The typical FWI calculation is performed using the relevant atmospheric fields at local noon-time . Since many RCMs do not provide subdaily scale information, daily fields need to be identified in such a way that they can replace the typical noon-time calculation with minimal bias. Figure  shows the distribution of the relative percentage bias in the 95th percentile of FWI calculated using the four different input combinations (given in Table S1) relative to the typical noon-time 95th percentile of FWI (FWIoriginal95) across Europe based on ERA5-Land reanalysis (note that the straight lines at 0 and 20° longitudes result from to the concatenation operation described in Sect. ). At the European scale, combinations that include mean relative humidity generally underestimate extreme fire weather danger (Fig. a and b), while those that include minimum relative humidity tend to overestimate it (Fig. c and d). Regarding the magnitude of the bias, using minimum relative humidity instead of mean relative humidity increases the absolute bias when the accompanying variable is maximum wind (Fig. a and c), whereas it decreases the bias when the accompanying variable is mean wind (Fig. b and d).

Since maximum temperature and daily precipitation are common in all combinations, using mean values for both relative humidity and wind leads to a substantial underestimation of FWIoriginal95 with a mean absolute relative bias of 35.1 % (Fig. b). In contrast, using daily extremes, i.e., minimum relative humidity and maximum wind speed, considerably overestimates FWIoriginal95, resulting in a mean absolute relative bias of 43.9 % (Fig. c). Thus, using the daily extreme for one variable and the mean for the other appears to offer a balanced compromise. The mean absolute biases for these combinations are similar: 21.2 % for Comb-1 and 20.8 % for Comb-4 (Fig. a and d, respectively). Sensitivity tests show that this spatial pattern of bias between the different combinations does not depend on the 95th percentile threshold and is qualitatively similar across other parts of the distribution (Figs. S1–5 in the Supplement). Selecting Comb-4 would significantly reduce the number of models in the EURO-CORDEX ensemble, as minimum relative humidity is not an available output for most simulations. Therefore, Comb-1 is selected to calculate the FWI projections using the EURO-CORDEX ensemble, namely, daily maximum temperature, daily accumulated precipitation, mean relative humidity, and daily maximum wind speed. The remainder of the analysis in this study uses these atmospheric fields, for both ERA5-Land reanalysis and EURO-CORDEX models to ensure consistency.

To better understand the influence of ongoing climate change on fire danger in Europe, this section explores the climatology and trends of two FWI metrics, namely FWI_mid and FWI_95 d, for the period 1950–2023 based on ERA5-Land data. We begin with FWI_mid, first presenting its climatology (Fig. a), followed by the corresponding trends (Fig. c). The FWI_mid climatology displays a latitudinal gradient, with the highest values found around the Mediterranean Basin due to hot and dry summers, while the lower values are found in northern Europe (Fig. a). The Iberian Peninsula and Turkey are mainly associated with FWI_mid values in the range of moderate and high levels of fire risk according to the scale developed by . FWI_mid shows a small but statistically significant positive trend in the central Iberian Peninsula, France, and Germany, with statistically significant trends observed in almost 30 % of the total burnable area across the study domain (Fig. c). Most of Europe shows an increase in the intensity of fire weather conditions, but in eastern Europe and Scandinavia the trends are generally not statistically significant. It is also important to note that some of these regions are historically not prone to burning due to limiting factors such as climate or lack of ignition sources, as in parts of Norway (Fig. b).

FWI⁹⁵ climatology (Fig. b) displays the highest values in the Mediterranean Basin, with a latitudinal gradient across Europe except for lower values in mountainous regions. FWI⁹⁵ values above 38 indicate very high fire danger , but this relationship varies regionally: higher values (∼ 50) usually signal extreme danger in warm and dry climates, while lower values (∼ 25) may indicate similar danger in cooler and moister climates . Therefore, FWI_mid and FWI⁹⁵ should be interpreted with caution, taking into account the local climate and biome.

