Introduction

Forests are significant for the protection of the natural balance besides being major and amongst the most precious sources providing a good number of services for the welfare and the sustainability of the environment, biodiversity and man (Aleemahmoodi Sarab et al. 2015; Bilici 2009; Gheshlaghi 2019; İşçi 2018). However, the number of the forest fires originating from a variety of different reasons, either human-caused (negligence, intentional acts) or naturally caused (lightening), is continuously increasing worldwide, and the destructions caused by forest fires are mostly irreversible (Kuter et al. 2011; Rowell and Moore 2000; Sazawa et al. 2018). Between 1950 and 1990, 50% of the world’s forests was ravaged due to a number of reasons (Ertuğrul 2005; Laurance 2010; Rudel and Roper 1997). On the contrary of this fact, the amount of the forests in Turkey is declared to have increased from 20.2 million ha to 22.3 million ha between 1972 and 2015 (Orman Genel Müdürlüğü 2015). Nevertheless, forest fires also show tendency to increase in the country. The rate of this increase is 58% for the last 30 years (1372 forest fires in 1988 and 20,167 forest fires in 2018) (Orman Genel Müdürlüğü 2018). According to the statistics of the General Directorate of Forestry, the annual number of forest fires for the last 5 years (2013–2018) in the country varies between 2149 and 3755, and the annual amount of the burned lands ranges from 3117 ha to 11,993 ha. For the same period, more than 4 million forest fire alerts were recorded worldwide (Brushlinsky et al. 2016; Global Forest Watch Fires 2019).

In case of forest fires, it is nearly impossible to stop the natural cycle. Still, some measures can be taken to mitigate and prevent the negative impacts and ravages on the environment (Akay and Erdoğan 2017; Bailey 2001; Calfee and Little 2003; Kaysis 2019; Majlingová 2012; Peker and Atılgan 2015; Yang et al. 2007). Development of forest fire risk/danger/hazard maps is a basic requirement for the planning and protection of the forest lands (Eugenio et al. 2016). These maps are also of great importance for the conservation of the forests during the production of the development plans (Oliveira et al. 2002). Besides, forest fire risk maps provide precious inputs for developing long-term strategic plans for fire prevention and mitigation activities (Akay and Şahin 2018). Thus, forest fire risk maps make useful assets for the determination of different risk mitigation measures in line with the risk ranges/classes, as well as the determination of the appropriate locations for fire towers, emergency service units, and water storage tanks (Goldammer et al. 2017). Moreover, risk data facilitate the planning of a variety of operations and allocation of human resources and support the protection of the critical infrastructure (FAO 2001).

Forest fire risk maps are produced using various data layers related to the risk factors that are essential for the detection of the risk zones (Jaiswal et al. 2002). There are a good number of parameters that affect the risk ranges such as forest stand types, topographic conditions, human factors and climate (Carmel et al. 2009; Gao et al. 2011; Gazzard 2012). The amount of oxygen and flare-up temperature, as well as the tree species, ages and canopy closure rate in a forest stand, are amongst the most significant factors affecting the starting, advancing and spreading of the forest fires (Bailey 2001; Bilgili 2003; Gazzard 2012; Kaysis 2019). From this perspective, there are various studies in the literature focusing on the determination of forest fires, fire behaviours, fire severity classes, risk classes and other forest management–related issues, most of which have been conducted with the utilization of advanced technologies such as geographical information systems (GISs) and remote sensing (RS). Today, no doubt that GIS is a unique and a widely used asset for obtaining, manipulating, analysing and presenting a good deal of spatial data that are required both for the determination of the fire risk zones and planning forest fire intervention operations (Değerliyurt and Çabuk 2015).

Accordingly, a good number of authors benefited from these technologies for the determination of the fire risks in different regions (Asri et al. 2017; Bingöl 2017; Erten et al. 2004; Jaiswal et al. 2002). Özşahin (2014) used AHP (analytic hierarchy process) method for the realization of forest fire sensitivity analysis in Antakya region. Between 1993 and 2009, kernel density estimation of forest fire events in Middle East Technical University (METU) Campus, Ankara, Turkey, was made by using different bandwidth choices (Kuter et al. 2011). Yassemi et al. (2008), on the other hand, developed GIS-based forest fire simulation models and thus prepared fire risk maps according to three different scenarios. Van Wagtendonk et al. (2004) utilized RS methods for the determination and comparison of the data related to fire severity, similar to that of Navarro et al. (2017), which was realized in Madeira. Kavgaci et al. (2016) and White et al. (1996) are amongst the other authors who benefited from RS technologies in their studies for the evaluation of fire severity. Meddens et al. (2016) used Landsat images in order to develop a model for the classification of the unburned regions around the fire areas, while Shafiei et al. (2010) evaluated the results of the forest fires in terms of varying reactions of different tree species against fires. Mirzaei (2016) focused on the effects of the human-caused fires on soil components in Zagros Mountains in western Iran.

