1. Introduction
Global change includes increased temperatures and prolonged droughts, which create the ideal conditions for increasing the likelihood of spreading extreme forest fires [
1]. Forests present microclimates that strongly contrast with the climate outside them. Integrating microclimates in ecological research will promote the better understanding of forests’ biodiversity and functions related to climate and climate change [
2]. A challenge in understanding the fire-prone landscapes is their approach as coupled human and natural systems [
3]. Anthropic activities, such as those in the recreational and agricultural sectors and human settlements, are associated with high wildfire risk [
4]. For example, farmers consider burning essential for clearing vegetation or weeds, reducing acidity, and nutrient availability [
5]. However, in high-intensity fires, these burning practises in agriculture affect microfauna, evapotranspiration, the reproduction of plant species, nutrients, and soil fertility [
6].
Extreme wildfires are highly destructive agents within the ecosystem that impact the environment, wildlife habitats, and surrounding communities [
7]. Moreover, wildfires destroy trees, shrubs, and other vegetation types, leaving the area barren, and changing hydrological processes (i.e., infiltration, interception) [
8]. Wildfires can release large amounts of smoke and other pollutants into the air, causing health problems for humans and animals [
9]. The intense heat caused by wildfires alters the chemical and physical properties of the soil, making it less fertile and unable to support plant growth [
10].
Worldwide, recorded fires over time affect numerous ecosystems. For instance, in 2018, Washington, Oregon, Idaho, Nevada, and California recorded a burned area of 4000 hectares (ha) and 100 deaths [
11]. Meanwhile, the Australian mega-fires in 2019 originated from natural conditions, consumed 70,000 ha. and affected regions such as Queensland, New South Wales, Victoria, and Tasmania [
12]. The mega-fires of 2017 occurred in Pedrógão Grande (Portugal) and Gois, causing 64 deaths, with 200 individuals sustaining injuries and leaving a burned area of 45,000 ha [
13]. In a local context, the Loja canton in Ecuador has experienced several forest fires, with a burned area of 12,848 ha from 2010 to 2022 [
14].
In Ecuador, fires are mainly widespread during the dry season (from June to September), because solar radiation increases and produces hydric stress in vegetation and dried soil and lowers precipitation [
15]. The leading causes of wildfires in this country are human activities, such as agricultural burning, campfires, and discarded cigarette butts, as well as the effects of climate change, which are leading to drier and hotter conditions [
16]. The case study of this article comprises Vilcabamba parish in Loja Province in Ecuador, an Andean region in altitudes over the 1400 m.a.s.l. [
17].
Remote sensing tools are widely used to understand wildfire dynamics. However, these techniques present some limitations due to weather conditions associated with cloud cover, which are challenges with observations on terrestrial surfaces [
18]. However, multi-temporal analysis and machine learning based on remote sensing and statistical analysis (i.e., Logistic Regression, Multivariate Adaptive Regression Spline, Random Forest) facilitates the understanding of spatial, social dynamics and the temporality of fire events on the land surface [
19]. The time series data of spectral indices could be inconvenienced due to cloudiness, aerosols, sun, ozone, and dust [
20]. In this way, the application of the Cloud-gap-filled (CGF) method through Google Earth Engine (GEE) allows for efficient cloud removal, especially in Andean regions with high cloud cover [
21].
Spectral indices qualify the phenological behaviour of plant species, showing their vigour, development, and dynamics [
22]. The Normalised Difference Vegetation Index (NDVI) is used for the analysis of vegetation recovery, photosynthetic activity, mapping fire severity, crop yields, climate change, drought mapping, and the impact of flooding on crops [
23,
24]. The Normalised Burn Ratio (NBR) is one of the most used indices to assess fire severity, and its primary function is the detection of burned areas and vegetation status [
25].
