Comparison Between Post-Fire Analysis and Pre-Fire Risk Assessment Produced from Various Geospatial Data

1. Introduction

Forest fires, prevalent in various biomes globally, alter the scope and structural characteristics of forested areas. Consequently, these changes disrupt the provision of ecosystem services and products, introducing uncertainties in the trajectory of these alterations. In the context of global climate change, the frequency of severe forest fires is inclined to rise, necessitating an accelerated adjustment in forest fire regimes [1]. Forest fires, as intricate natural disturbances, significantly influence ecosystem dynamics and have profound impacts on human societies. The continuous increase in fire danger and the extent of burned areas in Mediterranean ecosystems across diverse global regions is anticipated, attributed to factors such as global warming, population growth, and irregular land use [2,3]. This escalation underscores the growing significance assigned to forest fire risk and susceptibility [4], as well as post-fire studies [5,6]. To enhance firefighting success, risk-based possibilities and their management information should be seamlessly integrated into decision support systems through tactical and operational synthesis [7,8]. Achieving effective fire decision support necessitates a precise explication of uncertainties associated with the risk. Therefore, it is crucial to elucidate the uncertainties arising from spatial patterns, comprehend their origins, and make informed decisions regarding fire prevention and ignition probability [9,10].

With the increasing frequency of severe wildfires, the imperative to comprehend and predict fire behavior becomes pivotal for effective forest management. Pre-fire risk assessments assume a crucial role in anticipating fire behavior, directing resource allocation, and fortifying preparedness [11,12]. Pre-fire risk assessments encompass a myriad of factors contributing to fire behavior in forested areas, including forest type, fuel load, moisture content, weather patterns, and topography. Models such as the Fire Spread Probability Model and the Fire Potential Index leverage these inputs to estimate fire behavior and potential spread, forming the foundation for proactive fire management strategies.

In forest fire risk mapping, diverse methods are employed, including logistic regression [9], GIS-based decision support systems [13], maximum entropy [14], machine learning [15,16], artificial neural network analytics [17], analytical hierarchy process [18], and fuzzy analytic hierarchy process [19,20]. GIS-based Multi-Criteria Decision Analysis (MCDA) is an intelligent approach that transforms spatial and non-spatial data into information usable alongside the decision-maker’s judgment to facilitate critical decisions [21]. Analytical Hierarchy Process (AHP) and Fuzzy Analytical Hierarchy Process (FAHP) are commonly featured in the literature as GIS-based MCDA methods preferred for risk analysis in various sectors [22,23]. However, AHP’s susceptibility to subjectivity in determining criteria roles within the pairwise comparison matrix is acknowledged. To mitigate this issue, methods for improving AHP have been developed [24,25]. Fuzzy logic methods can be integrated into the post-AHP decision-making process to evaluate criteria and utilize fuzzy membership functions, enhancing result accuracy [21,23]. In FAHP fire risk analysis, despite numerous criteria constituting fire risk, it facilitates hierarchical weighting of criteria and their sub-criteria importance levels. This proves crucial for risk management, enabling the storage of a quantitative database essential for deciding on measures to minimize the impact of variables related to risk factors [19].

The mortality rate of vegetation is directly influenced by fire characteristics, which, in turn, are determined by the spatial distribution of varying weights assigned to forest fire risks. Determining tree mortality in various forested areas with different burn indexes [6] is crucial for establishing a sustainable foundation in forestry to support post-fire recovery [1]. The extent of mortality depends on several factors, including the physiological parts of the tree that have been damaged [26,27], burn ratio [1], burn severity [28], fire behavior factors, and specific tree characteristics [6,29]. Identifying distinct burning intensities across the burned area yields crucial data for strategically planning post-fire spatial treatments within the framework of sustainable forest management.

