1 Introduction

For centuries, forest fires, as natural disturbances, have posed a potential hazard all over the world (Lin and Rinaldi 2009; Zumbrunnen et al. 2011). Specifically, in recent decades, countries under a Mediterranean-type climate have become extremely fire-prone due to climate change and anthropogenic effects that have caused changes in the fire regime (Nunes 2012). In fact, this has resulted from the complex interactions of certain factors including ignition agents, flammable material (fuel), topography, and the weather (Flannigan et al. 2000). Among these factors, the weather (i.e., temperature, relative humidity, wind velocity, and the amount and frequency of precipitation) is the most important parameter influencing forest fires (Carvalho et al. 2011; Zumbrunnen et al. 2011). Despite increased efforts for fire prevention and suppression over the past decades, because of global warming, a significant increase in the number of fires as well as in the size of the areas affected has been observed worldwide (Flannigan et al. 2005; Moriondo et al. 2006). Globally, burned forested areas vary considerably from year to year (van Lierop et al. 2015). Even though only a small number of forest fires are defined as large- or mega-sized, they are responsible for most of the burned areas (Dimitrakopoulos and Mitsopoulos 2006). Globally, their increasing frequency and intensity of occurrence has turned these large forest fires into catastrophic events that are responsible for major impacts on forests and residential areas (Boer et al. 2020; Silva Junior et al. 2022).

Because of the latest advances in both techniques and related technologies such as sensors, remote sensing has become an essential observational and analytical tool that is fast, replicable, objective, efficient, scalable, and cost-effective (Li et al. 2019). Therefore, it is commonly applied for the operational monitoring of forest fires and assessment of the affected areas at varying spatiotemporal scales (Quintano et al. 2011). Google Earth Engine (GEE) is a cloud-computing platform that provides global-scale analysis capabilities and has several satellite data catalogues (such as Landsat, Sentinel-2, and MODIS) at different spatial resolutions. Recent publications have reported the widespread use of GEE, essentially as sequels to the first significant studies using it (Hansen et al. 2013). The GEE platform offers notable advantages such as being free and user-friendly and provides storing and mapping data, as well as high-speed parallel processing without downloading data (Zhao et al. 2021).

The current scientific literature shows that remote sensing together with geographical information systems (GIS) has been used for active fire detection (Higa et al. 2022; Rostami et al. 2022; Seydi et al. 2022), delineating/mapping of burned areas (Zhang et al. 2021; Pérez et al. 2022), mapping canopy damage and detecting tree mortality (Hamilton et al. 2021; Zhai et al. 2022), fire monitoring (Feng et al. 2022; Fischer et al. 2022), fuel (type/load) mapping (Heisig et al. 2022; Wilson et al. 2022), fire severity analysis/mapping (Luca et al. 2021; Yılmaz et al. 2022), fire management (Myroniuk et al. 2021; Nascente et al. 2022), forest-fire susceptibility mapping (Trucchia et al. 2022; Saim and Aly 2022), post-fire vegetation regeneration (Pérez-Cabello et al. 2021; Smith-Ramírez et al. 2022), and determining shortest and safest routes to forest fires (Akay et al. 2012). A significant number of recent studies in the literature have reported on using remote sensing to determine fire severity (Collins et al. 2020). Fire severity refers to the above- and belowground organic matter consumption from burning used to study ecosystem responses to fire (Keeley 2009). The accurate and rapid mapping of fire severity is essential for fire research and fire management (Eidenshink et al. 2007) since it is one of the main factors influencing ecosystem responses (Leverkus et al. 2018) while having spatial heterogeneity, depending on weather, terrain, and fuel properties (Collins et al. 2014a, b). Currently, different remote sensing indices have become available for determining fire severity, including the normalized difference vegetation index, enhanced vegetation index, integrated forest index, and the differenced Normalized Burn Ratio (dNBR) (Chen et al. 2011; Cansler and McKenzie 2012). Among these, dNBR has been verified as an accurate index of fire severity in different types of forests including boreal, temperate conifer, temperate deciduous, Afrotemperate, and Mediterranean forest types (Epting et al. 2005; French et al. 2008; Chen et al. 2011; Giddey et al. 2022).

For most European countries in the Mediterranean region, forest fires are the main destructive disturbance agent (San-Miguel-Ayanz and Camia 2010; Keeley et al. 2012). In Türkiye, as a country with some regions under the Mediterranean-type climate, forest fires are a major problem that shapes and converts forested areas. According to the statistics reported by the General Directorate of Forestry (GDF) in 2020, forests cover nearly 22.9 million hectares (ha) in Türkiye (GDF 2020). About 55% of forested areas (~ 12.5 million ha) are situated in 1st- and 2nd-degree fire-sensitive regions in Türkiye. These fires are mostly observed during the long fire season covering the period between 1 May and 31 October. The GDF records show that, in total, 115,000 forest fires burned approximately 1.85 million ha of forested areas over the 85-year period between the years 1937 and 2021. Accordingly, on average, 1380 fires burned 21,800 ha of forested areas per year, resulting in an average of 16 ha burned in each fire event. More than 95% of all fire events are induced by human activities, and over 90% of all burned areas have resulted from these fires (GDF 2022). Land use and the socio-economic situations of communities in urban areas located mainly in fire-sensitive regions are determinative in the occurrence of such fires (Bilgili et al. 2021).

