Introduction

Forest fires have cascading effects on biodiversity, climate change, and livelihoods (Kelly et al. 2020). Biodiversity is very important for all people and human societies to continue their lives (Adom et al. 2019). In addition to animals, many plant species form the basis of biodiversity and are important for economic prosperity and food safety for human beings. The richness of biodiversity keeps ecosystems in balance with support for the natural environment, and settlements provide economic gains. The protection of forested areas with high plant diversity is also important for the sustainability of biodiversity. A large part of the world's forests are used for many purposes, including economic reasons such as timber and paper production, as well as the protection of biological diversity (Bistinas et al. 2013). In addition, forested areas are also used for socioeconomic purposes such as animal husbandry and recreational activities. In general, the socioeconomic benefits of forest ecosystems are very diverse. Forests have a regulating effect on water basins in regions where water resources are limited. It helps to reduce air pollution by regulating the atmosphere (Balzter et al. 2007). Wood and timber obtained from forests are used as raw materials in many industries. The plants in the bush and herbaceous formations in the forest area are important for animal grazing. The United Nations (UN) Sustainable Development Goal (SDG Goal 15) emphasizes measures to "protect, restore, and promote sustainable use of terrestrial ecosystems, sustainably manage forests, combat desertification, and halt and reverse land degradation and biodiversity loss”. This is rather general; thus, the need to design a specific SDG that will adequately encapsulate fire disasters is of concern to many scholars (Cerdà 2019; Martin 2019).

The overexploitation of natural resources, environmental pollution, climate change, and especially forest fires, together with the pressure of rapid population increases, cause the extinction or damage of many plant species (Ager et al. 2014; Bistinas et al. 2013; Çolak and Sunar 2020; Ganteaume et al. 2013). Forest fires emit large quantities of gases and particulate matter to the atmosphere, e.g., gaseous emissions such as GHGs (CO2, CH4) and numerous air pollutants (e.g., CO, VOCs, SOx, NOx, and NH3). Forest fires have led to both changes in plant species and reductions in forest area (Atalay 2015). While these fires cause long-term changes in forestlands, they also cause ecological and socioeconomic losses. In addition to the ecological consequences of deforestation, forest fire activity has socioeconomic and sociocultural implications. For example, Statistica reported that in October 2003, a forest fire in the U.S. caused an economic loss of 3.5 billion U.S. dollars. Forest fires with a great degree of unpredictability may cause damage to the landscape in the forest area. Loss of biodiversity after wildfires in forest areas where animals are grazed negatively affects livestock activities. In addition, forested lands with dense tree species are areas where drinking water is provided. The burning of many trees in fires causes a significant decrease in both the quality and quantity of drinking water. Disruption of the function of the environment poses a threat both locally and globally. Forest fire activities constitute one of the most important causes of this environmental degradation. Forests are in danger of fire due to their large accumulation of flammable materials. Forest fire activities have resulted in many consequences in different locations, and fires endanger the continuity of forests in many parts of the world. Large or small forest fires may cause significant damage to forests in any region (Özden et al. 2012). In regions with hot and dry summers, the effects of fires on forest areas are greater. In Turkey, which is located in the Mediterranean climate zone where the summer months are very hot and dry, forest fires cause the destruction of thousands of hectares of forest area every year (Çolak and Sunar 2020). (Martinho 2019) emphasized that damage due to forest fires has increased worldwide. In recent years, the number of forest fires has increased throughout the world, and the incidence of forest fires has increased in different seasons (Nimmo et al. 2022)

