Fire Behavior and Propagation of Twin Wildfires in a Mediterranean Landscape: A Case Study from İzmir, Türkiye

Abstract

Twin wildfires burned over 9500 ha in Seferihisar, İzmir, western Türkiye, on 29—30 June 2025 under extreme fire weather conditions. This study reconstructs the spatiotemporal progression of the fires and examines the drivers of contrasting behaviors and burn severity. Multi-source datasets—Sentinel-2 imagery, VIIRS/MODIS thermal detections, MTG images and thermal detections, aerial photos, and ground data—were integrated to delineate progression polygons and compute rate of spread (ROS), fuel consumption (FC), and fire-line intensity (FI). Kuyucak fire showed rapid early growth, burning 3554 ha in 2.5 h (mean ROS of 5.0 km h⁻¹; mean FI of 37,789 kW m⁻¹), driven by strong northeasterly winds of 40–50 km h⁻¹, steep terrain, dense Pinus brutia fuels, and very low dead fine-fuel moisture (<6%). Kavakdere fire advanced more slowly (mean ROS of 1.6 km h⁻¹) across open grassland and cropland, yielding lower FC and FI. Synoptic analysis revealed a strong pressure-gradient-induced northeasterly wind regime linked to a mid-tropospheric geopotential height dipole between Central Europe and the Eastern Mediterranean, while WRF simulations indicated a dry boundary layer and enhanced low-level winds during peak spread. Sentinel-2 dNBR burn severity mapping showed substantial spatial variability tied to fuel and topography contrasts. Findings demonstrate how twin ignitions under similar weather conditions can produce divergent outcomes, underscoring the need for terrain- and fuel-aware strategies during extreme Mediterranean fire outbreaks.

1. Introduction

In Mediterranean ecosystems, hot and dry summers and flammable vegetation create conditions conducive to intense recurrent wildfire activity [1,2]. In recent decades, climate change, land-use alterations, and the accumulation of forest fuels have contributed to an increase in the frequency, intensity, and size of wildfires in the region [3,4]. These changes have raised concerns about the resilience of Mediterranean landscapes, as well as the capacity of fire management agencies to cope with extreme fire events [5].

In the Mediterranean Basin, large and fast-moving wildfires are often associated with periods of extreme fire weather [6], particularly strong, dry winds combined with low fuel moisture content [5]. Under such conditions, crown fires can develop rapidly, producing high rates of spread (ROS) and extreme fire-line intensities (FI) that challenge suppression efforts and threaten human lives and infrastructures [7].

Most wildfire studies in the region have focused on single large events, yet simultaneous or near-simultaneous ignitions, often referred to as “twin fires” or “multiple large fire outbreaks”, pose unique operational challenges. Multiple concurrent fires can strain suppression resources and produce overlapping disturbance patterns in the landscape [8]. Despite their potential impact, such events remain understudied, particularly in the Eastern Mediterranean region.

On 29 June 2025, two large wildfires ignited within nearly one and a half hours of each other in the Seferihisar district of İzmir Province, western Türkiye. The Kuyucak fire (first fire) originated near Orhanlı village, while the second (Kavakdere fire) ignited approximately 10 km to the west. This study aims to (i) reconstruct the spatiotemporal progression of both fires using multi-source remote sensing and ground data; (ii) quantify key fire behavior parameters, including rate of spread, fuel consumption (FC), and fire-line intensity (FI); (iii) assess burn severity using Sentinel-2-derived spectral indices; and (iv) evaluate the influence of weather, fuels, and topography on fire dynamics and severity. Although the twin wildfires ignited under broadly similar fire weather conditions, their fire behavior and resulting burn severity differed markedly. This study aims to elucidate the factors responsible for these differences. By documenting and analyzing this twin-wildfire event, we provide new insights into extreme fire behavior under Mediterranean conditions and offer valuable information that is relevant to both the scientific community and operational fire management.

2. Materials and Methods

2.1. Wildfire Events

On 29 June 2025, at 09:57 UTC (12:57 local time, UTC+3), the Kuyucak fire ignited in a forested area near (Kuyucak locality) the village of Orhanlı in the Seferihisar district of İzmir Province, western Türkiye. Later the same day, at 11:20 UTC, the Kavakdere fire ignited about 10 km west of the Kuyucak fire in the Kavakdere locality (Figure 1). Records from the General Directorate of Forestry (GDF) indicate that the Kuyucak fire was caused by a spark from power lines, while the cause of the Kavakdere fire remains unknown [9,10].

2.2. Topography

The two fires occurred in terrain with slopes ranging from 0% to approximately 101%, indicating notably steep sections, including cliffs. The mean slope of the Kuyucak fire is 23%, with a high standard deviation (23.1%), reflecting a highly variable topography. In comparison, Kavakdere fire’s slopes range from 0% to 58%, with a mean of 16% and a standard deviation of 9.3%, suggesting a more uniform and moderately inclined terrain (Figure 2).