The increase in both the number of forest fires and the corresponding affected areas is of great importance for some countries in terms of their forest-fire history. The year 2021 can be considered a milestone with regard to the forest-fire history of Türkiye. One of the most important negative effects of forest fires on a social scale is the loss of life. In large-scale forest fires, loss of life is generally inevitable, especially in intermingled Wildland Urban Interface (WUI) areas. Unfortunately, there were also loss of lives in the forest fires of 2021. During the extreme fire season of 2021, 11 employees of the GDF lost their lives in forest firefighting efforts (GDF 2022). The highest loss of life occurred on August 14, 2021, with 8 people perishing as a result of the crash of an amphibious firefighting aircraft (Gabbert 2021). Besides, the total financial loss resulting from the 2021 forest fires was approximately 1.32 billion Turkish liras (TL), and the total expenses incurred due to forest fires in the same year were approximately 2.97 billion TL (GDF 2022). The reasons made the 2021 summer forest fires in Türkiye unprecedented, destructive, and uncontrollable can be listed as: (i) the existence of fuel with very low moisture content because of high air temperature during the fire events, (ii) the existence of winds of varying speed and direction, (iii) the existence of intermingled wildland and urban interface areas, (iv) the organizational deficiencies in fire-fighting operations for successive massive forest fires, resulting in insufficient and unsuccessful fire interventions, and (v) the effects of climate change. This situation led to the need for researchers/experts to carry out comprehensive investigations of the forest fires in the summer of the year 2021 in Türkiye. By using remote sensing and GIS techniques, the present study aimed to document and investigate the large (affecting > 100 ha), unprecedented 2021 forest fires in Türkiye. Consequently, a comprehensive fire database was created that included information on the burned areas, fire severity, fuel types determined from forest-stand types, and topographical properties such as slope, elevation, and aspect (in eight directions). For obtaining fire-severity information, the remote sensing data from Sentinel-2 satellite images were used and analyzed in the GEE platform to calculate dNBR values. To create the fire database, three GIS-integrated Python scripts were developed to map the burned areas from the fire-severity map and to classify fuel types from forest-stand type information together with topographic properties derived from DEM data. In brief, the main objectives were: (i) to propose a remote sensing and GIS-based approach to create automatically a fire database to document the large fire events in 2021 for future investigations, (ii) to investigate the total area burned and to determine the residential and agricultural areas affected, (iii) to investigate how forested areas were affected with regard to fuel types and topographical properties, and (iv) to investigate the impacts of the forest fires depending on the spatial distribution of the fire severity.

2 Material and methods

2.1 Study area

The study area covered the three regions of Türkiye that had suffered from forest fires (Fig. 1). These large (affecting > 100 ha) fires were mostly located in the Mediterranean, Aegean, and the Eastern Anatolian regions. These large fire events affected 13 cities (Adana, Antakya, Antalya, Aydın, Bingöl, Burdur, Denizli, Isparta, İzmir, Kahramanmaraş, Mersin, Muğla, and Tunceli), and eight regional forest directorates (RDFs) (Adana RDF, Antalya RDF, Elazığ RDF, Isparta RDF, İzmir RDF, Kahramanmaraş RDF, Mersin RDF, and Muğla RDF). Türkiye’s geographic regions have different climates mostly due to topographic variability caused by its current landform, which was created by the convergence of Tethys Sea sediments located among the African, Eurasian, and Arabian plates (Bozkurt 2001). Türkiye, covering 769,471 km2 excluding lakes, has a mean altitude of 1141 m above sea level with a mean slope angle of 10° (Elibüyük and Yılmaz 2010). According to Turkish State Meteorological Service (TSMS) records, the mean temperature and precipitation in the last 30 years were 13.9 °C and 573.4 mm, respectively (TSMS 2022). In the Mediterranean and Aegean regions, summer seasons are hot and moderately dry, whereas winters are mild and rainy. However, the Eastern Anatolian region, with elevations exceeding 2500 m, has a longer winter, and summers are hot and dry. In Türkiye, where most forest lands are located in Black Sea, Mediterranean, and Aegean geographic regions, the year 2021 was the hottest and driest one with 14.9 °C annual mean temperature and 524.8 mm mean precipitation (TSMS 2022). This situation made these forest fires in the 2021 summer as unprecedented, destructive, and uncontrollable for Türkiye. Both the Mediterranean and Aegean regions, where the most fire-prone Turkish red pine and Anatolian black pine are two dominant tree species, have approximately 33% of forested areas, whereas the Eastern Anatolian region, where fire-prone oak species are dominant, has only 8% of forested areas in Türkiye (Boydak et al. 2006; GDF 2022). These two pine species also cover nearly 41% of forested areas in Türkiye (GDF 2022).