Major challenges of landscape–wildfire interactions in southern Europe can be seen in these key areas, as proposed by (Moreira et al. 2011). Many socioeconomic factors have contributed to changes in land cover over the past few decades, resulting in increased fire hazards and frequent large wildfires, which invariably increase the number of homogeneous landscapes (covered by shrubland), increasing the susceptibility to fires (Moreira et al. 2011). One of the best options to protect the landscape against fire disasters while considering their strong impact on ecology (e.g., vegetation disturbance, species extinction, pollution, etc.) and socioeconomic impact on livelihoods is an approach that provides a measure of fire severity, vulnerability, predictive ability, and risk reduction. The remote sensing approach for burned area detection for active and post-fire activities (Lentile et al. 2006) is now well recognized by the scientific community (Chuvieco 2003; Chuvieco et al. 2020; Chuvieco et al. 2010; Kurbanov et al. 2017). Multitemporal remote sensing data can provide spatiotemporal observations (before, during, and after) of wildfires, which can potentially be used for efficient and cost-effective fire detection, damage assessment, and mitigation planning (Attri et al. 2020).

Remote sensing data from optical passive (e.g., MODIS, Landsat, and Sentinel 2), radar and LiDAR data plus dedicated fire sensors are being used to provide information on pre, active-, and post-fire events (Chuvieco et al. 2002; Chuvieco et al. 2020). Despite the development of different remote sensing-based fire products, the demands of fire managers are not always met due to the varying spatial and temporal resolutions of the sensors. However, there are readily available products that can be used for a variety of analyses on a global scale (Giglio et al. 2009; Mouillot et al. 2014; Roy et al. 2005; Tansey et al. 2008). However, the low spatial resolution of the various existing fire products, coupled with their discrepancies in terms of their estimates of the total area burned, the location of burning, and the timing of burning, make them unsuitable for small and medium fires (Humber et al. 2019; Schroeder et al. 2016). Resource managers are increasingly looking for information from small fires. The use of high- and moderate-resolution data, such as multispectral Landsat and Sentinel data, provides opportunities for small-to-medium-scale fire severity detection and mapping.

Novel burnt area indices have been developed and are continually being developed from multitemporal satellite observations for assessing fire severity, risk, and vulnerability in different ecosystems (Chen et al. 2011; Chuvieco et al. 2002; Escuin et al. 2008; Kurbanov et al. 2017; Liu et al. 2020). The Normalized Difference Vegetation Index (NDVI) and Normalized Burn Ratio (NBR) and their temporal counterparts, the difference normalized difference vegetation index (dNDVI) and the differenced normalized burn ratio (dNBR), are among the most appropriate standard spectral indices for estimating fire and burn severity (Escuin et al. 2008; Key and Benson 2006; Lacouture et al. 2020; Liu et al. 2020; Mallinis et al. 2018). They are also reported to be less sensitive to atmospheric contamination, as most studies have used measurements of spectral reflectance within the visible and NIR regions to characterize vegetation conditions. This is because the spectral reflectance signatures of plants behave differently across regions of the electromagnetic spectrum (Chuvieco and Huete 2009). For example, the reflectance of visible (e.g., red band) and shortwave infrared (SWIR) spectra is low due to high chlorophyll absorption by plants, while the high reflectance in the near-infrared (NIR) region is due to individual leaves and whole plant canopies, which strongly scatter NIR energy (Ollinger 2011). Spectral indices derived from these bands have the potential to be used for fire detection because they can measure vegetation removal, charcoal deposition, canopy moisture reduction, and canopy shadows, which can reduce the NIR and increase the SWIR post-fire reflectance compared to those of healthy vegetation (Key and Benson 2006; Mallinis et al. 2018).

This study focused on forest fires in the Mediterranean region of Turkey because of the growing concerns raised by scientists (Çolak and Sunar 2020; Coskuner 2022; Yakupoglu et al. 2022) and decision-makers (DUVAR 2020; TRTWORLD 2021) on the impact of frequent fires and the need to devise impact assessment strategies for designing appropriate mitigation measures. The villages where intense forest fires occurred in the Kozan district in 2020 were chosen as the study area. These rural villages are located in the Mediterranean region, where forestry and livestock activities are the main sources of income. Forest fires negatively affect forestry and livestock activities in socioeconomic terms as well as their ecological effects. Assessing the ecological and socioeconomic impacts of forest fires in these villages is remarkably important in terms of rehabilitating burned areas and making future plans for the economic development of the local inhabitants/residents of the area. Thus, the objectives of this study were to 1) assess the spatial distribution of fire severity in burnt areas in 2020 by using multitemporal Landsat 8 imagery in rural settlements of the Kozan district, 2) evaluate the impact of fire severity on LULC, and 3) evaluate the influences of vegetation, topography and population density on the spatial pattern of fire severity.