Post-fire analysis involves evaluating fire severity, burn patterns, and the effectiveness of suppression efforts. By comparing these outcomes with predictions made in pre-fire assessments, we can identify discrepancies and assess the accuracy of the assessment models. Numerous case studies exemplify situations where pre-fire risk assessments failed to accurately predict fire behavior due to unexpected changes in weather, ignition points, or fire behavior feedback. Several factors contribute to disparities between pre-fire predictions and post-fire outcomes in forest ecosystems. Rapid shifts in weather conditions, anomalies in ignition sources, and changes in fuel availability can substantially alter fire behavior. Additionally, uncertainties in input data quality, model assumptions, and calibration procedures can introduce errors into pre-fire risk assessments. Improving the accuracy of pre-fire risk assessments in forested areas requires addressing the unique challenges of these ecosystems. Incorporating real-time weather data, high-resolution fuel maps, and advanced fire behavior models tailored to forests can enhance prediction reliability. Regular validation and updating of assessment models using post-fire data are crucial for refining predictions. The occurrence of fires in the investigated area is primarily attributed to climate anomalies characterized by exceptionally high air temperatures and drought. It is crucial to consider these factors in the context of sustainable development goals related to land use and cover [30], as well as the quality and quantity of ecosystem services [31].

According to the EFFIS 2021 database, Turkey recorded the highest total area burned. Specifically, the Manavgat Fire stands out with the largest burned area among the recorded incidents in Turkey, the European Mediterranean, North Africa, and the Middle East [32]. The study aims to assess the risk of fire spread based on pre-fire behavior during the Manavgat Fire in 2021 and examine the post-fire situation by detecting burn severity using satellite images. In addition, the study delves into the precision of pre-fire risk assessments in predicting post-fire outcomes within forest ecosystems.

2. Materials and Methods

2.1. Study Area

The study focused on the Manavgat region in Antalya, characterized by a typical Mediterranean climate. This area is subject to hot and dry summers along with mild and rainy winters. Utilizing digital elevation model data provided by the General Directorate of Maps, the average elevation within the study area is 488.85. This is accompanied by an average slope value of 21.71 degrees. The predominant forest type in the region is Turkish red pine (Pinus brutia Ten.) and evergreen Mediterranean shrubs. Agricultural activities are concentrated in areas with lower slopes [33]. The research encompasses the fire zone that ignited on July 28, 2021, and persisted for approximately 10 days. Additionally, it includes various land cover and land use classes such as settlements, industrial and agricultural areas, rocky terrains, and wetland areas in the vicinity (Figure 1).

The anticipation of potential wildfires in forested areas and the subsequent analysis of their impacts constitute significant aspects. While interventions during a fire event may influence the extent of its spread, prior knowledge of the degree to which forested areas will be affected by the fire provides a substantial input to risk analyses. The utilization of land cover information emerges as a critical parameter in the estimation and evaluation of these predictions. In extensive forested areas affected by wildfires, remote sensing data stands out as a crucial source for the rapid generation and updating of such information. Furthermore, satellite imagery is commonly employed in the assessment of the consequences following a fire incident. In this study, the outcomes of a risk analysis conducted prior to the 2021 Manavgat wildfire were juxtaposed with the results of burn severity derived from satellite imagery. Figure 2 represents the general framework of the methodology employed in this study.

2.2. Data

In this study, data obtained from various sources has been amalgamated. Due to the extensive geographical coverage of the study area, satellite images with varying resolution characteristics were employed to extract information from the field. MODIS satellite images were utilized to assess the spread of the forest fire that occurred in the region. A Landsat satellite image was employed for determining surface temperatures, while a Sentinel-2 image was used to ascertain burn severity. To verify the accuracy of the results, high-resolution SPOT 7 and Pleiades satellite patterns, along with aerial photographs specific to the region, were utilized. Codes from the CORINE classification was used in the analysis of land uses. Additionally, ground surveys were conducted to validate the results, involving an examination of land cover/use conditions and the implementation of spectral measurements (Figure 3). The general characteristics of the data utilized in the study are presented in Table 1.

2.3. Methodology

2.3.1. Burn severity

The degree of burn in the Sentinel-2 satellite image was determined through classification using the Random Forest method [34]. 11519 training points that belongs to different land use/cover classes and 400 test points belonging to areas affected by fire in three different classes, named “Low Burn Severity”, “Moderate Burn Severity” and “High Burn Severity”, were produced from the fire zone. SPOT, Pleiades, and Aerial Photos were used for visual interpretation and production of spectral signatures. This approach enabled the creation of reliable and accurate training and test points, as well as an understanding of the spectral differences between the classes. Spectral signatures and statistical parameters also took into account in order to determine burn severity class definitions and thresholds between classes[35].

The burn severity classification has reached 0.81 of overall accuracy and the obtained results have been utilized as input for risk assessment.

