A Multispectral Satellite-Based Integrated System for Monitoring Fire Disturbance and Recovery Dynamics in Forest Ecosystems

Abstract

Forest fires are an increasing environmental challenge in Southern Europe, requiring reliable tools for assessing both fire-induced disturbances and subsequent ecosystem recovery. This study presents an integrated satellite-based system for automated monitoring of post-fire forest dynamics. The system combines multispectral data from Sentinel-2 and Landsat (TM, ETM+, OLI, OLI-2) with thermal anomaly information from MODIS and VIIRS within a unified processing framework. It is structured into two modules: Post-Fire Disturbance (PFDMO) and Post-Fire Recovery (PFRMO). The methodology builds on a validated algorithm integrating the Disturbance Index (DI), Vector of Instantaneous Condition (VIC), and Direction Angle (DA), enabling automated multi-temporal analysis from fire detection to recovery assessment. The system was applied to three wildfire-affected areas in Bulgaria under different environmental conditions. Results reveal substantial spatial variability in disturbance and recovery, with PFDMO values ranging from −5.17 to +10.16 and PFRMO values from −2.25 to +7.40. The results demonstrate the applicability of the proposed system for monitoring post-fire forest dynamics and illustrate its potential to support informed decision-making in forest management, biodiversity conservation, and sustainable resource use. The main contribution of the system lies in the integration of disturbance and recovery assessment within a single automated and scalable workflow based on freely available satellite data.

1. Introduction

Forest fires represent a major global challenge, significantly affecting the sustainable management of forest ecosystems [1,2,3,4]. This issue is particularly relevant for Bulgaria, where forests cover approximately one-third of the national territory [5,6]. Over the past decade, forest fires in Bulgaria have become increasingly frequent and severe, affecting progressively larger areas, including protected zones [7,8,9,10,11]. Driven in part by climate change, these disturbances pose serious threats to biodiversity, carbon storage, and the provision of ecosystem services [12,13,14,15]. Their effective management requires timely and accurate information on both fire-affected areas and subsequent ecosystem recovery. In this context, remote sensing and multispectral satellite imagery have become essential tools for detecting forest disturbances, assessing fire impacts, and supporting post-fire recovery monitoring and planning [16,17,18,19,20,21,22,23,24].

Numerous studies have developed methodologies for detecting and monitoring forest disturbances [23,24,25,26,27,28,29,30]. Chuvieco et al. (2019) [15], for example, reviewed approaches for mapping burned areas using multispectral and thermal satellite imagery, emphasizing the value of time-series analysis for capturing both immediate fire effects and longer-term vegetation recovery. Pickell et al. (2016) [26] and White et al. (2016) [27] proposed frameworks based on Landsat time series for continuous monitoring of post-fire recovery over large forested regions. Miller and Thode (2007) [25] introduced the relative delta Normalized Burn Ratio (RdNBR) which has become a widely used index for quantifying burn severity. Other studies have explored the integration of multiple satellite sensors, including Sentinel-2, Landsat, and MODIS, to improve spatial and temporal coverage and enhance the accuracy of post-fire assessment [28,29,30].

More recently, attention has shifted from single-purpose methods toward integrated remote sensing-based systems for comprehensive post-fire assessment. Earlier studies, including Boschetti et al. (2015) [31], Mitri and Gitas (2004) [32], Veraverbeke et al. (2011) [33], Hawbaker et al. (2017) [34], and Giglio et al. (2018) [35] demonstrated the potential of combining multisensor satellite data, time-series analysis, and semi-automated workflows for burned-area mapping, fire severity assessment, and post-fire monitoring.

More recent contributions have advanced toward increasingly automated and decision-oriented frameworks. For example, Chen et al. (2025) [36] developed a regional wildfire mapping workflow based on Sentinel-2/Landsat integration, semi-automatic sample generation, GEOBIA, and machine-learning classification for burned-area and burn-severity assessment [36]. Prodromou et al. (2025) [37] proposed a remote-sensing-based spatial decision support tool for prioritizing post-fire restoration actions in Mediterranean ecosystems, while Cristal et al. (2025) [38] presented a post-fire environmental assessment framework integrating remote sensing and multi-criteria analysis to estimate soil erosion risk and vegetation recovery potential. Recent studies have further extended this direction by combining post-fire severity and recovery analysis based on Sentinel-2 time series and spectral indices, as illustrated by Cubas Sanchez et al. (2026) [39], or by linking post-fire monitoring to rehabilitation-oriented assessment frameworks, as in Kaloudis et al. (2025) [40]. The latest studies have increasingly focused on integrated post-fire monitoring frameworks that combine multisensor satellite data, machine-learning techniques, and decision-support systems to assess both disturbance and recovery processes [41,42,43,44,45]. These approaches demonstrate a shift toward automated, scalable, and operational systems that integrate fire detection, burn severity mapping, and vegetation recovery analysis within unified workflows.

These developments provide an important methodological basis for integrated post-fire monitoring and demonstrate the increasing relevance of automated, scalable, and decision-oriented remote sensing workflows.

The proposed system builds upon a previously introduced algorithm for post-fire forest regrowth monitoring [46]. This algorithm is based on the combined use of three complementary components: the Disturbance Index (DI) [47], the Vector of Instantaneous Condition (VIC) [48], and the Direction Angle (DA) [49]. The DI facilitates the identification of pixels deviating from average forest conditions toward disturbed states [47,48,49,50], while the VIC and DA, derived from the tasseled cap transformation (TCT) [51,52,53], characterize the current ecosystem condition and its deviation from the pre-fire state [54,55]. Their integration enables both the quantitative assessment of post-fire regrowth intensity and the detection and quantification of fire-induced disturbances, thus providing the methodological basis for the two modules of the system.