Materials and methods

Study area

Adana Province is situated in southern Turkey within the Mediterranean Sea region. The province has 15 administrative districts. Kozan is a local government district with borough status in northern Adana, Turkey, which has a population of 132,320 according to 2021 census surveys. This district sits centrally in the triangle formed by Adana, Osmaniye, and Kayseri Provinces. Kozan comprises an area of 1.872 km2, and the county has 103 neighborhoods. The study was conducted in the forest in the protected natural area of rural settlements (Boztahta, Camlarca, Curuklu, Eskimantas, Karahamzali, Kuyubeli, and Minnetli) of the Kozan district located between latitudes 370 27’ N and longitudes 350 48' E. The range of the Taurus Mountains rises sharply in the northern section of the district. This mountain range corresponds to the natural forest area. There are also village settlements on mountain slopes and valley bottoms. Forestry and livestock activities are the main sources of income for these settlements. The region of the study, like the whole Mediterranean region, has a warm climate with rainy winters and hot and dry summers. The average temperature in summer is approximately 27.9°C, and that in winter is approximately 10.2°C. The average annual rainfall is 848 mm (Turkish State Meteorological Service 2022). The area is therefore very rich in biodiversity and plays a crucial role in biodiversity conservation and protection.

Generally, the majority of Mediterranean-type vegetation has a long history of forest degradation due to human-caused fires (Naveh 1975; Türkmen and Düzenli 2011) and environmental factors (Attri et al. 2020; Ganteaume et al. 2013). The General Directorates of Forestry in Turkey have been recording wildfires since 1937. According to the Ministry of Agriculture and Forestry, 69,567 forest fires occurred between 1988 and 2020 in Turkey, which is one of the Mediterranean basin countries where forest fires are common. In the same period, 345,842 hectares of forest area were burned (Ministry of Agriculture and Forestry 2020). Specifically, the selected rural settlements experienced extensive forest fires in 2020 (Fig. 1).

Fig. 1
figure 1

The study area

Full size image

Data

Landsat 8 data

Landsat 8 was launched on February 11, 2013. The data are freely available and were downloaded from the United States Geological Survey (USGS, Earth Explorer) website (https://earthexplorer.usgs.gov/). The satellite carries Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) instruments. However, only OLI data were used for this study. Even within the 9 bands available for the OLI, only Band 4 Red (0.64 - 0.67 μm), Band 5 Near-Infrared (0.85 - 0.88 μm), and Band 6 SWIR 1 (1.57 - 1.65 μm) were used with a spatial resolution of 30 m for generating spectral indices (NDVI and NBR) to detect burn severity. Cloud-free images (pre-fire, 15/08/2020, and post-fire: 31/08/2020) were selected from the data. These are collection 2 of the Landsat 8 archives, which were improved by providers based on geometric, radiometric, and atmospheric processing (https://www.usgs.gov/landsat-missions/landsat-collection-2). Landsat data offer strong potential for fire detection (Schroeder et al. 2016).

Land use/land cover (LULC) data

The LULC data of the study area were obtained from the Republic of Turkey, Ministry of Agriculture and Forestry (Ministry of Agriculture and Forestry 2020). The LULC data are in the form of a shapefile. Therefore, the LULC for the area was a shapefile extracted from the original map based on the boundary of the study area. The study area covers high forest, coppice forest, degraded forest, rural settings, and cultivation areas. Natural forests occupy approximately 60% of the territory. The main forest-forming species are Pinus, Brutia and Maqius. The rest of the territory is filled with glades, slopes, olive, persimmon, Carob Wulnat, some citrus trees, arable land, pastures, etc. (Fig. 2).