2.3. Fuels

Stand-type maps, based on stereo interpretation of aerial photographs and ground measurements taken at 300 × 300 m sampling points, were obtained from the Menderes State Forest Enterprise’s forest management plans archive. These maps, part of the GDF forest inventory, included data on tree species, development stage (diameter at breast height (DBH)), and crown closure (%). Surface and available crown fuel loads were calculated using established models [11,12]. It is known that only the fine fuels involving needles and fine branches smaller than 0.6 cm in diameter are consumed in a crown fire [13,14]. Therefore, surface (SFL) and available crown fuel loading (CFL) (needles and fine branches smaller than 0.6 cm in diameter) were calculated and considered to be totally consumed. Stand types were grouped into 13 fuel types (Table S1), ranging from cropland and grasslands to various age classes of Pinus brutia, Pinus pinea, and mixed stands with Cupressus spp. (Figure 3).

The burned areas of the Kuyucak and Kavakdere fires in İzmir show distinct fuel compositions. The Kuyucak fire is dominated by forests, including mature and immature Pinus brutia and Pinus pinea, mixed with shrubland and smaller patches of cropland. In contrast, the Kavakdere fire contains a higher proportion of grassland and cropland, alongside shrubland and overmature Pinus brutia (Figure 3).

2.4. Fire Weather

Weather observations from the Turkish State Meteorological Service automatic weather station at Menderes–Gümüldür (ID: 18050; 38.072033° N, 27.002578° E—Elevation: 59 m) [15] (Figure 1) were collected. Hourly observations included air temperature (°C), relative humidity (%), precipitation (mm), wind speed (km h⁻¹), and wind direction during the fire events (29 June 23:00 to 1 July 00:00 (UTC)). These were also used to calculate the vapor pressure deficit (VPD) [16,17] and estimate hourly dead fine fuel moisture content (FMC) [16] for Pinus brutia, the dominant fuel type in the burned area.

We further supplemented our analysis with atmospheric dynamics derived from reanalysis and short-term forecast data. More specifically, we used data from the fifth-generation reanalysis dataset of ECMWF, namely ERA5 [18], and short-term forecast data from the Weather Research and Forecasting (WRF) model, which has been validated and utilized in various previous studies [19,20]. We opted for ERA5 data to analyze the synoptic-scale atmospheric dynamics using both upper-tropospheric (geopotential height, relative vorticity, wind, temperature, moisture, equivalent potential temperature, and potential temperature at various atmospheric pressure levels) and surface-level (wind, temperature, moisture) data. ERA5 reanalysis data span from 1940 to the present, available with an hourly timestep, and have a spatial resolution of 0.25° × 0.25°. WRF short-term forecast data are initialized every 12 h, with an hourly timestep and a spatial resolution of 2 km. For further details on the operational WRF model setup used in this study, the reader is referred to Giannaros et al. [19,21].

We quantified large-scale circulation departures using standardized anomalies of geopotential height at the 500 hPa isobaric level. Anomalies are referenced to a 5-day running daily climatology over 1991–2020. In addition, we used WRF short-term forecast data to analyze the mesoscale atmospheric dynamics utilizing cross-sections and vertical profiles.

2.5. Fire Progression

Wildfire progression describes the spatial and temporal development of burned areas during a wildfire. To reconstruct the progression of the twin fires, we integrated the most comprehensive data available from satellite, airborne (images from airplanes and helicopters), and ground sources to achieve convergence of evidence (Figure 4).

Wildfire progression was represented by a series of consecutive polygons illustrating the temporal evolution of fire growth. The number of polygons depended on the burned extent and the availability of data. These polygons were constructed using a wide range of complementary data sources to enhance spatial and temporal accuracy [22]. By combining all available information, we manually delineated the ignition point and tracked the fire’s progression over time. The reconstruction was carried out chronologically, starting from the ignition and continuing until the fire front reached specific locations.

2.5.1. Satellite Data

To reconstruct the progression of the two wildfires, both satellite imagery and thermal anomaly data were used. Satellite imagery data from the VIIRS (Visible Infrared Imaging Radiometer Suite) instrument [23,24] aboard the S-NPP (Suomi National Polar-orbiting Partnership), NOAA-20, and NOAA-21 (National Oceanic and Atmospheric Administration) satellites were utilized. These satellites collect data, on average, twice per day, with spatial resolutions of 375 m and 750 m. The Moderate Resolution Imaging Spectroradiometer (MODIS), onboard the TERRA and AQUA satellites, provides spatial resolutions ranging from 250 m to 1000 m, offering, on average, six daily revisits when combined.

Atmospherically corrected Level 2 satellite images were also used to create false-color composites that highlight burned areas (characterized by low near-infrared [NIR] and shortwave-infrared [SWIR] reflectance), active flaming zones (high SWIR reflectance), and unburned vegetation (high NIR reflectance). Typical SWIR composites use bands 12–8A–4 for Sentinel-2 (Figure 5b). The imagery was downloaded from the Copernicus Browser (https://browser.dataspace.copernicus.eu/ (accessed on 6 July 2025)) and NASA Worldview (https://worldview.earthdata.nasa.gov/ (accessed on 6 July 2025)).

To complement the satellite imagery, thermal anomaly products from VIIRS (VNP14IMGML-C1) and MODIS (MCD14ML) were used [25]. These provide data at 375 m and 1 km resolution at nadir, respectively, and are available from FIRMS (https://firms.modaps.eosdis.nasa.gov/ (accessed on 6 July 2025)). These products enable the estimation of the approximate location and timing of active wildfires and provide fire radiative power values, estimates for the radiant energy released per unit time (Figure 5c–f).