In the present study, this previously developed and validated algorithm is implemented within an integrated framework that automates the full workflow of post-fire monitoring. The system performs sequential tasks including wildfire detection based on thermal anomaly data, acquisition and filtering of multispectral satellite imagery, multispectral processing, and disturbance and recovery assessment. Based on predefined criteria, it automatically selects appropriate Landsat and Sentinel-2 imagery from local or remote repositories, ensuring that only suitable datasets are used for further analysis.

Despite the considerable progress in remote sensing-based approaches for wildfire monitoring, several limitations remain in existing integrated systems. Many frameworks focus predominantly on either disturbance assessment or post-fire recovery, rarely addressing both processes within a unified and operational architecture. In addition, current approaches often rely on semi-automated workflows, limited multisensor integration, or region-specific implementations, which constrain their scalability and practical applicability. These limitations highlight the need for automated systems capable of integrating multi-source satellite data and providing consistent, multi-temporal assessment of both disturbance and recovery dynamics.

In response to this gap, the present study implements a previously developed and validated algorithm within a unified operational framework for automated post-fire disturbance and recovery assessment. By combining disturbance detection and recovery monitoring within a single processing architecture, the proposed system addresses a key limitation of existing approaches. Its design emphasizes automation, scalability, and the use of freely available satellite data, enabling consistent and transferable application across diverse environmental conditions. The system is demonstrated in multiple fire-affected forest areas with varying vegetation types, topographic settings, and fire regimes, illustrating its applicability for post-fire monitoring and decision support in sustainable forest management.

2. Materials and Methods

2.1. Study Area

The proposed system for post-fire disturbance and recovery assessment was implemented in three forested study areas affected by wildfire, in order to demonstrate its applicability under different environmental conditions. Three test areas meeting predefined selection criteria were identified as appropriate for the implementation of the developed methodology. The selection criteria included the occurrence of a documented forest fire during the period 2020–2024; representativeness in terms of vegetation type, topography, and burn severity; data availability, including multispectral satellite imagery, vector fire perimeters, and orthophoto data; and spatial diversity, with study areas located in different geographical regions (Slavyanka Mountain, the Struma River valley, and the Sakar region).

For the purpose of study areas selection, the following datasets were utilized: a forest fire database covering the period 2020–2024 [7]; multispectral satellite imagery from Landsat 8/9 OLI and Sentinel-2 MSI, including multi-temporal scenes acquired before and after fire events; base maps and orthophoto imagery; a digital elevation model (DEM) used for slope and aspect analysis; and additional thematic layers, such as habitat types and Natura 2000 protected areas [56].

The application of these criteria led to the identification of three representative fire-affected areas, which were subsequently selected as case-study sites for the implementation of the proposed system and the demonstration of its applicability under different environmental conditions. The selected study areas include a wildfire on Slavyanka Mountain, a wildfire on Maleshevska Mountain in the vicinity of the village of Strumyani, and a wildfire on Sakar Mountain near the town of Svilengrad (Figure 1).

The selection of the three study areas is based on their representativeness of key environmental gradients, vegetation types, and fire regimes characteristic of forest ecosystems in Bulgaria and the broader Mediterranean region. The chosen sites capture variability in topography (mountainous, hilly, and low-relief terrains), dominant vegetation (coniferous, broadleaf, and mixed formations), and fire behavior, including differences in fire intensity and spatial extent. In addition, all study areas correspond to recent wildfire events (2024) with well-documented fire perimeters and sufficient availability of high-quality multispectral satellite data, ensuring consistency in data processing and analysis. The inclusion of areas located within or overlapping Natura 2000 sites [56] further enhances their ecological relevance. At the same time, the application of a uniform methodological framework and consistent input data (primarily Sentinel-2 imagery) ensures comparability of results across the selected cases. This combination of representativeness, data availability, and methodological consistency provides a sound basis for demonstrating the applicability and transferability of the proposed system under diverse environmental and fire-related conditions.

The main characteristics of each wildfire event and the associated study site are outlined in the following subsections.

2.1.1. Slavyanka Mountain

In August 2024, an extensive wildfire affected forested areas of Slavyanka (Ali Botush) Mountain, located along the Bulgarian–Greek border (Figure 1b,

Table 1. Key characteristics of the study areas, including location, fire occurrence period, burned area, dominant vegetation, and topographic features.

Study Area	Region	Fire Period	Burned Area	Dominant Vegetation	Topography
Slavyanka	SW Bulgaria, border with Greece	August 2024	2000 ha	Mixed coniferous and broadleaf forests	Mountainous, steep slopes
Maleshevska	SW Bulgaria, Blagoevgrad Province	July 2024	3000 ha	Drought-tolerant broadleaf species and shrublands	Hilly to pre-mountains
Sakar	SE Bulgaria Haskovo Province	July 2024	10,000 ha	Oak-dominated formations and shrublands	Low to mid-elevation mountainous

). The event was associated with prolonged hot and dry conditions and resulted in significant forest damage on both sides of the border, with several thousand decares impacted within Bulgarian territory [58]. The fire spread into remote and mountainous areas, including forest stands within protected zones. Affected areas overlap with Natura 2000 sites [56] designated under both the Birds (2009/147/EC) (Slavyanka—BG0002078) [59] and Habitats (92/43/EEC) Directives (Sreden Pirin—Alibotush—BG0001028) [60], highlighting the high conservation value of the impacted ecosystems.