Regarding the trends in FWI_95 d (Fig. d), more than 36 % of all burnable areas exhibit a significant trend, almost all of which indicate an increase. Notable positive trends were observed in the Iberian Peninsula, France, Germany and Ukraine, with trends exceeding 3 d decade⁻¹ in some areas, corresponding to more than 20 additional days of extreme fire weather over the 74-year study period. Some areas of the Iberian Peninsula show trends above 5 d decade⁻¹, amounting to almost 40 additional days of high fire danger during the study period. In contrast, a small region in the eastern Mediterranean shows a statistically significant negative trend, while parts of Scandinavia and the United Kingdom show either no trend or negative but non-significant changes. Other fire-prone regions, including Italy, the Balkans, and the eastern Mediterranean, show mixed and generally non-significant trends, reflecting large interannual variability in extreme fire weather conditions. We note that an analysis focusing on more recent decades (e.g., 1980–2023) results in a much higher proportion of significant trends, with up to 61 % of the burnable land area exhibiting statistically significant trends (results not shown).

GCMs are known to exhibit systematic biases for various reasons, and RCMS often inherit these biases because they use GCMs as boundary conditions for downscaling and may also introduce their own biases . Figure  evaluates the bias in the 95th percentile of the FWI for the EURO-CORDEX ensemble median before and after bias adjustment, relative to the ERA5-Land reanalysis. It should be noted that the time period for the evaluation here is the same as the calibration period for bias adjustment (1971–2000). Although the performance of bias adjustment methods is typically evaluated using a separate validation period e.g.,, we focus on evaluating model performance during the calibration period as the historical simulations are not long enough to allocate a separate validation period for most of the EURO-CORDEX models, considering that 30 years of data have already been used for calibration. In addition, it is important to note that a cross-validation strategy may not always provide reliable results when evaluating the bias adjustment performance, as the internal variability of the climate system can dominate the differences between calibration and validation periods, potentially leading to misleading evaluations, particularly in mid-latitudes .

Adjusting the biases of the individual model input fields prior to the calculation of the FWI results in a significant reduction in the bias of the FWI ensemble median (Fig. ). The spatial mean of the absolute relative percentage bias in the 95th percentile is 80 % before adjustment (Fig. a), with particularly large biases found south of the 50° latitude. After bias adjustment, this spatial mean bias of the ensemble median is reduced to less than 9 % (Fig. b), with significant improvements observed in regions where the raw ensemble median previously exhibited large biases. Sensitivity tests show that the observed spatial pattern of bias reduction is robust and largely independent of the choice of the 95th percentile threshold and exhibits qualitatively similar behavior across other parts of the distribution (Figs. S6–10). However, some small areas still exhibit high biases, especially mountainous regions, such as parts of the Alps, Carpathians, and Caucasus Mountains. However, it should be noted that the FWI values over the mountainous regions are already very low, so even large percentage biases are also usually low in terms of absolute values in these regions. For the purposes of this study, the substantial improvement in model performance by adjusting the input fields with the univariate QDM approach is considered sufficient; hence, the bias-adjusted ensemble is used for the rest of the study.

As significant trends in extreme fire weather have been observed in many regions of Europe, and since all regions are projected to experience warmer summer climates, with a particular tendency towards increased drying in the southern and western regions , it is reasonable to expect further changes in a warming climate. Figure  shows the patterns of the frequency-based FWI metrics (left panels for FWI_fwsl and right panels for FWI_95 d) over Europe during the reference period (1971–2000) and the changes relative to the reference period at 2 and 3 °C GWL based on the ensemble median of 33 bias-adjusted EURO-CORDEX models. The climatological patterns of FWI_fwsl (Fig. a) show longer fire seasons in southern Europe, particularly in the Iberian Peninsula, southern Italy, Greece and Turkey, with more than 60 d yr⁻¹. In contrast, central and northern Europe experience shorter fire seasons. Since these values may seem lower than expected, it is important to distinguish between different methods for calculating the duration of the fire season. define the fire season based on a temperature threshold, and it is used for overwintering the DC . However, this definition may lead to an overestimation of fire season length, especially in southern Europe due to warmer temperatures throughout the yr. In contrast, define FWI_fwsl as the number of days when FWI exceeds its mid-range value. Here, the definition of is followed for FWI_fwsl, as in some other studies . However, this approach possibly underestimated the actual FWI_fwsl in some regions, because overwintering the DC reduces the number of days with an available FWI value throughout the yr due to the condition of .