High-temporal-frequency geostationary imagery (1 km at nadir) from the Flexible Combined Imager (FCI) onboard Meteosat Third Generation (MTG) was used to characterize the temporal evolution of fire activity [26,27]. MTG acquires data every 10 min. With that image, the thermal anomaly and associated fire radiative power (FRP) are estimated. Fire Radiative Energy (FRE) was calculated by integrating Fire Radiative Power (FRP) values over consecutive 30 min intervals. FRE was computed as (1)

$$\mathrm{F}\mathrm{R}\mathrm{E}=\sum {FRP}_i\times {∆}_t$$

(1)

where FRPi is the radiative power at time step i and Δ_t is the sampling interval (600 s for 10 min data). The resulting energy values were expressed in terajoules (TJ), allowing the estimation of both peak half-hour releases and cumulative totals across the fire’s duration.

Additionally, EUMETSAT’s MTG Cloud Phase RGB imagery, which is generated by using NIR1.6, NIR2.25, and VIS0.6 bands, with 10 min intervals, was also used for validation (Figure 6). The MTG satellites feature the Flexible Combined Imager (FCI), which advances from the SEVIRI instrument of the previous generation. FCI scans the full Earth disk every 10 min across 16 spectral channels, offering a spatial resolution from 2 km to 0.5 km. We supplement our analysis of fire progression with MTG data; however, a more detailed comparison and comprehensive exploitation of these datasets falls outside the scope of the present study.

2.5.2. Airborne and Ground Data

Airborne and ground-collected data, including photographs, videos, and post-fire interviews with field personnel, provided valuable spatiotemporal information on wildfire ignition and spread. Air suppression platforms and ground vehicles were equipped with RGB cameras; images captured at specific times by fire-suppression aircraft and helicopters (Figure 5a) were incorporated into the analysis. The locations of active areas and fire fronts were manually geolocated using satellite imagery and incorporated into manual polygon delineation. Photographs and videos recorded by suppression teams were likewise geolocated and used to refine the polygon delineation and reconstruct fire progression.

2.6. Fire Behavior

Based on the fire progression and fuel dataset, we calculated the fire behavior descriptors as forward ROS, FC (kg m⁻²), and FI (kW m⁻¹). The ROS was calculated by dividing the longest fire spread distance along the head-fire direction by the time elapsed between the pair of polygons and was expressed in kilometers per hour (km h⁻¹). The FC was calculated with the combination of stand type maps and fuel models for burned area. Fuel consumption was computed along the head-fire direction within ±100 m buffers of the fire-line. After calculating ROS and FC, Byram’s FI (2) computed using a low heat of combustion for each progression of the two fires [28]. The fire behavior metrics (e.g., ROS, FC, and FI) were calculated for each fire progression polygon, and these polygons were used to analyze the relationships between fire behavior metrics and environmental and weather data throughout the fire progression. For the analysis, hourly weather measurements and FMC values were aggregated by computing the mean over the duration of each progression polygon.

FI = H × w × r

(2)

Here, FI is the fire-line intensity (kW m⁻¹), H is the low heat of fuel combustion (kJ kg⁻¹), w is the amount of fuel consumed in the active flaming front (kg m⁻²), and r is the linear rate of fire spread (m s⁻¹). The H-values were taken from similar fuel types in the Portuguese fuel model [29].

Fire growth rate (FGR) was also calculated for each progression polygon by dividing the burned area of each polygon by its corresponding duration (ha h⁻¹) [22]. FGR is an important metric for describing fire behavior [30].

2.7. Burn Severity Classification and Assessment

Spectral indices derived from satellite imagery have long been utilized to detect burned areas. Among these, the Normalized Burn Ratio (NBR) is one of the most widely used indices in wildfire monitoring and assessment [31,32]. In this study, NBR was calculated using Sentinel-2 satellite imagery, which provides high-resolution multispectral data well-suited for fire impact analysis. Specifically, the near-infrared (NIR, Band 8) and shortwave infrared (SWIR, Band 12) bands of Sentinel-2 were used, as they are sensitive to vegetation condition and moisture content, respectively. The NBR is defined by the following equation (Equation (3)):

NBR = (NIR − SWIR)/(NIR + SWIR)

(3)

NBR values range between −1 and +1. Healthy vegetation typically produces high NIR and low SWIR reflectance, whereas burned areas show the opposite, allowing for effective identification of fire-affected zones. To assess burn severity, the differenced Normalized Burn Ratio (dNBR) was calculated by subtracting the post-fire NBR from the pre-fire NBR (4).

dNBR = NBRpre-fire − NBRpost-fire

(4)

For the analysis, Sentinel-2 images acquired on 25 June and 5 July 2025 were used to represent pre-fire and post-fire conditions, respectively. The dNBR serves as a proxy for assessing fire-induced damage to vegetation and soil (e.g., loss of organic matter, structural degradation). Higher positive dNBR values indicate more severely burned areas, and vice versa. Burn severity was classified using standard dNBR thresholds [32] (

Table 1. Burn severity classification [32].

Burn Severity Class	dNBR Range
Unburned	dNBR < 0.1
Low Severity	0.1 ≤ dNBR < 0.27
Moderate–Low Severity	0.27 ≤ dNBR < 0.44
Moderate–High Severity	0.44 ≤ dNBR < 0.66
High Severity	dNBR ≥ 0.66

).

