Determination of the critical threshold for the occurrence of fires in the Parque Nacional de Brasília (Brazil) through temporal analysis using spectral índices

Abstract

The Brazilian Savannah is one of the largest biomes in Brazil. Unfortunately, human pressure is aggravating the degradation processes of the biome and, together with the drought processes, fire events are a major concern. Monitoring tools must be designed, especially those involving the concepts of remote sensing, in anticipation of the fire phenomenon, mitigating the devastating effects. Therefore, the present research aims to determine the critical threshold for the occurrence of fires through the diagnosis of the conditions of greenness, humidity and senescence of the vegetation using the temporal analysis of Sentinel-2 images in the Parque Nacional de Brasília. Therefore, it is necessary to quantitatively determine each condition through the analysis of past events, determining the criticality threshold that conditions the region to present conditions that favor the onset of a fire, as well as the construction of an algorithm, in the Google Earth Engine. The study area is the Parque Nacional de Brasília, with the image data collection Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-1C inside the Google Earth Engine. The algorithm aims to build a masking system to remove materials from the scenes, as well as the calculation of the NDVI, NDII, PSRI and dNBR indices, extracting the data through the mask of burned pixels. It was possible to identify six periods of fire occurrence, the data extraction allowed the statistical determination of the threshold, which was 0,580 for the NDVI, 0,015 for the NDII and 0,150 for the PSRI.

Keywords:
Analysis; Biome; Google Earth Engine; Fire; Monitoring

Resumo

O Cerrado brasileiro é um dos maiores biomas do Brasil. Infelizmente a pressão antrópica está agravando os processos de degradação do bioma e juntamente com os processos de estiagem, os eventos de fogo surgem como grande fator de preocupação. Devem ser concebidos ferramentas de monitoramento, principalmente os que envolvem os conceitos de sensoriamento remoto, antepondo-se ao fenômeno de fogo, mitigando os efeitos devastadores. Sendo assim, a presente pesquisa tem como objetivo a determinação do limiar crítico à ocorrência de incêndios por meio do diagnóstico das condições de verdor, umidade e senescência da vegetação utilizando da análise temporal de imagens “Sentinel-2” no Parque Nacional de Brasília, para isso, faz-se necessário a determinação quantitativa de cada condição por meio da análise de eventos passados, determinando o limiar de criticidade que condicione a região apresentar condições que favoreçam o início de um incêndio bem como a construção de um algoritmo, no “Google Earth Engine”. Á área de estudo é o Parque Nacional de Brasília, com o acervo de dados de imagens “Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-1C” dentro do “Google Earth Engine”. O algoritmo visa construir um sistema de máscaras para remover materiais das cenas, bem como o cálculo dos índices NDVI, NDII, PSRI e dNBR, extraindo os dados por meio da máscara de pixels queimados. Foi possível identificar seis períodos de ocorrência de fogo, a extração de dados permitiu a determinação estatística do limiar, sendo esse 0,580 para o NDVI, 0,015 para o NDII e 0.150 para o PSRI.

Palavras-chave:
Análise; Bioma; Google Earth Engine; Incêndio; Monitoramento

INTRODUCTION

The Savannah Biome is considered one of the largest existing biomes in the Brazilian territory, with an approximate extension of 2 million square kilometers and is located in the central region of the Brazilian territory, as seen in Figure 1, being a natural border of the Amazon Rainforest and Caatinga biomes (COLLI et al., 2020).

Figure 1
Map of Brazilian Biomes

The biome has a rich biodiversity and an extensive list of fauna and flora species, however, it suffers from increasing human pressure, marked by the expansion of agricultural activities and forest fires, which can occur both naturally or by human action (HOFMANN et al., 2021).

The Cerrado has dry seasons, which provide an environment conducive to the occurrence of fire events, raising questions about the fundamental role of prevention, surveillance, detection and firefighting tools (HOFMANN et al., 2021; VERNOOIJ et al., 2021). Such events also function as maintenance of ecosystems, but management and monitoring are essential in order to keep the impacts under control, avoiding the spread of flames and the danger to biodiversity (RIBEIRO et al., 2019).

Fire monitoring techniques and systems should be designed to take precedence over the event, providing support for decision-making, so a good monitoring system aims to help manage efforts, producing specific intervention solutions in such events (BARMPOUTIS et al., 2020).

Considering that territorial monitoring systems usually involve large dimensions, remote sensing emerges as a powerful science and tool, allowing the creation of mechanisms with great potential for data mass analysis. (BARMPOUTIS et al., 2020).

Fire events demonstrate the environmental fragility of ecosystems, in particular those of native vegetation, and studying the severity of these events is important for verifying the ecological, economic and physical consequences of the affected region, strategically coordinating mitigation actions (FASSNACHT et al., 2021; FERNÁNDEZ-MANSO; QUINTANO, 2020).