The study area is characterized by complex mountainous terrain, steep slopes, and limited accessibility, which complicate both fire suppression efforts and post-fire field assessments. The region supports predominantly coniferous and mixed forest stands adapted to mountainous conditions, making it particularly sensitive to high-intensity fire events. These characteristics make the area well suited for demonstrating the applicability of remote sensing-based approaches to post-fire disturbance and recovery monitoring.

2.1.2. Maleshevska Mountain

In July 2024, a large wildfire affected forested areas of Maleshevska Mountain, in the vicinity of the village of Strumyani, southwestern Bulgaria (Figure 1c, Table 1). The fire occurred during a period of elevated temperatures and prolonged drought, leading to rapid fire spread across heterogeneous terrain. The affected area includes a mosaic of forested and shrub-dominated ecosystems distributed over complex topography, characterized by varying slope and aspect conditions.

The fire impacted both managed forest stands and semi-natural habitats, some of which are located within or adjacent to protected areas of high ecological value. Parts of the burned area overlap with the Natura 2000 sites Kresna (BG0002003), designated under the Birds Directive (2009/147/EC) [59], and Kresna–Ilindentzi (BG0000366), designated under the Habitats Directive (92/43/EEC) [60]. Both sites are part of the Natura 2000 ecological network [55] and are designated for the conservation of valuable bird species and priority habitats of European importance. Therefore, assessing post-fire disturbance and recovery dynamics in this region is essential for supporting biodiversity conservation and sustainable ecosystem management.

2.1.3. Sakar Mountain

In July 2024, a wildfire affected extensive areas of the Sakar region near the town of Svilengrad, southeastern Bulgaria (Figure 1d, Table 1). The fire developed under hot and dry summer conditions and spread rapidly across low-altitude terrain dominated by mixed forest, shrubland, and grassland ecosystems. Compared to the mountainous study areas, the Sakar region is characterized by gentler relief and more open landscapes, which contributed to different fire behavior and spatial burn patterns.

The affected area includes a heterogeneous mosaic of natural and semi-natural habitats, some of which are subject to conservation and land-use regulations. The entire burned area falls within the Natura 2000 site Sakar (BG0000212), designated under the Habitats Directive (92/43/EEC) [60], while approximately half of the territory also overlaps with the Natura 2000 site Sakar (BG0002021), designated under the Birds Directive (2009/147/EC) [59]. The presence of these protected Natura 2000 sites further highlights the ecological significance of the area and the importance of reliable methods for monitoring wildfire impacts and subsequent ecosystem recovery. The contrasting environmental conditions of the Sakar region provide an important comparative case for demonstrating the applicability of the proposed integrated system across diverse forest types, topographic settings, and fire regimes.

The selected study areas provide a representative basis for demonstrating the applicability of the proposed integrated system under diverse environmental and fire-related conditions. Their inclusion supports comparative analysis across different forest types, topographic settings, and fire regimes.

2.2. Data Used

The integrated system is designed to operate with multiple satellite data sources, including Sentinel-2, Landsat, and thermal anomaly products such as MODIS and VIIRS, depending on data availability, temporal coverage, and the specific requirements of the monitoring task. In the present study, however, the system is demonstrated using Sentinel-2 multispectral imagery as the primary input dataset for post-fire disturbance and recovery assessment in the selected test sites. This choice was motivated by the availability of suitable cloud-free observations, the high spatial resolution of Sentinel-2 data, and their temporal correspondence with the investigated fire events. Therefore, the present implementation represents a case-specific demonstration of the broader system architecture, while the system itself remains applicable to additional multisensor inputs depending on data availability and monitoring requirements.

In the present implementation, post-fire forest disturbance and recovery were assessed across the three selected sites using high-resolution, multitemporal Sentinel-2 multispectral imagery, which enables consistent monitoring of fire-induced changes and subsequent vegetation dynamics. The assessment of post-fire forest disturbance and recovery was carried out for the period 2024–2025, focusing on multi-temporal satellite observations acquired during the forest growing season and ensuring minimal cloud and shadow contamination. Multispectral Sentinel-2 imagery (

Table 2. Satellite datasets used for post-fire forest disturbance and recovery assessment.

	Slavyanka	Maleshevska	Sakar
Sentinel 2 A		10 July 2024	8 July 2024
Sentinel 2 A		31 July 2024	15 June 2025
Sentinel 2 B	16 July 2024	26 July 2024	23 June 2024
Sentinel 2 B	15 August 2024	10 July 2025
Sentinel 2 B	10 August 2025

), with spatial resolutions ranging between 10 and 60 m, was obtained from the Copernicus Open Access Hub [57] to support the assessment of post-fire forest disturbance and recovery processes. The Sentinel-2 imagery used in this study was acquired as Level-2A products, providing atmospherically corrected surface reflectance data. Additional preprocessing included the selection of cloud-free scenes and the exclusion of images affected by cloud and shadow contamination to ensure data quality and consistency.

2.3. Methodology

The methodology is implemented through two modules: the Post-fire Disturbance Module Output (PFDMO) and the Post-fire Recovery Module Output (PFRMO), which support the assessment of fire-induced disturbances and post-fire recovery dynamics in forest ecosystems. It builds upon a previously developed algorithm [46], implemented here within an integrated system for automated post-fire monitoring. The workflow combines multispectral satellite imagery, thermal anomaly data [61,62], and forest cover information, including the Copernicus High Resolution Layers (HRL) Forest Dominant Leaf Type (DLT) raster [63]. In the present implementation, the methodology is demonstrated primarily using Sentinel-2 imagery together with the corresponding supporting datasets.

Figure 2 presents a schematic representation of the workflow, highlighting the main processing steps of the PFDMO and PFRMO modules. The specific operations involved are described in the subsequent sections.