Fig. 2
figure 2

LULC map of the area

Full size image

Digital elevation model (DEM)

In this study, the freely available Shuttle Radar Topographic Mission’s 1 s (30 m) digital elevation data (Farr et al. 2000) were downloaded from the EarthExplorer of the United States Geological Survey (https://earthexplorer.usgs.gov/) based on this product, and the elevation, slope, and aspect were generated for further analysis.

Population data

The projected population data for 2021 for the seven villages of the Kozan districts were downloaded from the Turkish Statistical Institute. The population density for each village was determined based on the area in km2 using population data (Turkish Statistical Institute 2022).

Method

Landsat color infrared composite

This study employed a Landsat 8-color infrared composite to visually interpret pre-fire and post-fire events. NIR (band 5), red (band 4), and green (band 3) Landsat 8 data are used to create a representation of the Earth's surface, which is otherwise known as the NIR composite. Specifically, healthy vegetation appears red, while unhealthy vegetation appears dark in color. The purpose of this analysis is to enable us to visually differentiate between pre- and post-fire events.

Spectral indices

NDVI index

The NDVI is a proxy of vegetation greenness. The index makes use of data from red and near-infrared (NIR) reflectance values or radiances. The NDVI is calculated as the ratio between the NIR and red, as shown in equation 1. The index is frequently used in the literature to identify burn areas since it produces plausible results. (Lacouture et al. 2020) Morante-Carballo et al. 2022).

$$\textrm{NDVI}=\frac{NIR- RED}{NIR+ RED}$$
(1)

NBR Index

The NBR is also one of the most widely used indices for fire assessments (Key and Benson 2006). (Liu et al. 2020) designed the Landsat 8 OLI index. This index utilizes the NIR and SWIR and is recommended as a standard burn severity index for assessments of medium and large fires (García and Caselles 1991; Key and Benson 2006; Liu et al. 2020). The index is derived as a ratio between the NIR and SWIR, as shown in equation 1

$$\textrm{NBR}=\frac{NIR- SWIR2}{NIR+ SWIR2}$$
(2)

Burn severity/area mapping

In this study, burn severity was estimated in three fundamental steps:

  1. i.

    The burn spectral indices were calculated using pre-fire and post-fire images. By subtracting the post-fire image from the pre-fire image, differences between the NDVI (dNDVI) and NBR (dNBR) were obtained. This was done for both indices, as shown in equations 3 and 4.

$$\textrm{dNDVI}=\textrm{prefire}\ \textrm{NDVI}-\textrm{postfire}\ \textrm{NDVI}$$
(3)
$$\textrm{dNBR}=\textrm{prefire}\ \textrm{NBR}-\textrm{postfire}\ \textrm{NBR}$$
(4)
  1. ii.

    Maps of burn severity were produced by extracting the fire-affected areas where subtle changes in the phenology of vegetation were observed. The differences between the pre- and post-fire events indicate the magnitude of change caused by the fire. As explained earlier, the fire event occurred on 23/08/2020, and post-fire imagery was acquired on 31/08/2020. The pre-fire image was acquired a week before the fire, while the post-fire image was acquired a week after the fire. We assumed that most of the impacts of fire on the land surface will manifest after one week. Based on this development, we assumed that the changes in the phenology of the forest ecosystem in this area may not be subtle in two weeks except under a prevailing disturbance (e.g., fire). Since the difference in the image acquisition period between the pre-fire and post-fire events is only two weeks, any significant changes resulting in the phenology of vegetation in the study area could be referred to as those occurring due to fire. Based on the magnitude of the burn severity as quantified from the burn spectral indices, the severities were characterized as high or low, and unburned areas

  2. iii.