Fig. 1
figure 1

Location of large (affecting > 100 ha) forest fires in the summer of 2021 in Türkiye. Colors ranging from green to red symbolically represent fire severity within each fire event, increasing from unburned areas to those with high severity (Bottom)

Full size image

2.2 Data collection and processing

The methodology was divided into two parts: (1) remote sensing data analysis in the GEE platform, and (2) automatic GIS database creation for each fire using Python scripts. The data-processing workflow is given in Fig. 2.

Fig. 2
figure 2

Workflow of data processing and database creation

Full size image

Sentinel-2 satellite images were filtered from ImageCollection using the band information, location, cloud coverage, and timestamp metadata properties. The Near Infra-red (NIR) and Short-wave Infra-red (SWIR) bands were filtered because they were used for the Normalized Burn Ratio (NBR) indices, which are useful to distinguish the burned areas and are calculated using the following formula:

$${\text{NBR}}=({\text{NIR}}-{\text{SWIR}})/({\text{NIR}}+{\text{SWIR}}$$
(1)

The NBR indices, with values varying between − 1 and 1, were calculated for both pre-, and post-fire imageries, and were used to calculate the differenced Normalized Burn Ratio (dNBR), which was used to map burn severity classes via the following formula:

$${\text{dNBR}}={{\text{NBR}}}_{{\text{pre}}-{\text{fire}}}-{{\text{NBR}}}_{{\text{post}}-{\text{fire}}}$$
(2)

For this reason, images were filtered separately for each fire event according to the start and end dates obtained from GDF records. These timespans corresponded to 1 month before the starting date (i.e., pre-fire) and 1 month after the ending date of a fire (i.e., post-fire). Cloud coverage percentage was defined as 10% in the filtering images from the ImageCollection. Geometry was imported in the “GeoJSON” format representing a rectangle-shaped polygon covering the extent of each fire. The extent of each fire was determined from post-fire Sentinel-2 images. Following their creation, dNBR data were exported together with the pre-fire and post-fire Sentinel-2 images in “.tif” format for use in further analysis in ArcGIS. All exported data used the “EPSG:4326—World Geodetic System 1984” coordinate system.

For further data analysis and to create a GIS database, three scripts working in ArcMap were developed: (1) the “Burned Area Extractor” for delineating burned areas from dNBR data, (2) the “Sieving Algorithm” for classifying fire severity by eliminating small groups of pixels, and (3) the “Database Creator” for create a GIS database. All scripts were created using the Python 2.7 programming language, which provided certain modules such as “arcpy”, “numpy”, and many others. The first script takes raw dNBR data in raster format and automatically extracts the burned areas as polygon features in ArcMap. This script incorporates both raster algebraic calculations and feature data editing steps with functions including re-scaling pixel values, filtering, and segmenting, as well as slicing an algorithm for conversion from raster format into a polygon and then editing it to achieve a final polygon representing the burned area. The second script takes both the dNBR and the burned area created by the first script as input. When the dNBR data were re-classified into several fire severity classes, some single or small groups of pixels representing other classes could be seen inside a severity class. We considered the classification scheme proposed by United States Geological Survey (USGS) to obtain severity classes. In order to get classes that were more regular, by scripting this algorithm, we aimed to eliminate these single pixels or small groups having fewer than a certain number of pixels (5 pixels for this study) (see Fig. 3). The script was developed as a part file because its use is optional.

Fig. 3
figure 3

Example showing severity classes obtained with and without sieving algorithm: A without sieving, B with sieving. The legend in the image denotes different severity levels of burning, with 3 representing unburned areas, 4 for low severity, 5 for moderate-low severity, 6 for moderate-high severity, and 7 indicating high severity