In Turkey, where the number of the forest fires increased by 58% for the last 30 years, it has also become extremely necessary to determine the fire risks and adopt proactive approaches and strategies for the protection of the forests (Orman Genel Müdürlüğü 2018). In the country, Forest Management Units are assigned with the determination of precious and risky areas in line with the related legislation (Orman Genel Müdürlüğü 1995; Orman Genel Müdürlüğü 2017). Moreover, the Ministry of Agriculture and Forestry develops forest fire risk maps on national scale and realizes various projects for the determination of fire risks. Considering a number of studies, generally 5 classes are used for defining fire risk ranges varying from ‘very low’ to ‘extreme’ risk classes (Corona et al. 2013; Dong et al. 2005; Ghimire et al. 2014; Ghorbanzadeh et al. 2019; Rowell and Moore 2000; Vanha-Majamaa 2006). The General Directorate of Forestry in Turkey also classifies forest fire risks in 5 groups, in which ‘1st-degree risk’ refers to the highest risks and ‘5th-degree risk’ is the lowest. The factors affecting the severity of the risk are also determined by the Directorate (Orman Genel Müdürlüğü 1995).

According to the records of the General Directorate of Forestry, Kütahya province, comprising 6 regions, is amongst the highest risk areas (1st-degree risk areas) in terms of forest fires in the country. Considering the fire sensitivity classes of these regions, Simav, Tavşanlı, Gediz and Kütahya regions take the first place (high sensitivity), while Emet has 2nd-degree sensitivity and Domaniç region is of 3rd-degree sensitivity.

Since the determination of forest fire risks provides significant benefits such as effective and productive use of manpower, time and resources, as well as realization of efficient interventions against fires, it becomes a major task to develop forest fire risk maps especially for the high-risk areas such as Kütahya province. However, it is not possible to find forest fire risk maps prepared on local areas/scales in Turkey. According to our investigations, no spatial records of previous fires or forest fire risk maps were either detected for Ören Forest Sub-district Directorates territory.

From this point, the main target of this paper is to produce the forest fire risk map in Kütahya province using GIS capabilities. The selected study area comprises the Ören Forest Management Office. Within the context of the study, the location of the existing fire observation towers in the study area was also evaluated with GIS analyses for the determination of their visual accessibility over the area. Moreover, the fire in Çamdibi region, which burned 4 hectares in 2001, was examined, and the results were compared with the risk map via remote sensing (RS) and GIS methods.

Materials and Methods

Study area and material

Study area comprises Kütahya-Ören Forest Sub-district Directorates, which is one of the 10 subunits operating under Kütahya Forest Management Directorate. The Directorate was established in 1943, serving an overall area of 413,716 ha on the inner parts of Aegean Region within Kütahya provincial territories. Moreover, 51% (209,052 ha) of the territory is covered with forests, while 49% (204,124 ha) of it is mostly made up of open spaces. Furthermore, 58% (121,897 ha) of the forested areas are classified as normal woodlands and the ratio of the disturbed woodlands is 42% (87,155 ha).

Ören Forest Sub-district Directorates is located 980 metres above the sea level between northern latitudes of 39° 0′ 0.25″ and 39° 10′ 26.60″ and eastern longitudes of 29° 40′ 39.90″ and 29° 55′ 24.00″ (Fig. 1). The surface area of the study area is 237 km2. The majority of its territories is located within Aslanapa province, and some parts of it spreads in Gediz and Altıntaş provinces (Kültür ve Turizm Bakanlığı 2019). According to the 2018 statistics of Turkish Statistical Institute (TUİK), the population of Aslanapa province is 9224 (TUİK 2018). The climate of the region has Aegean, Marmara and Central Anatolian characteristics. Summers are hot and dry. Winters are mostly cold and rainy. July and August are the hottest and January and February are the coldest months of the year (OGM 2019). Moreover, 13.759 ha of the study area is covered with forests. The main tree species in Ören forests are black pines, junipers and scrubs. The most dominant species is black pine (8.539 ha). The other land use types within the study area are agriculture areas (8684 ha), forest soils (1068 ha), settlement areas (224 ha), grasslands (38 ha), cemeteries (5 ha) and water surfaces (2 ha). The number of settlement areas in the study is 11.