    The severity of the burn area for distinct LULC was quantified using GIS data (polygons) on LULC. The LULC types found across the study area included high forest, coppice forest, rural setting/cultivation zones, and degraded forest. These LULC polygons were used to calculate the zonal statistics of each burn spectral index (dNDVI and dNBR). For each land cover type, the impact of fire on LULC was calculated as a function of burn area. Welch's t test was used to assess the significance of the difference in the burn areas between the dNDVI and dNBR (Welch 1938)

  3. iv.

    Based on the spatial distribution of burn areas, the total burn areas were also extracted for each village to quantify the magnitude of fire occurrence in those villages. The total burn area for each village was expressed in hectares. Similarly, Welch's t test was used to assess the significance of the difference in the burn areas between the dNDVI and dNBR (Welch 1938)

Assessing the relationships between burn severity and vegetation, topography, and population density

The spread of forests and its associated fire risk susceptibility, for example, can be influenced by fuel availability (Keane et al. 2001; Szpakowski and Jensen 2019), surface configuration (Adab et al. 2013), and anthropogenic disturbance (Turkish Statistical Institute 2022). The relationships between the fire severity (derived from the dNDVI and dNBR) and the vegetation indices of the pre-fire event, topographic variables (elevation, slope, and aspect), and population density were investigated using simple linear regression to evaluate whether these factors could influence fire occurrence. The values of one thousand (1000) randomly selected burnt pixels and their corresponding values for these factors were extracted to assess these relationships. Here, the fire severity is the response variable. Data analysis was carried out in R statistical software (R Core and Team 2022). Note that for the assessment of the relationship between population density and burn severity, only villages with predominant fires were used.

Results

Pre- and post-fire events

Figure 3 shows the color infrared image composite derived from Landsat 8 data for the study area during pre-fire and post-fire events (Fig. 3a/b). The healthy vegetation in Fig. 3a, which depicts the scene before the fire, is all red. However, in Fig. 3b, the burned-over forested area appears dark in color, suggesting the effects of the fire.

Fig. 3
figure 3

Color infrared composite maps of pre- and post-fire events

Full size image

Spatial distribution of burn area severity

The spatial distribution of the burn severity for our study area is shown in Fig. 4. Geographically, fire events are concentrated in the most central part of the area. Some villages are more affected than others. Camlarca village is most affected, as it has an area of 189.3 hectares based on the dNDVI and 178.5 hectares based on the dNBR. The village is probably the most vulnerable in terms of ignition points and the availability of fuel due to the presence of high forests.

Fig. 4
figure 4

Spatial distribution of burn severity in the Kozan district: (a) dNDVI and (b) dNBR

Full size image

Fire severity on LULC types

In the burned area of rural settlements, the fire severity was assessed. Table 2 indicates the burn severity according to the LULC of the area. The impact of fire on these LULC types is computed from Landsat 8-based burn spectral indices (dNDVI and dNBR). The burn area computed using the two indices does not vary significantly. The high forest area was the most affected, with 566.5 and 528.6 hectares burned during the fire events based on the estimations from the dNDVI and dNBR, respectively. The rural setting and cultivation area are next after the high forest area, with 163.6 and 167.8 hectares of land burned for dNDVI and dNBR, respectively. Fewer than 10 hectares of coppice forest were recorded for dNDVI, and approximately 15 hectares were recorded for dNBR. The degraded forest was the least affected, with burnt areas of 1.1 and 3.8 hectares for the dNDVI and dNBR, respectively.

Relationships between burn area data and population, NDVI, and topography

To assess whether forest fires can be influenced by vegetation type, topographic variables, and population density, we further analyzed the linear relationships between them (Fig. 5). The pre-fire vegetation (NDVI/NBR) versus burn severity showed a moderate positive relationship, R2 =0.48 and R2 =0.48 for dNDVI and dNBR, respectively. The topographic variables (elevation, R2 =0.04 (dNDVI) and R2 =0.00005 (dNBR); slope, R2 =0.00008 (dNDVI) and R2 = 0.01 (dNBR); aspect, R2 =0.001 (dNDVI) and R2 <=0.19 (dNBR)) showed a weak positive relationship with fire severity. Similarly, population density showed a weak positive relationship (R2= 0.11 and R2 = 0.09 for dNDVI and dNBR, respectively, with burn severity).

