Fires are global widespread natural, dynamic, and periodically disturbing phenomena (Whitman et al., 2020; Fernández-Guisuraga et al., 2023b; Kurbanov et al., 2022; Jain et al., 2020), with more than half of the surface land at risk of being affected (Alonso-González and Fernández-García, 2021). Fires expose terrestrial ecosystems, resulting in various impacts on forest ecology and structure, soil erosion, loss of biodiversity, and endangerment of human life and infrastructures. Over the last 20 years, fires have burned, on average, 0.4 × 109 ha of land annually, with a cumulative total of 7.2 × 109 ha of burned area globally (Kurbanov et al., 2022).

According to statistics provided by the European Commission (2018), approximately 50 000 fires have burned from 1980 to 2018, with an annual average of 0.5 × 106 ha, especially in five Mediterranean European member states: Spain, Portugal, Italy, Greece, and France (Fernández-Guisuraga et al., 2023a). In these countries, the occurrence of extreme fires is getting more frequent and more intense, with larger burned areas, as their fire regime has shifted from “fuel-limited” to “drought-driven” (Pausas and Fernández-Muñoz, 2012). The fire regime shift has heterogeneous extent, seasonality, and frequency (Morresi et al., 2022). Its two main causes are the accumulation of flammable fuels and the consequences of global warming such as prolonged and more frequent droughts and heatwaves (Fernández-Guisuraga et al., 2023a). The accumulation of flammable fuels is caused by land use change, agricultural farming abandonment in rural zones, and the lack or absence of adaptive management (Moreira et al., 2020). Moreover, increasing global warming will likely lead to prolonged fire seasons, which may contribute to an increase in the number, frequency, and area of fires (Moreira et al., 2020; Miller et al., 2023; Holsinger et al., 2021; Fernández-Guisuraga et al., 2021). However, it is still unclear whether this predicted increase will lead to an increase in burn severity (Soverel et al., 2011; Morresi et al., 2022; Fernández-Guisuraga et al., 2023b; Parks et al., 2016).

Burn severity can be defined as the extent to which fire induces ecological and visible changes in soil, vegetation, and fuels (Key and Benson, 2006; Key, 2006; Lentile et al., 2006; Veraverbeke et al., 2010). The estimation of burn severity provides insights into forming better pre- and post-fire management strategies, including fuel treatments and post-fire recovery plans (Chu and Guo, 2014; Miller et al., 2023; García-Llamas et al., 2019). Burn severity estimates are highly time-sensitive since post-fire conditions depend on pre-fire conditions (Miller et al., 2023). A delayed estimation of burn severity will most likely lead to its poor estimation due to environmental responses such as forest recovery, tree and/or seedling recruitment, resprouting, vegetation regrowth, and ash removal by wind or precipitation (Miller et al., 2023; Dos Santos et al., 2020; Keeley et al., 2008; Keeley, 2009; Chu and Guo, 2014; Key, 2006). Hence, burn severity can be assessed during three periods based on different time lags, which are the difference between the dates of fire occurrence and burn severity estimation, categorized as rapid assessment (less than 2 weeks), initial assessment (1 to 8 weeks), and extended assessment (2 to 12 months). Beyond timeline differences, each assessment captures distinct stages of burn severity. For instance, delayed mortality and survivorship of vegetation are generally undetected during rapid assessment. Vegetation may still be senescent, stressed, or dying at this stage, leading to underestimation or overestimation of burn severity. Exclusively, burn severity estimation during rapid assessment is normally performed to assist post-fire responses to large fires. However, during the rapid assessment, the influence of environmental responses to burn severity is minimized. Estimates obtained during the initial assessment are considered to be the first opportunity to have a complete ecological evaluation as, for instance, some signs of delayed mortality and survivorship of vegetation can be detected slightly during initial assessment while vegetation may still be senescent. During the extended assessment, environmental responses are the most influential in relation to burn severity estimates while vegetation survivorship and delayed mortality are more easily detectable and when vegetation has normally returned to its stress-free green state. Overall, burn severity estimates obtained during the extended assessment are normally considered to be suitable as the final reference (Key, 2006).