Baptista et al. (2018) and Bento-Gonçalves et al. (2019) suggest that the evaluation, through remote sensing, of the indicators of greenness, humidity and senescence can be an important milestone to determine the methodologies for monitoring areas that may have fuel potential, since these areas have different pixels from the areas that did not burn, indicating fire susceptibility.

The occurrence of fires, whether of anthropic causes or not, represent a potential danger to fauna, flora and human lives, as can be seen in events in Portugal, during the year 2017 (BAPTISTA et al., 2018) or in several countries such as Australia, Brazil, Canada, the D.R. Congo, Greece, Indonesia, Russia, Spain and the USA (BENTO-GONÇALVES, 2022). The adoption of tools and policies for the prevention and prediction of fires in a solid and efficient way becomes necessary for the correct targeting of intervention measures in order to minimize the impact generated by the spread of a fire.

Management strategies are usually done in a timely manner and the adoption of the use of spatial data can expand actions related to integrated fire management, so it is possible to raise the hypothesis that it is possible to evaluate the fire events that occurred in the National Park from Brasília, starting in 2015 (the year in which Sentinel-2 images began) and stipulating critical limits of greenness, humidity and senescence in these areas.

Therefore, the present work has as its general objective the determination of the critical threshold for the occurrence of fires through the diagnosis of the conditions of greenness, humidity and senescence of the vegetation using the temporal analysis of Sentinel-2 images in the Parque Nacional de Brasília. Therefore, it is necessary to quantitatively determine each condition through the analysis of past events, determining the threshold of criticality that conditions the region to be a potential ignitor for fire events, as well as the construction of an algorithm, in Google Earth Engine - GEE, based on the proposed methodology.

METHODS

Study area

The study area in question is the Parque Nacional de Brasília, located in the northwest of the Distrito Federal. The park is a Unidade de Conservação de Proteção Integral, wich is a law created territory to protect the circumscribed region, and was created on November 29, 1961, under Decree No. 241 and has approximately 42,389 hectares, encompassing the administrative regions of Plano Piloto, Sobradinho, and Brazlândia, as well as a portion of the municipalities people from Padre Bernardo and Planaltina, as seen in Figure 2.

Figure 2
Location of Parque Nacional de Brasília

The cover and use of the soil, as it is an integrally protected Conservation Unit, is predominantly made up of forest, savannah and grassland plant formations, as observed in the Figure 3.

Figure 3
Coverage and Land Use of Parque Nacional de Brasília

GIS

The GEE will be used, which is an online platform for processing, viewing and manipulating geographic data (GORELICK et al., 2017). Due to its great processing capacity, it will be possible to develop a script capable of performing the necessary analyzes for the periods under study, which mitigates the large data manipulation time required in conventional means of processing in remote sensing.

Data Collection

It was used the collection of multispectral images from the Copernicus program called Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-1C was used. The images of this segment, according to the European Space Agency, have already been systematically processed by ground control stations, which in this case is the “Payload Data Ground Segment”. This collection slides the pixel values of images taken after January 25, 2022, to respect the range of old data values.

The products of this collection are already sent from the Payload Data Ground Segment with the analyzed telemetry, decompressed, with the necessary corrections and radiometric and geometric calibrations and with resampled values for reflectance data. The product does not have a solution for atmospheric calibration, which will be performed by applying the Sensor Invariant Atmospheric Correction, present in a library in the GEE, developed and validated by MarcYin (2019a) and MarcYin (2019b), being used by Nursaputra et al. (2021) and Carella et al. (2022).

The normal products Level-1C and Level-2A were not used because, for the former, the values of the old images are not in the same ranges as the new images, which could harm automatic processes within the platform, and for the second, there are no corrected images for the entire period you want to search.

Research methodological procedures

The procedures used to obtain the results and objective of the research will be described in this section.

Determining event periods

For the development of the methodology, it was necessary to identify periods in which fire events occurred within the Parque Nacional de Brasília. Therefore, with the help of the Instituto Nacional de Pesquisas Espaciais (which is a research unit of the Brazilian Ministry of Science, Technology and Innovations) systems, six periods of occurrence of events were identified, which, for each event, it was necessary to identify two images, one pre- and one post-fire.

Spatial Filtering of scenes

Aiming to improve the automatic processes built within the platform, mitigating the negative effects arising from the use of spectral indices, spatial masks were used to filter the elements in question, allowing only the regions with plant material to be analyzed. For this, it used spectral indices to identify clouds, cloud shadows, water and finally, as a way to guarantee that the analyzed regions do not suffer from non-vegetable interference. Figure 4 shows how the general mask was obtained.

Figure 4
Methodology for creating the general event mask

To identify the clouds, the Sentinel Cloud Probability product was used, provided by the Sentinel Hub, developed on the “Sentinel2-Cloud-Detector” algorithm, together with the “LightGBM” library. The value of 50% was used, which indicates that the pixels that were 50% sure that the content adjacent to the pixel was really a cloud were filtered.