Fig. 5
figure 5

Relationships between vegetation, topographic variables, and population and burn severity: (a) pre-fire vegetation versus burn severity, (b) elevation versus burn severity, (c) slope versus burn severity, (d) aspect versus burn severity, (e) population density versus burn severity

Full size image

Discussion

This study explored the spatial distribution of fire severity in the Kozan district of Adana Province in Turkey based on spectral burn indices (NDVI and NBR) of pre- and post-fire events from Landsat 8 data (Fig. 3). We quantified the magnitude of fire at the spatial extent and observed that the fire spread in seven villages across the district, although the magnitude of the severity varied (Fig. 4). Some villages are more affected than others (Tables 1 and 2). For example, Camlarca village has been largely affected, with burnt areas of 189.3 hectares and 178.5 hectares, as measured by dNDVI and dNBR, respectively. Karahamzali was next largest, with burn areas for dNDVI and dNBR of 155.1 and 146 hectares, respectively. However, for Kuyubeli, burn areas covered 4.5 and 4.1 hectares for the dNDVI and dNBR, respectively. Note that none of the villages have less than 50 hectares of burn area. The fire has been observed to progress toward both the north and south, but it did not significantly expand farther north, where Kuyubeli is located. This might be due to the rescue effort by the government and other concerned stakeholders to contain the fire in that direction. Authorities have ordered the evacuation of locals due to massive forest fires in the Kozan district in the southern province of Adana. Bekir Pakdemirli, the minister of agriculture and forestry, stated on August 24, 2020, that 750 personnel, 35 heavy machinery, two planes, and 19 helicopters were mobilized in an effort to extinguish the fire (DUVAR 2020).

Table 1 Spatial distribution of fire severity in the Kozan district
Full size table
Table 2 Fire severity (burn area) based on LULC type
Full size table

This study also attempted to evaluate the effects of fire on the socioeconomic and ecological conditions of the study area. For example, in terms of ecological conditions, high forests accounted for more than 70% of the affected land cover types, while other forested and degraded forest areas were least affected. Socioeconomic effects were found within the rural setting and cultivated areas, accounting for 22% of the overall burnt areas. Forest fires, whether deliberate or accidental, are among the enormous problems for countries located in the Mediterranean region. Wildfires have been recorded by the General Directorates of Forestry in Turkey since 1937. According to the Ministry of Agriculture and Forestry, 69,567 forest fires occurred between 1988 and 2020 in Turkey, which is one of the Mediterranean basin countries where forest fires are common (Ministry of Agriculture and Forestry 2020). Due to the extremely hot weather conditions in the summer months, the majority of the forests in the Mediterranean region are among the first-degree sensitive areas (Özkazanç and Ertuğrul 2011). During the same period, 345,842 hectares of forest area were burned. When the causes of these fires are examined, natural causes are also identified, as are intentional and unintentional causes.

Recently, fires in the Mediterranean region have become disasters. The Adana Regional Directorate of Forestry covers a wide area extending from the wide plains in the south to the upper boundaries of the settlements located in the foothills of the Taurus Mountains. According to the data from the Ministry of Agriculture and Forestry, 41% of the surface area of Adana consists of forest areas. Red pine is the dominant species in this region, where the coniferous tree species group is concentrated in the forest area of the Mediterranean region. There are permanent and temporary settlements within this forest area. In addition, villages established in forest areas are places where recreation and tourism activities are widely carried out within the provincial borders of Adana. In particular, forest areas close to city centers are easily accessible and are being used extensively in this context. Wildfires have both direct and indirect socioeconomic consequences (Bivolarski 2019). In this context, forest fires negatively affect sustainable forest management and tourism activities.

