The synergistic use of AHP and GIS to assess factors driving forest fire potential in a peat swamp forest in Thailand

Narissara Nuthammachot; Beomgeun Jang; Dimitris Stratoulias

doi:10.1515/geo-2025-0888

Article Open Access

The synergistic use of AHP and GIS to assess factors driving forest fire potential in a peat swamp forest in Thailand

Narissara Nuthammachot , Beomgeun Jang and Dimitris Stratoulias

Published/Copyright: October 9, 2025

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information

From the journal Open Geosciences Volume 17 Issue 1

Abstract

Wildfires are a main reason of environmental changes in natural ecosystems. While prevention and suppression of forest fires are crucial components of a forest management plan, it is of paramount importance to support decisions with accurate and tangible information. Geospatial data have been a useful source in this aspect, and various techniques have been proposed to define the zonation of forest fire risk in fire-prone areas. In this study, we classify and map forest fire risk in a peat swamp forest in Khreng sub-district, Thailand. Geospatial, climatic, and topographic data (namely elevation, slope, rainfall, land use/land cover, normalized difference vegetation index, and proximity to rivers, settlements, and roads) are fed into the analytic hierarchy process (AHP) and Geographical Information System (GIS) methods to map the area into five fire risk categories. The results are validated based on 70 historic fire events. About 88.57% of the ground truth fire incidents occurred at locations characterized by our assessment as very high risk class and 11.43% in the high risk class. It is suggested that the synergy of geospatial data and GIS and AHP techniques yields useful and accurate fire zone maps, which may aid addressing future fire prevention measures and forest management of fire-prone areas.

Keywords: fire risk; geospatial data; Sentinel-2; analytic hierarchy process; geographical information systems; peat swamp forest

1 Introduction

Fires are among the most destructive natural disasters, posing significant threats to ecosystems, biodiversity, air quality, and human health. In tropical regions, such as Southeast Asia, fires are also a major contributor to land degradation and deforestation [1], often causing irreversible environmental damage. Understanding the spatial and temporal drivers of fire ignition and propagation is thus critical for developing effective fire prevention, mitigation, and land management strategies.

Therefore, breaking down the contribution of factors involved, such as climatic conditions, topography, land use/land cover (LU/LC) characteristics, human interference, etc., and managing the fire effects are of crucial importance [2]. Fires may initiate from a natural phenomenon, such as lightning, or accidentally (as well as incidentally) by humans worldwide with notable places being Siberia, the Mediterranean, North America, and East and Southeast Asia (Vidal; Keeley et al.). In Thailand, fire events have been reported in forested and agricultural areas in the northern, southeast, and southern parts of the country and trigger environmental and human health concerns [3,4]. Among the ecosystems at risk, peat swamp forests are particularly vulnerable due to their high carbon content, flammability, and limited natural regeneration after fire. Yet, despite their ecological significance and susceptibility, fire risk in peat swamp forests remains understudied, especially in southern Thailand where LU pressure and fire incidence are increasing.

A combination of environmental conditions may drive fire dynamics, notably the topography, weather, fuel availability, and human activities. These factors bring ecosystem changes such as increased fuel accumulation and temporal anomalies [5,6,7,8]. For instance, it has been suggested that high and very high fire risk levels are found near settlements because of human unintentional or intentional activities [9,10], and consequently fires are a threat to human health and well-being. Additionally, the analysis of fire risk in regard to the road network is important in the sense that roadside forest management is a fundamental task for wildfire prevention [11].

Although fire is an event difficult to predict, it is of paramount importance to elaborate solutions for fire prevention and forest management. Fire risk and fire severity are an important source of information when conducting pre-fire plans and assessing the suitability for prevention measures. Hence, the existence of a fire risk zone map is crucial when putting in place related decision-making actions [12]. Geoinformation technology, and specifically remote sensing (RS) and Geographic Information System (GIS), is a useful tool for producing fire risk models and consequently detect, classify, and map fire risk. Moreover, the analytical hierarchy process (AHP) is a multiple-criteria decision-making (MCDM) that has gained attention lately in RS and has been used synergistically with GIS to build fire models. For example, Adab et al. [13] used earth observation data and GIS to derive several fire-related indices and eventually identified high-risk locations in a forested area in northeastern Iran. Similarly, Akbulak et al. [14] selected 11 criteria to classify fire risk (namely elevation, slope, aspect, vegetation type, stand crown closure, distance to settlement, population density, distance to agricultural land, distance to previous forest fire spot, normalized difference vegetation index [NDVI], and Forest Fire Weather Index) in Çanakkale, Turkey. They found that extreme, very high, high, and moderate risk categories were 3.87, 63.46, 32.13, and 0.53%, respectively. Jia et al. [15] used moderate resolution imaging spectroradiometer (MODIS) and vegetation, climatic, topography, and anthropogenic characteristics and concluded that over 90% of the fire hotspots and burned patches have taken place in medium- and high-risk areas in Inner Mongolia. Nuthammachot and Stratoulias [16] applied the AHP technique to classify a fire-prone area in South Thailand; LU/LC, elevation, slope, aspect, precipitation, and proximity to settlements and rivers were considered to produce a fire risk map. The result was validated against 705 historic fire events, and they found that 83% of the fires occurred in locations characterized by the highest fire risk zone. Nasiri et al. [17] mapped a fire-prone area in Paveh city, Iran by selecting six factors, namely NDVI, elevation, slope, aspect, LC, and evaporation using AHP and GIS methods. The resulting risk map showed strong agreement with historical fire occurrences, highlighting that about 64.7% of the area falls into high risk and very high-risk zones. In another study, El Mazi et al. [18] simulated a forest fire risk model in Mediterranean forests in Morocco. Seven potential criteria were selected in this study related to forest structure, geomorphology, and human environment using AHP and GIS methods. Validation with burned area data from 2012 to 2022 showed that high risk and very high-risk zones dominate, covering 62 and 26% of the forest, respectively, with lower-altitude forests being more vulnerable due to fuel accumulation. The findings highlight cork oak and pine forests as most fire-prone and offer a valuable tool for decision-makers to implement targeted fire prevention and mitigation strategies. The literature reveals several similar examples where satellite images, GIS, and AHP are used synergistically to model fire risk accurately.