2.3.1. Satellite Data Selection and Filtering

The initial stage of the processing algorithm involves querying and acquiring data from external satellite data repositories to identify wildfire events, using sources such as MODIS, VIIRS, or alternative datasets. Subsequently, information on the spatial coordinates and temporal occurrence of detected thermal anomalies is extracted and utilized to query Landsat and Sentinel-2 image archives, stored either locally or in remote repositories, for further analysis.

The retrieved Landsat/Sentinel-2 imagery is then filtered according to a predefined set of criteria in order to select only those scenes suitable for subsequent processing. Image selection is based on the following conditions:

• Acquisition within the forest growing season, typically defined between 1 July and 30 August;
• Minimal atmospheric interference, ensuring the use of cloud-free imagery with negligible cloud shadow effects over the study areas;
• Accurate delineation of the area of interest (AOI), defined by the fire-affected area (fire scar), which is identified through the calculation of spectral indices such as the Normalized Burn Ratio (NBR) or the DI.

2.3.2. Forest Area Delineation and Filtering

This step aims to spatially intersect the identified wildfire events with forest cover data and to subsequently filter the resulting dataset based on forest type, distinguishing between broadleaved and coniferous formations. For this purpose, the Copernicus HRL Forest DLT product [63] is employed to generate a forest mask and to exclude non-forest areas from the PFDMO and PFRMO processing workflow.

2.3.3. Data Resampling and Spatial Harmonization

In this step, the spectral bands used as input data are resampled, where necessary, to a common spatial resolution (10, 20, or 30 m) in order to ensure consistency and compatibility for subsequent processing.

2.3.4. Input Satellite Data and Spectral Band Configuration

The PFDMO and PFRMO products are generated using multi-temporal, stacked multispectral imagery acquired from Landsat 5 TM, Landsat 7 ETM+, Landsat 8 OLI, Landsat 9 OLI-2, and Sentinel-2 MSI (A and B platforms). For Landsat 5 TM and Landsat 7 ETM+, the input data consist of stacked spectral bands B1, B2, B3, B4, B5, and B7. In the case of Landsat 8 OLI and Landsat 9 OLI-2, the selected bands include B2, B3, B4, B5, B6, and B7. Sentinel-2 MSI imagery incorporates a broader spectral configuration, utilizing bands B1 through B12, including the red-edge and shortwave infrared bands (B5, B6, B7, B8A, B11, and B12), which are particularly relevant for vegetation condition and burn severity assessment.

2.3.5. Spectral Transformation Using the Tasseled Cap Approach

The objective of this processing step is to apply the TCT to the input multispectral imagery. Since the transformation is sensor-specific, different coefficient matrices are employed for each dataset, including Landsat 5 TM, Landsat 7 ETM+, Landsat 8 OLI, Landsat 9 OLI-2, and Sentinel-2 MSI.

The orthogonal transformation coefficients applied to Landsat 5 TM are based on the formulation by Crist and Cicone, 1984 [52], while those for Landsat 7 ETM+ follow Huang et al., 2002 [64]. For Landsat 8 OLI and Landsat 9 OLI-2, the coefficients proposed by Baig et al., 2014 [65], are used, whereas the Sentinel-2 transformation relies on the coefficients derived by Nedkov, 2017 [66].

The transformation results in a set of orthogonal components representing key biophysical characteristics of the land surface, namely Tasseled Cap Brightness (TCB), Tasseled Cap Greenness (TCG), and Tasseled Cap Wetness (TCW). In the subsequent step, each of these components is further decomposed for use in the disturbance and recovery analysis.

2.3.6. Normalization of Tasseled Cap Components

The objective of this step is to derive normalized values of the Tasseled Cap (TC) components in order to ensure radiometric consistency across the dataset. Initially, the mean values and standard deviations of each TC component are computed. Subsequently, normalization is performed to standardize the spectral response and reduce the influence of radiometric variability.

The normalization is implemented as follows:

$${}_{n}TCB={\displaystyle \frac{TCB-\mu TCB}{\sigma TCB}}$$

(1)

$${}_{n}TCG={\displaystyle \frac{TCG-\mu TCG}{\sigma TCG}}$$

(2)

$${}_{n}TCW={\displaystyle \frac{TCW-\mu TCW}{\sigma TCW}}$$

(3)

where µTCB, µTCG, and µTCW denote the mean values of the respective TC components, and σTCB, σTCG, and σTCW represent their corresponding standard deviations, calculated for a reference class of mature forest. Accordingly, _nTCB, _nTCG, and _nTCW express the normalized TC Brightness, Greenness, and Wetness components relative to the statistical characteristics of this reference class.

The mature forest class used for normalization is delineated based on forest cover information derived from the Copernicus HRL, specifically the Forest DLT raster [62].

2.3.7. Calculation of DI for Disturbance Assessment

The purpose of this step is to compute the DI as an integrated indicator of post-fire forest disturbance. Following the normalization procedure, the three TC components are linearly combined to derive DI [46], according to the following expression:

$$DI={}_{n}TCB-({}_{n}TCG+{}_{n}TCW)$$

(4)

This formulation captures the contrast between brightness and the combined greenness and wetness components, providing a quantitative measure of disturbance intensity.

2.3.8. VIC Estimation

The objective of this step is to estimate the VIC based on the normalized TC components. The VIC is calculated using the established formulation [47,53,54], which integrates the spectral information into a single metric representing the instantaneous state of the ecosystem:

$$VIC=\sqrt{{{}_{n}TCB}^2+{{}_{n}TCG}^2+{{}_{n}TCW}^2}$$

(5)

The computation of VIC constitutes an intermediate step required for the subsequent derivation of the DA.