The most reliable approach to estimate burn severity is via field assessment (Key and Benson, 2006) by measuring the observable fire-induced changes such as the extent of fire-consumed vegetation, stems of vegetation being charred, soil being exposed, and loss of chlorophyll in leaves (Keeley, 2009). These fire-induced changes correspond to structural, thermal, and spectral alterations in soil and vegetation (Miller et al., 2023). One of the most used metrics is the Composite Burn Index (CBI) (Key and Benson, 2006; García-Llamas et al., 2019), which visually ranks the burn severity from 0 (unburned) to 3 (high severity) (Parks et al., 2018; Fernández-Guisuraga et al., 2023a; Addison and Oommen, 2018). The reason for the CBI's popularity is due to its rapid protocol and its overall estimation of fire-induced damage to vegetation and soil, especially when assessing the burn severity of large fires. However, burn severity field assessments have multiple drawbacks as they are intensive; logistically challenging; highly resource-dependent, especially in inaccessible and/or remote burned areas; and have limited capability in capturing the burn severity heterogeneity over large burned areas. Moreover, the impacts of historical fires and the evolution of burn severity cannot be measured via field assessment (Miller et al., 2023).

The emergence of remote sensing (RS), especially via application of satellite imagery, over the past decades has facilitated obtaining free-of-charge remotely sensed burn severity estimates as an alternative option to expensive and time-consuming field severity observations (Miller et al., 2023; Holsinger et al., 2021; Fernández-Guisuraga et al., 2021; Miller and Thode, 2006; Parks et al., 2014; Fernández-García et al., 2018). The capability of capturing spectral information in the visible, near-infrared (NIR), and shortwave infrared (SWIR) parts of the electromagnetic spectrum (Key and Benson, 2006) has enabled the detection of fire-induced structural, thermal, and spectral changes to the land surface (Miller et al., 2023). Satellite sensors have different optical bands, spatial resolutions, temporal revisiting frequencies, and time spans (Lentile et al., 2006). Thus, there are tradeoffs in the application of different satellite RS sensors and in the availability of clear-sky imagery (Miller et al., 2023). Moreover, caution should be taken when using RS products as they may acquire top-of-the-canopy reflectance, with limited capability to estimate the burn severity of the understory strata (García-Llamas et al., 2019; Mihajlovski et al., 2023). Last but not least, RS-derived burn severity estimates “must be linked to ground-truth data” (García-Llamas et al., 2019; Miller et al., 2023; Chu and Guo, 2014).

Multiple studies have found moderate correlations between satellite-derived RS and ground burn severity indices, providing higher confidence in RS-derived burn severity estimates. However, the strength of these correlations varies from one region to another and can be influenced by environmental factors such as fuel, vegetation type, and topography (Fernández-Guisuraga et al., 2021). In this context, mono- (only post-fire image) and bi-temporal (both pre- and post-fire images) indices derived from the normalized burn ratio (NBR), such as the differenced normalized burn ratio (dNBR), the relative differenced normalized burn ratio (RdNBR), the relativized burn ratio (RBR), and a burn severity index that combines the dNBR with the enhanced vegetation index (EVI) (Gao et al., 2000) (dNBR-EVI) (Fernández-García et al., 2018), have been applied. The dNBR (Key and Benson, 2006) is considered to be the “standard” index for burn severity quantification (Alonso-González and Fernández-García, 2021), specifically in the Mediterranean regions (Fernández-García et al., 2022; Miller and Thode, 2006; Picotte et al., 2016; Fernández-Guisuraga et al., 2023a; Chu and Guo, 2014; Keeley et al., 2008; Fernández-García et al., 2018). The RdNBR provides a relative measurement of burn severity based on the pre-fire state of vegetation (Miller and Thode, 2006) and has proven to be more sensitive than the dNBR, especially in areas with low vegetation cover density (Parks et al., 2014). However, the calculation of the RdNBR presents some difficulties due to its formula and its “numerically unstable” range resulting from pre-fire NBRs with very low values (Fernández-Guisuraga et al., 2023a). Another relative measure of burn severity without the calculation difficulties is the RBR (Parks et al., 2014). According to Fernández-Guisuraga et al. (2023a), the RBR showed better correlation with the CBI in Mediterranean ecosystems in comparison to the dNBR. Additionally, according to Fernández-García et al. (2018), their proposed index, known as the dNBR-EVI, exhibits the best correlation with the CBI in Mediterranean regions in comparison to the NBR, dNBR, RdNBR, and RBR. Moreover, the dNBR-EVI has been claimed to show no signal saturation in high-severity areas as saturation is a known issue for NBR-derived indices in regions with high burn severity (Fernández-García et al., 2018; Santis et al., 2010; Fernández-Guisuraga et al., 2023a).