For the identification of cloud shadows, the concepts obtained in Magno et al. (2021), the “Cloud Shadow Index” - CSI, systematized in Equation 1, and filtered at the thresholds of interest for identifying shadows, thanks to the specific behavior of the “Shortwave Infrared” band of the “Multispectral Instrument” sensor

$CSI=\frac{{\left(1-{\rho }_{2.2}\right)}^2}{2\times 2_{2.2}}$ Equation 1

The Normalized Difference Water Index - NDWI, proposed by McFeeters (1996) and the Sentinel Water Mask, proposed by Milczarek et al. (2017). The indices, as they can be obtained, respectively, by Equation 2 e Equation 3, emphasize water bodies.

$NDWI=\frac{{\rho }_{0.55}-{\rho }_{0.65}}{{\rho }_{0.55}+{\rho }_{0.65}}$ Equation 2

$SWM=\frac{{\rho }_{0.45}-{\rho }_{0.55}}{{\rho }_{0.85}+{\rho }_{1.60}}$ Equation 3

The pre- and post-fire masks were obtained by calculating the aforementioned indices for the images in question. After calculating both, the overall mask of the event was calculated for each period, consisting of pre and post-fire masks. It served to ensure that the analysis within the platform worked for any period in question, considering that pixels not present in an image and that were not in another would not be analyzed, thus guaranteeing the homogeneity of the analysis.

Still aiming at the platform procedures, it was decided to create a buffer of 120 meters (or 12 pixels of the sensor) to mitigate the material transition effects.

Spectral Indices

In this section, the central methodology of the research will be demonstrated.

Normalized Difference Vegetation Index (NDVI)

The “Normalized Difference Vegetation Index” - NDVI shows the greenness of vegetation based on the absorption of red and near infrared through simple math of the band between the red (0.65 µm) and near-infrared (0.85 µm) layers) which intends to create representativeness between the targets of greater emission of red and infrared (ROUSE et al., 1974). To calculate the NDVI, we used the Equation 4.

$NDVI=\frac{{\rho }_{0.85}-{\rho }_{0.65}}{{\rho }_{0.85}+{\rho }_{0.65}}$ Equation 4

The index varies from -1 to 1, where values below 0.2 show the absence of photosynthetic activity, that is, the index shows the degree of greenness of the vegetation through the gain of the red band over the near infrared band, highlighting the vegetation with greener aspect (DUAN et al., 2017), and values in the range from 0.2 to 1 represent, gradually, the level of greenness of vegetation according to its health standards (SILVA; BAPTISTA, 2015).

Normalized Difference Infrared Index (NDII)

The “Normalized Difference Infrared Index” - NDII explains the humidity of the tree canopy by means of band mathematics similar to the NDVI, however, in this case, the NIR information layers are used together with the SWIR of 1.60 µm (HARDISKY et al., 1983). To calculate the NDII, Equation 5 was used.

$NDII=\frac{{\rho }_{0.85}-{\rho }_{1.60}}{{\rho }_{0.85}+{\rho }_{1.60}}$ Equation 5

The index ranges from -1 to 1, whose values below 0 indicate that the NIR reflectance is greater than the SWIR reflectance, which may indicate water stress in the vegetation (SRIWONGSITANON et al., 2016, SRIWONGSITANON et al., 2015).

Plant Scenescence Reflectance Index (PRSI)

The Plant Scenescence Reflectance Index - PRSI measures the senescence and health of the vegetation through the dynamics of pigmentation influenced by the chemistry of the plant which produces different spectral responses (MERZLYAK et al., 1999).

The index demonstrates the sensitivity of chlorophyll in relation to carotenoids, and negative values represent the green plant mesh with high chlorophyll composition, the range from -0.1 to 0.2 indicates the beginning of the aging stage and the final intervals indicate the final stage of canopy senescence, therefore, is a contrary analysis of the numerical scale, when compared to the other indices. Equation 6 demonstrates the calculation of the PSRI.

$PSRI=\frac{{\rho }_{0.65}-{\rho }_{0.45}}{{\rho }_{0.75}}$ Equation 6

Normalized Burn Ratio (NBR)

The Normalized Burn Ratio - NBR highlights the burned areas (KEY; BENSON, 1999; KEY, BENSON, 2006), through band mathematics similar to the previous indices and can also serve as a basis for another calculation that is “ Differenced Normalized Burn Ratio” - dNBR for the pre- and post-fire periods, more accurately assessing the severity of the event (ROY et al., 2006). The Equation 7 e Equation 8 demonstrate how the data referring to the traverses of the fire events were obtained.

$NBR=\frac{{\rho }_{0.85}-{\rho }_{2.20}}{{\rho }_{0.85}-{\rho }_{2.20}}$ Equation 7

$dNBR={NBR}_{Pr\mathrm{é}-fogo}-{NBR}_{P\mathrm{ó}s-fogo}$ Equation 8

Table 1 demonstrates the intervals considered by Miller and Thode (2006), considering the Differenced Normalized Burn Ratio by Roy et al. (2006).