2.3.9. DA Calculation

Following the estimation of VIC, the DA is calculated to quantify the angular deviation between the greenness component and the overall spectral condition vector. The DA is defined as follows [53,54]:

$$DA={\cos }^{-1}({\displaystyle \frac{{}_{n}TCG}{VIC}})$$

(6)

This metric provides an additional geometric interpretation of spectral changes associated with disturbance and recovery processes.

2.3.10. Classification of Post-Fire Disturbance and Recovery Patterns

The objective of this step is to integrate the DI and DA through raster-based calculations in order to generate a classified output representing post-fire ecosystem dynamics. The PFDMO and PFRMO products are derived by combining the values of DI and DA, resulting in a composite indicator that captures both the magnitude and the structural characteristics of disturbance and recovery processes.

The resulting raster is subsequently classified into thematic categories reflecting different levels of disturbance severity and vegetation recovery, forming the outputs of the disturbance and recovery modules (PFDMO and PFRMO). This classification enables a spatially explicit interpretation of post-fire ecosystem conditions and facilitates comparative analysis across study areas.

The classification thresholds applied for PFDMO and PFRMO are based on values previously derived and validated within the framework of the underlying algorithm. Specifically, the threshold ranges were established in a prior study [46] through empirical analysis and accuracy assessment of post-fire disturbance and recovery classes. These thresholds reflect the statistical behavior of DI- and DA-based metrics and their relationship to vegetation condition. Higher PFDMO and PFRMO values correspond to increased disturbance severity, while lower or negative values indicate low or negligible disturbance. Their use in the present study ensures methodological consistency and supports the transferability and reliability of the classification results.

2.3.11. Workflow Automation and User-Defined Settings

From an operational perspective, the proposed system is designed to minimize user intervention once the analysis has been initialized. The end user is required to define the AOI and the temporal context of the wildfire event, which determine the spatial and temporal scope of the analysis.

After initialization, the workflow follows the predefined processing logic illustrated in Figure 2. This includes selection of suitable input products, filtering according to image suitability criteria, masking of non-forest areas, spatial harmonization of the input data, multispectral transformation, normalization of the derived components, calculation of the DI, VIC, and DA metrics, and classification of the final PFDMO and PFRMO outputs. In this way, the system supports a standardized processing sequence for post-fire disturbance and recovery assessment based on multispectral satellite imagery.

The image suitability criteria are applied during the filtering stage and include correspondence to the temporal context of the fire event, acquisition within the forest growing season, minimal cloud and cloud-shadow contamination, and sufficient spatial coverage of the AOI. Additional filtering is performed through the application of the forest mask, which excludes non-forest areas from further processing. These conditions ensure that only scenes appropriate for subsequent analysis are included in the workflow.

In its standard configuration, the system operates with predefined filtering logic and processing steps, while user intervention is limited to initialization of the analysis and, where necessary, adjustment of the spatial or temporal search extent. Thus, the workflow should be regarded as automated in its internal processing sequence, while remaining dependent on a limited set of user-defined inputs that specify the application context.

2.4. Validation Background of the Algorithm Implemented in the System

The algorithm implemented in the integrated system has been previously developed and validated in earlier studies [46], based on an accuracy assessment framework originally introduced by Avetisyan et.al. (2023) [18]. In the Remote Sensing study [46], the validation procedure was applied to vegetation recovery intensity ranges in three separate case-study sites distinct from those analyzed in the present paper, with an average overall accuracy of 62.1%. Producer’s accuracy was particularly high at the upper and lower ends of the regrowth range, reaching 93.5% for the high regrowth intensity and 75.5% for the low regrowth intensity. In the Fire study [18], the disturbance-related performance of dDI was evaluated in comparison with dNBR and dNDVI, showing better overall accuracy for post-fire disturbance assessment than the compared indices. In the present study, these previously published validation results are not reproduced in detail; instead, they are cited to document the previously established performance of the algorithm implemented in the integrated system.

In this context, the present implementation should also be considered in relation to the characteristics of currently available external fire products. Burned-area products such as European Forest Fire Information System (EFFIS) provide important harmonized regional-scale information, but they are also fundamentally based on remote-sensing workflows using MODIS, VIIRS, and Sentinel-2 imagery, and are intended primarily for European-scale burned-area estimation [67]. As a result, comparison with such products may support the interpretation of broad spatial agreement in disturbance patterns, but does not necessarily provide a fully independent benchmark for a system based on similar multispectral observations [67]. More generally, differences among burned-area products may arise not only from the burned signal itself, but also from differences in sensor characteristics, spatial resolution, temporal aggregation, and mapping logic [67,68].

For post-fire recovery, direct comparison is further constrained by the absence of a single harmonized classification framework with universally accepted recovery categories and thresholds [68]. Existing recovery-oriented approaches differ in whether they represent short-term post-fire greenness increase, spectral return toward pre-fire conditions, vegetation-condition trajectories, or broader ecological and structural regeneration [68]. This conceptual heterogeneity is evident even in operational post-fire frameworks such as Monitoring Trends in Burn Severity (MTBS), where “increased greenness” denotes a short-term vegetation response rather than a direct equivalent of forest regrowth or ecosystem recovery in a broader sense [69]. Consequently, direct category equivalence across recovery-oriented products and studies remains methodologically limited [68,69].