The estimation of the burn severity of historical fires can be performed using satellite-derived RS indices. This feature can enable the evaluation of changes or trends in burn severity patterns over a specific period (Lutz et al., 2011; Picotte et al., 2016). The first project devoted to the creation of a burn severity atlas was the Monitoring Trends in Burn Severity (MTBS), which provided dNBR and RdNBR maps for large fires from 1984 to the present in the USA using Landsat imagery (Eidenshink et al., 2007; Picotte et al., 2020). There have been records of burn severity atlases created for parts of some countries such as Canada (Picotte et al., 2016; Whitman et al., 2020; Guindon et al., 2021). Moreover, the MOSEV dataset provides 20 years of burn severity maps using Moderate Resolution Imaging Spectroradiometer (MODIS) imagery and products (Alonso-González and Fernández-García, 2021). Although this dataset provides daily global coverage, it has considerable limitations such as the spatial resolution (500 m); the limited capability of mapping burn severity heterogeneity, especially at the regional scale (Alonso-González and Fernández-García, 2021); the absence of burn severity estimates for fires before the year 2000; and low burned-area mapping accuracy (Moreno-Ruiz et al., 2020).

To the best of our knowledge, detailed long-term estimates of burn severity are missing for European countries, such as Portugal, the most “fire-prone” country in the European Mediterranean basin (Oliveira et al., 2011). Portugal is characterized by a Mediterranean-type climate (Ermitão et al., 2023; Parente et al., 2023) and a drought-driven fire regime (Pausas and Fernández-Muñoz, 2012) and is dominated by shrubs and pines, eucalypts, and evergreen oaks forests (Fernandes et al., 2016). During the last few decades, Portugal has been significantly affected by fires, such as catastrophic fires in 2003, 2005, and 2017 (Nitzsche et al., 2023; Beighley and Hyde, 2018). Thus, the main objective of this study is to create a high-resolution multidecadal burn severity atlas for mainland Portugal, entitled the Portuguese Burn Severity Atlas, aimed at providing estimates of immediate fire impacts.

The study area consists of mainland Portugal (37° N to 42° N latitude and 6° W to 10° W longitude) (Parente et al., 2016), covering around 90 000 km² of southern Europe (Rego and Bacao, 2010), generally with a Mediterranean climate consisting of warm, dry summers and cold, wet winters (Nunes et al., 2016), with elevation ranging from sea level to approximately 2000 m (Mora and Vieira, 2020), and with the domination of different vegetation types within its extent (e.g., farmland and evergreen oak (Quercus suber, Q. rotundifolia) woodlands in the south, forests of pine (Pinus pinaster) and eucalypt (Eucalyptus globulus) in the north, and shrublands and deciduous oak forests in the center) (Tonini et al., 2017).

To estimate the burn severity of historical fires in Portugal (1984–2022), it is necessary to primarily gather fire data containing the start and end dates (Sdate and Edate), the burned perimeters, and the extents (Sect. 2.1). Then, we selected an RS sensor or a family of sensors to have coherency over the study period (Sect. 2.2). The next steps were to select burn severity indices well correlated with ground observations, specifically in the Mediterranean regions (Fernández-Guisuraga et al., 2023a) (Sect. 2.3); to assign a sampling period to select RS imagery (Sect. 2.4); to quantify the influence of time lag on burn severity estimates and accordingly apply the necessary adaptation to the sampling period (Sect. 2.5); and, finally, to calculate the burn severity estimate for each of the fires with the most suitable pair of images (Sect. 2.6).