Table 1
dNBR intervals

Severity level	Range	Color
High Regrowth	-500 a -250
Low Regrowth	-249 a -100
Not Burnt	-99 a 100
Low Severity	101 a 270
Low Moderate Severity	271 a 440
Moderate High Severity	441 a 600
High Severity	661 a 1.300

• Source: Miller and Thode (2006). Adapted by the authors.

The delimitation of the burned regions was based on the regions considered burned, that is, from the value of 101.

Determination of the critical threshold

By means of the burn severity mask, obtained by dNBR, it was possible to segregate the regions of interest for the extraction of pixel values. For each pixel centroid, the values of NDVI, NDII and PSRI were extracted. The PSRI, NDVI, NDII and burn severity values are obtained by calculating the spectral indices of the methodology.

The extracted values were submitted to average statistical calculations for each event. These values served to determine the general average of the analyzed period, quantitatively determining the value that represents the beginning of the criticality stage for the occurrence of fires.

RESULTS AND DISCUSSION

Determining event periods

Six periods of fire occurrence were determined, which were named Event 1, Event 2, Event 3, Event 4, Event 5 and Event 6. The dates of the images are arranged in Table 2.

Table 2
Date of event images

Event	Pre-fire	Post-fire
Event 1	11/08/2017	15/09/2017
Event 2	26/08/2018	10/09/2018
Event 3	13/05/2019	12/07/2019
Event 4	02/04/2020	12/05/2020
Event 5	05/08/2021	09/09/2021
Event 6	02/04/2022	02/05/2022

• Source: The authors (2022).

Spatial Filtering of scenes

Filtering through the masks together with the “buffering” process showed a very satisfactory result. It was possible to calculate the overall mask for each event, from the pre and post-fire masks.

The Sentinel Clound Probability product showed an excellent degree of cloud identification, even using the 50% threshold (recommended for a more homogeneous selection of the product).

The product developed with the expertise of Magno et al. (2021) provides, within the threshold of interest, excellent accuracy in the selection of shaded regions. As found by Magno et al. (2021), the shaded regions could be identified, by recording the spectral behavior of the dry and wet threshold.

The NDWI index showed great efficiency for identifying massive bodies of water, with established thresholds. The literature by Kaplan and Advan (2017) recommends, for Sentinel-2, the value of 0.1, which was used in the processing.

It was possible to observe, as in Kaplan and Advan (2017), that mountainous regions that produced extensions were mistakenly identified, indicating shaded regions or dark material.

Corroborating with the ideas of Milczarek et al. (2017), with the SWM it was possible to discretize and specialize the water surface with a high degree of precision, also allowing the identification of small bodies of water and rivers with a short border length.

It was observed that the water indices have difficulties to differentiate water bodies submitted to eutrophication processes, due to the presence of biological material in the structure of the water, as well as the dynamics of the water surface can generate a propensity to specular reflection phenomena, which is the accentuated reflection of the incidence of solar rays, impairing the recognition of the water surface as a water body.

Figure 5 demonstrates how the process identified the materials that should be removed from the scene. It is possible to observe that most of the unwanted material for processing was discretized, and the “buffer” process allowed treating the pixel transition region.

Figure 5
Mask demonstration

As observed in Santos et al. (2023), the use of masks reduced the confusion and classification of pixels not belonging to the class of vegetation and burned, which were being analyzed in the study. The coverage and use of classification of exposed soil, urban area and water, for example, were well characterized and segregated from the automatic classification for extracting index values.

Spectral Indices

The results obtained with the spectral indices explained in the research methodology will be demonstrated.

Normalized Difference Vegetation Index (NDVI)

The NDVI index could be calculated for all the studied periods, and in general, it presented behavior consistent with the research objects. The index presented a problematic behavior in the dry periods, since the absence of rain generated great hydric stress in the vegetation, impairing the photosynthetic activity. Figure 6 shows the index calculation result.

Figure 6
Example of calculated NDVI

Baptista et al. (2018) and Bento-Gonçalves et al. (2019) found mean values of 0.889. This value is considered very high in terms of verdure (considering different climatic and vegetation contexts), and by itself does not affect the region's susceptibility to fire. This research found values close to 0.582. This value already indicates an alert to the degree of stress, since the acceptable threshold for the index is [0.2, 1.0].

The values found for this index had a high degree of amplitude, indicating a heterogeneous behavior of the vegetation that burned, that is, regions with low and high rates of greenness burned. It is hard to say that the NDVI has a great impact on determining the fuel potential of vegetation.

Normalized Difference Infrared Index (NDII)

The NDII index could be calculated for all the studied periods, and in general, it presented behavior consistent with the research objects. The index showed great susceptibility to periods of drought, generating a great overestimation of the critical condition of the vegetation, mainly the low and medium size vegetation. Figure 7 shows the index calculation result.