3. Results

3.1. Post-Fire Disturbance and Recovery Assessment on Slavyanka Mountain

The results for Slavyanka Mountain case study illustrate the spatial patterns and intensity of post-fire disturbance, as well as the subsequent vegetation recovery dynamics derived from the application of the proposed integrated system. Figure 3 presents the spatial distribution and intensity of post-fire disturbance (PFDMO) immediately after the fire event on Slavyanka Mountain (Figure 3a), together with the subsequent vegetation regrowth one year after the wildfire (Figure 3b). As shown in Figure 3, the values of the post-fire classified thematic raster exhibit substantial variability, ranging from −1.46 to +6.85 for PFDMO and from −1.28 to +6.26 for PFRMO, indicating a pronounced heterogeneity in both disturbance intensity and recovery response across the study area.

3.2. Post-Fire Disturbance and Recovery Assessment on Maleshevska Mountain

The following section presents an analysis of post-fire disturbance (PFDMO) and subsequent vegetation recovery (PFRMO) on Maleshevska Mountain, highlighting the spatial patterns and temporal dynamics observed in the affected area. Figure 4 illustrates the spatial distribution and magnitude of post-fire disturbance (PFDMO) in Maleshevska Mountain immediately following the fire event (Figure 4a), alongside the subsequent patterns of vegetation regrowth observed one year after the wildfire (Figure 4b). As illustrated in Figure 4, the PFDMO and PFRMO values display considerable variability, ranging from −1.00 to +3.89 for PFDMO and from −1.98 to +7.40 for PFRMO, reflecting marked spatial heterogeneity in both disturbance intensity and vegetation recovery across the study area.

3.3. Post-Fire Disturbance and Recovery Assessment on Sakar Mountain

The following subsection examines post-fire disturbance (PFDMO) and subsequent vegetation recovery (PFRMO) on Sakar Mountain, with particular emphasis on the observed extremes and their spatial expression across the test sites. Figure 5 depicts the distribution and magnitude of PFDMO immediately after the fire event (Figure 5a), together with the corresponding vegetation recovery patterns one year later (Figure 5b). Notably, PFDMO values exhibit the widest range among the analyzed case studies, varying from −5.17 to +10.16, while PFRMO values range between −2.25 and +5.49, underscoring pronounced spatial contrasts in disturbance severity and subsequent recovery dynamics within the study area.

To complement the visual interpretation of the PFDMO and PFRMO maps, a quantitative assessment was performed for all study areas. Statistical metrics, including minimum, maximum, mean, and standard deviation values, were calculated, together with area proportions of different disturbance and recovery classes (low, moderate, and high). Furthermore, comparative indicators between disturbance and recovery were derived to quantify post-fire dynamics. These results are summarized in

Table 3. Quantitative statistics of post-fire disturbance (PFDMO) and recovery (PFRMO) across study areas.

Study Area	Indicator	Min	Max	Mean	Std. Dev.	Low Level (%)	Moderate Level (%)	High Level (%)
Slavyanka	PFDMO	−1.46	6.85	2.31	1.74	28.4	46.7	24.9
	PFRMO	−1.28	6.26	2.05	1.52	35.2	44.1	20.7
Maleshevska	PFDMO	−1.00	3.89	1.87	1.02	31.6	52.3	16.1
	PFRMO	−1.98	7.40	2.64	1.95	29.8	40.5	29.7
Sakar	PFDMO	−5.17	10.16	3.95	2.86	22.7	38.4	38.9
	PFRMO	−2.25	5.49	2.18	1.63	33.5	47.2	19.3

and

Table 4. Comparative metrics between disturbance and recovery.

Study Area	ΔMean (PFRMO–PFDMO)	Recovery Ratio (%)	Interpretation
Slavyanka	−0.26	88.7	Limited recovery
Maleshevska	+0.77	141.2	Strong heterogeneous recovery
Sakar	−1.77	55.2	Partial recovery with persistent disturbance

and provide a more comprehensive basis for cross-site analysis and interpretation.

The quantitative analysis confirms substantial variability in disturbance and recovery across all study areas (Table 3 and Table 4). Sakar Mountain exhibits the highest disturbance intensity (mean PFDMO = 3.95), while Maleshevska shows the most pronounced recovery dynamics (recovery ratio > 140%). Slavyanka demonstrates comparatively limited recovery within the first year after fire.

4. Discussion

This section discusses the key findings of the study, interpreting the observed patterns of post-fire disturbance and vegetation recovery in the context of the applied methodology and the specific environmental conditions of the study areas. In this regard, the results reveal a broad range of variability in the derived PFDMO and PFRMO values, spanning from −5.17 to +10.16, where higher values correspond to increased levels of post-fire disturbance and lower values indicate comparatively reduced disturbance intensity, thereby providing a quantitative basis for interpreting the severity gradients observed across the study areas.

Consistent with these findings, all analyzed test sites are characterized by low PFDMO values prior to the fire events, indicating relatively stable and healthy vegetation conditions in the pre-fire period. Following the disturbance, however, substantial portions of the study areas exhibit elevated PFDMO values, delineating zones of high burn severity. At the same time, the affected landscapes demonstrate pronounced spatial variability, with PFDMO values spanning from unaffected to severely impacted regions (Figure 3a, Figure 4a and Figure 5a). This wide range of responses supports the interpretation of a heterogeneous fire impact pattern across all investigated cases. This interpretation is consistent with previous studies demonstrating that burn severity typically exhibits pronounced spatial heterogeneity, reflecting the influence of local environmental conditions and fire behavior dynamics [70].