Full size image

The third script takes four inputs: the extracted burned area, fire severity map, forest-stand map, and digital elevation model (DEM). Two of the inputs (burned area and forest severity map) were created with the two previous scripts. The last two inputs were obtained from GDF. The DEM was obtained from GDF at 15-m spatial resolution and consisted of raster data used to extract information on topographic features such as slope, aspect, and elevation for each fire event. The forest-stand map consisted of polygon feature data. These forest-stand maps have been created by the GDF in Türkiye for responsible Forest Enterprise operations, and their attribute table (or GIS database) includes certain detailed information such as the names of directorates, enterprises, section numbers, forest-stand types, forest functions, residential and agricultural areas etc. However, there is no column providing the information needed by fire researchers about fuel types. In the forest-stand maps used in Türkiye, stand types are stored in a database using codes (e.g., Çzab2, ÇzMbc3, BM, İs, Z, etc., in original form) created with certain guidelines determined by the GDF. In fact, these codes occupy one column in the database and include all information about stand types: (i) forested or non-forest area; (ii) if forested, tree species; (iii) pure or mixed forest; (iv) if mixed forest, the dominant tree species; (v) degraded or not; (vi) its development stages; (vii) its canopy closure. All this information preserved in the one code allowed us to determine fuel types. Therefore, in the first module of the third script, the aim was to classify forest-stand types as fuel types and add this information into the database as a new column. In the present study, fuel types were determined in three main categories: (I) according to tree species and their composition (for a forest), (II) according to stand-development stage, and (III) according to canopy closure. In total, 21 fuel types were defined according to tree species and their composition (Table 1). Five fuel types were defined depending on both the stand development stage and canopy closure of each forest stand, (Table 2). In Turkish forestry, forest stands are categorized into five stand-development stages, depending on average diameter at breast height (DBH): “a”—juvenile density stage (< 8 cm), “b”—sapling/mast and pole stage (8–19.9 cm), “c”—thin-trunk stage (20–35.9 cm), “d”—medium-trunk stage (36–51.9 cm), and “e”—mature/thick-trunk stage (> 52 cm). Forest stands are also categorized in five canopy closure classes depending on the percentage of ground covered by tree crowns: “0 (degraded)”—0–10%, “1 (low)”—11–40%, “2 (medium)”—41–70%, and “3 (normal)”—> 70%. As previously stated, the forest stand codes, indeed, include all this information. For clarity, for the simple forest stand-type code “Çzd2”, the last character “2” represents the canopy closure class, the next-to-last character “d” represents the stand-development stage, and the remaining characters represent the tree species composing the forest stand. Some of this information included in the code is also given in separate columns in the of GDF GIS database. However, the first module of the script uses only these codes in the classification of fuel types, even though a complete database should include all this information in separate columns. Because some data were missing, the forest stand-type codes in the database provided by GDF were accepted. The second module of the script applies the DEM to extract the mean slope (°), the aspects (eight directions), and the mean elevation (m a.s.l.) for each fuel type, together with fire-severity information. In addition to existing information on forest stands, the output database included fire severity classes, fuel types (in three categories), and some topographic information such as the mean slope, mean elevation, and aspects.

Table 1 Fuel type classes according to tree species and their composition in the stands
Full size table
Table 2 Fuel types depending on stand-development stages and canopy closures
Full size table

3 Results

3.1 General results from the analysis of fire events

In Türkiye, a total of 30 forest fires, each affecting more than 100 ha, occurred in the summer of 2021. The general results present the size of the burned areas caused by fire events in terms of the regions of Türkiye, provinces, and Regional Forest Directorates (RFDs). In addition, the study determined the amount of forested and non-forest areas (especially residential and agricultural areas) affected by fire events. A list of the forest fires and some summary information about them are given as Supplementary Material (Table S1). The sizes of the burned areas were calculated from the output of the “Burned Area Extractor” algorithm. According to these results, 164,658 ha had been burned due to these large fires. These forest fires occurred in only three regions of Türkiye: 13 fires (43%) in the Mediterranean region, 13 fires (43%) in the Aegean region, and 4 fires (14%) in the Eastern Anatolian region. Most of the total burned area (59%) was located in the Mediterranean region, with only 3% in the Eastern Anatolian region. These 30 fires affected 13 provinces: Adana (4), Antakya (1), Antalya (3), Bingöl (1), Burdur (1), Denizli (2), Isparta (1), İzmir (1), Kahramanmaraş (1), Mersin (2), Muğla (9), and Tunceli (3). In terms of the burned areas, in Antalya, although three fire events occurred, 74,410 ha (i.e., 45% of the total burned area) were affected (Table S2). With regard to RFDs, the fire events affected eight RFDs: Adana RFD (4), Antalya RFD (3), Elazığ RFD (4), Isparta RFD (2), İzmir RFD (1), Kahramanmaraş RFD (2), Mersin RFD (2), and Muğla RFD (12). Most of the burned areas (83%) were located in two RDFs (Antalya and Muğla). The burned areas in RDFs are given in Table S3. The minimum and maximum size of the areas burned in these events was 150 ha in the HONZ fire and 58,798 ha in the MATA fire event, respectively. Seven fire events burned areas larger than 10,000 ha, and three of them were in the Mediterranean region, whereas the remaining fires were in the Aegean region. The affected areas for 12 fires were less than 500 ha, whereas the affected areas for 11 fires were between 500 and 10,000 ha. The mean of the areas burned due to fires was 5489 ha.