Figure 7
Example of calculated NDII

Baptista et al. (2018) and Bento-Gonçalves et al. (2019) found, for Portugal, mean values of 0.35 and those of the present research found values in the mean of 0.014. The value found indicates that there is an average value close to the vegetation water stress threshold, according to Sriwongsitanon et al. (2015).

According to Baptista et al. (2018), it is expected that the burned regions have lower humidity compared to the unburned areas. The NDII values showed less dispersion than those of the NDVI, indicating that the pixels of the burned regions present a certain degree of similarity among themselves, when the fuel potential is contextualized. It is possible that regions with a high greenness content also burn, but it is expected that regions with a high moisture content burn less and are more persistent in the passage of flames.

Also observed by Santos et al. (2023), scars from burns are sometimes confused with water surfaces. The technique used by Santos et al. (2023) involves machine learning, and the one used in the present research involves the combination of water surface identification mask, to remove them from the image to be analyzed.

Plant Scenescence Reflectance Index (PSRI)

The PSRI index could be calculated for all periods studied, and in general, it showed behavior consistent with the research objects, as we see the Figure 8.

Figure 8
Calculated PSRI exaple

There was a tendency towards a high average of the values related to the PSRI in the regions that burned, indicating that these regions had a certain degree of senescence. Baptista et al. (2018) and Bento-Gonçalves et al. (2019) found mean values of 0.26 in the burned regions analyzed in their research. This value is within the region considered to be of senescent vegetation.

The values extracted in this study indicated means close to the senescence stage onset interval, which encompasses the interval [0.1, 0.2].

Normalized Burn Ratio (NBR)

The NBR was calculated in the pre and post-fire images, allowing the calculation of the dNBR for each event. The dNBR was used for the automatic creation and extraction of polygons from the burned areas. Within these polygons, pixel values were extracted, necessary for statistical calculations.

Although these are different regions, it was observed by Santos et al. (2023) the NBR was very efficient in identifying regions burned by thermal anomaly, as well as suggesting that this index represents the best proposal for temporal analyzes involving thermal anomalies. It was possible to identify the burned regions with a certain degree of precision, including in the temporal analysis, using the same threshold proposed for the index.

The calculation of dNBR and RdNBR were performed from NBR. It was observed that the dNBR proved to be quite efficient for determining the mask of the burned region, however, when segregated into the severity classes, it showed a certain homogeneous degree of classification of the severity classes, and the RdNBR overestimated such intervals, identifying and mixing the classes of “moderate” and “high severity”, observed in the fires of Gerais Vieira, Mucugê and Gerais Machobongo in Santos et al. (2020).

As dNBR showed a better ability to separate the severities, mainly used in homogeneous vegetation, the mask of this index was used, discretizing the burned regions from the unburned ones.

Determination of the critical threshold

The dNBR polygon allowed the extraction of values, as shown in the Figure 9.

Figure 9
Extraction of points in burned regions

Six tables were extracted from the GEE platform, as exemplified in Table 3. Each table refers to an analyzed event, so it was possible to perform statistical averages for the 6 events.

Table 3
Excerpt from the extracted mass of data

Point	NDII	NDVI	PSRI	Severity
1	0,189	0,714	-0.086	459
2	0.194	0.721	-0.095	471
3	0.180	0,707	-0,081	474
4	0,185	0,716	-0,089	456
5	0,124	0,617	-0,081	460

• Source: The authors (2022).

To determine the general critical threshold, we chose to average the values obtained in the 6 events. Systematically demonstrates the values calculated in the survey, for each year, as well as the suggested final average.

Table 4
Systematization of the values obtained

Event	NDVI	NDII	PSRI	Extracted Points
Event 1	0,416	-0,024	0.124	98.095
Event 2	0.504	0.024	0.140	119
Event 3	0,646	0.017	-0,033	68.272
Event 4	0,646	0,023	-0,089	33.968
Event 5	0,670	0,050	0,093	81.928
Event 6	0.612	0.000	0.102	72.950
Critical Threshold	<=0.580	<=0.015	>=0.150	-

• Source: The authors (2022).

The aforementioned values show a high degree of coherence, as in the studies by Baptista et al. (2018) and Bento-Gonçalves et al. (2019) and the averages, for the burned regions, of 0.26 for the PSRI, 0.85 for the NDVI and 0.35 for the NDII.

It should be noted that the climatic and vegetation context of the study region of both researches is completely different. The difference in values does not invalidate the applied methodology, since the average values found demonstrate a slightly more pessimistic perspective compared to the scenario found by Baptista et al. (2018) and Bento-Gonçalves et al. (2019).

As observed by Santos et al. (2023), the greenness values show high numbers in the rainy season. The values undergo radical changes after fire events, and indicate, before the occurrence of the event, the susceptibility to the occurrence of fires. Although Santos et al. (2023) presents a much more refined methodology, when using machine learning techniques, the essence of the research is similar to that of the present article, since they are closely related themes and with several by-products, such as spectral indices, used in both.