The projected changes in FWI_fwsl at 2 °C and 3 °C GWLs relative to the reference period are shown in Fig. c and e, respectively. At 2 °C, robust increases in FWI_fwsl are already evident in some regions, with a magnitude of up to 100 % in areas like parts of the Balkans, but the larger changes are mostly limited to regions with smaller climatological values. The increase in FWI_fwsl becomes more widespread and intense at 3 °C, particularly in France and the Balkans, with a relative increase exceeding 150 %. Almost all regions show positive changes in FWI_fwsl at both GWLs (except for a small area in northern Poland at 2 °C). The projected mean relative change in FWI_fwsl for the grids where the models agree on both the sign and significance of the change almost doubles between the two GWLs: 52 % at 2 °C and 94 % at 3 °C. The signal is largely confined to regions south of 50° latitude, with a robust signal simulated in 25 % and 48 % of the burnable land area, at 2 °C and 3 °C, respectively.

The right column of Fig.  shows the number of days per yr that exceed the 95th percentile FWI (FWI_95 d) during the reference period (Fig. b), along with the relative changes compared to the reference period at 2 °C (Fig. d) and 3 °C (Fig.  f) GWLs. Here, it is important to note that the spatial distribution of FWI_95 d during the reference period would be nearly uniform with around 18 d yr⁻¹ across the domain if FWI was calculated continuously, since the 95th percentile is calculated locally for each grid. However, this is not the case here, as overwintering of the DC interrupts the continuous FWI calculation. Therefore, the latitudinal differences in FWI_95 d are a natural consequence of this calculation. More days exceed the local 95th percentile at lower latitudes, primarily because the fire season is longer as defined by . In the Iberian Peninsula and some regions in southern Europe, there are about 18 d yr⁻¹ of FWI_95 d, indicating that the fire season defined by the temperature threshold endures almost all yr.

The projected changes in FWI_95 d relative to the reference period at 2 °C and 3 °C GWLs are shown in Fig. d and f. At 2 °C, the changes are mainly confined to southern Europe, with only 23 % of the total burnable land area in the domain exhibiting robust signals (Fig. d). In contrast, at 3 °C, significant and robust signals are projected to extend into central Europe, covering 46 % of the total burnable area (Fig. f). In addition, the strength of the signal becomes much more pronounced at 3 °C, with relative increases exceeding 150 % in many regions, such as the Iberian Peninsula, southern France, the Balkans, and Turkey. Even regions with historically low frequencies of extreme fire weather, such as northern Europe, show increases of up to 50 %. However, these increases are not statistically significant or model agreement is not established. In areas where the change signal is robust, the mean relative increase in FWI_95 d is 97 % and 153 %, respectively, at 2 and 3 °C GWLs. Overall, a general increase in the frequency of fire weather conditions is projected across Europe, with the magnitude and spatial extent of the changes becoming much more pronounced at 3 °C compared to 2 °C. However, robust changes are still mainly limited to areas south of 50 °C latitude even at 3 °C.

In addition to projections of the frequency of extreme fire weather, it is also important to quantify the projected changes in the magnitude of extreme fire weather. Figure  shows the patterns of the magnitude-based FWI metrics (left panels for FWI_fs and right panels for FWI_max) over Europe during the reference period (1971–2000) and the changes relative to the reference period at 2 °C and 3 °C GWLs based on the ensemble median of 33 bias-adjusted EURO-CORDEX models. Similar to frequency metrics, a latitudinal gradient is apparent for FWI_fs, with higher values concentrated mainly in southern Europe due to hot and dry summer conditions (Fig. a). In regions such as the Iberian Peninsula, southern Italy, Greece, and Turkey FWI_fs values exceed 30, approaching the very high fire danger threshold of 38, as accepted by .