When the affected residential and agricultural areas were evaluated, 16 of the 30 forest fires had affected a total of 679 ha of residential areas. These affected residential areas corresponded to 0.4% of the total area burned by fire events. The residential area affected the most was 246 ha in the MATA fire event. Agricultural areas were affected in all fire events. In total, 22,601 ha of agricultural land was burned, which corresponds to 14% of the total burned area. The largest burned agricultural area (12,031 ha) also resulted from the MATA fire event.

3.2 Results on analysis of fire severity and fuel types

Fire-severity analyses of the forest fires in the summer of 2021 were conducted by applying dNBR indices using Sentinel-2 images. The Sieving Algorithm Python script was used. This algorithm classifies dNBR data into fire-severity classes by eliminating small groups of pixels that can be assessed as noise. The fire-severity distribution for each fire is given as Supplementary Material (Fig. S1). The percentage of area affected at different severity classes can be seen for each fire event. The largest area affected at high-severity was observed in the HONZ fire event (28%). In four fire events (KADR, SALB, HONZ, and GZMR), the High-Severity class covered more than 20% of the affected area. In eight fire events (CEYH, HASS, MATA, GENC, BUSG, KVKD, MRMS, and MILS), the High-Severity class covered 10–20% of the affected area. The highest percentage of unburned area was observed in the KOZN fire event (48%), whereas the lowest was in the HASS fire event (2%).

The developed “Database Creator” algorithm also provided the information on slope, elevation, and aspect for each fire event. The mean slope distribution for the fire severity classes of each fire event according to aspects is given as Suppelementary Material (Fig. S2). When the largest fire event (MATA) was evaluated, the mean slope values of the severity classes ranged from 2° to 22°. In the MATA fire event, a clear increase in mean slope values can be seen for each fire severity class at the N, NE, and NW aspects. According to this figure, the lowest maximum and mean slope values were seen in the KADR fire event, whereas the highest were in the GZPS and NZMY fires. The mean and maximum values of slope and elevation for the fire-severity classes of all fire events are shown in Fig. 4. In terms of mean slope values, the Low-Severity class was 18.5°, and the Mid-High, Mod-Low, and High-Severity classes were similar. In terms of elevation, the lowest maximum elevation value (below 2000 m a.s.l.) was observed in the High-Severity class. The highest maximum elevation was found in the Low-Severity class.

Fig. 4
figure 4

Mean and maximum slope and elevation values for fire severity classes of all fire events

Full size image

In the present study, 21 fuel types were determined (see Table 1), and their areal distribution for each fire event is given as Supplementary Material (Table S4) and in Fig. 5. In terms of fuel material types, the class “Non-forested areas (Non-FrA)”, including residential, agricultural, stony, and dune land-use types, covered 14.99% of the total area affected by all fire events. The largest “Non-FrA” area covered 13,244 ha in the MATA fire event. This was 3.4 times larger than the same class in the MRMS fire event (3849 ha). The class of the most fire-prone pine species (“Pure Turkish pine species”—Pr-Çz) covered 59.56% of the total area affected (99,516 ha), whereas the class “Turkish pine species-dominated other broadleaved species (Çz-d-O-BrS)” covered 4.71% (7873 ha). Except for six fires (CEYH, GENC, HONZ, HOZT, MUNZ, and NZMY), the fuel-type class “Pr-Çz” was present in the affected areas of the remaining 24 fire events (80% of all events). This fuel-type class covered more than 50% of the affected areas for 19 fires. In the fire event HASS, 99.43% of the affected area was covered by fuel type “Pr-Çz”. However, the largest area of fuel type “Pr-Çz” (37,492 ha) was burned in the MATA fire event. This was 3.5 times larger than the second largest area for same fuel type burned in the MRMS fire event. Another important fire-prone pine species, that of “Pure Black pine species (Pr-Çk)”, covered 7.67% (12,811 ha) of all burned areas. The fuel type “Pr-Çk” was effective in seven fire events, covering the largest area percentage (85.48%) in the KRCS fire. However, the largest affected area was found in the KVKD fire event (7333 ha). The “Other pure coniferous species (O-Pr-CnS)” fuel type, including juniper, fir, Aleppo pine, stone pine, cedar, and cypress, covered 0.6% of all burned areas and its largest affected area was found in the MATA fire event (569 ha). The fuel-type classes “Pure broadleaved species (Pr-BrS)” (oak, hornbeam, beech, olive, poplar, bay, white birch, and sweetgum) and “Mixed broadleaved species (Mxd-BrS)” covered 3.67% and 3.01%, respectively. The fuel type “Pr-Brs” covered the largest area percentage in the NZMY fire event (93.34%), but its largest area was burned in the GENC fire event (2168) ha. The “Mxd-BrS” covered the largest area percentage in the CEYH fire event (97.44%), whereas its largest area was burned in the MILS fire event (2039 ha). All remaining classes, except “Turkish pine species-dominated other coniferous species (Çz-d-O-CnS)” and “Open forested areas (Open-FrA)”, covered less than 1% of the total area affected by fire events. In the CEYH event, 97% of the burned area consisted of maquis species.