This study focuses on the need to understand and prevent the effects of forest fires, a phenomenon that represents a critical threat to ecosystems, biodiversity and human communities. Fire severity levels provide a fundamental perspective for assessing environmental impact while developing susceptibility models to identify fire-prone areas. However, in Vilcabamba parish, there is no evidence of studies that analyse the severity of fires or government reports documenting the scars these events left. Likewise, no multi-criteria model has been developed to identify areas susceptible to forest fires. The available scientific literature is limited to quantifying fire frequency [
14]. Additionally, forest fire monitoring usually includes the application of quantitative techniques based on remote sensing tools, statistical algorithms or machine learning methods. However, a holistic approach considering both qualitative and quantitative approaches is needed for a better explanation of fire events and successfully establishing an action plan oriented to support decision-making processes. Therefore, the research question of this study is as follows: How does assessing fire severity and susceptibility contribute to establishing a proposed action plan for fire monitoring and prevention?
The aim of this study is to assess the severity and the susceptibility of a fire event in Vilcabamba parish by calculating spectral indices such as the NDVI, NBR, and Normalised Difference Moisture Index (NDMI) and applying the Analytic Hierarchy Process (AHP) method. We developed fire severity models with the delta Normalised Burn Ratio (dNBR), the delta Normalised Difference Vegetation Index (dNDVI), and a wildfire susceptibility model using the AHP approach. The integration of these remote sensing tools and the GIS-based approach allowed, as the development of the proposal of a fire action plan and the establishment of a conceptual model, for wildfire management and prevention to support decision making in Vilcabamba parish. To accomplish this aim, we used geoprocessing tools such as the Geographic Information System (GIS) and GEE. We used the AHP method for the fire susceptibility model, selecting the most representative fire triggers in the study area. We chose this method because it allows us to include experts and community criteria for decision making and has been widely applied for fire susceptibility analysis. The versatility of the AHP method lies in integrating multiple layers of environmental information, meteorological conditions, and even local socio-economic conditions for forest management in hard-to-reach areas [
26]. However, there are other methods, such as the Analytic Network Process (ANP), that are mainly applied for risk evaluations rather than susceptibility [
27]. Finally, we validated the models with the field data of fires recorded in 2019 in the fire perimeter and statistical algorithms.
We selected the fire event of 3 September 2019, which occurred southwest of the parish, and it is considered the most representative background of fire dynamics in the region. By including the entire parish in this study, we can establish escape routes, considering the proximity to health centres, water bodies, fire stations, and fires that occurred in the sector in the last decade. Human activities provoked the fire of 3 September 2019, and it is recognised as the most representative in this region because it spread across 1837.64 ha and was extinguished in three days by firefighters [
28]. Additionally, the Podocarpus National Park is located northeast of Vilcabamba parish, which has been declared a protected area of Ecuador according to the Ministry of Environmental, Water and Ecological Transition in Ecuador (MAE, Spanish acronym), a biodiversity hotspot in southern Ecuador with more than 560 bird species recorded [
29].
This article is of great relevance for both environmental management and territorial planning because it provides a solid scientific basis for understanding the patterns and effects of forest fires, allowing the identification of the most vulnerable areas and the factors that contribute to their occurrence. Furthermore, implementing an action plan promotes effective preventive strategies, reduces the risks of future fires, and promotes the sustainable management of natural resources, especially in regions of high ecological value and vulnerability, such as Vilcabamba parish.
2. Study Area
In this case study, we analysed the severity of the wildfire that occurred on 3 September 2019, in the southwest part of Vilcabamba parish in the Andean region of Ecuador (
Figure 1a) [
30]. This parish belongs to Loja Province (79°15′38.31″ W and 4°13′8.38″ S), which is located in the west part of the Catamayo-Chira basin (see
Figure 1b). The Podocarpus National Park lies in the northeastern part of Vilcabamba parish, and, at the southwest area, there is the wildfire perimeter (
Figure 1b. According to the National Secretary of Risks and Emergencies (SNGRE, Spanish acronym), 79 fires were recorded in Vilcabamba parish during the period 2011–2019 (pre-fire scene) [
31].