However, few studies have explored the application of such integrated methods, specifically in tropical peatland environments, where fire dynamics differ significantly from forests due to persistent soil moisture and subsurface burning. The current study addresses this gap by applying a combined AHP–GIS approach to assess fire risk zones in the Khreng sub-district of Cha-uat district, located within a peat swamp forest ecosystem in Southern Thailand. By integrating environmental, climatic, topographic, and human-related factors, the study aims to generate a fire risk map that supports evidence-based land management and policy decisions in this sensitive and under-researched region.

2 Study area

The Khreng sub-district is located within the Cha-uat district of Nakhon Si Thammarat province at approximately 7°57′54″N latitude and 99°59′54″E longitude (Figure 1). This area forms part of the Kuan Kreng peat swamp forest, the second largest protected wetland ecosystem in Southern Thailand, covering 161.20 km². The region is characterized by flat terrain, with elevations ranging from 0 to 60 m above the sea level and is recognized as one of Thailand’s most fire-prone peatland zones.

Figure 1

Location of the study area in Southern Thailand. The Khreng sub-district is depicted in the right side map, while the larger administrative units of Cha-uat district and the Nakhon Si Thammarat province are visualized in the left side maps.

The Kuan Kreng peat swamp forest is ecologically unique and encompasses a variety of vegetation species such as Melaleuca cajuputi, Syzygium spp., and Stenochlaena palustris. These species form dense, waterlogged forest canopies during the wet season but are also capable of producing large amounts of combustible surface and ground-level biomass. In peatlands, the accumulation of surface organic material over thousands of years provides a highly flammable and persistent fuel bed, particularly during prolonged dry periods.

The area experiences a distinct climate regulated by monsoons. The rainy season lasts from May to January and is characterized by high humidity levels and waterlogged soils that inhibit fire ignition [19]. In contrast, the dry season extends from February to April and brings about declining water tables, intense heat, and increased evapotranspiration, which all add to creating favourable conditions for fire ignition and propagation. More information about the climate and landscape characteristics of the broader Kuan Kreng peat swamp forest can be found in Nuthammachot and Stratoulias [16].

Historical records indicate frequent fire events in this region, with many attributed to land clearing, agricultural expansion, and traditional burning practices. Recurrent fires have degraded large portions of the forest, reducing peat thickness, and altering the composition of native vegetation. Moreover, LU in and around the Khreng sub-district is dynamic and heavily influenced by agricultural encroachment, illegal logging, and aquaculture development. These anthropogenic pressures, in combination with the favourable conditions for fire ignition, have led to concerns on environmental degradation. Hence, the local authorities have established a forest fire control centre and carried out significant fire preventive and mitigation measures such as construction of check dams, patrolling, and firefighting support unit preparation. Despite these efforts, fire prevention remains a complex challenge due to overlapping environmental, land-use, and socio-ecological drivers, and it is of paramount significance to manage and prevent fire occurrences frequently during the dry season.

3 Methodology

This study adopts a multi-criteria decision analysis (MCDA) framework, supported by a diverse set of geospatial data layers. MCDA is particularly well suited for complex spatial decision-making processes, where multiple, often conflicting criteria must be evaluated simultaneously. In the context of fire risk assessment, this approach enables the integration of heterogeneous spatial datasets (such as LU, vegetation type, topography, climatic variables, and proximity to human infrastructure) into a single, interpretable output. By synthesizing fire-related geospatial layers through the MCDA process, the study produces a comprehensive spatial representation of fire risk. This risk map is instrumental for informed decision-making in fire prevention, preparedness, and mitigation planning. It allows stakeholders to identify high-risk zones, prioritize resource allocation, and implement targeted prevention strategies.

The methodological framework applied in this research has been previously tested and validated in several studies (e.g. Akbulak et al. and Van Hoang et al. [14,20]), demonstrating its robustness and adaptability across different geographic and environmental contexts.

A schematic overview of the methodological workflow, detailing the sequential steps from data acquisition to final risk map generation, is provided in Figure 2.

Figure 2

Flowchart of the overall methodology carried out in this study.

3.1 Spectral vegetation index

An image from the Sentinel-2A satellite is used in this study. The data are available and free for download from the European Space Agency (ESA) web portal (https://sentinel.esa.int). Sentinel-2A carries the high-quality MultiSpectral Instrument (MSI) capable of recording information in 13 spectral bands at three nominal spatial resolutions (i.e. 10, 20, and 60 m) and has 10 days revisiting time (Table 1). The near-infrared (NIR) and short-wave infrared (SWIR) bands are especially useful for agriculture, natural disasters, and forestry-related applications. The satellite image used in this study was acquired on November 13, 2023. The Sen2Cor processor was used to perform atmospheric correction and derive top-of-atmosphere data. Atmospheric correction was judged necessary since the NDVI is derived based on the red and NIR bands, which are significantly affected by atmospheric effects. The free software Sentinel Application Platform was used to further pre-process the data and remove the cloud-contaminated pixels. Last, the related spectral indices were derived.