The projections show a minor relative increase in FWI_fs at 2 °C GWL across Europe, with the exception of regions such as Poland and northern Scandinavia (Fig. c). However, many areas still lack robustness, with only 22 % of the burnable land area is projected to show robust signals. At +3 °C, FWI_fs is projected to become more widespread and intense, following the same pattern seen in frequency-based metrics (Fig. e). It is projected to increase by more than 50 % (with robust signals) not only in southern and eastern Europe, but also in regions that historically exhibited lower fire danger, such as eastern France. The percentage of burnable land area with model agreement also increases to 41 % at 3 °C. The spatially averaged relative increase in FWI_fs for regions with model agreement on robustness of the change signal is also projected to increase as the GWL increases: 25 % at 2 °C and 47 % at 3 °C.

The climatological distribution of FWI_max displays higher values exceeding 40 in southern Europe and lower values, usually below 20 in central to northern Europe (Fig. b). Mountainous regions, such as the Alps and Carpathians, exhibit very low FWI_max values, reflecting their colder and moister climates compared to the surrounding areas. In southern Europe, regions like the Iberian Peninsula and Turkey have FWI_max values greater than 50, indicating that these areas experience extreme fire weather conditions annually on average.

The projected changes in FWI_max at 2 and 3 °C GWLs relative to the reference period are shown in Fig. d and f, respectively. At 2 °C, the simulated positive changes are relatively small (around 10 %–20 %). Central and northern Europe show spatial variability, including negative changes in some regions (e.g., Poland), although these are not robust signals (Fig. d). Only 8 % of the burnable land area shows model agreement on the direction and significance of the change at 2 °C, and these areas are mostly limited to regions with very low climatological FWI_max values. In contrast, at 3 °C, the percentage of burnable land area where the majority of models project robust climate change signals increases sharply to 40 % (Fig. f). However, areas with higher relative changes are still primarily the regions with low climatological values (e.g., mountainous regions), with the exception of some parts of Southern Europe, such as Italy and the Balkans, where projected increases range from 20 % to 40 %. Overall, in regions where change signal is robust, the simulated mean increase in FWI_max is 26 % and 33 % relative to the reference period, respectively, at 2 and 3 °C.

In this section section, we show the same fire weather metrics analyzed throughout the study (FWI_95 d, FWI_fwsl, FWI_max, FWI_fs), but aggregate and average them over the PRUDENCE regions to highlight differences between the European subregions and show the spread across the model ensemble. Using the absolute values, we also aim to provide context for the relative changes presented in Sect. . The results are displayed as box plots for each PRUDENCE region for the reference period (1971-2000), and at 2 and 3 °C GWLs (Fig. ). The regions are ordered in a way that roughly corresponds to the latitude-longitude orientation (North-West to South-East) of the regions.

FWI_95 d shows a quasi-uniform distribution across models in all PRUDENCE regions for the reference period, due to the definition of the metric and the bias adjustment of temperature, which effectively determines the start and end of the fire season based on the temperature-threshold used for overwintering DC (Fig. a). Under both warming scenarios, the median FWI_95 d is projected to increase across all regions with a stronger signal at 3 °C. In the Iberian Peninsula, for example, FWI_95 d is projected to increase rapidly from a median of ∼ 17 d yr⁻¹ during 1971–2000 to 31 d yr⁻¹ at 2 °C and to 46 d yr⁻¹ at 3 °C (Fig. a). A similar trend is projected for France, with the median increasing from 14.5 d yr⁻¹ during 1971–2000 to more than 23 d yr⁻¹ at 2 °C and to 35 d yr⁻¹ at 3 °C. In some regions, such as Eastern Europe, the model spread is large, with some models projecting a negative change, and some projecting a very extreme positive change. The model agreement is stronger in southern European regions (IP, MD, TR), where all models project a positive trend under both 2 and 3 °C GWLs. In addition, the spread among models increases from 2 °C to 3 °C in all regions, indicating a growing uncertainty with higher levels of warming.