Fig. 5
figure 5

Percentage of burned areas for fuel types in all fire events (see Table 1 for the fuel type codes)

Full size image

Fuel types were also evaluated by considering both forest-stand development stages and canopy closure (Fig. 6). With regard to forest-stand development stages, the largest area percentage burned was the Mature class (26.48%), with the fuel-type classes Very Young and Young corresponding to 25.57% of all burned areas. The Old forests covered 9.64% of all burned areas. In terms of canopy closure, the fuel-type class NoClosure covered the largest percentage of all burned areas (27.17%). Whereas the Medium and Normal canopy closure classes covered 46.78% of all burned areas, the Degraded and Low canopy closures covered 11.01%.

Fig. 6
figure 6

Percentage of burned areas for fuel types in terms of stand-development stages (left) and canopy closure (right)

Full size image

4 Discussion

In the summer of 2021, Türkiye suffered from fire events unprecedented in its forest-fire history. These fires occupied the public agenda from the time of their ignition and that of their suppression to nearly the end of that year. Hence, many questions have been raised and are in need of answers: How great an area was affected? How much forested area was burned? Were residential or agricultural areas affected? Why did such a terrible situation emerge?—and so on. Although many different factors such as land-use and land-cover trends that reconfigure the landscape (Modugno et al. 2016) play a role in the increase in forest fires, both in number and frequency, it is clear that climate change has been the most important factor (Flannigan et al. 2000; Carvalho et al. 2011; Zumbrunnen et al. 2011). Acar and Gönençgil (2023) conducted a comprehensive examination of the atmospheric conditions surrounding the 2021 fire events in Türkiye’s Muğla and Antalya provinces. Their study delved into the heatwave and wind characteristics, identifying them as pivotal factors influencing the intensity and spread of the fires. They observed that the heatwave, characterized by consecutive days of high temperatures, preceded the outbreak of the forest fire in Manavgat on July 28, 2021. Notably, the heatwave engulfed the entire southern region of Türkiye, affecting multiple weather stations. The researchers emphasized the critical role of wind parameters, including speed, direction, and frequency, in shaping the climatological environment during the heatwave period. Specifically, they noted that the escalation of wind speed, combined with the occurrence of foehn wind, exacerbated the fire situation, making firefighting efforts challenging. Acar and Gönençgil (2023) underscores the intricate relationship between heatwave dynamics and wind patterns in predisposing regions to forest fires, highlighting the importance of understanding these factors for effective fire management and mitigation strategies in vulnerable areas such as Muğla and Antalya provinces.

With regard to the number and size of the areas burned by forest fires, a trend toward escalation has been observed in Türkiye, according to the European Forest Fire Information System (EFFIS) Annual Country Statistics (https://effis.jrc.ec.europa.eu/apps/effis.statistics/estimates). Between 2010 and 2019, Türkiye ranked seventh after Croatia in the EFFIS records with regard to the annual mean burned areas. The 2021 fire events affected an area 22.5 times higher than the annual mean area. According to EFFIS Annual Country Statistics for Türkiye, in 2021, 291 fire events caused 202,361 ha to be burned. From this study it was determined that the burned areas totaled 164,658 ha in the 30 large fire events of 2021, corresponding to 81% of all the areas burned in whole country. As Im et al. (2006) stated, between 1999 and 2004, there were 25 fires in Türkiye that were larger than 200 ha. A similar situation was observed in our results. The number of 2021 forest fires larger than 200 ha was found as 26. The largest forest fire event in 2021 was the MATA forest fire, covering 58,798 ha. This was 3.7 times larger than the area burned (15,795 ha) in Manavgat / Antalya in 2008, which was at that time regarded as the largest fire event in Turkish forest-fire history (Bilgili et al. 2010). In addition, an area similar to that burned in the 2008 Manavgat forest fire was observed in the 2021 MILS, KVKD, and GUND fire events.