2.3.2. Risk Assessment

Weather conditions, especially the daily and seasonal drying/cooling based on surface fuel moisture, have short and long-term effects [36]. Daily fire danger rating systems based solely on meteorological variables have even been developed [37]. Increasing high temperatures enhance evaporation (transpiration), rapidly reducing the moisture content of fuel types, creating suitable conditions for ignition temperature as a result [38].

The fuel-moisture content (FMC), defined as the ratio between the ten-hour-time-lag fine-fuel moisture and the extinction moisture, plays a crucial role in forest fires [39]. The flammability of the vegetation cover is largely dependent on the moisture content of fine dead fuels. It is the fuel type where moisture change occurs most rapidly. Therefore, the accuracy of moisture content in such fuels is a crucial component of fire management and the fire danger rating system, contributing to the shortening of ignition time [40] and controlling fire behavior, which is one of the key factors [41,42].

Topography interacts with climate in terms of fire behavior characteristics. In dry weather conditions, topography has less effect than in wet weather conditions. Regional or temporal climate variations that reduce fuel moisture make much of the land in various topographic positions susceptible to fire spread [43].

The distribution of flammable material characteristics in the field varies significantly, directly impacting fire risk and behavior in different spatial areas [44]. A crucial precondition for successful fire management, including risk assessment studies, is the accurate determination of fuel (flammable material) load and associated fuel characteristics [45]. In the risk assessment part of this study, forest structural characteristics and flammable material load data were used to incorporate flammable material characteristics. Flammable indices for surface and canopy fuels were developed using the amounts of dead surface and crown flammable material. Data from flammable material studies conducted in pine and hardwood forests throughout Antalya were used to determine flammable material indices [19,46]. The flammable material loads of existing pine and hardwood stands in the study area were calculated by taking the average loads of the same stands in the study conducted for Antalya (estimated). Flammable material quantities for pine and other tree and shrub species stands [47,48,49,50] were derived by interpolation from similar distinctive features, such as leaf biomass, stand density, developmental stage, rocky growing environment, of pine stands and maquis vegetation types. Coefficients ranging from 0.7 to 1.4 were used in the interpolation of flammable material quantities for mixed forest structures of needle-leaved and broad-leaved tree species in comparison to pine forests. It is assumed that broad-leaved forests have a higher flammable material load than needle-leaved forests.

2.3.2.1. The fuzzy AHP model (FAHP)

The criteria, sub-criteria, and their respective intervals for the FAHP risk assessment were established in a hierarchical manner. The weights assigned to the criteria, sub-criteria, and factors were computed through pairwise comparison matrices utilizing Chang’s extension analysis method [51] as cited in [19]. The fuzzy scale of relative importance, developed by [52], was employed to quantify the relative weights (see Table 2).

Figure 4. The weights of the criteria, sub-criteria and factors used in FAHP.

Figure 4. The weights of the criteria, sub-criteria and factors used in FAHP.

3. Results

In this study, a comparison was conducted between the risk assessment based on pre-fire data and the post-fire situation, specifically examining the analysis of outcomes from 266 selected test points located within the forest boundaries in the study area (Table 3). All primary factors demonstrated a heightened impact in areas with elevated burn severity. Notably, climate emerged as the most influential factor, constituting 67% of the overall effect. In prior risk determination using the FAHP method, climate had already been assigned the highest weight value, accounting for 32.2% among other factors. Topography emerged as the second-most influential risk factor in burn severity, securing the second-highest value at 27% among other factors in the FAHP assessment. The risk weights of additional environmental and stand structure factors, contributing to burn severity, followed a similar ranking pattern in the FAHP risk assessment.

Given that the 2021 fire in question was the largest in both the country and the European Mediterranean region, it is evident that climatic factors played a significant role in its expansion. Among the climatic risk factors, the cumulative effect of annual average temperature and flammable substance content contributed to 66% of the total weight, significally influencing the fires expansion. In terms of topography, terrain aspect held the highest weight at approximately 46%, while elevation had the highest weight at approximately 32%. In the FAHP risk assessment, Stand Fuel Loading Index (SFLI) and Canopy Fuel Loading Index (CFLI) were subsumed under stand structure, collectively representing 66% of the weight within this primary factor. When compared to burn severity levels, these indices exhibited higher accuracy in areas characterized by low and moderate burn severity.