Table 1

Spectral and spatial specifications of the MSI instrument of Sentinel-2A

Band #	Spectral region	Central wavelength (nm)	Bandwidth (nm)	Spatial resolution (m)
1	Coastal aerosol	442.7	21	60
2	Blue	492.4	66	10
3	Green	559.8	36	10
4	Red	664.6	31	10
5	Red-edge	704.1	15	20
6	Red-edge	740.5	15	20
7	Red-edge	782.8	20	20
8	NIR	832.8	106	10
8A	NIR	864.7	21	20
9	Water vapour	945.1	20	60
10	SWIR (cirrus)	1373.5	31	60
11	SWIR	1613.7	91	20
12	SWIR	2202.4	175	20

First, the data were spatially re-sampled and subset to the area of interest. Thereafter, the NDVI, a common index used in RS [21], was estimated. NDVI has been frequently applied in forest fire risk studies as it is an indication of the greenness of an area; high NDVI values associate with dense green vegetation (forest, grasslands, and crops), while low NDVI values with sparse vegetation and are, therefore, related to higher fire risk [22]. NDVI also strongly correlates with soil moisture, especially over homogeneous vegetation cover, which is indicative of drought conditions and consequently of fire susceptibility [23]. The theoretical range of the NDVI value is between −1 and 1; however, in our study, the range is between −0.35 and 0.81. The data were then classified into five bins according to the Jenks natural breaks optimization method. This range was divided into five equal parts and was used for the categorization of the risk index. Band 4 and Band 8A of Sentinel-2A were used to calculate the NDVI based on the following equation:

(1) NDVI = ( Band 8 A – Band 4 ) / ( Band 8 A + Band 4 ) .

Figure 3 presents the thematic maps of the eight factors considered in the study.

Figure 3

Thematic maps of the eight factors considered in the current study. Values indicate the fire rating from 0 to 5 for each factor separately.

3.2 Topographic factors

Elevation and slope are attributes of the topography of the area and were taken into account in this study. Elevation is a critical physiographic characteristic that is related to wind intensity and fire distribution [24]. Moreover, areas at relatively higher altitudes are receiving larger amounts of rainfall and, therefore, inherit a lower opportunity to encourage fire behaviour. Slope is the second criterion affecting fire spread. Fires occurring at steep slopes can move faster than down slopes [25]. The two topographic factors were calculated from the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model by NASA delivered at 1 arc-second (30 m) spatial resolution. The SRTM collected radar data at near-globe coverage in February 2000; however, the 17 years of time difference between the SRTM and Sentinel-2A data is not of great concern for the integrity of our dataset, as the topography of an area does not change significantly during such short time frames, especially in areas with smooth topography and lack of seismic activity as is the case for the current area of interest. The SRTM data were downloaded from the EarthExplorer website in GeoTIFF file format, the tiles were stitched together, the result was clipped to the region of interest, and finally projected onto the WGS 84 (UTM zone 47N) map coordinate system. Finally, the elevation and slope values were derived. The processing was carried out with the Geospatial Data Abstraction Library. The elevation values were then classified into five categories, namely <30 m (very high), 30–60 m (high), 60–90 m (moderate), 90–120 m (low), and >120 m (very low); while the slope (in percentage) was classified as follows: 5 classes: <5 (very low), 5–15 (low), 15–25 (moderate), 25–35 (high), and >35 (very high), and these results are presented in Figure 3.

3.3 Climatic factor

Climatic and environmental factors are foundational factors for fire ignition and propagation and define the fire weather of a locale. In the context of climate change, there is strong evidence that global warming is associated with increased forest fire frequency, duration, and size [26]. This is because climate change alters the patterns of extreme temperatures, duration of dry seasons, and precipitation patterns, which are all defining environmental variables in occurrences. The IPCC’s Sixth Assessment Report (AR6) estimates that in the case scenario of a global warming of 2°C, the global land area burned by wildfire is projected to increase by 35% (medium confidence) [27].

The annual precipitation was considered as the climatic variable in this study. Rainfall data from 22 meteorological stations distributed within the southern provinces of Thailand during 2020–2022 were provided by the Thai Meteorological Department. The Climate Data Management System was used to estimate monthly rainfall (January–December). Thereafter, the annual precipitation was calculated with the kriging method by interpolating the 22 point data. The map was classified into five risk classes, namely <2,500 (very high), 2,501–3,000 (high), 3,001–3,500 (moderate), 3,501–4,000 (low), and >4,001 (very low). Temperature is also an important climatic determinant in fire ignition and propagation as it directly affects the possibility of fire ignition and indirectly the humidity of the vegetation and soil. Nevertheless, in tropical areas such as Southern Thailand, the seasonal temperature fluctuation is small, and the mean temperature values are stable. Wind speed and prevailing direction are also determining the fire propagation direction and speed; however, our study area is relatively small, and temperature and wind are relatively homogeneous over the whole area. For this reason, temperature and wind have not been considered as important factors for fire occurrence in this study, and instead precipitation has been assumed as the defining environmental variable for fire occurrence. In Figure 4, it becomes apparent that from January to September, there is a clear inverse relationship: as rainfall decreases or remains low, fire events increase significantly, peaking in September. Conversely, during the months with the highest rainfall (October to December), fire events drop to zero. This pattern suggests that dry conditions, particularly in summer months, significantly contribute to the rise in fire occurrences.

Figure 4

Correlation between monthly rainfall and the number of fire events throughout 2019, showcasing the inverse relationship between these two variables. Note: for March and April, there are no rainfall data.

3.4 Proximity to anthropogenic and natural elements

Anthropogenic activities are frequently a cause of intentional or accidental fire ignition events, as well as one of the main factors of fire occurrence [28]; hence, it needs to be accounted for in a multi-criteria model. Most studies based on GIS-MCDA consider the anthropogenic element by integrating GIS layers related to human activities. In addition, more advanced approaches of incorporating anthropogenic influences into fire probability models exist, as demonstrated, for example, by Mann et al. [29] and discussed in a meta-analysis by Costafreda-Aumedes et al. [30]. It is worth noting that the human influence depends on several factors and is variable across space [31].