The Portuguese Burn Severity Atlas includes estimates for 66 % of the total burned area between 1984 to 2022. The remaining 34 % of burned area without burn severity estimates was mainly due to the exclusion of fires <100 ha, the exclusion of fires merged into one, and the lack of satellite imagery to estimate burn severity. When considering only the fires with start and end dates, only 3 % of the burned area did not have burn severity estimates due to either a lack of Landsat imagery with clear pixels within our sampling period or the limitations in GEE performing heavier processing and stopping the code after the fourth iteration. Within this study, the focus was on large fires (≧ 100 ha); however, this can be considered to be a limitation of our atlas. As our developed methodology can be applied to smaller fires (<100 ha), the exploration of the burn severity of small fires could be a research opportunity for future studies.

As can be seen in Fig. 2b, 9 % of the dNBR pixel values are lower than 0.1. According to the EFFIS (European Commission, 2018) burn severity discrete classification (European Commission, 2018; Llorens et al., 2021), dNBR pixel values less than 0.1 can be considered to be in the class of unburned or regrowth of vegetation. This can be caused by either commission errors in the burned-area mapping or the high regrowth potential of Portuguese vegetation cover, especially within the first month following the fire occurrence (Neves et al., 2023).

Results show that, in almost 40 years, Portugal has, on average, experienced high burn severity. Mateus and Fernandes (2014) detailed the following reasons behind the high-severity fire regime in Portugal: (1) the dominance of highly flammable Eucalyptus globus, especially in the Centro and Norte regions, accounting for 77 % of all forest fires; (2) the fact that 90 % of all forest fires occur within the months of June to September as the result of droughts; (3) dry conditions of dominant vegetation types due to seasonal weather patterns; (4) high productivity of understory plants, specifically shrubs; (5) the dominance of stand-replacing and crown fires; and, most importantly, (6) prioritizing fire suppression over prevention.

No visible trend was observed regarding the annual burned-area extent over the years studied (Fig. 3); however, the three largest fire seasons were in 2003, 2005, and 2017. Many fire research studies focused on these 3 years, not only because of their huge magnitude of burned extent but also due to their different drivers (Beighley and Hyde, 2018) and consequences (Nitzsche et al., 2024). Oliveira et al. (2021) also found no trend regarding the burned extent over the years in Portugal while highlighting that 2017 was “the worst year” in terms of the burned area. For these 3 years, more than 98 % of valid fires had burn severity estimates, although imageries from different sensors – Landsat-5 TM and Landsat-8 OLI – were applied due to different satellite availability dates (Table 1).

More reliable fire data since 2001 are available for Portugal as better means of data record technologies from various sources have been applied (Nunes et al., 2016). For recent years – 2013 onwards – higher percentages of burned area with severity estimates (on average, ≈ 99 %) and with no drastic variations were obtained, for which only images from Landsat-8 OLI sensor were used. Hence, our omission errors for fires following 2001 are less than the ones prior to this year. Although both Landsat-5 TM and Landsat-8 OLI sensors have a similar temporal resolution, it was observed that, via Landsat-8 OLI, more images with clear pixels – minimum reflectance contamination such as cloud, cirrus, shadow, and smoke – were available during the sampling period of ±110 d. This can be explained due to the difference in terms of the “daily image acquisition rate” of Landsat-8 OLI and Landsat-5 TM. The image acquisition rate is defined as the number of images acquired by each sensor on a daily basis. Landsat-5 TM acquires 225 to 250 images per day, which varies based on various factors such as the amount of sunlit land (Loveland and Dwyer, 2012), while Landsat-8 OLI acquires 725 images per day (Loveland and Irons, 2016). This point can be confirmed by the time series boxplot of fire-by-fire confidence (%), i.e., the average of the suitability of pre- and post-fire images, showing smaller range variations with high average values (>80 %) for recent years (Fig. 5). Higher confidence is the result of a greater suitability value, which occurs when time lag decreases (Eq. 1). Thus, in recent years, the time lag variation range should also be smaller, which is confirmed by Fig. B1 (mentioned in Sect. 3.2).