Table 3. Main Factors, SFLI, CFLI, and complete Pre-Fire Risk Analysis Accuracy Analysis Scores.

Evaluated Factors	F-Score			Overall Accuracy
	Low Burn Severity	Moderate Burn Severity	High Burn Severity
Topographic Factors	0.2581	0.4110	0.5089	0.4167
Climatic Factors	0.4634	0.3710	0.6705	0.5288
Environmental Factors	0.0779	0.4184	0.4800	0.3936
Stand Structure Factors	0.3648	0.2074	0.4660	0.3640
SFLI	0.4953	0.3846	0.2439	0.4131
CFLI	0.4129	0.4533	0.2105	0.3846
Pre-Fire Risk Analysis	0.4959	0.2920	0.5432	0.4476

Table 3. Main Factors, SFLI, CFLI, and complete Pre-Fire Risk Analysis Accuracy Analysis Scores.

Evaluated Factors	F-Score			Overall Accuracy
	Low Burn Severity	Moderate Burn Severity	High Burn Severity
Topographic Factors	0.2581	0.4110	0.5089	0.4167
Climatic Factors	0.4634	0.3710	0.6705	0.5288
Environmental Factors	0.0779	0.4184	0.4800	0.3936
Stand Structure Factors	0.3648	0.2074	0.4660	0.3640
SFLI	0.4953	0.3846	0.2439	0.4131
CFLI	0.4129	0.4533	0.2105	0.3846
Pre-Fire Risk Analysis	0.4959	0.2920	0.5432	0.4476

To perform a more comprehensive evaluation of the results, both visually and statistically, two regions with distinct canopy characteristics were chosen within the study area (Figure 5).

In Case 1, the F-score, particularly for severe and moderate fire severity levels, demonstrated greater accuracy compared to low severity (Table 4). Case 1 encompasses the initial outbreak point of the fire (or the closest distance). The implication is that the fire advanced significantly from the ignition point. In Case 2, a reduction in high fire severity was noted, with a prevalence of more moderate-severity fires. Case 2 includes the Oymapınar Dam area and larger agricultural fields, which appear to have functioned as barriers in mitigating high fire severity.

Figure 5. Pre-Fire Risk Assessment Result within Case1,2 Boundaries and Post-Fire Burn Severity.

Figure 5. Pre-Fire Risk Assessment Result within Case1,2 Boundaries and Post-Fire Burn Severity.

One of the pivotal factors influencing the progression of fire in forested areas is the external interventions made to control the fire. The direction of spread during the days when the fire continued was evaluated in conjunction with the results obtained from MODIS images, as depicted in Figure 5, The figure illustra the course and severity of the fire from its initial outbreak until it was brought under control. The fire commenced on July 28^th, advancing with high intensity towards the south in the mountainous area within the first 24 hours. On the third day, as the fire predominantly moved towards the north in sporadic locations, it also made a move towards the east on the same day. By the 4^th day, the fire reached the Case 2 section, located further east. In this section, although the intensity decreased due to land use and the Oymapınar Dam, it remained active over the forested areas, progressing over the northeastern mountainous region of this section for an additional 3-4 days. During the same period, the fire also moved north. In this region, there are deep valleys and long mountain slopes. Generally, the fire remained active in the mountainous region during the last 3-4 days of the fire duration. It is understood that during these final 3-4 days, the fire was also effective in the region between the Case 1 and Case 2 sections. Overall, the variable topography in the northern regions facilitated the formation of local winds during the fire, and it was observed that the progression of the fire frequently changed according to the wind directions. The higher elevation in the northern regions, the prevalence of low-vegetation rocky areas, and the absence of Scots pine and hard-soiled vegetation cover contributed to easier control of the fire. The fire initially progressed towards the south in the mountainous area with strong northerly winds. The higher density of settlements and agricultural areas further south can be explained by lower flammable materials. On the third day, when the fire turned towards the north and then to the east, it could be attributed to finding new flammable material sources and the more effective topography. The threat posed by the fire to settlements by reaching south in the first two days shifted the firefighting organization to this region. On the 3^rd day, as the fire weakened and moved towards the north and then to the east on the same day, it was due to the presence of a weaker firefighting organization in this region.

Figure 6. Fire Directions based on MODIS Data.

Figure 6. Fire Directions based on MODIS Data.