Therefore, the images of the periods before the fire of the aforementioned spectral indices were manipulated, in order to isolate the values present in the critical interval found. A new image was generated, for each year, containing the superposition of each critical interval of the year with the burned pixels of the image after the fire period. Figure 10 demonstrates an example of calculating regions with fuel potential in Event 5.

Figure 10
Critical zones in Event 5

It was then observed that the indication of the critical zones generated by the combination of the indexes filtered by the thresholds showed satisfactory agreement with the regions that actually burned, thus allowing the mapping of regions with fuel potential, corroborating the proposal listed by Baptista et al. (2018).

However, it should be noted that the mere spatialization of these regions can present problems such as indication of inert zones as potential fuel, indication of already burned areas as potential fuel and overestimation of critical zones.

The inert zones, such as exposed soil, are highlighted by the method due to the fact that they present a favorable spectral behavior in each index. This is due to the unfavorable valuation it receives in the indexes thanks to the type of material being observed, but it is necessary to understand that, although these areas are critical, they do not present risks of fire propagation, such as sandbanks or gravel. for example.

The burned regions also present problems in the images, as there is an unfavorable valuation in the indices. This valuation is due to the fact that these are regions that do not present values of greenness and humidity compatible with plant structures and the values of senescence are misinterpreted, generated by the burn scar.

The years 2017, 2019, 2021 and 2022 were chosen to perform a spatial quantitative analysis using the data obtained by the thresholds, as they were the periods with the largest amounts of data extracted. Table 5 shows, in hectares, the analysis from the perspective of the area actually burned (Burned Area), the critical area obtained by the thresholds (Critical Area) and the critical area actually burned (Effective Area), obtained by the intersection of the burned area with the critical área.

Table 5
Quantitative analysis of critical zones

Analysis	2017	2019	2021	2022
Burned Area	3357.355	692.616	865.019	797.285
Critical Area	1742.854	19176.207	2104.648	2411.144
Effective Area	9.814	671.423	4.682	14.83
(Effective Area) / (Burned Area)	0.56%	3.50%	0.22%	0.62%
(Effective Area) / (Critical Area)	0.29%	96.94%	0.54%	1.86%

• Source: The authors (2022).

The relationship “(Effective Area) / (Burned Area)” is observed, which indicates the percentage of the area that effectively burned, that which is within the critical region and burned at the same time, and the “(Effective Area) / (Area Critical)”, indicating how much of the critical region actually burned. The year 2019 was able to identify the critical regions well, as the burn spots were largely located within the critical regions.

FINAL CONSIDERATIONS

It is concluded that the hypothesis previously launched was confirmed, that is, it is possible to stipulate physical thresholds that determine the range of criticality to fires by means of spectral indices. The burned regions showed a pattern of values in the indices, which explained the criticality of the region.

Greenness showed slightly more varied values, indicating that senescence and moisture are factors of greater weight in determining criticality. Moisture in the burned regions was also at the threshold of the beginning of the water stress interval. It was also observed that the average senescence of the burned region is concentrated in the region of values close to the interval that is considered the beginning of the senescence stage. The vegetation that burned had varied verdure, but had moisture and senescence favorable to fire criticality.

It is suggested to deepen the study of the determination of the critical threshold for the occurrence of fires, seeking to establish mathematical models that aim to reduce the negative effects found in the results of the spatialization of the critical regions.

It is suggested to deepen the GHG platform, with the techniques of this research, aiming to develop a monitoring system of potential fuel regions, using free platforms and free data collection.