Some studies in the literature have evaluated and/or analyzed large fire events. For example, Siegert and Hoffmann (2000) evaluated the 1998 forest fires quantitatively using ERS-2 SAR images and NOAA-AVHRR data. Similarly, Kasischke et al. (2002) analyzed patterns of large fires (1990–1999) in the boreal forest region of Alaska. Stocks et al. (2002) developed a large-fire (> 200 ha) database for Canada covering fire events between 1959 and 1997 that included information on fire location, start date, final size, cause, and suppression action. In another example, Dimitrakopoulos et al. (2011) analyzed the large forest fires (> 1000 ha) occurring between 1990 and 2003 in Greece. According to Dimitrakopoulos et al. (2011), the statistical analysis considered the meteorological factors (air temperature, humidity, wind speed, etc.) and topographical (elevation, sleep, aspect, etc.), vegetation (density, duff cover), and suppression times as variables. Also, they used fire records documented by trained foresters immediately after every fire event. For Türkiye, Ertuğrul et al. (2019) investigated the large forest fires in the Çanakkale region in terms of the relationship between the areas burned and climate factors. They included forest fires that affected areas larger than 100 ha occurring in eight fire seasons from 1969 to 2007. Similarly, in this study, we covered the large fire events that affected areas larger than 100 ha, but only in 2021. We aimed to document and analyze the post-fire situation with regard to fire severity together with some topographic factors and fuel types. The documentation and analysis of the large forest fires were mainly carried out using data collected via field work, remote sensing, and GIS. The primary aim of this study was to create an automatic fire database using remote sensing and GIS that could be used to enable further statistical analysis. In this regard, the present study has aspects both similar to and different from related studies in the literature. The main difference is the consideration of the fuel types obtained from the GDF forest-stand types in Türkiye. In total, 21 fuel types were determined and recorded in the GIS database. The fuel types were classified depending on stand development stages and canopy closure, which are important forest parameters that should be considered, especially when evaluating the forest fire-related studies in Türkiye that used remote sensing and GIS (Ateşoğlu 2014; Akbulak et al. 2018; Çolak and Sunar 2018; Yavuz et al. 2018; Gülçin and Deniz 2020; Iban and Şekertekin 2022). It was noted that none had considered fuel types, topographic conditions, and fire severity comprehensively together in detail as we did. Recent studies in particular have focused mainly on remote sensing-based methodologies in the detection of burned areas and fire severity (Nasery and Kalkan 2020; Çavdaroğlu 2021; Tonbul et al. 2016, 2022).

The large forest fires were documented and analyzed by applying a two-phased method using remote sensing and GIS. In the first phase, a fire severity map was created for each event from Sentinel-2 satellite images in the GEE platform. Sentinel-2 has a high potential for forest fire research because of its high temporal and spatial resolution. Thus, it has been used more frequently for both burned area mapping and severity analysis, either alone or with other comparable optical and radar satellite imageries (Gibson et al. 2020; Konkathi and Shetty 2021; Tariq et al. 2021; Luca et al. 2021; Morresi et al. 2022; Seydi et al. 2022). For example, Luca et al. (2021), because of the high accuracy of the severity map they obtained, proposed the combined use of Sentinel-1 (radar) and Sentinel-2 (optical) satellites to analyze fire severity using a combined burn index. However, they also attained a similar accuracy when they tested Sentinel-2 separately. Moreover, different approaches have been proposed in the literature to detect and/or map burned areas. For example, Hu et al. (2021) tested the capability of deep learning models to map burned areas automatically. The fire severity of each fire event was calculated by using the dBNR index because it enables results to be obtained quickly and easily and to be verified via different applications in different types of forests around the world (Epting et al. 2005; French et al. 2008; Chen et al. 2011; Giddey et al. 2022). In the second phase of the present study, we extracted the burned areas automatically using the GIS-integrated Python script called the “Burned Area Extractor”, which delineates burned areas from dNBR data by making certain algebraic raster calculations and feature data editing steps. This algorithm enabled the fast and automatic extraction of the burned area sections. The output of the burned areas extracted via the algorithm was evaluated by visual interpretation, and the results were obtained at satisfactory accuracy levels. Indeed, in the second phase, three GIS-integrated Python scripts were developed to delineate the burned areas from the outputs of the first phase. This aided in the classification of the fire-severity maps by eliminating small pixel groups (< 5), and in the creation of a GIS database including forest information, classified fuel types, and some topographic information such as slope, aspect, and elevation. This study, in fact, proposed the creation of a fast and accurate forest-fire database using the advanced developments of remote sensing and GIS techniques.

The literature reveals that fire severity is an important metric that should be considered in forest fire studies. It is especially important to know its spatial pattern in order to better understand the ecological effects of fire and to comprehensively analyze the driving factors of fire behavior (Collins et al. 2014a, b), as well as to develop better fuel treatments in fire risk reduction (Tubbesing et al. 2019). Thus, fire severity was included in the fire database created in this study. Fire severity can be mapped easily and accurately because of the present availability of remote sensing data. The resulting fire severity classes showed a heterogeneous areal distribution for each fire event, especially in forested areas, because factors such as fuel-type class (i.e., composition of vegetation), topography, weather conditions, and climate have a critical influence on the intensity of a fire (Keeley and Syphard 2016). The database created having fire severity information together with fuel types and topography will be available for use in further quantitative analyses.