In the current study, we consider human settlements, road network, and river network as human-related fire risk factors. It has been suggested that with the decrease of distance from settlements, the number of fire ignition increases since fires may be initiated by human activities in forested areas [32]. The distance from roads is a potentially equally important parameter because humans access forested areas through the road infrastructure and therefore areas nearby roads have a higher risk of a forest fire ignition. The same holds for the proximity to rivers; however, proximity to roads and rivers has an inverse effect; the closer to a road or river a fire occurs, the easier it is for emergency units to access and suppress the fire. Therefore, the distances from roads and rivers were considered proportional to the fire risk (i.e. proximity to a road or a river yields lower risk), while the distance from settlements is reversely related. The distances from roads, rivers, and settlements were calculated as the Euclidean distance from each raster cell to the nearest respective feature, and five categorical values were created for each attribute, as presented in Figure 3.

3.5 LU/LC

The vegetation volume and type are critical factors for fire spreading as they represent the total fuel available for combustion. The LU/LC data were given by the GeoInformatics Research Center for Natural Resource and Environment and Southern Regional Center of GeoInformatics and Space Technology, Thailand. Five LU/LC categories were defined based on the following classes: forest (very high), marsh and peat swamp (high), agriculture (moderate), built-up areas (low), and water bodies (very low). The classification scheme was adapted from previous relevant studies (i.e. Khampeera and Aiemlaaor [33,34]), and the corresponding outcome is provided in Figure 3. Vegetated areas inherit a high fire risk while built-up areas and water bodies have a very low fire risk.

3.6 Weight of factors

The AHP, a method developed by Saaty [35], was used to assess the contribution of each individual layer and derive the corresponding weights. The AHP is a decision rule falling in the domain of MCDM; for a review of the latter scientific domain, the reader is directed to the studies of Malczewski [36], Diaz-Balteiro and Romero [37], and Mardani et al. [38]. In the AHP, the following steps are usually applied: breaking down of the decision-making problem into a hierarchy of criteria and alternatives, effective criteria and pairwise comparison, calculation of the relative weight of comparison elements and the combination weight of each layer element, and finally consistency evaluation. The weights for each factor were calculated by dividing the sum of each rows with the total number of factors (Table 2). The consistency index (CI) representing the deviation and the consistency ratio (CR) were based on equations (2) and (3).

(2) CI = λ max − n n − 1 ,

(3) CR = CI RI ,

where λ _max = sum (consistency vector)/n, consistency vector = weight sum/criteria weights, and RI is the random CI (the appropriate consistency index).

Table 2

Rank, weight, and fire risk classes for the criteria considered during the AHP analysis

Criteria/factor	Class	Rank	Weight	Fire risk classes
Elevation (m)	<30	5	0.05	Very high
	30.01–60	4		High
	60.01–90	3		Moderate
	90.01–120	2		Low
	>120.01	1		Very low
Slope (%)	<5	1	0.05	Very low
	5–15	2		Low
	15–25	3		Moderate
	25–35	4		High
	>35	5		Very high
Rainfall (mm)	<2,500	5	0.13	Very high
	2,501–3,000	4		High
	3,001–3,500	3		Moderate
	3,501–4,000	2		Low
	>4,001	1		Very low
Road (m)	0–250	1	0.06	Very low
	250–500	2		Low
	501–750	3		Moderate
	751–1,000	4		High
	>1,001	5		Very high
River (m)	0–250	1	0.15	Very low
	251–500	2		Low
	501–750	3		Moderate
	751–1,000	4		High
	>1,001	5		Very high
Settlement (m)	0–200	5	0.1	Very high
	200–1,000	4		High
	1,000–2,500	3		Moderate
	2,500–5,000	2		Low
	>5,000	1		Very low
LU/LC	Forest	5	0.26	Very high
	Marsh and swamp	4		High
	Agriculture	3		Moderate
	Built-up	2		Low
	Water body	1		Very low
NDVI	0.58–0.81	5	0.21	Very high
	0.34–0.58	4		High
	0.11–0.34	3		Moderate
	–0.12 to 0.11	2		Low
	–0.35 to –0.12	1		Very low

The random index (RI) table (Table 3) provides consistency index values for matrices of different sizes. It was used when calculating the CR to ensure that the pairwise comparisons in the AHP are logically consistent, as proposed by Saaty [35].

Table 3

RI table for n = 15

Matrix size (n)	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
Random index (RI)	0	0	0.58	0.9	1.12	1.24	1.32	1.41	1.45	1.49	1.51	1.48	1.56	1.57	1.59

The inconsistency is considered acceptable if CR ≤0.10. In the current study, λ _max = 8.42, n = 8, CI = 0.06, and RI = 1.41; consequently, CR = 0.043, which is lower than the critical threshold of acceptance and therefore is considered statistically acceptable.

3.7 Fire risk model

The forest risk model was developed based on the weights assigned to each variable from the AHP, which are provided in Table 2. Eventually, the risk map was calculated by a linear combination of factors considered and using the weights for each factor, as estimated by the AHP (equation (4)).

(4) Risk map = ( 0.05 × Elevation + 0.05 × Slope + 0.06 × Road + 0.13 × Rainfall + 0.15 × River + 0.10 × Town + 0.26 × LU / LC + 0.21 × NDVI ) .

3.8 Validation

About 70 instances of ground truth data on fire events were collected from the Department of National Parks, Wildlife and Plant Conservation, Nakhon Si Thammarat province, Thailand. The points were scattered across the study area and recorded between 2017 and 2022 with a Global Positioning System device during recurring field trips. The validation method was performed by co-locating and establishing the spatial relationship between the ground truth data and the fire risk map. The confusion matrix (Table 4) presents how well the model predicted fire occurrences. Out of 70 actual fire cases, the model correctly predicted 61 of them, meaning the number of true positives is 61. However, it failed to identify nine actual fires, which are false negatives, and it made no false fire predictions. The number of false positives is zero. Since there were no cases where fire did not occur, the true negatives are also zero. From this matrix, we calculated three key performance metrics. The precision is 1.00, meaning every time the model predicted a fire, it was correct. The recall is approximately 0.871, indicating that the model detected about 87.1% of all actual fire cases. Finally, the accuracy is also 0.871 because there were only fire cases in the dataset, so the accuracy matches the recall.