2.8. Statistical Analysis

Pearson correlation analysis was conducted to examine the relationship between weather conditions and associated fire behavior parameters. The analyses were based on fire progression polygons. Mean values of weather and fire behavior variables for each fire progression polygon were calculated and used in the analysis. Before the analyses, a Shapiro–Wilk normality test was conducted on all variables. Statistical analyses were performed using SPSS version 26.0 for Windows (IBM SPSS Statistics for Windows. Version 26.0, IBM Corp, Armonk, NY, USA).

3. Results

3.1. Rate of Spread, Fuel Consumption, and Fire-Line Intensity

3.1.1. Kuyucak Fire

Kuyucak fire was analyzed in nine progression polygons using the most reliable available datasets (Figure 7). The fire propagation was divided into two distinct burn periods. The first burn period began at the time of ignition (09:57) and continued until 12:38 on 29 June, during which the head fire reached its final extent at the southwest, covering approximately 2.5 h of active spread (polygons 1–4) (Figure 8a). During this period, the fire spread predominantly in a southwesterly direction with an estimated indicated mean rate of spread (ROS) of 5.00 km h⁻¹ (minimum and maximum of 4.27 km h⁻¹ and 5.71 km h⁻¹, respectively). The extent of the burned area reached 3554 ha, accounting for 76% of the total area burned in the Kuyucak fire (Figure 7).

The mean fuel consumption (FC) during this period was 1.27 kg m⁻², with a minimum of 0.66 kg m⁻² and a maximum of 1.62 kg m⁻² (Figure 8a). The calculated fire-line intensity (FI) reached up to 52,495.02 kW m⁻¹, with a mean of 37,789.13 kW m⁻¹ and a minimum of 17,216.12 kW m⁻¹ (Figure 8c).

Following the spread of the fire to its southwestern extent, it continued spreading laterally, reaching its final perimeter at 13:18 on 30 June in the second burn period (Phases 5–9) (Figure 8a,c), lasting approximately 26 h. During this period, the mean ROS was much lower than in the first period, at 0.24 km h⁻¹ (minimum and maximum of 0.02 km h⁻¹ and 0.70 km h⁻¹, respectively). The fire burned an additional 1148 ha, corresponding to 24% of the total burned area. The mean FC was 1.28 kg m⁻² (minimum and maximum of 0.89 kg m⁻² and 1.60 kg m⁻², respectively). The mean calculated FI was 1901.41 kW m⁻¹ (minimum and maximum of 179.87 kW m⁻¹ and 6345.88 kW m⁻¹, respectively) during this second burn period (Figure 8c).

3.1.2. Kavakdere Fire

The Kavakdere fire was also analyzed in nine progression polygons and divided into two distinct burn periods (Figure 7). The first burn period began at the time of ignition (11:20) and continued until 22:57 on 29 June, during which the head fire reached its final extent, covering approximately 11 h of active spread (polygons 1–4) (Figure 8b). During this period, the fire spread predominantly in a southwesterly direction, similarly to the Kuyucak fire. Results indicated a mean ROS of 1.99 km h⁻¹ (minimum and maximum of 0.87 km h⁻¹ and 3.25 km h⁻¹, respectively).

The burned area reached 3322.72 ha, accounting for 68% of the total area burned in the Kavakdere fire (Figure 7). The mean fuel consumption (FC) during this period was 0.38 kg m⁻² (minimum and maximum of 0.18 kg m⁻² and 0.94 kg m⁻², respectively) (Figure 8b). The calculated fire-line intensity (FI) reached up to 4779.59 kW m⁻¹, with a mean of 3173.17 kW m⁻¹ and a minimum of 1866.67 kW m⁻¹ (Figure 8d).

The Kavakdere fire then continued to spread laterally until 10:17 on 30 June, reaching its final perimeter at 23:56 on the same day. This period is considered the second burn period (Phases 5–9) (Figure 8b,d), lasting approximately 25 h. During this phase, the mean ROS was lower, at 0.41 km h⁻¹, with a minimum of 0.06 km h⁻¹ and a maximum of 0.81 km h⁻¹. The fire burned an additional 1542.97 ha, corresponding to 32% of the total burned area. The mean FC was 0.82 kg m⁻², with a minimum of 0.39 kg m⁻² and a maximum of 1.33 kg m⁻². The mean calculated FI was 2367.53 kW m⁻¹, with a minimum of 135.71 kW m⁻¹ and a maximum of 6004.50 kW m⁻¹ during this second burn period (Figure 8d).

3.2. Fire Radiative Energy

The Kuyucak fire exhibited a rapid escalation in fire radiative energy (FRE) output from its first detection at 09:58 on 29 June to a pronounced maximum at 13:08, reaching a peak FRE of 61.83 TJ per 30 min (peak half hour) (Figure 9a). Over the full observation window (09:58, 29 June to 18:58, 30 June; 33 h), the integrated energy release totaled 434.27 TJ (Figure 9b). FRE remained elevated through the early afternoon, with secondary bursts (e.g., 47.66 TJ at 12:30; 35.77 TJ at 13:30; 31.24 TJ at 14:30), followed by a pronounced decay toward evening and intermittent low-level activity overnight (Figure 9a). Day partitioning of FRE shows that 95.27% of the total energy was released on 29 June, with a modest 4.72% contribution on 30 June.