With regard to fuel types, the existence of Turkish pine tree species played a critical role in the transformation of a fire event into a large, uncontrollable phenomenon, especially depending on the weather conditions. Of all areas affected by fire events, the largest was covered, as expected, by forests comprised of Turkish pine because this pine species is among the tree species in Türkiye that are highly sensitive to forest fires (Küçük and Bilgili 2008). However, a considerable amount of the black pine forests was also affected by these large fire events. It is to be expected that both “Turkish pine” and “black pine” forests could be threatened by such fires in the near future because of the increasing effects of anthropogenic factors and climate change (Şahan et al. 2022). These fire-sensitive forests are complicatedly intermingled with residential and agricultural areas. This situation has created a significant problem in terms of forest-fire risk. Forest fires pose an especially significant threat to residential and agricultural areas. With regard to forest-stand development stages, the Very Young, Young, and Mid-Aged forests together covered the largest amount (52%) of the total affected areas. This means that in the areas mostly dominated by Turkish pine, the progress of the fires continued at a severe pace as crown fires (Bilgili et al. 2006) due to weather conditions and topographic properties (Bilgili et al. 2021). In addition, immature forests areas, when greatly affected by these fires, also pose a great threat to continuous wood production (Martell 1994; Savage et al. 2010) and sustainable regulated age-class distribution (Bilgili and Baysal 2013; Baysal 2014). With regard to canopy closure, the “NoClosure” fuel-type covered the largest area. These areas are not completely bare, but mostly covered with maquis vegetation, which is rated at the highest fire intensity among the fuel types (Salis et al. 2016). Moreover, the maquis vegetation that had increased in forested areas with decreased canopy cover (Ertuğrul et al. 2019) led to crown fire outbreaks (Bilgili et al. 2010), especially in unproductive forested areas (23.28%) and those with “Low” crown closure (38.18%) (Fig. 6).

As in the study of Dimitrakopoulos et al. (2011), the topographic parameters (i.e., slope, elevation, and aspect) in the present study were calculated and added to the fire database. We applied a very simple method to evaluate the relationship between the fire severity classes and these parameters only, using the mean and maximum slope and elevation values from all fire events. In addition, we presented the mean slope of the fire-severity classes within each aspect class for each fire event. The comprehensive fire database created with this study will enable future studies to carry out a more detailed analysis by applying different methods. Thus, our approach did not consider any meteorological parameters such as temperature, humidity, or wind speed and direction. In the present study, we focused mainly on post-fire documentation and evaluation by aiming to determine the extent of the areas burned due to especially large fire events, the distribution of fire-severity classes, and the role of fuel types in the affected forested areas.

5 Conclusions

In Türkiye, the summer of 2021 was a very critical fire season in which the large fires that occurred were unprecedented in its forest-fire history. In total, 291 forest fires affected more than 200 thousand ha. The total size of the area affected by these fires was 22.5 times larger than the area annually burned in Türkiye. In addition to their catastrophic damage to the environment and society, these fire events elicited intensive public debates, especially in Turkish social media. These discussions covered different aspects of the issue, with some expressing environmental anxieties from the clearly felt effects of climate change in recent years, and others reflecting differences in political views and the perceived organizational deficiencies in fire suppression. It can be concluded that having a comprehensive database is very crucial. This study investigated 30 large (affecting > 100 ha) forest fires using remote sensing and GIS techniques with the aim of creating an automatic GIS-based forest fire database. Although the GDF has its own fire database including some information, no comprehensive and detailed database is accessible to researchers. With this study, we aimed to lay the foundations for the creation of a GIS-based database for further forest fire studies. In future, this study will allow us to create a WebGIS-based platform having a database that will include all forest fires, from the recent past to the present, that will be updated each year, and that will be accessible to researchers. However, within the scope of this study, we presented only 30 large 2021 fire events. In this respect, the study provided a database that includes information on the total extent of the burned area, the amount of residential and agricultural areas affected, and most importantly, considering the severities of the fires, the extent of the forested areas affected with regard to fuel types and topographic properties. The latter is crucial because the information most required for comprehensive and advanced forest fire research considering climate change should include a database that provides complete information on fire severity and forest types with regard to fuel types as well as topographical and meteorological data. In the extraction and determination of both the burned area and fire severity, Sentinel-2 satellite data were used in the GEE platform, which allowed the researchers fast, free, and easy access in analyzing the data. The GIS-based fire database was supported by the powerful Python programming language. Thanks to developed scripts such fire database can be created easily and fast. These scripts are also suitable for using other forested areas and regions. When the recent effects of climate change emerged, it was considered advantageous to integrate both remote sensing and GIS techniques, not only as essential tools for forest fire-related issues, but also for another research. In addition, especially after the 2021 fire season, the number of large forest fires and their related studies are foreseen to increase in Türkiye.