Fig. 1
figure 1

Location of study area – Kütahya Forest Sub-district Directorates

Full size image

The main material of the study is the vector and the raster data of Ören Forest Sub-district Directorates. A variety of the vector data, such as stand types, settlement areas, roads and dry streams, was obtained from the Kütahya Regional Directorate. Slope and aspect maps, as well as 15-m resolution digital elevation model (DEM), which was used for the visibility analysis, were provided from NASA EARTHDATA database, and 30-m resolution LANDSAT images captured on October 2, 2001 (pre-fire) and October 18, 2001 (post-fire) were downloaded from USGS database. ArcGIS 10.6 software was used for spatial analyses.

Methods

In this study, GIS techniques were applied to develop forest fire risk map for the study area, Kütahya Forest Sub-district Directorates. Also, the risk map was compared with the past forest fires in the region. Remote sensing methods were used to detect the past forest fires within the study area. Figure 2 illustrates the main workflow.

Fig. 2
figure 2

Main workflow of the study

Full size image

Weighted overlay analysis

The main method of the study, which was utilized for the development of the forest fire risk map within Ören Forest Sub-district Directorates, is weighted overlay analysis. For the production of sub-data layers such as slope and aspect, topographic analyses were performed with ArcGIS 10.6 software. Buffer analysis was used for the production of multi-buffer zones around roads, dry streams and settlement areas.

The weighted overlay process was performed in two phases. In the first phase, data layers (tree species, canopy closure and tree ages) in stand types map (Fig. 3) were classified in terms of their risk categories to produce the tree hazard class map. Tree hazard class map was then used as one of the main data layers/inputs for the second-phase weighted overlay process.

Fig. 3
figure 3

Stand types map

Full size image

During the second phase of the weighted overlay analysis, topography, settlement areas, roads and dry streams were used as main data layers besides tree hazard class maps. Each layer was appointed with influence weights and reclassified for the weighted overlay analysis. Figure 4 illustrates the method that was applied for the development of forest fire risk map.

Fig. 4
figure 4

Method for forest fire risk map

Full size image

In weighted overlay analysis, only raster data with particular values are used as input, followed by the ordering of the attributes of each layer from the most suitable to the less, in accordance with the goal of the study. The most suitable class is usually assigned with the highest score/point in the selected working scale (ESRI 2016). Each layer of data used as input for weighted overlay is also given weights in relation with their influence on the study.

In this study, with regard to the main goal, the ‘most suitable class’ was referred as ‘highest risk’ and a scale of 9 was used, where 9 points was determined as the reference for the highest risk class. Data, such as settlement areas, grasslands, cemeteries, forest soils, water surfaces and agricultural lands, were determined as ‘restrictions’ and extracted from the overlay layers. Influence weights were assigned in parallel with the opinions of the professionals in the field working at Ören Forest Sub-district Directorates. Table 1 summarizes the weighted overlay layers, risk classes and the weights of the data, and Table 2 gives the similar information for tree hazard class data layer.

Table 1 Weighted overlay information
Full size table
Table 2 Weighted overlay information for tree hazard classes
Full size table

Four risk classes were determined in this study as follows:

  • 1, 2 and 3 points: low risk

  • 4 and 5 points: moderate risk

  • 6 and 7 points: high risk

  • 8 and 9 points: very high risk

Visibility analysis

The shape of a surface is an important factor determining the amount and the location of the lands visible to a man standing on a certain spot in the landscape. Visibility of specific locations or visibility between two different items/points is essential for both design/planning and military works (ESRI 2018). At this point, visibility analysis is a widely benefited method used for the detection of the visible areas from certain points or vice versa. Land surface, slope, aspect, elevation and dimensions of the objects are regarded to perform an efficient visibility analysis (Popelka and Vozenilek 2010).

In this study, visibility analysis was used for the determination of the visible lands/forests from the fire towers existing in and around Ören Forest Sub-district Directorates.