Table 4

Confusion matrix of validation with ground truth data

	Actual fire	No actual fire
Predicted fire	61 (TP)	0 (FP)
Predicted no fire	9 (FN)	0 (TN)

	Formula	Value
Precision	TP/(TP + FP)	1
Recall	TP/(TP + FN)	0.871428571
Accuracy	(TP + TN)/total	0.871428571

In addition, a sensitivity analysis was conducted to assess the effect of variations of the input factors on the final risk classes. This approach has been suggested from several similar studies in the literature (e.g. Nikolić et al. and Horvat and Karleuša [39,40]).

4 Results and discussion

4.1 Correlation between factors and ground truth data

LU/LC is the most important individual factor in the fire risk model developed; from the results depicted in Figure 5, it is apparent that forests are the main LC type accommodating fire occurrences (60 fire spots), while a few fires occur in marsh/swamp and agriculture (nine and one fire spots, respectively), and none in built-up and water bodies. In regard to the NDVI, the second most important contributor (weight: 0.21), there are 49 recorded fire spots coinciding with very high and high risk level classes, 20 fire spots in the moderate class, and only one fire spot in the low class; consequently, we conclude that forest and green areas (NDVI: 0.34–0.81) are the main cause of fire, while in non-forested areas (NDVI: between 0.11 and −0.35) only one fire incident occurred. Evidently, the documented fire occurrences are found in areas with concurrently high NDVI and vegetated landscape. Hence, the role of satellite imagery in developing fire risk models is paramount as extracting information related to vegetation characteristics, primarily based on vegetation indices, is an established application of RS science [41].

Figure 5

Relation between the fire risk class and number of fire spots for each factor individually (NDVI, elevation, slope, rainfall, road, settlement, river, and LU/LC).

In regard to topography, 70 fire spots are found in the very high risk class category of elevation (<30 m) and in the lowest rank for slope (<5%); it is apparent that flat areas influence the fire behaviour. The climatic variable, annual rainfall, is at a moderate level in the whole study area, and therefore 70 fire spots are recorded in this category. The vicinity to anthropogenic elements presents less biased statistical distributions; it is found that the number of fire spots in regard to the distance from roads is approximately the same for all 5 classes: 14 spots (very high), 9 spots (high), 15 spots (moderate), 16 spots (low), and 16 spots (very low). Regarding the distance from settlements, the number of fires is larger in the very high class, high, and moderate classes (5, 33, and 29 fire spots, respectively), while only three fire spots are found in the low class and no fire spots in the very low class. Finally, the distance from rivers is inversely related to the fire spot occurrences; the distribution follows an exponential curve (R ² = 0.89) with most fires occurring closer to rivers: 41 fire spots in very low class, 12 fire spots in low class, 9 fire spots in moderate class, 5 fire spots in high class, and 3 fire spots in very high class. However, it is worth noting that if the river and settlement are in close proximity, there is a lower chance for fire occurrence.

4.2 Forest fire risk map and validation

The final results for forest fire risk assessment (Figures 6 and 7) indicate that 71.55% of the covered study area is characterized by very high risk class and 24.41% by high risk class in the Khreng sub-district in Cha-uat district, Nakhon Si Thammarat province, Thailand. Moderate, low, and very low risk classes represent only 3.86, 0.16, and 0.02% of the total area, respectively (Figure 7). The attribution of a higher than moderate class in the majority of the image might be explained by the fact that the study area is located in a tropical environment; during the AHP weight derivation, LU/LC and NDVI were the most important factors accounting cumulatively for 0.47 of the relative weight of the factors. Hence, in a tropical environment, which is typically characterized by extensive and lush vegetation cover, a large value for both of these factors is anticipated, and a skew towards the larger categorical values of fire risk is expected.

In the validation analysis, the degree of co-occurrence between the risk levels extracted from the GIS-AHP approach and the historic data was investigated. It is found that 62 (88.57%) of fire spots coincided with the very high risk class and 8 (11.43%) in the high risk class, while no fire events occurred in moderate, low, or very low risk areas, as presented in Figure 8. The existence of fire spots only in the two highest risk classes has also been reported by Maric et al. [42] in a similar study conducted in the Mediterranean environment. In general, this high degree of agreement between fire zonation prognosis and actual fire occurrence has been reported by other similar studies which employ AHP as the MCDM in their methodology (e.g. Akbulak et al., Jia et al., Nuthammachot and Stratoulias, Eskandari and Miesel, Sachin and Prabin [14,15,16,43,44,45].

Furthermore, based on the visual evaluation of the product map in Figure 6, it is clear that most ground truth data (i.e. recorded fire incidents) are marked in the areas of very high and high fire potential zones. It is also worth noting that fire incidents are generally found neither adjacent nor far from settlements and anthropogenic activity; for instance, the large forested area in the southeast part of the study area does not have a large number of fire spots recorded. Moreover, the peat swamp area in the west part of the study area, as indicated in the LU/LC subfigure of Figure 3, also does not exhibit significant fire incidents; it is the forested areas within 2 km from settlements and roads that encompass the fire hotspots (Figure 6). While all the available ground truth data were used in this study (i.e. 70 reported fire events), a higher number of historic fire spots may yield more representative results [46].

Figure 6

Fire risk categorical map of the Khreng sub-district based on the weights derived from the AHP. The recorded fire events from the validation data are overlaid (green points).