Burn severity estimates were obtained from different Landsat sensors to ensure “spectral consistency” over the long study period (Fernández-Guisuraga and Fernandes, 2024). This objective was achieved for all years. For 2012, aside from Landsat-7, estimates were also provided by MODIS Terra imagery. According to the Landsat sensors' availability dates (Table 1), no images from this family of sensors were available for this year, aside from Landsat-7 ETM⁺ with SLC failure. One possible alternative would be to use images from the Earth Observation-1 Advanced Land Imager (hereafter EO-1 ALI), available from November 2000 to March 2017 with a spatial resolution of 30 m (Chander et al., 2009). Although EO-1 had the capability of imaging with global coverage, it was an experimental and mission-based satellite (Hoang and Koike, 2018), and, according to the USGS website (EarthExplorer, 2025), there were no images available for Portugal in 2012. Hence, in addition to MODIS-derived estimates for 2012, burn severity estimates via Landsat-7 ETM⁺ are included in the second version of the atlas, providing approximately 46 % of the area of valid fires for 2012 with estimates. The rest (54 %) do not have estimates due to either gaps caused by the SLC failure or a lack of cloud-free imagery. Users can now choose between having estimates for the year 2012 with

Landsat-7 data, with a high resolution (30 m), covering about 37 % of total burned-area extent, or

We define RS images with quality as images with small time lags and, preferably, with little to no reflectance contaminations (e.g., cloud, cirrus, shadow, and smoke). Via the application of images with high quality, the most reliable burn severity estimates can be obtained (Miller et al., 2023; Dos Santos et al., 2020; Keeley et al., 2008). However, the number of high-quality images over each individual fire perimeter is scarce, especially when considering the older sensors (Gao et al., 2006). To increase the coverage of burn severity estimates, we considered an extensive sampling period (±110 d) and incorporated an iteration process, improving the coverage of our atlas by 12 %. To the best of our knowledge, no other studies in the literature have applied such a process before. However, mean compositing of different scenes to obtain burn severity or other types of atlases has been practiced (Parks et al., 2018; Whitman et al., 2020; Neves et al., 2023).

Due to our relatively long sampling period and because no specific scene acquisition row and/or path were determined when applying Landsat imagery, the occurrence of seasonal variations and “mismatched phenology” between the pre- and post-fire images could have happened (Parks et al., 2018; Key, 2006; Storey et al., 2005; Lutes et al., 2006) as phenology is not constant throughout the year (Balata et al., 2022). On average, the worst pre- and post-fire time lags of this atlas were lower than ±50 d for the first iterated burn severity estimations (Fig. B1); thus, on average, the seasonality influences have been minimized, although no further assessment was performed in this regard. More research related to seasonality and phenology analysis is necessary (Key, 2006; Howe et al., 2022; Parks et al., 2018). In this atlas, we have calculated the offset values of burn severity indices of individual fires according to their corresponding formulas (Table 2). As indicated by Parks et al. (2018), offset values are accounted for to differentiate between phenology variations between pre- and post-fire images. Hence, they can be applied to minimize the impacts caused by any possible “mismatched phenology” between pre- and post-fire images. As a suggestion for future studies, the lack of knowledge regarding the degree of seasonality influencing the quality of burn severity estimates can be highlighted. Such analysis can be conducted by using the offset values and time lags provided by this atlas.

The definitions related to sampling periods are ambiguous. Our sampling period is ±110 d to avoid and minimize capturing environmental and ecological responses. In many studies related to the “trend and evolution analysis of burn severity”, such as the one performed by Dillon et al. (2006), the burn severity estimates calculated within 6 months or ±180 d from the ignition date were excluded. Dillon et al. (2006) classified this sampling period as the “initial assessment”. On the other hand, by the definition provided by Key (2006), our sampling period is categorized as an extended assessment. Thus, we have called our sampling period a rapid to extended assessment.