The Kavakdere fire was first detected at 11:28 on 29 June and similarly intensified to an early afternoon maximum at 15:30 (peak FRE 27.97 TJ per 30 min) (Figure 9a). The fire persisted through 02:58 on 1 July (39.5 h), with sustained high FRE between 15:00 and 17:00, before tapering into the evening and continuing with lower-amplitude pulses into the next day (Figure 9). The total FRE was 193.65 TJ, apportioning 80.72% on 29 June, 19.13% on 30 June, and a negligible 0.09% on 1 July. The Kuyucak fire was the more energetic, releasing ~2.25× the total FRE of the Kavakdere fire (434.27 TJ vs. 193.65 TJ) and attaining a ~121% higher peak 30 min FRE (Figure 9b).

3.3. Synoptic and Mesoscale Fire Weather Conditions

An analysis of the large-scale atmospheric dynamics based on ERA5 data, both prior to and during the wildfires, illustrates that synoptic-scale fire weather conditions were characterized by the gradual establishment of a notable dipole of positive (>2.5 sigma) and negative (<−1 sigma) geopotential height anomalies between Central Europe/the northwestern Balkans and the Eastern Mediterranean, respectively (Figure 10).

More specifically, two days prior to the ignitions (Figure 10a,b), a mid-tropospheric ridge gradually built up over Central Europe and the northwestern Balkans that reached its highest positive anomalies on the day of the ignitions (Figure 10c). Further downstream, over the Eastern Mediterranean, because of the amplified atmospheric circulation and a wavy jet stream over eastern Europe (Figure 11), this ridge was accompanied by a trough that reached its lowest negative anomalies on the day of the ignitions (Figure 10c). One day after the ignitions, the mid-tropospheric dipole weakened, while the mid-tropospheric ridge shifted eastwards (Figure 10d).

This synoptic-scale pattern was associated with an enhanced surface pressure gradient between the Balkans and the Eastern Mediterranean and the development of strong northerly–northeasterly winds across Western Türkiye, the Aegean Sea, and eastern parts of Greece, a well-known atmospheric pattern that is associated with adverse fire weather conditions and elevated fire danger across the region (Figure 10c and Figure 11c) [33,34].

To investigate the mesoscale dynamics associated with the most active period (i.e., the interval of largest ROS on 29 June) of the most extreme wildfire (i.e., the Kuyucak fire), we analyzed WRF cross-sections along the Kuyucak fire axis at 12:00, 15:00, 18:00, and 21:00 UTC on 29 June (Figure 12). The sequence reveals a deepening, very dry boundary layer beneath an isentropic cap (i.e., stable layer) and persistent low-level northeasterly winds. From midday to evening, isentropes descend, and the dry layer expands upward (Figure 12), consistent with subsidence associated with negative vorticity advection over the area (Figure S1). At 18:00 UTC (Figure 12c), winds at about 300–500 m above the ground reached maximum intensity over the area of the fire below the inversion layer, with speeds on the order of 90 km h⁻¹ (Figure 12 and Figure S2). By 21:00 UTC, humidity remains low between the surface and 750 hPa, and the capping inversion persists, with slightly weaker winds near the surface compared to the afternoon hours.

An analysis of observed surface fire weather conditions from a nearby weather station (Figure 1) shows that on 29 June, both fires ignited under adverse fire weather conditions characterized by elevated atmospheric aridity with VPD between 30 and 40 hPa, and strong northerly–northeasterly winds of 40–60 km h⁻¹, with gusts exceeding 80 km h⁻¹ (Figure 13).

Observed winds maximized at 19:00 UTC, in good agreement with the simulated WRF low-level jet (i.e., low-level wind maximum), peaking at 18:00 UTC (Figure 12c and Figure S2c). The co-occurrence of this low-level jet with enhanced downward motions at the same time (Figure S2c) indicates an increased potential for the vertical transport of higher-momentum air from the lower troposphere to the surface. Furthermore, during the ignition time, the estimated dead fuel moisture content (FMC) of fine dead fuels was at critical levels with FMC between 5 and 6%, which, under these critical fire weather conditions, was favorable for the development of long-range spotting and rapid rates of spread [5,35]. Fire weather conditions remained adverse until 21:00 UTC (10–12 h after ignition times), when winds weakened to 20 km h⁻¹, VPD decreased to 15 hPa, and fine dead fuel moisture slightly recovered to values between 9 and 10.5%. On 30 June, conditions were characterized by drier conditions compared to 29 June, with VPD exceeding 40 hPa and FMC of 5–6% during the afternoon hours, while winds were blowing from northerly directions at a speed of 15–35 km h⁻¹ (Figure 13).

3.4. Burn Severity

The burned area of the Kuyucak fire was 4701.9 hectares, with 19% classified as low severity, 29% as moderate–low, 47% as moderate–high, and 5% as high burn severity. Kuyucak fire exhibited a contiguous core of high and moderate–high severity, with the most severe areas concentrated centrally and extending to the southwest (Figure 14), which was the head fire direction during the first progression period (Figure 7). Severity decreased radially toward the margins, transitioning to moderate–low and low classes near discontinuous fuels. This spatial pattern indicated sustained, high-severity burning across continuous conifer stands with high fuel loads and likely wind-driven head fire behavior.