Figure 7

Area coverage for each of the fire risk classes.

Figure 8

Number of ground truth data coinciding with each fire risk zone developed from the model in this study.

Last, the sensitivity analysis is presented in Table 5. The table summarizes the contribution of each factor to the final risk map, as the result of the test of the impact of change in the weights of the decision criteria. Specifically, the analysis shows a high total-order index for the LU/LC and NDVI parameters. This is followed by moderate contributions from town and rainfall, indicating a relatively less biased distribution of fire spots related to settlement distance. In contrast, the road distance presents a very low index, which is consistent with the nearly uniform distribution of fire spots across its classes.

Table 5

Sensitivity analysis matrix

Parameter	S1	S1_conf	ST	ST_conf
LU/LC	0.521	0.055	0.521	0.047
NDVI	0.171	0.033	0.171	0.017
Town	0.105	0.025	0.105	0.011
Rainfall	0.103	0.027	0.103	0.01
River	0.043	0.016	0.043	0.004
Elevation	0.042	0.016	0.042	0.005
Road	0.014	0.01	0.014	0.001
Slope	0.002	0.001	0.002	0

5 Conclusions

The success of preventing forest fires and eliminating the associated loss caused depends on studying thoroughly the potential fire risk and developing preventive measures. In this research, the application of RS, GIS, and AHP method was developed to determine fire risk zones in a fire-prone area in Khreng sub-district, Thailand. The derivation of the fire risk assessment was carried out based on eight criteria (namely, LU/LC, NDVI, elevation, slope, annual precipitation, and proximity to roads, rivers, and settlements), and the output was validated with historic fire events. The results indicate that the combined use of RS, GIS, and AHP techniques yields useful information suitable for fire zoning development. The resultant fire risk map can be helpful for addressing efficient fire prevention and suppression. Future steps will be the incorporation of Sentinel-1 images in order to estimate biomass and improve the LU/LC layers as the integration of radar data has been demonstrated to improve the LU/LC classification in the tropics [47,48] and allows for dynamic monitoring of fire risk and LC changes. In addition, the AHP is a technique that has proven to provide useful results when applied on geospatial data; however, research and comparison with other techniques, such as fuzzy AHP [49,50], is needed to manifest the potential of MCDM in the context of fire risk assessment. In fuzzy AHP, improving the expert weight calibration through consistency checks and sensitivity analysis can increase decision robustness. Combining radar-derived indices (e.g. backscatter coefficient or coherence changes) with fuzzy AHP criteria may provide a more comprehensive and data-driven fire susceptibility map.

Funding information: This research was supported by the National Science Research and Innovation Fund (NSRF) and Prince of Songkla University (Grant No. ENV6801040S).
Author contributions: Narissara Nuthammachot: review & editing, funding acquisition, conceptualization. Beomgeun Jang: validation, formal analysis. Dimitris Stratoulias: writing original draft, review & editing, methodology, conceptualization.
Geolocation information: The study area is the Khreng sub-district, Cha-uat district, Nakhon Si Thammarat province, Thailand, and can be found at the geographic coordinates 7°57′54″N latitude and 99°59′54″E longitude.
Conflict of interest: Authors state no conflict of interest.

References

[1] Hernandez-Leal PA, Arbelo M, Gonzalez-Calvo A. Fire risk assessment using satellite data. Adv Space Res. 2006;37:741–6.10.1016/j.asr.2004.12.053Search in Google Scholar

[2] Mitsopoulos I, Chrysafi I, Bountis D, Mallinis G. Assessment of factors driving high fire severity potential and classification in a Mediterranean pine ecosystem. J Environ Manag. 2019;235(1):266–75.10.1016/j.jenvman.2019.01.056Search in Google Scholar PubMed

[3] Vidal J. Fire, Fire Everywhere: The 2018 Global Wildfire Season Is Already Disastrous; 2018. https://www.huffingtonpost.com/entry/fire-fire-everywhere-the-2018-global-wildfire-season-is-already-disastrous_us_5b5a1271e4b0de86f494ed28. accessed on 05 December 2019.Search in Google Scholar

[4] Keeley J, Bond W, Bradstock R, Pausas J, Rundel P. Fire in Mediterranean Ecosystems: Ecology, Evolution and Management. vol. 515, Cambridge: Cambridge University Press; 2012.10.1017/CBO9781139033091Search in Google Scholar

[5] Fearnside PM. Deforestation in Brazilian Amazonia: history, rates, and consequences. Conserv Biol. 2005;19(3):680–8.10.1111/j.1523-1739.2005.00697.xSearch in Google Scholar

[6] Bond WJ, Keeley JE. Fire as a global “herbivore”: the ecology and evolution of flammable ecosystems. Trends Ecol Evol. 2005;20:387–94.10.1016/j.tree.2005.04.025Search in Google Scholar PubMed

[7] Bowman DMJS, Balch JK, Artaxo P, Bond WJ, Carlson JM, Cochrane MA, et al. Fire in the earth system. Science. 2009;324:481–4.10.1126/science.1163886Search in Google Scholar PubMed

[8] Ahmad F, Uddin MDM, Goparaju L. Fire risk assessment along the climate, vegetation type variability over the part of Asian region: a geospatial approach. Model Earth Syst Environ. 2019;5:41–57.10.1007/s40808-018-0517-ySearch in Google Scholar

[9] Jaiswal K, Mukherjee S, Mukherjee KD, Saxena R. Forest fire risk zone mapping from satellite imagery and GIS. Int J Appl Earth Obs Geoinf. 2002;4:1–10.10.1016/S0303-2434(02)00006-5Search in Google Scholar

[10] Ajin RS, Loghin AM, Vinod PG, Jacob MK. Forest and wildland fire risk assessment using geospatial techniques: A case study of Nemmara forest division, Kerala, India. J Wetl Biodiversity. 2015;5:29–37.Search in Google Scholar