The Vilcabamba parish has a population of 5516 inhabitants [
32] and an area of 15,932.61 ha. In this parish, there are outcrop cineritic volcanic rocks. An amount of 46.53% of the territory has a mountainous relief with slopes ranging from 70% to 100%. On the other hand, 56% of the parish has entisol soils, 36% inceptisol soils, and only 1.86% alfisol soils [
33]. Vilcabamba parish has a temperate semiarid climate and distinct seasonal variations in altitudes from 1400 to 3750 m.a.s.l. [
17]. According to Köppen climate classification [
34], this sector has a mesothermal environment with a dry winter, considered a variety of the mesothermal humid climate type. This area experiences two main climate seasons (rainy and dry season). The rainy season typically occurs from October to May, while the dry season spans from June to September. According to the local weather stations (Malacatos (code M143), Quinara (code M145), Yangana (code M137), San Jose, and Porvenir stations), the average annual precipitation is around 875 mm, and the average yearly temperature is 19 °C. Finally, evapotranspiration is around 2268.1 mm/year.
The study area is characterised by its economy based mainly on agriculture and tourism. Agriculture is centred on cultivating coffee, sugar cane, fruits, vegetables, and products for local consumption and nearby markets. About 87% of the population is engaged in agriculture, forestry, hunting and fishing [
33]. However, these agricultural activities involve the removal of vegetation through the traditional use of fire, which, on numerous occasions, triggers forest fires [
35]. During the period 22–25 August 2024, a fire was recorded in Quilanga canton, Loja province. In this fire, 738 ha of pine forest and grassland was affected, adding to the 8600 ha recorded in 2024 in Loja [
36]. On the other hand, the tourism boom has encouraged the creation of inns, restaurants, and wellness centres, generating employment and, at the same time, contributing to changes in land use in the sector [
37].
5. Discussion
GIS-based models (e.g., AHP method) have been widely applied for fire susceptibility [
26,
80]. However, an integral approach has not been recorded to consider these models when developing a fire action plan. Therefore, this study highlights the importance of using remote sensing tools (e.g., spectral indices) and the AHP method to establish a fire action plan in the Andean region of Vilcabamba parish. The proposed fire action plan considered fires recorded in 2019 (pre-fire scene), fire hotspots from satellite data (VIIRS) in the fire perimeter of the study area, and a fire susceptibility model to delineate evacuation routes for firefighters to attend fire events within the study area and refuge zones with lower fire propagation. These methodologies allow us to establish a conceptual model for fire management and monitoring adapted to the particularities and needs of the Vilcabamba parish. The purpose is to contribute to decision making by the authorities and, at the same time, to be considered a preliminary study but one that can be applied to other regions with similar geographic–climatic conditions. Remote sensing tools are crucial in analysing preconditions, which can lead to a significant forest fire in Vilcabamba parish. Daily satellite scans also contribute to a vast dataset, enabling vegetation monitoring through time series analysis.
This study analysed the fire severity of the event of 3 September 2019. We analysed specifically the fire perimeter within the delineation of the Vilcabamba parish, located in the Andean region of Ecuador for several reasons: (i) the SNGRE recorded 87 historical events in the study area in the period 2011–2019 and 148 fire hotspots from VIIRS data, (ii) the availability of 78 fire validation points according to the VIIRS and SNGRE data in the year 2019 (pre-fire scene), (iii) the presence of the Podocarpus National Park and protected forest, in the northeast region of the parish, (iv) wind patterns that influence its mountainous topography and the prevailing regional weather patterns throughout the southern region, and (v) the drought conditions related to the increase in urban areas and the reduction in shrub and herbaceous vegetation evident in the post-fire scene. This area has evidence of a dry austral rainy season (from June to September), with rainfall lasting from December to April, extending into May.
In Ecuador, some analyses of wildfires used remote sensing techniques. For instance, in Imbabura province, a study provides comprehensive insights into the vegetation dynamics in a forest fire by applying NDVI values that show a significant decline in post-fire scenes [
81]. Meanwhile, another study developed a replicable approach for semi-automatically detecting forest fires and evaluating vegetation recovery post-fire through Landsat 8 (OLI) imagery in Quilanga canton (Loja province) [
82]. Moreover, in Chilla canton (El Oro province), a high performance of the NBRI (burned normalised index) identified the scar of the forest fire. Meanwhile, the NDVI determined a mean value of 0.15 showing low vegetation productivity [
83]. Finally, a study in Chimborazo found that dNBR using Landsat imagery delimited the fire perimeter and showed a 55% similarity with the delineation of the national database [
84].