Kavakdere fire covered around 4865.8 hectares, of which 31% was classified as low severity, 45% as moderate–low, 23% as moderate–high, and 1% as high burn severity. Kavakdere fire showed a more heterogeneous severity mosaic, with high-severity patches interspersed among moderate–high and moderate–low areas rather than forming a single contiguous core (Figure 14). High-severity zones are localized and fragmented, suggesting variable fuel continuity or local topographic influences. Lower-severity classes were more frequent along the periphery and near landscape features that interrupted fire spread, producing a patchy distribution of burn severity. The Kuyucak fire was particularly severe, with nearly 53% of the burned area classified as moderate to high severity, compared to 24% for the Kavakdere fire.

3.5. The Relations Between Weather, Topography, and Associated Fire Behavior Parameters

The correlation analysis showed that wind variables were important drivers of fire behavior during the twin wildfire events. Wind speed (WS) was significantly correlated (Figure 15) with ROS (r = 0.724, p < 0.01) (Figure 16a), FI (r = 0.526, p < 0.05), fire radiative energy (FRE) (r = 0.640, p < 0.01), and growth rate (r = 0.597, p < 0.01), while wind gust (WG) exhibited even higher correlations with ROS (r = 0.760, p < 0.01), FI (r = 0.627, p < 0.01), FRE (r = 0.646, p < 0.01), and fire growth rate (FGR) (r = 0.656, p < 0.01).

FRE was significantly correlated with ROS (r = 0.821, p < 0.01) (Figure 16c), FI (r = 0.699, p < 0.05), and FGR (r = 0.953, p < 0.01) (Figure 16e). FGR was significantly correlated with ROS (r = 0.823, p < 0.01) (Figure 16b). Burn severity (BS) was significantly correlated with FC (r = 0.584, p < 0.05), FI (r = 0.567, p < 0.05), and had a significant moderate correlation with FGR (r = 0.469, p < 0.05) and FRE (r = 0.484, p < 0.05).

Slope (SLP) was significantly correlated with, ROS (r = 0.713, p < 0.01), FC (r = 0.680, p < 0.01), FI (r = 0.828, p < 0.05) (Figure 16c), FGR (r = 0.641, p < 0.01), FRE (r = 0.805, p < 0.01), and BS (r = 0.755, p < 0.01 (Figure 16f). Overall, the results highlight the dominant role of wind and topography in driving extreme fire behavior, while fuel characteristics and topography play a key role in determining burn severity patterns (Figure 15 and Figure 16).

4. Discussion

4.1. Meteorological Drivers of Extreme Fire Behavior

The twin wildfires in Seferihisar, İzmir, occurred under extreme fire weather conditions with low fuel moisture content (<6%) and strong northeasterly winds exceeding 40 km h⁻¹ with gusts above 70 km h⁻¹. These surface conditions were associated with an anomalous mid-tropospheric geopotential height dipole between Central Europe and the Eastern Mediterranean. Such synoptic weather patterns and wind conditions are a well-recognized driver of extreme fire spread in Mediterranean-type ecosystems, particularly when coupled with atmospheric aridity, prolonged drought, and heat stress on vegetation [1,4,5,36]. Comparable wind-driven fire behavior has been documented in the 2007 Peloponnese fires in Greece [37], the 2018 Attica fire [38], and other fires in Europe [39,40,41], all of which exhibited rapid forward spread and extreme fire-line intensity within the first hours of progression. These examples underscore the recurring hazard posed by seasonal (e.g., Etesian winds) and katabatic/foehn-like winds, which can generate highly erratic and uncontrollable fire behavior within minutes, particularly when these winds are combined with hot and dry conditions.

Our analysis confirms the dominant role of wind speed and gusts in driving fire behavior, aligning with studies from Portugal, Spain, and Greece that have demonstrated how high wind events drastically shorten suppression windows and promote extreme ROS [7]. Strong wind not only accelerates head fire propagation but also facilitates spotting, resulting in rapid perimeter expansion and increased suppression complexity [8,42,43,44]. Under such conditions, topographic effects and moderate variations in fuel moisture become secondary to wind velocity in determining fire growth rates, and suppression efforts are often rendered ineffective during peak wind activity.

4.2. Differences in Fire Spread and Intensity Between Kuyucak and Kavakdere Fires

The Kuyucak fire exhibited a substantially higher mean rate of spread (5.00 km h⁻¹ in the first burn period) and fire-line intensity (37,789 kW m⁻¹) than the Kavakdere fire. This discrepancy can be attributed to differences in wind alignment, fuel continuity/load (Figure 3) and topography (Figure 2). The dominance of Pinus brutia in the Kuyucak fire’s progression area is significant, as this species’ high resin content, dense crown, and ladder fuel structure are known to favor intense crown fire behavior once thresholds of wind speed and dead FMC are exceeded [16,45,46]. In contrast, the Kavakdere fire’s slower initial spread and lower fuel consumption may reflect heterogeneity in vegetation structure (Figure 3), a greater proportion of lower-flammability fuel types, and possible differences in fire suppression effectiveness.