As burn severity estimates are highly time-sensitive, with the increase in time lag, the accuracy of RS estimates decreases and is degraded as environmental responses are cumulated over fire impacts (Key, 2006). To have an understanding of this degradation, we modeled the variation in the dNBR as the standard burn severity index, caused by the increase in time lags (Fig. 4). The dNBR tends to increase with the increase in the pre-fire time lag, which leads to an overestimation of burn severity. In Portugal, as in all Mediterranean-climate areas, vegetation vigor is lower before fire occurrence as a result of high temperatures and solar radiation and low water availability (Verbyla et al., 2008; Chu and Guo, 2014; Pascolini-Campbell et al., 2022; Fernández-Guisuraga et al., 2023b). Generally, the pre-NBR tends to decrease as the fire season approaches (Alonso-González and Fernández-García, 2021). Thus, with the increase in pre-fire time lag, burn severity is overestimated as the amount of vegetation considered to be burned is overestimated. The dNBR tends to decrease with the increase in the post-fire time lag, which leads to an underestimation of burn severity. This can result from the fire scars becoming less visible due to environmental and ecological responses such as resprouting, especially in Portugal, where vegetation tends to regrow within the first month after the fire (Neves et al., 2023) and/or via ashes being removed by rain and wind (Key, 2006; González-Pelayo et al., 2023, 2024).

Burn severity estimates via different indices reveal varying degrees of post-fire impacts across the landscape (Parks et al., 2014; Miller and Thode, 2006; Fernández-García et al., 2018). Combining information from multiple indices could offer a more comprehensive understanding of burn severity. Thus, in this atlas, we provided burn severity estimates using four dNBR-derived indices. For instance, as can be observed from the examples provided in Fig. 6, for the same fires, each of the used indices can contribute to a better interpretation of burn severity. In summary, the RdNBR is more sensitive to vegetation type than the dNBR; its tendency towards higher pixel values can be interpreted to be due to the low fuel load of vegetation cover (Miller and Thode, 2006) for the chosen fires. However, more in-depth analysis is needed to confirm this point. Within these examples, the RBR was observed to be more prone to signal saturation. Signal saturation of dNBR-derived indices is defined as an incapability of indices to measure very high burn severity, with their values reaching a certain point where they are no longer capable of discerning subtle differences in terms of burn severity (Veraverbeke et al., 2012). Thus, caution should be taken when interpreting the dNBR-derived burn severity maps as they are subjected to signal saturation (Fernández-García et al., 2018; Santis et al., 2010; Fernández-Guisuraga et al., 2023a), especially the RBR. Moreover, as stated by Fernández-Guisuraga et al. (2023a), any interpretation of RS burn severity estimates must be accompanied by and confirmed with ground truth data, specifically for relative forms like the RdNBR and RBR. This is crucial because burn severity often varies across vertical strata, and satellite-derived reflections are differently sensitive to the impacts at each layer. Aggregating these impacts into a single metric can obscure important ecological details (Fernández-Guisuraga et al., 2023a; Miller and Thode, 2006; Parks et al., 2014; Cansler and McKenzie, 2012).

In this atlas, no data regarding the ground burn severity assessment are included and analyzed. The provision of the means of interpretation for the burn severity degrees or classes of our maps is not within the scope of this study. Thus, no classification thresholds for any of the burn severity indices are proposed, and the means of interpretation of burn severity must only align with users' objectives. As an example of means towards the interpretation of burn severity, in this study, the thresholds assigned by the EFFIS are mentioned. The thresholds of the EFFIS are assigned only for the dNBR index and not for other indices, and they are obtained from the comparison between dNBR pixel values and ground burn severity estimates considering the dominant environmental and climatic conditions within the Mediterranean regions (Llorens et al., 2021). In other studies, based on the correlations between the CBI and dNBR and the RdNBR and RBR, different thresholds have been introduced but only for specific parts of the USA (Alonso-González and Fernández-García, 2021; Parks et al., 2018). In Spain and specifically in the province of Valencia, by comparing the CBI and dNBR, the RdNBR, and the RBR, the classification thresholds for the interpretation of these indices were introduced (Botella-Martínez and Fernández-Manso, 2017) and have been furthered utilized for the interpretation of burn severity estimates by the ICNF for fires which burned in Monchique and Portimão in Portugal (ICNF, 2021). Moreover, there is a study that was conducted by Fernández-García et al (2022) which considers a total number of 23 fires, among which only 4 fires were located in Portugal (Fernández-García et al., 2022). Although there are studies comparing burn severity observations with estimates, they are isolated and limited, and, hence, they cannot be incorporated at a large scale, which, in our case, is the mainland of Portugal. Hence, our maps have all been presented in their continuous raw forms, and no classifications have been applied to them. We acknowledge the lack of validation as a limitation of our atlas, and we encourage future studies to validate the burn severity estimates of the Portuguese Burn Severity Atlas.