Band ratio methods

Remote sensing methods were utilized for the determination of the areas that were burned in the study area in 2001. To fulfil this aim, NDVI (normalized difference vegetation index) and NBR (normalized burn ratio) methods were used. NDVI is basically the ratio of the radiation reflected from healthy vegetation to the total reflected radiation over a surface, which is calculated per pixel in a satellite image. The results range between −1 and +1, where +1 is the proof for the existence of healthy plant cover and −1 refers to bare surfaces (Sabuncu and Özener 2019).

The other band ratio method, namely NBR, is mostly applied for the determination of the burned areas through the detection of the difference of the images captured from an area before and after a fire. In NBR, 7th and 4th bands of the images are used (Key and Benson 2006). NBR ratio is particularly sensitive to live plant cover, humidity content and specific changes in soil after the fires. Therefore, NBR is a highly preferred method for burned green area detection studies (Sabuncu and Özener 2019). Table 3 summarizes the band ratio methods of NDVI and NBR. In this study, the images obtained via NDVI and NBR methods were overlaid with the forest fire risk map, so as to determine the risk classes of the burned areas and make evaluations.

Table 3 NDVI and NBR band ratio methods (Meddens et al. 2016; Navarro et al. 2017)
Full size table

Results

This section may be divided by subheadings. It should provide a concise and precise description of the experimental results, their interpretation and the experimental conclusions that can be drawn.

Development of fire risk map

As the main goal of this study, forest fire risk map for Ören Forest Sub-district Directorates was developed via the weighted overlay analysis of a variety of spatial data explained in previous sections of the study (Table 1 and Table 2). In accordance with the method illustrated in Fig. 4, tree hazard class map was produced. Moreover, a variety of spatial maps were also developed as data layers/inputs and reclassified with regard to their risk classes for the weighted overlay analysis. Figures 5, 6, 7, 8, 9, 10 show the weighted overlay data layers. Figure 11 gives the resulting weighted overlay analysis map that shows the forest fire risk classes of Ören Forest Sub-district Directorates.

Fig. 5
figure 5

Reclassified slope map

Full size image
Fig. 6
figure 6

Reclassified aspect map

Full size image
Fig. 7
figure 7

Reclassified distance to road map

Full size image
Fig. 8
figure 8

Reclassified distance to dry stream map

Full size image
Fig. 9
figure 9

Reclassified distance to settlement area map

Full size image
Fig. 10
figure 10

Tree hazard class risk map

Full size image
Fig. 11
figure 11

Ören Forest Sub-district Directorates forest fire risk map

Full size image

According to the forest fire risk map given in Fig. 11, 36.86% of the study area comprises very-high and high-risk zones. The rate of the moderate-risk areas is 60.39% and low-risk lands cover 2.7% of the study area.

Evaluation of existing fire towers

For the determination of the visibility of Ören Forest Sub-district Directorates territories from the towers, 19 fire towers located in and around the study area were used to perform the visibility analysis with ArcGIS 10.6 software. As elevation data is vital for visibility analyses, both DEM data of the study area and the tower heights were taken as main parameters. Figure 12 shows the locations of the fire towers and the resulting visibility map is illustrated in Fig. 13.

Fig. 12
figure 12

Locations of fire towers in and around the study area

Full size image
Fig. 13
figure 13

Visibility analysis result map

Full size image

According to the visibility map, 82.8% of the study area is visible from the towers. However, 17.2% of the land is out of towers’ sight.

Determination of burned areas using band ratio methods

On October 3, 2001, a forest area of 4 ha within Ören-Çamdibi region was burned as recorded by the official records. Within the aim of this study, pre- and post-fire LANDSAT 7 ETM images were obtained from USGS for performing NDVI and NBR band ratio methods with ArcGIS 10.06 software. Figures 14, 15, 16, 17, 18, 19, 20, 21, 22 show the images produced throughout the process.

Fig. 14
figure 14

543 band combination pre-fire (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 15
figure 15

543 band combination post-fire (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 16
figure 16

d543 difference (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 17
figure 17

NDVI pre-fire, 2001 (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 18
figure 18

2001 NDVI post-fire, 2001 (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 19
figure 19

dNDVI – difference (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 20
figure 20

NBR pre-fire, 2001 (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 21
figure 21

NBR post-fire, 2001 (spectral index maps for Ören-Çambidi region – 2001)

Full size image
Fig. 22
figure 22

dNBR – difference (spectral index maps for Ören-Çambidi region – 2001)

Full size image

Determination of fire risk classes for burned areas in 2001

This phase of the study was conducted for the determination of the fire risk classes of the areas burned in 2001 in Ören-Çamdibi region, as detected via NDVI and NBR methods. To fulfil this aim, the burned areas located on dNDVI and dNBR images (Figs. 14, 15, 16, 17, 18, 19, 20, 21, 22) were digitized using ArcGIS tools and overlaid with the forest fire risk map (Fig. 23).