The Kuyucak fire’s higher rate of spread and fire intensity can also be attributed to its steeper and more variable terrain (Figure 2). Slopes up to 101% and a mean of 23% created conditions that favor rapid upslope fire spread by tilting flames toward unburned fuels, enhancing preheating and ignition. Steep slopes also increase convective heat transfer and flame attachment, leading to greater energy release and fuel consumption. In contrast, the Kavakdere fire’s more moderate and uniform slopes (up to 58%, mean 16%) limited these effects, resulting in less extreme fire behavior. These findings align with previous studies showing that slope significantly amplifies fire spread and intensity through topography-driven heat transfer and spotting mechanisms [47,48,49,50].

4.3. Temporal Dynamics of Fire Radiative Energy

MTG/FCI FRP-Pixel analysis revealed that both fires showed a rapid ramp-up to an early-afternoon maximum (time-to-peak ≈ 3.17 h from first detection), followed by a multi-peak pattern and evening decay (Figure 9). The Kuyucak event released ≈ 2.25 × more total FRE than Kavakdere, driven by longer sustained high-FRP intervals and stronger bursts around the peak window. The 10 min MTG/FCI cadence is key to resolving such sub-hourly energy pulses of FRE [26], improving near-real-time wildfire assessment and emissions inference.

Across the twin wildfire events, FRE increased with wind speed and wind gusts and with fire behavior variables (ROS, FI, FGR) (Figure 15). Physically, higher winds tilt flames and enhance convective/radiative pre-heating of unburned fuels, accelerating spread and elevating intensity, thereby increasing FRE. Short-lived gusts amplify these effects and produce brief surges in radiative output. Similarly, slope raises FRE via upslope pre-heating and increased flame attachment [51], similar to the wind effect and documented in classic and modern modeling and experiments [49,52]. Because FRE is proportional to combustion rate and fuel consumed [53], stronger spread, intensity, and growth translate directly into larger energy release, explaining the observed correlations.

4.4. Burn Severity Patterns and Vegetation Impacts

Despite burning under the same synoptic wind conditions, the Kuyucak fire recorded a substantially higher proportion of moderate-to-high severity area (53%) compared to the Kavakdere fire (24%) (Figure 14). This discrepancy reflects differences in fuel type, vertical and horizontal continuity, and canopy structure. In Mediterranean pine forests, attributes such as canopy bulk density, ladder fuel abundance, and lack of vertical discontinuity are strongly correlated with crown ignition probability and the persistence of high-intensity fire fronts [11,31].

In the Kuyucak fire, high-severity patches were concentrated in closed-canopy stands (Figure 3). This pattern mirrors broader findings that unmanaged, dense pine forests in the Mediterranean Basin are more susceptible to stand-replacing crown fires [11,54,55]. In contrast, the Kavakdere fire burned through more fragmented and partially managed stands, grasslands, and croplands, producing a more heterogeneous burn mosaic with larger areas of low severity and unburned patches (Figure 14).

Positive correlations between severity, FC, FI, and slope indicate that both pre-fire fuel loading and in-fire combustion dynamics were major determinants of post-fire ecological impact (Figure 15). dNBR has proven to be a robust indicator of burn severity across diverse fuel types, as also reported in global assessments [32,56,57]. High-severity patches can have lasting ecological consequences, including delayed regeneration and soil erosion risk [58], making these metrics valuable for post-fire recovery planning.

4.5. Operational Challenges of Simultaneous Large Wildfires and Implications for Future Wildfire Management in the Eastern Mediterranean

The near-simultaneous ignition of two large wildfires presented significant operational challenges, including resource allocation trade-offs, reduced suppression capacity, and potential atmospheric interactions between the two fire plumes that can alter fire behavior. The concurrent Kuyucak and Kavakdere fires imposed significant operational challenges on the General Directorate of Forestry (GDF), necessitating the division of suppression resources and complicating strategic decision-making [9,10].

Multi-fire outbreaks are known to strain firefighting systems, particularly under extreme weather conditions [5], and the concurrence of large, wind-driven fires can force difficult decisions about where to commit suppression teams and equipment [59]. These constraints reduce overall suppression effectiveness and increase the likelihood of fire growth and spotting, while plume interactions may further complicate fire behavior forecasting and suppression planning. During the first burn period of the Kuyucak fire (Figure 7 and Figure 8c), fire-line intensity reached up to 52,495.02 kW m⁻¹, classifying it as an extreme wildfire [59], in which fire behavior overwhelmed suppression capacity. Therefore, the most appropriate strategy for dealing with these fires is to focus on prevention and to implement measures that safeguard critical locations and resources. In this context, measures to protect life and property should be a priority [5].

In the Eastern Mediterranean, projections of increased drought frequency, stronger synoptic wind events, and prolonged heatwaves indicate that such compound events are likely to become more common, elevating the risk of clustered large fires that challenge conventional response models [3,4,36,60]. Addressing these risks requires adaptive operational strategies that combine the pre-positioning of firefighting resources, coordinated multi-incident command systems, and real-time tactical fire analysis that integrates all fire-behavior drivers. High-resolution, real-time fire-spread modeling and dynamic risk mapping are essential to inform rapid allocation decisions during multi-fire incidents and to reduce the trade-offs inherent in simultaneous large-fire suppression.