Regarding multispectral imagery, this study showed that data obtained from Sentinel-2A are highly beneficial for identifying burned areas. For instance, a study on an island in the Aegean Sea (Greece) demonstrated the accuracy shared by Sentinel-2A and Landsat-8 by showing the high exactitude of the Sentinel-2A to map fire scars and the excellent performance of the NIR and mid-infrared bands of both sensors [
85]. On the other hand, Lo Conti et al. [
86] found a high bias in the surface analysis concerning cloudy and water body pixels in Sicily, Italy.
In this study, we use the cloud-gap-filled (CGF) method to fully fill data gaps in Sentinel-2A imagery of the NDVI in the period 2016–2022. This analysis allowed a better understanding of the dynamics of vegetation recovery in pre-fire and post-fire scenes by considering monthly data that allow us to relate this to the sector’s climatology and fire frequency incidence. The CGF method for NDVI time series is a widely used tool worldwide because it provides a bi-temporal analysis using satellite data that encompass several factors such as phenology, climate change, and geological-natural risk assessments [
87]. According to Karlsen et al. [
20], the time series vegetation index accurately depended on the atmospheric conditions instead of being highly structured, as shown in their study in Central Spitsbergen, Svalbard. Their findings showed extreme bias in 2018, where the algorithm had low detection. In the same way, the 2017 scene for Vilcabamba parish showed a low detection problem due to cloudiness. Still, the rest of the time series had an exceptional performance.
The outcomes in the vegetation recovery in the wildfire perimeter using the NDVI showed 9.63% in the low vegetation productivity category on the post-fire scene in 2019 and 0.99% for the post-fire scene, after a two-year recovery period (2021), showing the vegetation recovery in the region. However, the NBR index in the post-fire scene (2019) showed that the Vilcabamba parish presented 3.48% in high vegetation regrowth and 16.58% for the post-fire scene (2021). The effects of climate change in the study area include the predominant frequency of fires in dry periods, showing that, as precipitation decreases, temperatures rise significantly.
The Analytic Hierarchy Process (AHP) method applied in the Vilcabamba parish revealed that most settlers exhibit moderate fire probability, verified by SNGRE data, which indicates that the wildfire perimeter exhibits a high susceptibility to fire. Similarly, other regions, such as the states of Kashmir (JK), Himachal Pradesh (HP), and Uttarakhand (UK) of India demonstrated a moderate wildfire probability [
88]. On the other hand, a study in the Andean region in Ibarra (Ecuador) delineated areas within the paramo ecosystem exhibiting low fire severity, attributable to low temperatures and challenging accessibility [
65].
5.1. The Main Trends in the Results
This study identified the safety zones through the action plan for wildfire management that designates strategic areas according to the proximity to rivers, streams, water bodies (i.e., lakes, lagoons), and roads for impeding the kinetic energy of fire propagation. We used historical records of wildfires in Vilcabamba parish from the SNGRE and VIIRS data to define the shortest escape routes given the probability of fire occurrence. For escape route identification, we used the Road graph plugin that uses Dijkstra’s shortest path algorithm. The safety zones could enhance the effectiveness of fire management and containment efforts, considering the strategic position of firefighting resources to ensure a swift and efficient response to any wildfire within the region. Similarly, in western New Mexico, western Wyoming, and Northern California they used a GEE tool to delineate safety zones, allowing the users to draw a polygon of potential safety zones based on slope, wind, and fire conditions [
89]. In the same way, Debnath, P.A. [
90] used the shortest path tool to aim for the best route to respond to a medical emergency in India, indicating that the Road graph plugin shows that the shortest route is 41.75 km long, considering information about roads and junctions. The forest fire warning system should include surface water and groundwater (aquifer) management. Stable isotope techniques and the study of sociohydrology makes it possible to monitor water quality and consider the human–environment system, which is vital for the sustainability of natural resources [
91].