Fig. 23
figure 23

Comparison of burned areas with fire risk map

Full size image

According to the findings, the digitized areas cover 5.6 ha, although the official records declare a total burned area of 4 ha. It was concluded that 83% (4.65 ha) of the burned forests were classified as of moderate risk, while 17% (0.95 ha) of it was on very-high and high-risk zones.

Conclusions

The main goal of this study was the determination of forest fire risk classes in the selected study area, namely Kütahya-Ören Forest Sub-district Directorates, with weighed overlay analysis method using GIS capabilities. As Kütahya province is listed within the highest risk areas (1st-degree risk areas) in terms of forest fires by the General Directorate of Forestry, the main motive for the selection of the study area was this critical fact, as well as the availability of the spatial data and the absence of the related works or researches in this territory. The risk maps produced by the General Directorate is on national scale, and therefore, more detailed risk determination studies on local scales are necessary for the prevention and efficient management of the forests in the country.

Besides the development of the forest fire risk map, visibility analysis, NDVI and NBR band ratio methods were also utilized respectively for the evaluation of locations of the existing fire towers and spatial determination of a previous forest fire in the study area, which took place in 2001. When the band combinations used in this study were evaluated, it is concluded that Landsat 543 and combination alone was insufficient, while NDVI and NBR methods provided more accurate results in the determination of the burned areas. Especially, NBR method stands out with its ability to classify forest fires according to their fire severity.

The forest fire risk map revealed that areas with young trees and low canopy closure caused moderate or low fire risks, and fire risk increased in parallel with the increase in canopy closure rate (Hu et al. 2017). In terms of stand types, it was found that productive black pine and juniper vegetation was more sensitive to fire risks in comparison with the disturbed forests. These facts show that stand types, tree species and canopy closure are of great importance for the determination of fire risk classes. Forest fire risk map also showed that the fire risk was more on the peripheries of the settlement areas and roads, which is not surprising at all due to the existence of human factors. Zones around dry streams also had higher fire risks as a result of the scrubby vegetation and human activities.

Considering the locations of the existing fire towers, the visibility analysis results put forward that they had sight over the majority of the area, which is accepted suitable in the detection of smokes.

During the study, it was detected that information related to forest fires is archived as non-graphical records/data by the authorities, and therefore, it is not possible to accurately locate the burned areas spatially unless other methods are used. This is also why the area of the burned areas on October 3, 2001, which is evaluated as the sample forest fire in this study, was calculated 40% more (5.6 ha) than the officially declared records (4 ha). To sum up, the spatial data for forests fires are unavailable in most of the Forest Sub-district Directorates, and hence, remote sensing methods are necessary for the determination of the burned area territories.

Besides the detection of the risk classes of the burned areas in 2001, which were from moderate- (83%) and very-high and high-risk (17%) areas, it was also concluded from the satellite images that the burned areas were rehabilitated after the fire.

When the literature is examined, it is seen that besides the data groups also used in this study, a number of different layers such as surface temperature, vegetation humidity, agricultural lands and population density are utilized by some of the researchers (Akbulak et al. 2018; Başaran et al. 2007; Fanhua et al. 2005; Karabulut et al. 2013; Orozco 2008; Sabuncu and Özener 2019; Siachalou et al. 2009). Moreover, climate has an eminent role in the determination of the forest fire risks and taking rapid actions/cautions against risks (Banja 2019). Nevertheless, climatic factors are ignored in this study since the major target was to produce a general forest fire risk map, rather than a dynamic model operating with seasonal and/or hourly climatic criteria.

One of the other important issues to highlight in terms of the methodology used in this study is the approach adopted for the determination of the influence weights, suitability/risk points and scale. It is necessary to underline that such an approach can differ depending on the geographical characteristics of the study area, previous experiences gained, aim and strategies of the study and applicable legislation. Therefore, it is significant to examine the geographic context comprehensively and select the most appropriate method in line with the aim and context.