There is a clear regional separation in both the reference period and projected values of FWI_fwsl, with southern European regions (IP, MD, TR) showing higher levels of fire danger than the rest of Europe (Fig. b). However, a positive change signal is evident across almost all regions, especially at 3 °C GWL. In the Alps, a region historically characterized by shorter fire seasons, the median FWI_fwsl is projected to more than double: from 14.4 d yr⁻¹ during 1971–2000, to almost 22 d yr⁻¹ at 2 °C, and to more than 31 d yr⁻¹ at 3 °C. The Iberian Peninsula is projected to experience the highest absolute increase in multi-model median FWI_fwsl, rising from 43 d yr⁻¹ during 1971–2000 to a projected 64 d yr⁻¹ at 2 °C and almost 78 d yr⁻¹ at 3 °C.

A clear latitudinal gradient is also observed in the magnitude of FWI_max during the reference period and at both 2 °C and 3 °C GWLs with southern European regions exhibiting higher fire danger due to their warm and dry summer climates (Fig. c). The main exception to this latitudinal pattern is the Alps, which exhibits lower values due to its colder and moister climate conditions. Overall, the median FWI_max is projected to increase consistently across all regions as the GWL increases. However, the projected relative changes in FWI_max (Fig. c), a magnitude-based metric, are not as pronounced as those seen in frequency-based metrics (Fig. a and b). The Mediterranean, the Iberian Peninsula, and Turkey clearly stand out from other regions. For instance, Turkey shows a simulated median FWI_max of 49 during 1971–2000, which is projected to increase to 54.6 at 2 °C and to 58 at 3 °C.

The FWI_fs (Fig. d) is also projected to follow a similar pattern to FWI_max. However, the distribution ranges for FWI_fs are narrower than those for FWI_max, particularly in northern and central European regions due to the time averaging. The intensity of prolonged fire weather, as represented by FWI_fs, is projected to increase in all regions. For example, the median FWI_fs in the Mediterranean is projected to increase from 15 during 1971–2000 to 18 and 20 at 2 and 3 °C GWLs, respectively.

In summary, all four fire weather metrics are projected to increase in all regions with increasing temperature, and more pronounced changes are expected at 3 °C GWL compared to 2 °C. In general, southern European regions (IP, MED, TR) are projected to experience more intense and prolonged extreme fire weather conditions, consistent with their baseline climatologies of dry and warm summers. Overall, the relative change signals are stronger for the frequency-based metrics (FWI_95 d and FWI_fwsl) than for the magnitude-based metrics (FWI_max and FWI_fs). A clear latitudinal gradient is also observed, with southern European regions projected to experience more severe and frequent fire weather conditions. Furthermore, the ensemble spread increases from 2 to 3 °C GWL, particularly for the frequency-based metrics, indicating greater uncertainty at higher warming levels.

Given the projected changes in extreme fire weather conditions across Europe, it is important to explain possible drivers. To provide insights into this question, we examine the two main subcomponents of FWI: BUI, which represents the effects of longer-term atmospheric conditions on fuel dryness, and ISI, which reflects the influence of short-term atmospheric conditions, namely the role of wind patterns and FFMC . Figure  illustrates FWI as a function of ISI and BUI across PRUDENCE regions, based on spatially aggregated fields from the bias-adjusted EURO-CORDEX models. The ISI–BUI pairs correspond to the averaged values for the days when FWI exceeds its 99th percentile during the reference period (1971–2000), as well as at 2 and 3 °C GWLs for each model in the ensemble.