5.2. Methodological Limitations
We identified the following limitations: (i) the high cloudiness, due to its Andean region, affected index results within the study area and proved to be a significant challenge, adversely impacting the time series and vegetation index analysis; (ii) the proposal plan focused on a specific point of wildfire probability; and (iii) practical challenges in the proposal action plan such as land use conflicts, funding constraints, and community engagement issues.
5.3. Implications of the Results and Avenues for Future Research
For future studies, it is recommended that automatised fire georeferenced points are implemented and the closest escape route is drawn for the action plan. This study also suggests effective stakeholder coordination and adaptive management approaches for overcoming the practical challenges in the development of the action plan for fire management. This study recognises the implementation of machine learning tools that exhibit significant potential as a future research topic, contributing to identifying high-probability areas with unparalleled accuracy. However, it is crucial to note that the efficacy of these methods is contingent upon the quality of data obtained from satellites, a factor often hindered by cloud cover in most cases.
6. Conclusions
This study evaluated forest fire severity in Vilcabamba parish, Ecuador, on 3 September 2019, using the NBR, NDVI, dNDVI and dNBR indices, revealing that vegetation recovery depends on fire severity and anthropogenic activities in the area. Complementarily, we implemented a multi-criteria model for fire susceptibility, considering slope angle, slope aspect, elevation, distance to roads, distance to rivers, land use, isotherms and isohyets. The AHP approach compared fire severity models (dNDVI and dNBR) for mapping burned areas and vegetation recovery, providing a robust fire prevention and planning framework. These models were validated with statistical algorithms, proving their accuracy and functioning as critical criteria for the proposed action plan for fire management. The integrated approach of this article highlights the importance of implementing specific restoration strategies that provide a framework for post-fire management. The methodology applied can be replicated in regions with similar climates and socio-economic conditions.
The dNDVI and dNBR models explain wildfire dynamics over time for a more nuanced assessment of fire severity and its impacts on natural ecosystems by considering bi-temporal scenarios. This study implements GEE’s cloud-based platform to calculate spectral indices and streamline the data processing and analysis workflow. Its rapid computation of indices facilitates timely decision making and management actions. We combined the fire severity models and the GIS-based approach (AHP model) to establish a fire action plan that enhances the effectiveness of wildfire management efforts. This article provides a framework for decision making in preventing and mitigating wildfires, facilitating stakeholder coordination, and implementing targeted interventions.
The NDVI and NBR showed a notable correlation, offering technical insights into conditions preceding fire events and encompassing vegetation regrowth patterns. The wildfire perimeter exhibited 25.7% of vegetation regrowth. Pixels derived from the pre-fire NDVI covered 3076.21 ha (
Figure 6), while the NBR index amounted to 1018.56 ha, indicating burned areas characterised as moderate-to-low severity (
Figure 5). For the time series of the NDVI, we used the cloud-gap-filled method for filling gaps in data for a more precise analysis of the wildfire dynamics (
Figure S2). The NDMI showed that 55% of Vilcabamba parish shows a high water stress and only 1.26% with a very high canopy cover or no water stress (
Figure 6).
The wildfire susceptibility model provides valuable insights into the spatial distribution of the fire, enabling targeted interventions in high-risk areas (
Figure 11). The AHP method integrates diverse factors and prioritises adaptation efforts, enhancing the resilience of communities and ecosystems against the threat of wildfires, and it was verified with SNGRE field data, capturing the temporal evolution of fire events and their associated environmental variables.
The proposed action plan included the optimal escape routes and five refuge zones that help the group of firefighters as it enables them to identify the nearest river or water body (lake or lagoon) for decision-makers that could provide water for fire mitigation by air or land transport (
Figure 12). This analysis considered climate, topographic conditions (altitudes from 1400 to 3750 m.a.s.l.), and less densely populated areas conducive to mitigating the wildfire spread. This methodology offers governmental and institutional authorities the preliminary conceptual model (
Figure 4) to protect these designated areas, as an approach adaptable to places with similar geographic conditions.