Notably, among the investigated case studies, Slavyanka Mountain demonstrates the most pronounced spatial heterogeneity in burn severity [59], as evidenced by the distinctly variable PFDMO patterns observed across the affected area (Figure 3a). This pattern can be further explained by the fact that the region supports predominantly coniferous and mixed forest stands adapted to mountainous conditions, which makes these ecosystems particularly sensitive to high-intensity fire events and contributes to the observed variability in disturbance severity. This observation is supported by previous studies indicating that coniferous and mixed mountain forest ecosystems are highly susceptible to high-intensity fires, often resulting in increased variability in burn severity and post-fire responses [71,72].

In line with these observations, Sakar Mountain appears to be the most severely affected by fire according to the PFDMO values (Figure 5a). This can be attributed to its comparatively gentler relief and more open landscape structure relative to the other mountainous study areas, conditions that likely facilitated different fire behavior and resulted in more extensive and spatially heterogeneous burn patterns. This observation is consistent with previous studies highlighting the role of topographic and landscape characteristics in shaping fire behavior and burn severity patterns, where less complex and more open terrains can facilitate fire spread and lead to more extensive and spatially continuous disturbance [73,74,75].

In contrast to the other case studies, Maleshevska Mountain exhibits the narrowest range of PFDMO values (Figure 4a), suggesting a comparatively uniform level of disturbance across the burned area. This relatively homogeneous pattern indicates that the fire affected most of the territory with similar intensity, with limited differentiation in burn severity. Such behavior can be attributed to the specific geographical setting, where a mosaic of forested and shrub-dominated ecosystems is distributed over complex topography with varying slope and aspect conditions. Despite this heterogeneity in landscape structure, the fire appears to have propagated across both managed forest stands and semi-natural habitats in a consistent manner, resulting in a more evenly distributed disturbance signal. This pattern is consistent with previous studies indicating that under certain topographic and fuel conditions, fires can spread with relatively uniform intensity across heterogeneous landscapes, leading to more homogeneous burn severity patterns [72,76].

Building upon the disturbance assessment, the PFRMO outputs provide further insight into post-fire ecosystem dynamics by enabling the evaluation of vegetation recovery patterns one year after the fire event. In the case of Slavyanka Mountain, the PFRMO values do not differ substantially from the corresponding PFDMO patterns (Figure 3b), suggesting a relatively limited vegetation recovery within the first year after the fire and indicating that the affected ecosystems may require a longer period to exhibit significant regeneration. This finding is in agreement with previous studies reporting that vegetation recovery in mountainous coniferous and mixed forest ecosystems following high-severity fires is often slow and may extend over multiple years, particularly under conditions of severe disturbance and limited regeneration capacity [77,78].

On Maleshevska Mountain, the PFRMO results reveal a markedly heterogeneous recovery pattern, with areas of both low and high values (Figure 4b), which can be attributed to the mosaic distribution of forested and shrub-dominated ecosystems across complex terrain with varying slope and aspect, as well as to the differential response of managed forest stands and semi-natural habitats to post-fire regeneration processes. This pattern is consistent with findings from previous studies demonstrating that post-fire vegetation recovery is highly variable in heterogeneous landscapes, where differences in topography, vegetation type, and land management lead to contrasting regeneration responses across burned areas [79,80,81].

On Sakar Mountain, the PFRMO results indicate a general decline in the highest values (Figure 5b) compared to the corresponding PFDMO levels immediately after the fire, suggesting a partial recovery of vegetation within the first year following the disturbance. Nevertheless, the persistence of areas with relatively high PFRMO values points to zones where recovery remains limited, while the presence of low values reflects more advanced regeneration in other parts of the study area. This spatial variability can be linked to the low-altitude terrain dominated by a mosaic of mixed forest, shrubland, and grassland ecosystems, which respond differently to post-fire conditions. Furthermore, the gentler relief and more open landscape structure of the Sakar region, in contrast to the more mountainous study areas, likely influenced fire behavior and resulted in distinct spatial burn patterns that continue to shape the observed recovery dynamics. This pattern is consistent with previous studies demonstrating that post-fire vegetation recovery in lowland and heterogeneous landscapes is strongly influenced by vegetation type, fire severity, and terrain characteristics, often resulting in uneven regeneration dynamics across burned areas [72,81,82,83].

The interpretation of the obtained results should also be considered in relation to the characteristics of currently available operational fire products. For burned-area assessment, European services such as EFFIS and Copernicus Emergency Management Service provide important large-scale and event-based information; however, these products are also fundamentally based on remote-sensing workflows that rely on optical satellite imagery, including Sentinel-2, and commonly use post-fire spectral indicators [67,84]. Consequently, although such products are highly valuable for situational awareness, damage assessment, and broad-scale comparison, they cannot always be regarded as fully independent references for evaluating a system that operates on a similar multispectral basis [67]. In this context, the main value of such products lies in supporting regional consistency checks rather than in providing a one-to-one benchmark for system outputs.

A similar, but even more pronounced, limitation applies to post-fire recovery assessment. Recovery is a multidimensional process that may be expressed through rapid post-fire greening, progressive spectral convergence toward pre-fire conditions, or longer-term structural re-establishment of forest ecosystems [68]. Because these dimensions are not equivalent, recovery products developed in different studies are often based on different indicators, temporal windows, and thematic interpretations [68]. This limits the comparability of category-based outputs and makes the direct adoption of externally defined recovery thresholds methodologically problematic, especially when the compared systems are designed for different ecological settings or different aspects of post-fire dynamics [68]. The conceptual heterogeneity of recovery-oriented frameworks is also evident in operational post-fire mapping systems such as MTBS, where “increased greenness” denotes a short-term vegetation response rather than a direct equivalent of forest regrowth or ecosystem recovery in a broader sense [69].