As global warming increases, there is a projected shift in BUI towards higher values in all regions (Fig. ). However, the signals in BI and SC are mixed, with a slight tendency towards higher fire danger (Fig. a and b). In FR, ME, AL and EA regions, there is a notable shift in BUI, along with an increase in ISI, both of which contribute to the resulting increase FWI (Fig. c–f). In particular in France, the median BUI is projected to increase by almost 50 % at 3 °C GWL relative to the reference period, while ISI is projected to increase only by around 25 %, which is an indication of the worsening of longer-term fuel drying. The increase in BUI is also apparent in the southern European regions (Fig. g–i). However, this does not directly translate into a significant increase in FWI in these regions because the contribution of BUI to FWI saturates at higher BUI values, consistent with the principle that there is a limit to the amount of fuel that can be used in fires . Instead, in these regions, the projected increase in FWI is driven by a projected increase in ISI values, which may result from an increased surface layer dryness (reflected by FFMC) or stronger winds.

To disentangle the contributions of wind speed and dryness conditions, Fig.  shows VPD plotted against maximum wind speed across PRUDENCE regions. The VPD-wind speed pairs correspond to the averaged values for the days when FWI exceeds its 99th percentile during the reference period (1971–2000), as well as at 2 and 3 °C GWLs for each model in the EURO-CORDEX ensemble. The medians of maximum wind speeds are projected to either decrease or remain largely unchanged across all PRUDENCE regions, except in southern Europe, where a very slight increase (less than 2 %) is projected, but likely not significant. In contrast, the median VPD, which reflects the thermodynamic effects of temperature and relative humidity through atmospheric drying, is projected to increase across all regions, with particularly strong increases in central and southern Europe. In southern European regions, the projected increase in the median VPD on days when FWI exceeds its 99th percentile at 3 °C GWL relative to the reference period is almost 40 %, with some models projecting changes greater than 70 % (Fig. g–i).

To further analyze the key drivers responsible for changes in fire weather extremes, we calculate the simulated changes in 30 d accumulated antecedent precipitation conditioned on days with extreme FWI values (Fig. S13). There is a trend towards decreasing precipitation totals as GWL increases across regions for these conditional days in the ensemble median (except for Scandinavia). The decreasing precipitation totals seem more critical for regions such as France, in contrast to southern Europe, where the baseline climatology during extreme fire weather days is already very dry. Therefore, further drying in these regions might not increase fire weather danger as much as increases in temperature, as also reported in . However, to better discern the role of precipitation in intensifying fire weather extremes, a more targeted analysis is needed, as there is no model agreement in the change signal in many regions (e.g., Eastern Europe).

These results suggest that projected changes in FWI in southern Europe are primarily associated with an increase in fuel aridity due to amplified atmospheric moisture demand rather than changes in wind speeds. This may also indicate that the projected changes in ISI in southern Europe (Fig. ) are largely influenced by increased surface layer fuel dryness (FFMC), rather than wind speed. Precipitation is also an important contributor in regions such as France, but its role is more difficult to assess in many other regions due to the higher uncertainty.

In addition to the distribution of model statistics across the PRUDENCE regions, we also present the spatial pattern of the ensemble median changes in the composites of the variables of interest at 2 and 3 °C GWLs relative to the reference period conditioned on days when FWI exceeds its 99th percentile (Fig. ). The pattern of changes in extreme FWI (mean FWI on days when FWI > FWI⁹⁹) shows a high spatial correlation with changes in VPD on these days, higher than the individual correlations with maximum temperature and relative humidity. Specifically, the ensemble median field correlation between changes in VPD and FWI composites is 0.85 at 2 °C GWL and 0.87 at 3 °C GWL, with ensemble ranges of 0.51–0.91 and 0.71–0.91, respectively (at 2 °C, only two models show correlations below 0.67). The decline in 30 d accumulated precipitation is also evident in regions such as France and suggests that these events occur under drier antecedent conditions, especially at 3 °C GWL, but this is less relevant for southern regions. Changes in maximum wind speed are weak and spatially heterogeneous. Overall, the northward expansion of more extreme fire weather danger in a warming climate in Europe is associated with stronger atmospheric moisture demand through increasing VPD (i.e., increasing temperature and/or decreasing relative humidity), with compounding effects from declining precipitation in regions such as France. However, these relationships should be interpreted with caution, as they represent spatial co-variations rather than formal attribution statements.