Comprehensive risk mapping and fire station optimization for forest fire management: An application in Antalya

Abstract

Forest fires are a significant environmental hazard, exacerbated by climate change, prolonged droughts, and increasing human activity in forested regions. These fires not only threaten biodiversity and ecosystems but also pose severe risks to human life, infrastructure, and natural resources. Efficient forest fire risk (FFR) mapping and strategic fire station placement are critical to mitigating these risks and ensuring rapid response in high-risk areas. However, traditional methods of fire risk assessment and resource allocation often fail to capture the complexity of large-scale forested regions, where environmental variables and fire occurrence probabilities vary widely. This thesis addresses these challenges by developing an integrated framework that combines advanced multi-criteria decision-making techniques and optimization models to enhance forest fire management. The thesis focuses on the Antalya region, offering both theoretical insights and practical applications to improve fire risk mapping and fire station deployment strategies in similarly vulnerable areas. This thesis presents a comprehensive framework for FFR mapping and fire station placement optimization, focusing on large, forested regions. By integrating multi-criteria decision-making techniques with advanced optimization models, the thesis addresses critical challenges in fire management, providing both theoretical advancements and practical applications in resource allocation. The thesis is structured into three main parts: a detailed literature review and taxonomy of FFR assessment methodologies, the selection and application of the Analytic Hierarchy Process (AHP) and Ordered Weighted Averaging (OWA) methods for the Antalya region, and the development and implementation of optimization models for fire station placement using Set Covering Model (SCM) and Maximal Covering Model (MCM). The first phase of this study involves a comprehensive literature review of existing FFR assessment methodologies. The review examines FFR mapping studies published between 2020 and 2022 and follows a systematic analysis process. As a result of this process, 170 studies are classified into 8 main categories and 23 subcategories. This detailed classification reveals both the application areas and methodological approaches of existing studies. Through this review, it becomes clear which methods are more commonly used in the literature such as AHP, and where significant gaps exist like optimization methods in FFR mapping. Techniques frequently used in FFR assessment studies include the AHP, which is chosen for this study due to its widespread use in risk mapping. Additionally, the less commonly used OWA method is included to provide a comparative analysis. These two methods are selected as the primary tools for developing FFR maps in this thesis. The review also identifies key factors, such as wind speed, temperature, and vegetation density, which are incorporated into the risk maps for the Antalya region. These factors are selected based on their relevance to both the literature and the local environmental conditions, ensuring that the risk maps accurately reflect the specific characteristics of the study area. In the second phase, the thesis develops detailed FFR maps for the Antalya region using both AHP and OWA. The region is divided into five distinct risk categories, and the accuracy of these maps is validated using Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC) metric. AHP highlights clear high-risk zones, particularly in coastal and central areas, while OWA distributes the risk more conservatively, identifying moderate risk across a broader area. These findings provide valuable insights into the strengths and limitations of each method for FFR assessment. Based on the validation results, the OWA-generated risk values are selected for the fire station placement optimization phase, as they provide a more balanced risk distribution for long-term planning. The third phase of the thesis focuses on fire station placement optimization. A literature review reveals that most existing optimization models focus on urban areas, leaving a significant gap in the literature for forested regions. To address this, SCM and MCM are selected for their effectiveness in facility location problems and their widespread use in fire station placement optimization. However, since these models are deterministic, they cannot fully capture the uncertainties inherent in forest fire occurrences. To overcome this, stochastic models are developed, incorporating scenario-based approaches to account for the variability and unpredictability of fire events in large-scale regions. A key finding of this thesis is that the MCM achieves similar coverage to the SCM, but with fewer fire stations. This result demonstrates that MCM is more efficient, requiring fewer resources without compromising coverage. Therefore, we recommend MCM for fire station placement in large forested areas where resource constraints are critical. Additionally, an algorithm is developed to identify expected fire locations using inverse transform sampling, focusing on high-risk areas identified by the risk maps. This approach reduces computational complexity by limiting the evaluation to high-risk zones, rather than analyzing every pixel in the study area. A clustering algorithm is then introduced to group expected fire locations, minimizing distances between them on two-dimensional risk maps. This step ensures that fire scenarios are both realistic and manageable, a key challenge in large-scale forested regions. To validate the effectiveness of the stochastic models, they are compared with deterministic solutions using expected fire locations. The Value of Stochastic Solution (VSS) is calculated to further validate the robustness of the models. Additionally, two approaches are used to determine the optimal number of scenarios: first, by evaluating the coverage results of the stochastic models, and second, by using the SCM to ensure that the scenario count is sufficient to achieve robust solutions. In the final phase, the optimization models are applied to the Antalya region using a 1 km by 1 km pixel resolution. The application of these models demonstrates that MCM efficiently achieves desired coverage with fewer fire stations compared to SCM, reinforcing the model's utility in large-scale planning. The stochastic models, in particular, show a high level of adaptability to the unpredictable nature of forest fires, highlighting their value in real-world fire management scenarios. These results underscore the importance of scenario-based planning for enhancing the robustness of fire station placement strategies, especially in regions with significant uncertainties related to fire occurrences. To further enhance the practical application of the optimization models, a Decision Support System (DSS) is developed. The DSS allows fire management agencies to simulate different fire station placements based on varying risk levels and fire scenarios. This tool provides strategic planning capabilities, enabling decision-makers to optimize resource deployment and improve response times in forested areas prone to wildfires. The DSS can also be adapted to other regions, demonstrating the scalability and flexibility of the proposed framework. The contributions of this thesis are multifaceted. First, it offers a comprehensive literature review that identifies key gaps in FFR assessment and fire station placement optimization, highlighting the need for more advanced methodologies. To address these gaps, the thesis introduces a robust framework that integrates both deterministic and stochastic elements, enhancing the flexibility and realism of fire station placement models in large forested areas. Scenario-based approaches are incorporated to improve traditional models like SCM and MCM, making them more applicable to the unpredictable nature of forest fires. The development of FFR maps for the Antalya region, validated through AHP and OWA, provides a strong foundation for fire management strategies. The introduction of clustering and scenario-generation algorithms further enhances the practicality and efficiency of fire station placement. Additionally, the DSS developed in this study offers actionable insights for policymakers, enabling more adaptive and effective strategies for fire prevention and response. This framework is particularly valuable for fire-prone regions like Turkey, where the Mediterranean climate, extensive forests, and limited resources pose significant challenges. The methodology developed, especially MCM, optimizes resource allocation under these constraints, making it suitable for countries with similar environmental and budgetary conditions. The user-friendly DSS allows fire management agencies to implement these strategies more efficiently, improving response times and resource deployment. Moreover, the scalability of this approach makes it applicable not only to Turkey but also to other Mediterranean regions and beyond, contributing to global fire management efforts. In summary, the thesis applies AHP and OWA for risk mapping, with OWA selected for its conservative risk distribution after validation. Optimization models like SCM and MCM are developed for fire station placement, with a stochastic approach supported by an algorithm for expected fire locations and clustering. Ultimately, MCM is recommended for the Antalya region due to its superior performance with fewer fire stations, demonstrating both resource efficiency and scalability.