Complementary management measures are also needed to reduce landscape flammability and strengthen resilience to extreme events. Beyond the specific case study, this work provides a transferable methodology for robust post-fire analysis that relies on tangible fire behavior descriptors (e.g., ROS, FC, FI) and can therefore be applied to wildfires in diverse regions and fuel types. By enabling the consistent reconstruction of what occurred and why, the approach can serve as an operationally relevant template for agencies seeking to evaluate extreme-fire episodes and compare outcomes across events.

Importantly, the knowledge derived from such post-fire analyses can be translated into improved decision-making over the long term by informing training, operational protocols, and strategic planning. For example, insights on the conditions and fuel configurations associated with high burn severity can guide targeted fuel reduction in high-risk pine stands and reinforce the role of pre-fire treatments—such as pruning and prescribed burning—in moderating fire effects even under extreme meteorological conditions. Similarly, evidence on ignition patterns and initial growth dynamics can support strengthened prevention programs and enhanced early-detection capability to reduce the likelihood and potential severity of large fires. Finally, lessons learned regarding resource constraints and simultaneous incidents can be used to refine cross-institutional resource sharing agreements and justify regular multi-agency exercises, improving preparedness for multi-incident scenarios. Collectively, these operational, landscape, and institutional actions illustrate how linking scientific fire reconstruction with management practice can inform adaptive fire-management strategies that enhance preparedness, resilience, and the protection of communities under future extreme wildfire conditions.

4.6. Limitations and Future Improvements in Fire Reconstruction

Reconstructing fire progression inevitably involves uncertainties arising from both spatial and temporal limitations in the data sources used [61] and the progression delineation method applied. Different sensors provide different levels of accuracy; for example, Sentinel-2 multispectral imagery provides 10–20 m spatial resolution, whereas VIIRS active-fire detections operate at a coarser 375 m resolution [24]. As a result, fire progression maps should be interpreted as approximate rather than exact representations of the fire front [61].

Temporal uncertainties in fire progression polygons represent a major limitation: satellite detections (e.g., Sentinel-2, VIIRS, MODIS) capture conditions only at specific acquisition times, and fires may have already stalled or changed direction by the time an image is collected. Fires can also ignite and extinguish between satellite overpasses, or remain undetected under cloud or smoke [62]. These temporal gaps, together with spatial uncertainties in manually delineated fire perimeters, can propagate into derived fire behavior metrics such as ROS, FI, and FGR [22,61]. Even small discrepancies in polygon boundaries may alter calculated spread distances, while variations in spatial resolution can smooth or exaggerate extreme fire-run behavior.

To reduce these uncertainties, a convergence of evidence from multiple sensors is essential. When independent systems such as Sentinel-2 and VIIRS provide consistent information, confidence in reconstructed fire-behavior metrics (e.g., ROS, FI, FGR) is strengthened, even if uncertainties cannot be quantified precisely [61]. High-frequency geostationary platforms like the Meteosat Third Generation Flexible Combined Imager (MTG-FCI), with ~10 min refresh intervals, offer the potential to decrease temporal uncertainty by capturing fire dynamics more continuously [63]. In this study, MTG-FCI was not directly employed, but it is highlighted as a promising resource for future applications where improved temporal fidelity is required.

Future improvements should emphasize the expansion of automated fire-perimeter extraction approaches, such as unsupervised edge-detection algorithms applied to aerial thermal infrared imaging [64]. Additional gains may come from improved multi-sensor data fusion and the incorporation of high-resolution, high-frequency observations from UAV platforms and also high-frequency observations from MTG [63] which have demonstrated high value near-real-time monitoring [65].

5. Conclusions

This study provides a detailed reconstruction and analysis of the twin wildfires that occurred in Seferihisar, İzmir, on 29–30 June 2025, offering valuable insights into the drivers, dynamics, and impacts of extreme fire events in Mediterranean ecosystems. The findings highlight the dominant influence of atmospheric dynamics, particularly strong winds and critically low fuel moisture, on fire spread and intensity. Wind speed and gust emerged as significant predictors of rapid rate of spread, high fire-line intensity, and elevated fire radiative energy, underscoring the need to integrate advanced fire weather forecasting and monitoring into operational fire management for both preparedness and real-time tactical fire analysis and decision-making.

The Kuyucak and Kavakdere fires were influenced by broadly comparable synoptic-scale weather patterns; however, they exhibited distinct fire behavior and severity due to local-scale differences in fuel structure and slope. Pinus brutia-dominated stands in the Kuyucak fire’s path contributed to extreme crown fire behavior, rapid early expansion, and higher proportions of moderate-to-high severity burn. In contrast, the Kavakdere fire progressed more slowly and exhibited lower severity, reflecting the complex interplay of vegetation and topography.

The near-simultaneous ignition of two large wildfires created operational challenges by dividing suppression resources and complicating strategic decision-making. As climate change is expected to increase the frequency of extreme fire weather days and the likelihood of concurrent large fires in the Mediterranean Basin, proactive preparedness measures, including dynamic resource allocation, enhanced early warning systems, and fire weather monitoring, as well as integrated multi-fire management protocols, will be critical.

Overall, this case study underscores the value of combining multi-source remote sensing, ground observations, atmospheric data, and details of fire behavior characteristics to better understand extreme wildfire dynamics. Such integrative approaches can inform both scientific understanding and practical fire management in the face of an increasingly challenging fire regime.