The Portuguese Burn Severity Atlas can be used as the foundation of many future research projects and presents numerous research opportunities. With 97 % of burn severity estimates for valid fires from 1984 to 2022 within the sampling period of ±110 d, we can confidently state that the characterization of long-term burn severity patterns (Singleton et al., 2019; Gale and Cary, 2022), burn-severity-heterogeneity-related analyses (Lutz et al., 2011; Buonanduci et al., 2023), analyses isolating burn severity environmental and climate variables (Miller et al., 2009), post-fire recovery studies (Oliveira et al., 2011; Alonso-González and Fernández-García, 2021; Whitman et al., 2020), and fire consequence studies (Wells et al., 2021; Petratou et al., 2023; Amerh et al., 2022; Vieira et al., 2023; Singh et al., 2022) can be conducted. Moreover, as suggestions for future research and via this atlas, we can highlight the gap of knowledge in the interpretation of RS burn severity estimates, especially the relative forms in the Mediterranean regions. Secondly, the most influential burn severity drivers, both environmental and climatic, need to be distinguished, and, accordingly, more informed pre- and post-fire management plans should be formed to minimize future fire impacts in Portugal. Last but not least, we encourage the execution of trend analysis of burn severity evolution to assess whether the burn severity in mainland Portugal has changed over the years.

The maps of the Portuguese Burn Severity Atlas can be accessed at 10.5281/zenodo.12773611 (Jahanianfard et al., 2025) (version 2, with corrections to fires before 2001 and inclusion of Landsat-7 estimates for 2012). The annual maps are provided in subfolders entitled as the corresponding year. Annual fire perimeters (shapefile), along with a table of details for the pairs of images used for each iteration, the confidence, and the offset values of burn severity indices, are also stored within these subfolders.

The GEE code can be accessed at https://code.earthengine.google.com/042de010edb5abdd14247f65a23a6193?noload=true (last access: 15 April 2025), with the fire data already having been shared as assets; however, it is necessary to have a GEE account to access the code. The code can also be accessed at https://github.com/DinaJahanianfard/Portuguese-Burn-Severity-Atlas_v2/commit/7aee76ea5b3df0db8cd047a4b8cf6624bf965d50 (10.5281/zenodo.17160468, Jahanianfard, 2025), with no account or registration needed.

In this study, a comprehensive Portuguese Burn Severity Atlas was developed, spanning from 1984 to 2022, derived from Landsat satellite imagery (30 m resolution). This atlas contains burn severity estimates for 66 % of the 4.85 × 106 ha burned between 1984 and 2022 and for 97 % of the valid fires, with a total burn area of 3.18 × 106 ha. The atlas illustrates that Portugal has, on average, experienced high burn severity over the study period.

Through an iteration process, we expanded the coverage of the atlas, providing burn severity estimates for an additional 12 % of the valid fire area, totaling 0.38 × 106 ha. Furthermore, we developed a semi-automated code in Google Earth Engine (GEE) that can be easily updated and modified by users to generate burn severity estimates for any region worldwide using any desired sampling period, with only the requirement of fire data.

Our findings regarding the relationship between the dNBR and time lag indicate that increasing pre-fire time lags results in the overestimation of burn severity, while increasing post-fire time lags leads to its underestimation.

The analysis of burn severity estimates using different indices reveals distinctive characteristics. Notably, the RdNBR tends to indicate higher burn severity, while the RBR shows a tendency towards signal saturation compared to other indices. However, further investigation is necessary to validate these findings, particularly in the context of defining thresholds for interpreting burn severity classes in Mediterranean regions.

Ultimately, the burn severity maps provided in this atlas offer numerous opportunities for research across various disciplines. They enable investigations into burn severity heterogeneity, trend analysis, studies of environmental and climatic drivers of burn severity, and studies related to air and water quality and soil erosion.