The combination of fast winds, high temperatures, and extreme dry conditions quickly influences the spread of flames in the region, which can lead to serious disasters. For this reason, most villagers, residential areas, and summer cottages were evacuated during this fire. The forest area in this region constitutes the backbone of the rural community’s economy, thus, when wildfires spread to these villages it does result to killing of livestock. Forest fires are common in the Mediterranean region of Turkey, but this recent wildfire has covered a large area and has affected the socioeconomic lives of inhabitants; as a result, the region has been negatively affected their livelihoods. For example, many farmers have lost their animals to forest fires, and pasture areas for the grazing of surviving livestock have resulted in large losses in productivity. Similar findings in Özden et al. 2012 on the destruction of forests, soil, flora, and seed reserves caused by wildfires and the loss of organic substances have been reported. Wildfires have resulted in reduced food availability for both humans and animals, reduced growth rates of pasture and vegetation, and the burning of houses in the region. Additionally, Keane and Karau (2010), in a study on wildfires leading to destruction of residential houses, mosques, and animals, were also highlighted. In this context, fires, which are effective at changing natural life (Keane and Karau 2010), negatively affect the efficiency of future pastures. In summary, these wildfires have a substantial impact on both the ecological and socioeconomic activities, including the loss of pastures, forests, livestock and personal effects.

Thus, there are needs for effective methods for fire risk assessments because wildfires have a significant impact on socioeconomic and environmental aspects, even at a spatial scale. This will allow for the use of proactive control measures for environmental management (Attri et al. 2020; Moreira et al. 2011; Sari 2022). This study also evaluated the influence of vegetation, topographical factors (elevation, slope, and aspect) and population density on burn severity (Fig. 5a/e). Our findings confirm that of all the variables considered, the pre-fire vegetation indices (NDVI and NBR) were found to have the strongest relationships with burn severity. Although we observed that there was no significant difference between the two burnt spectral indices, dNBR estimates appeared to be more accurate in this case (Fig. 5a). This finding is comparable to that of a recent analysis by (Fernández-García et al. 2022), which predicted the potential severity of wildfires over southern Europe. Similarly, we found a distinct trend of increasing severity as forest conditions increased (Fig. 5a). Previous research revealed that the same behavior occurs in other parts of the world, showing that the amount of fuel is the main driver of burn intensity during the fire season in the ecosystems of southern Europe. Moreover, Zhang et al. (2023) provides valuable insights into the spatiotemporal patterns of forest fire occurrence in the province of Anhui, China. The study highlights the significance of socioeconomic factors, particularly nighttime light, and emphasizes the importance of considering both spatial and temporal heterogeneity when analyzing forest fire patterns. The authors emphasize proper utilization of remote sensing data and robust modeling tools to characterize forest fires and therefore provide options for managing them effectively.

The topographic factors show a weak positive relationship with burn severity (Fig. 5b/d). However, fire behavior can be influenced by the topography of the area. The occurrence of fire is mostly at an elevation of 400 to 800 m (Fig. 5b), at a moderate slope (Fig. 5c), and in all directions (Fig. 5d). Theoretically, fires tend to spread more rapidly upslope, probably due to fuel availability (early fuel drying at lower elevations), prevailing winds, and a propensity for more lightning strikes and subsequent ignitions at higher elevations. The occurrence of fires on lower and moderate slopes may be a result of the somewhat homogeneous nature of the forested area. A recent study in which machine learning was used to map wildfire susceptibility using remotely sensed fire data and GIS data in the Adana and Mersin Provinces, Turkey, revealed that elevation, temperature, and slope were the main contributing factors (Iban and Sekertekin 2022). In a recent study, Purnama et al. (2024) used machine learning to estimate the risk of forest fires in the Mediterranean region of Turkey to assess the vulnerability of forest fires. This study examined various machine learning algorithms, including support vector machines, naive Bayes, random forests, decision trees, and artificial neural networks. Several variables, such as precipitation, soil moisture, temperature, humidity, wind speed, land cover, elevation, aspect, slope, proximity to roads and electricity networks, and population density, were used to train and test the models. The random forest (RF) algorithm performed the best, highlighting its critical role in proactive approaches to managing fire risk. The selection of feature importance revealed that land cover type and temperature consistently ranked as the most important variables across the different methods.