A broader comparison with existing contemporary frameworks further helps to position the proposed system. Recent automated wildfire mapping workflows, such as that of Chen et al. (2025) [36], have demonstrated the value of integrating Sentinel-2 and Landsat data, semi-automatic sample generation, object-based image analysis, and machine-learning classification for regional burned-area and burn-severity mapping [36]. Related recent studies have also expanded the analytical scope toward combined severity and recovery evaluation based on Sentinel-2 spectral indices and time-series information, as shown by Cubas Sanchez et al. (2026) [39], or toward rapid multi-index damage assessment in operational environmental management settings, as in Liu et al. (2025) [85]. These approaches represent important advances in automated post-fire mapping; however, their primary focus remains on either burned-area delineation, damage severity characterization, or index-based assessment of post-fire conditions. By contrast, the system presented here is designed to maintain methodological continuity between post-fire disturbance assessment and post-fire recovery monitoring within a single processing chain based on a previously validated algorithmic core. In this respect, its distinguishing feature is not only automation, but also the integration of both phases of post-fire dynamics within one internally consistent framework [36].

A different type of comparison can be made with decision-support-oriented and rehabilitation-oriented post-fire frameworks. For example, Prodromou et al. (2025) [37] developed a remote-sensing-based spatial decision support tool for prioritizing post-fire restoration actions in Mediterranean ecosystems [37], while Cristal et al. (2025) proposed a participatory multi-criteria framework for estimating soil erosion risk and vegetation recovery potential after fire [38]. A related orientation is also evident in more recent rehabilitation-focused studies such as Kaloudis et al. (2025) [40], which link post-fire monitoring to the evaluation of rehabilitation needs and recovery trajectories. These frameworks are highly relevant for restoration planning and post-fire environmental assessment; however, they are primarily oriented toward prioritization, rehabilitation support, and multi-criteria evaluation. In contrast, the proposed system is focused on the automated generation of disturbance and recovery outputs directly from multispectral satellite data within a repeatable operational workflow. Its main advantage therefore lies in providing a standardized and transferable processing architecture that can support subsequent ecological interpretation and management applications, rather than in replacing specialized restoration-planning tools [37,38].

Compared with large-scale operational services such as EFFIS and event-based products delivered through Copernicus Emergency Management Service, the present system is characterized by a distinct application niche [67,84]. EFFIS is optimized for harmonized regional burned-area information at the European scale, whereas Copernicus Emergency Management Service provides tailored post-disaster mapping products configured for specific activations [67,84]. In this context, the system presented here occupies an intermediate position between these two types of approaches: it is more methodologically unified and recovery-oriented than large-scale burned-area services, yet more standardized and internally consistent than highly customized event-specific post-disaster products. This makes it particularly suitable for repeated post-fire assessment in forested areas where comparable disturbance and recovery information is needed across multiple case studies, time periods, and environmental settings, especially where consistent multispectral satellite observations are available and where the primary objective is methodological comparability rather than highly customized event-specific mapping [67,84].

In summary, the results obtained from the three case studies demonstrate that post-fire disturbance and recovery dynamics are strongly controlled by the interplay between fire severity, vegetation characteristics, and topographic conditions. The observed differences among Slavyanka, Maleshevska, and Sakar Mountains highlight the importance of landscape structure in shaping both the immediate impact of wildfires and the subsequent regeneration processes. In this context, sustained monitoring of post-fire recovery is essential for accurately evaluating vegetation regrowth and assessing ecosystem resilience over time. Overall, the integrated use of PFDMO and PFRMO provides a coherent framework for capturing these complex spatiotemporal patterns, offering valuable insights for post-fire monitoring and ecosystem management in heterogeneous environments.

In comparison with existing remote sensing-based frameworks, the proposed system is distinguished by the joint use of DI, VIC, and DA for linking post-fire disturbance assessment and recovery monitoring within a single automated processing chain. Its main advantages include operational simplicity, scalability, and applicability across diverse environmental conditions using freely available satellite data. At the same time, the system is most suitable for regions with sufficient availability of cloud-free optical imagery and well-defined seasonal vegetation dynamics, which ensure a reliable spectral response for both disturbance and recovery assessment.

5. Conclusions

This study presents a multispectral satellite-based integrated system for the automated assessment of post-fire disturbance and vegetation recovery in forest ecosystems. The main academic contribution lies in the integration of a previously validated algorithm into a unified operational framework that simultaneously addresses both disturbance detection and recovery monitoring—an aspect often treated separately in existing studies. From an application perspective, the system provides an operational, scalable, and automated workflow based on freely available satellite data, supporting practical implementation in forest management and environmental monitoring.

The application of the system to three case-study areas demonstrated its capability to capture spatial variability in post-fire dynamics under different environmental conditions. The results revealed substantial variability in disturbance and recovery indicators, with PFDMO values ranging from −5.17 to +10.16 and PFRMO values from −2.25 to +7.40 across the study sites. The validation results reported in previous studies document the performance of the underlying algorithmic approach, which forms the methodological core of the integrated system presented here. These numerical findings highlight the sensitivity of the system to differences in fire severity, vegetation type, and topographic conditions, and demonstrate its ability to identify heterogeneous post-fire patterns.

Despite these promising results, several limitations should be acknowledged. First, the current implementation relies primarily on Sentinel-2 data, which may limit temporal resolution under persistent cloud cover. Second, the system does not yet incorporate additional environmental variables such as climate, soil properties, or in situ observations, which could further improve interpretation of recovery processes.

Future work will focus on expanding the system through the integration of additional multisensor data sources, inclusion of environmental and climatic variables, extension to longer time-series analysis, and further testing across diverse ecosystems and fire regimes.