptation and mitigation strategies are crucial to address these challenges.