REFERENCES

• BAPTISTA, G. M. M.; BENTO-GONÇALVES, A.; VIEIRA, A. Avaliação das condições de verdor, umidade e de senescência da vegetação queimada no incêndio de Braga, Portugal, em outubro de 2017. In: II Encontro Luso-Afro-Americano de Geografia Física e Ambiente - Desafios para Afirmar a Lusofonia na Geografia Física e Ambiente, 2018, Guimarães, Portugal. Anais (on-line). Guimarães: CEGOT-UMinho, Centro de Estudos de Geografia e Ordenamento do Território da Universidade do Minho, 2018. Disponível em: http://repositorium.sdum.uminho.pt/bitstream/1822/60165/1/II_ELAAGFA_Baptista_et_al.pdf Acesso em: 22 abr. 2020.
  » http://repositorium.sdum.uminho.pt/bitstream/1822/60165/1/II_ELAAGFA_Baptista_et_al.pdf
• BARMPOUTIS, P.; PAPAIONNOU, P.; DIMIITROPOULOS, K.; GRAMMALIDIS, N. A review on early forest fire detection systems using optical remote sensing. Sensors, v. 20, 2020. https://doi.org/10.3390/s20226442
  » https://doi.org/10.3390/s20226442
• BENTO-GONÇALVES, A. Incêndios rurais - O triste fado português? Fundação Francisco Manoel dos Santos, 2022. Disponível em: https://www.ffms.pt/pt-pt/atualmentes/incendios-rurais-o-triste-fado-portugues?fbclid=IwAR2hstJhcdvec8oAhMs2vG1bZp2J2DDmoFlMorSIKrCacWCywtlj7QqrvIY, acessado em 30 mar. 2023.
  » https://www.ffms.pt/pt-pt/atualmentes/incendios-rurais-o-triste-fado-portugues?fbclid=IwAR2hstJhcdvec8oAhMs2vG1bZp2J2DDmoFlMorSIKrCacWCywtlj7QqrvIY
• BENTO-GONÇALVES, A.; VIEIRA, A.; BAPTISTA, G.; ROCHA, J.; MOURA, S. The 2017 Large wildfire of braga - evaluation of the different conditions of the burned vegetation. Geo-Eco-Trop, v. 43, n. 4, p. 627-640, 2019.
• CARELLA, E.; ORUSA, T.; VIANI, A.; MELONI, D.; BORGOGNO-MONDINO, E.; ORUSA, R. An integrated, tentative remote-sensing approach based on ndvi entropy to model canine distemper virus in wildlife and to prompt science-based management policies. MDPI: Animals, v. 12, n. 1049, 2022. https://doi.org/10.3390/ani12081049
  » https://doi.org/10.3390/ani12081049
• COLLI, G. R.; VIEIRA, C. R.; DIANESE, J. C. Biodiversity and conservation of the cerrado: recento advances and old challanges. Biodiversity and Conservation, v. 29, p. 1465-1475, 2020. https://doi.org/10.1007/s10531-020-01967-x
  » https://doi.org/10.1007/s10531-020-01967-x
• DUAN, T.; CHAPMAN, S. C.; GUO, Y.; ZHENG, B. Dynamic monitoring of NDVI in wheat agronomy and breeding trials using an unmanned aerial vehicle. Field Crop Research, v. 210, p. 71-80, 2017. https://doi.org/10.1016/j.fcr.2017.05.025
  » https://doi.org/10.1016/j.fcr.2017.05.025
• EARTH RESOURCES OBSERVATION AND SCIENCE (EROS) (USA). Washington, United States of America: 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997 . Washington: EarthExplorer, 2023. Disponível em: https://earthexplorer.usgs.gov/. Acesso em: 05 nov. 2020
  » https://earthexplorer.usgs.gov
• EUROPEAN SPACE AGENCE (ESA) (EUROPE). Europe: Copernicus Open Access Hub, s.d., 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997. Disponível em: https://scihub.copernicus.eu/dhus/#/home Acesso em: 05 nov. 2022
  » https://scihub.copernicus.eu/dhus/#/home
• FASSNACHT, F. E.; SCHMIDT-RIESE, E.; KATTENBORN, T.; HERNÁNDEZ, J. Explaining sentinel 2-based dnbr and rdnbr variability with reference data from the bird’s eye (uas) perspective. International Journal of Applied Earth Observations and Geoinformation, n. 1, 2021. https://doi.org/10.1016/j.jag.2020.102262
  » https://doi.org/10.1016/j.jag.2020.102262
• FERNÁNDEZ-MANSO, A.; QUINTANO, C. A synergetic approach to burned area mapping using maximum entropy modeling trained with hyperspectral data and viirs hotspots. Remote Sensing, v. 12, n, 858, 2020. https://doi.org/10.3390/rs12050858
  » https://doi.org/10.3390/rs12050858
• GEOPORTAL (Brasil). Brasília, Distrito Federal. Secretaria de Estado de Desenvolvimento Urbano e Habitação, Brasília: dados vetoriais de Cidades Administrativas, Vias Rodoviárias e Unidades de Conservação, 2022. Disponível em: https://www.geoportal.seduh.df.gov.br/geoportal/. Acesso em: 02 fev. 2022
  » https://www.geoportal.seduh.df.gov.br/geoportal
• GOOGLE (USA) . Google Earth Engine, 2023. 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997. Disponível em: https://code.earthengine.google.com/. Acesso em: 30 jan. 2023
  » https://code.earthengine.google.com
• GORELICK, N.; HANCHER, M.; DIXON, M.; ILYUSHCHENKO, S.; THAU, D.; MOORE, R. GEE: planetary-scale geospation analysis for everyone. Elsevier: Remote Sensing Enviroment, v.202, p. 18-27, 2017. https://doi.org/10.1016/j.rse.2017.06.031