We also found a weak positive relationship between population density and burn area. Our findings confirm that burned area initially increases with population density and then decreases when population density exceeds a threshold, as observed in (Bistinas et al. 2013). The intentional or unintentional ignition of fire can be explained by variables related to agricultural activity or the unemployment rate, depending on the socioeconomic background of the people living in a particular area (Ganteaume et al. 2013). The study area is surrounded by small villages whose livelihoods are highly dependent on the primary sector of the economy. Their proximity to the forest area is also another potential ignition factor. Recent studies have shown that fire-prone areas in Turkey are driven by land use patterns, such as deforestation owing to livestock density, forest interfaces, and distance to villages (Viedma et al. 2017). Parente et al. (2024), who extensively covered the literature on fire hazards in relation to socioeconomic and regional issues, emphasized the importance of urgency in addressing the societal need to manage wildfires due to expected intensification and spread of wildfires under global change. Similarly, Sayedi et al. (2024) analysed the impact of human activities on wildfire regimes has witnessed a significant escalation in the past two centuries. The study further highlighted that it is imperative to incorporate the insights gained from previous fire incidents into land and fire management approaches. However, it is important to acknowledge that the altering patterns of fire behavior can be expected due to the unparalleled human interference with plant communities, climate conditions, and various other contributing factors. It is highly likely that forthcoming fire regimes will lead to the deterioration of essential ecosystem services, unless urgent and effective measures are taken to mitigate the effects of climate change.

Our study demonstrates the use of remote sensing techniques for evaluating the impact of forest fire severity on ecological and socioeconomic aspects. The implications of our findings can be extended to forest management practices, encompassing fire prevention strategies and policy development aimed at mitigating the ecological and socioeconomic consequences of forest fires within the region. Ultimately, this study contributes to the scientific understanding of how forest fires affect both natural ecosystems and human communities, with a specific emphasis on the Kozan district in Turkey. Furthermore, our method can be replicated in other regions with similar physical and environmental characteristics.

Conclusion

Forests in the mountainous areas of some rural settlements within the boundary of the Kozan district burned extensively in 2020. Forest fires negatively affected plant communities and animal habitats in rural areas located on the slope of the Taurus Mountain Range. The main forest species of recent high-severity burned areas are Pinus brutia and Maqius. In the degraded forest areas and grasslands of these rural settlements, livestock activities are carried out both as settled and seminomadic livestock. Therefore, catastrophic fires are extremely effective in shaping both plant habitats and animal communities and the ecological processes of nutrient and water cycling in forested areas. In this context, assessing the spatial distribution of fire severity in the burnt area of the Kozan district is important for the evaluation of ecological and socioeconomic impacts. This study evaluated the potential of remote sensing to detect fire severity by using pre- and post-fire events from multitemporal Landsat 8 imagery covering the Kozan district. Our results confirm the large burning of high forests, cultivation areas, and rural settings. High forest areas are more highly vulnerable to burn severity (R2 = 0.48 (dNDVI) and R2 = 0.49 (dNBR)) than other vegetation types as severity increases with increasing vegetation density. Topography characteristically influences fire behavior, as most burning occurs in upslope areas. The study also revealed a weak positive relationship with burn severity, suggesting the impact of prevailing anthropogenic factors. This approach to determine the effects of wildfire severity on ecology and socioeconomics has demonstrated the potential of dNDVI and dNBR from Landsat 8 data for spatial analysis of fire severity in the Kozan district of Adana Province in Turkey. The limitation of this approach is that it is unable to assess the specific impact of fires on socioeconomic and ecological conditions. This is due to insufficient data on socioeconomic variables in particular. Therefore, we recommend that, in addition to the remote sensing approach, soicioeconomic data of the communities affected be estimated to reveal an accurate impact analysis. This will go a long way providing information on the mitigation of socioeconomic impacts.