[11] Novo A, Fariñas-Álvarez N, Martínez-Sánchez J, González-Jorge H, Fernández-Alonso JM, Lorenzo H. Mapping forest fire risk-A case study in Galicia (Spain). Remote Sens. 2020;12(22):3705.10.3390/rs12223705Search in Google Scholar

[12] Eskandari S, Ghadikolaei JO, Jalilvand H, Saradjian MR. Forest risk modeling and predoction in district three of Neka-Zalemroud forest, using Geographical Information System. Iranian. J For Poplar Res. 2013;21(14):203.Search in Google Scholar

[13] Adab H, Kanniah KD, Solaaimani K. GIS based probability assessment of fire risk in grassland and forested landscapes of Golastan province, Iran. 2011 International Conference on Environmental and Computer Science IPCBEE. vol. 19, Singapore: 2011. p. 170–5.Search in Google Scholar

[14] Akbulak C, Tatlı H, Aygün G, Sağlam B. Forest fire risk analysis via integration of GIS, RS and AHP: The Case of Çanakkale, Turkey. J Hum Sci. 2018;15(4):2127–43.10.14687/jhs.v15i4.5491Search in Google Scholar

[15] Jia X, Gao Y, Wei B, Wang S, Tang G, Zhao Z. Risk assessment and regionalization of fire disaster based on analytic hierarchy process and MODIS Data: A Case Study of Inner Mongolia, China. Sustainability. 2019;11(22):6263.10.3390/su11226263Search in Google Scholar

[16] Nuthammachot N, Stratoulias D. A GIS-and AHP-based approach to map fire risk: a case study of Kuan Kreng peat swamp forest, Thailand. Geocarto Int. 2019;36(2):212–25.10.1080/10106049.2019.1611946Search in Google Scholar

[17] Nasiri V, Sadeghi SM, Bagherabadi R, Moradi F, Deljouei A, Borz SA. Modeling wildfire risk in western Iran based on the integration of AHP and GIS. Environ Monit Assess. 2022;194(9):644.10.1007/s10661-022-10318-ySearch in Google Scholar PubMed

[18] El Mazi M, Boutallaka M, Saber ER, Chanyour Y, Bouhlal A. Forest fire risk modeling in Mediterranean forests using GIS and AHP method: case of the high Rif forest massif (Morocco). Euro-Mediterr J Environ Integr. 2024;9(3):1109–23.10.1007/s41207-024-00591-3Search in Google Scholar

[19] Office of the Official Information Commission (2020). http://www.oic.go.th/. (accessed on 25 January 2020).Search in Google Scholar

[20] Van Hoang T, Chou TY, Fang YM, Nguyen NT, Nguyen QH, Xuan Canh P, et al. Mapping forest fire risk and development of early warning system for NW Vietnam using AHP and MCA/GIS methods. Appl Sci. 2020;10(12):4348.10.3390/app10124348Search in Google Scholar

[21] Bonneau LR, Shields KS, Civco DL. Using satellite images to classify and analyze the health of hemlock forests infested by the hemlock woolly adelgid. Biol Invasions. 1999;1:255–67.10.1023/A:1010021629127Search in Google Scholar

[22] Tatli H, Akbulak C, Aygün G, Çekmek M, Sağlam B. Determining risk of the forest fires in çanakkale via geographic information systems and analytic hierarchy process. International 8th Atmospheric Sciences Symposium – Atmos 2017, İstanbul, Turkey, vol. 1, 2017b. p. 381–92.Search in Google Scholar

[23] Gu Y, Hunt E, Wardlow B, Basara JB, Brown JF, Verdin JP. Evaluation of MODIS NDVI and NDWI for vegetation drought monitoring using Oklahoma Mesonet soil moisture data. Geophys Res Lett. 2008;35(22):L22401.10.1029/2008GL035772Search in Google Scholar

[24] Xiangwei G, Xianyun F, Hongquan X. Forest fire risk zone evaluation based on high spatial resolution RS image in Liangyungang Huaguo Mountain Scenic Spot. IEEE International Conference on Spatial Data Mining and Geographical Knowledge Services (ICSDM). Fuzhou, China: 2011. p. 593–6.10.1109/ICSDM.2011.5969116Search in Google Scholar

[25] Adab H, Kanniah KD, Solaaimani K. Modeling forest fire risk in the northeast of Iran using remote sensing and GIS techniques. Nat Hazards. 2013;65:1723–43.10.1007/s11069-012-0450-8Search in Google Scholar

[26] Calder WJ, Parker D, Stopka CJ, Jiménez-Moreno G, Shuman BN. Medieval warming initiated exceptionally large wildfire outbreaks in the Rocky Mountains. Proc Natl Acad Sci. 2015;112(43):13261–66.10.1073/pnas.1500796112Search in Google Scholar PubMed PubMed Central

[27] IPCC. Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, ed. Pörtner H-O, Roberts DC, Tignor M, Poloczanska ES, Mintenbeck K, Alegría A, et al., Cambridge, UK and New York, NY, USA: Cambridge University Press; 2022. p. 3056. 10.1017/9781009325844.Search in Google Scholar

[28] Dong XU, Shao G, Limin DAI, Zhanqing HAO, Tang L, Wang H. Mapping forest fire risk zones with spatial data and principal component analysis. Sci China Technol Sci. 2005;49:140–9.10.1007/s11434-006-8115-1Search in Google Scholar

[29] Mann ML, Batllori E, Moritz MA, Waller EK, Berck P, Flint AL, et al. Incorporating anthropogenic influences into fire probability models: Effects of human activity and climate change on fire activity in California. PLoS One. 2016;11(4):e0153589.10.1371/journal.pone.0153589Search in Google Scholar PubMed PubMed Central

[30] Costafreda-Aumedes S, Comas C, Vega-Garcia C. Human-caused fire occurrence modelling in perspective: a review. Int J Wildland Fire. 2017;26(12):983–98.10.1071/WF17026Search in Google Scholar

[31] Fusco EJ, Abatzoglou JT, Balch JK, Finn JT, Bradley BA. Quantifying the human influence on fire ignition across the western USA. Ecol Appl. 2016;26(8):2390–401.10.1002/eap.1395Search in Google Scholar PubMed

[32] Jaiswal P, Prasanna R, Singh PK. Isolation and screening of rice field cyanobacteria exhibiting algicidal activity. Asian J Microbiol Biotechnol Environ Sci. 2005;7:367–73.Search in Google Scholar

[33] Khampeera A. Spatial Analysis of drought in kuan kreng peat swamp for peatland fire management using geo-informatics technology. Faculty of Environmental Management. Prince Songkla University; 2017. p. 1–356. http://kb.psu.ac.th/psukb/bitstream/2016/11421/1/417019.pdf.Search in Google Scholar

[34] Aiemlaaor S. An analysis of forest fire effect on tourism in Nam Nao National Park. Faculty of Agriculture Natural Resources and Environment. Thailand: Naresuan University; 2015. p. 1–140.Search in Google Scholar

[35] Saaty TL. A scaling method forms priorities in hierarchical structures. J Math Psychol. 1977;15(3):234–81.10.1016/0022-2496(77)90033-5Search in Google Scholar

[36] Malczewski J. GIS and Multicriteria Decision Analysis. 392, New York: John Wiley and Sons; 1999.Search in Google Scholar

[37] Diaz-Balteiro L, Romero C. Making forestry decisions with multiple criteria: A review and an assessment. For Ecol Manag. 2008;255(8–9):3222–41.10.1016/j.foreco.2008.01.038Search in Google Scholar

[38] Mardani A, Jusoh A, Nor K, Khalifah Z, Zakwan N, Valipour A. Multiple criteria decision-making techniques and their applications–a review of the literature from 2000 to 2014. Econ Res Ekonomska Istraživanja. 2015;28(1):516–71.10.1080/1331677X.2015.1075139Search in Google Scholar

[39] Nikolić G, Vujović F, Golijanin J, Šiljeg A, Valjarević A. Modelling of wildfire susceptibility in different climate zones in montenegro using GIS-MCDA. Atmosphere. 2023;14(6):929.10.3390/atmos14060929Search in Google Scholar

[40] Horvat B, Karleuša B. Conceptual model for integrated meso-scale fire risk assessment in the coastal catchments in croatia. Remote Sens. 2024;16(12):2118.10.3390/rs16122118Search in Google Scholar

[41] Xue J, Su B. Significant remote sensing vegetation indices: A review of developments and applications. J Sens. 2017;2017:1353691.10.1155/2017/1353691Search in Google Scholar

[42] Maric I, Siljeg A, Domazetovic F. Derivation of wildfire ignition index using GIS-MCDA from high-resolution UAV imagery data and perception analysis in settlement sali, Dugi Otok Island (Croatia). In GISTAM; 2021. p. 90–7.10.5220/0010465000900097Search in Google Scholar

[43] Eskandari S, Miesel JR. Comparison of the fuzzy AHP method, the spatial correlation method, and the Dong model to predict the fire high-risk areas in Hyrcanian forests of Iran. Geomatics Nat Hazards Risk. 2017;8(2):933–49.10.1080/19475705.2017.1289249Search in Google Scholar

[44] Rasooli SB, Bonyad AE, Pir Bavaghar M. Forest fire vulnerability map using remote sensing data, GIS and AHP analysis (Case study: Zarivar Lake surrounding area). Casp J Environ Sci. 2018;16(4):369–77.Search in Google Scholar

[45] Nuthammachot N, Stratoulias D. Multi-criteria decision analysis for forest fire risk assessment by coupling AHP and GIS: Method and case study. Environ Dev Sustain. 2021;23(12):17443–58.10.1007/s10668-021-01394-0Search in Google Scholar

[46] Sachin KC, Prabin K. Manifestation of an analytic hierarchy process (AHP) model on fire potential zonation mapping in kathmandu metropolitan city, Nepal. ISPRS Int J Geo-Inf. 2015;4:400–17.10.3390/ijgi4010400Search in Google Scholar

[47] Nuthammachot N, Stratoulias D. Use of SAR and optical satellite data for land use and land cover classification in the Songkhla Lake Basin, Thailand. Int J Appl Eng Res. 2017;12(24):14358–64.Search in Google Scholar

[48] Nuthammachot N, Stratoulias D. Fusion of Sentinel-1A and Landsat-8 images for improving land use/land cover classification in Songkla province, Thailand. Appl Ecol Environ Res. 2019;17(2):3123–35.10.15666/aeer/1702_31233135Search in Google Scholar

[49] Sinha A, Nikhil S, Ajin RS, Danumah JH, Saha S, Costache R, et al. Wildfire risk zone mapping in contrasting climatic conditions: an approach employing AHP and F-AHP models. Fire. 2023;6(2):44.10.3390/fire6020044Search in Google Scholar

[50] Vujović F, Valjarević A, Vila-Subirós J, Šiljeg, A, and Lukić T. Geospatial modeling of wildfire susceptibility on a national scale in Montenegro: A comparative evaluation of F-AHP and FR methodologies. Open Geosci. 2024;16(1):20220694.10.1515/geo-2022-0694Search in Google Scholar

Received: 2025-02-14

Revised: 2025-07-23

Accepted: 2025-08-22

Published Online: 2025-10-09

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

https://doi.org/10.1515/geo-2025-0888

Keywords for this article

fire risk; geospatial data; Sentinel-2; analytic hierarchy process; geographical information systems; peat swamp forest

Creative Commons

BY 4.0