  » https://doi.org/10.1016/j.rse.2017.06.031
• HARDISKY, M.A.; KLEMAS, V.; SMART, R.M. The Influences of Soil Salinity, Growth Form, and Leaf Moisture on the Spectral Reflectance of Spartina Alterniflora Canopies. Photogrammetric Engineering and Remote Sensing , v. 49, p. 77-83, 1983.
• HOFMANN, G. S.; CARDOSO, M. C.; ALVES, R. J. V.; WEBER, E. J.; ALEXANDRE, A. B.; TOLEDO, P. M.; PONTUAL, F. B.; SALLES, L. O.; HASENACK, H.; CORDEIRO, J. L. P.; AQUINO, F. E.; OLIVEIRA, L. F. B. The brazilian cerrado is becoming hotter and drier. Global Change Biology, v. 27, p. 4060-4073, 2021. https://doi.org/10.1111/gcb.15712
  » https://doi.org/10.1111/gcb.15712
• IBGE (Brasil). Brasília, Distrito Federal. Instituto Brasileiro de Geografia e Estatística, Brasília: dados vetoriais de biomas brasileiros. Disponível em: https://www.geoportal.seduh.df.gov.br/geoportal//. Acesso em: 02 fev. 2022
  » https://www.geoportal.seduh.df.gov.br/geoportal/
• KAPLAN, G.; ADVAN, U. Object-based water body extraction model using Sentinel-2 satellite imagery. European Journal of Remote Sensing , v. 50, n.1, p. 137-143, 2017. https://doi.org/10.1080/22797254.2017.1297540
  » https://doi.org/10.1080/22797254.2017.1297540
• KEY, C. H.; BENSON, N. C. Landscape Assessment: Sampling and Analysis Methods. In: Lutes, D.C.; Keane, R.E.; Caratti, J.F.; Key, C.H.; Benson, N.C.; Sutherland, S.; Gangi, L.J. (eds). FIREMON: Fire Effects Monitoring and Inventory System. Edition: Gen. Tech. Rep. 2006.
• KEY, C. H.; BENSON, N. C. Meansuring and remote sensing of burn severity. Proceedings Joint Fire Science Conference and Workshop. v. II. 1999.
• MAGNO, R.; ROCHI, L.; DANIELLI, R.; MATESE, A.; GENNARO, S. F. D.; CHEN, C.; SON, N.; TOSCANO, P. Agroshadow: a new sentinel-2 cloud shadow detection tool for precision agriculture. Remote Sensing , v. 13, 2021. https://doi.org/10.3390/rs13061219
  » https://doi.org/10.3390/rs13061219
• MAPBIOMAS (Brasil). Dados de Mapas das Coleções - Cobertura e Uso do Solo. Disponível em: https://mapbiomas.org/download Acesso em: 02 fev. 2020.
  » https://mapbiomas.org/download
• MARCYIN. MarcYin/SIAC: for doi. Belgium: Bruxelles, 2019a.
• MARCYIN. MarcYin/SIAC_GEE: To Publish the Code. Belgium: Bruxelles , 2019b.
• McFEETERS, S.K. The use of the normalized difference water index (NDWI) in the delineation of open water features. International Journal of Remote Sensing , p. 1425-1432, 1996. https://doi.org/10.1080/01431169608948714
  » https://doi.org/10.1080/01431169608948714
• MERZLYAK, M. N.; GITELSON, A. A.; CHIVKUNOVA, O.B.; RAKITIN, V.Y. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiologia Plantarum, v. 106, p. 135-141, 1999. https://doi.org/10.1034/j.1399-3054.1999.106119.x
  » https://doi.org/10.1034/j.1399-3054.1999.106119.x
• MILCZAREK, M.; ROBAK, A. Sentinel Water Mask (SWM) - new index for water detection on Sentinel-2 images. Conference: 7th Advanced Training Course on Land Remote Sensing , 2017.
• MILLER, J. D.; THODE, A. E. Quantifying burn severity in a heterogeneous landscape with a relative version of the delta normalized burn ratio (dnbr). Remote Sensing , v. 109, p. 66-80, 2006. https://doi.org/10.1016/j.rse.2006.12.006
  » https://doi.org/10.1016/j.rse.2006.12.006
• NURSAPUTRA, M.; LAREKENG, S. H.; HAMZAH, A. S. The ndvi algorithm utilization on the GEE platform to monitor changes in forest density in mining areas. 2021. 2nd Biennal Conference of Tropical Biodiversity: IOP Conf. Series: Earth and Environmental Science, 2021. https://doi.org/10.1088/1755-1315/886/1/012100
  » https://doi.org/10.1088/1755-1315/886/1/012100
• RIBEIRO, N.; RUECKER, G.; GOVENDER, N.; MACANDZA, V.; PAIS, A.; MACHAVA, D.; CHAUQUE, A.; LISBOA, S. N.; BANDEIRA, R. The influence off fire frequency on the structure and botanical composition of savanna ecosystems. Ecology and Evolution, v. 9, p. 8253-8264, 2019. https://doi.org/10.1002/ece3.5400
  » https://doi.org/10.1002/ece3.5400
• ROUSE, J.W.; HAAS, R.H.; SCHELL, J.A.; DEERING, D.W. Monitoring vegetation systems in the great plains with erts. 1974. In: Proceeding Of Erts‐1 Symposium. 1974. Anais ... NASA: United States.
• ROY, D. P.; BOSCHETTI, L.; TRIGG, S. N. Remote sensing of fire severity: assessing the performance of the normalized burn ratio. Geoscience and Remote Sensing Letters, v. 3, n. 1, 2006. https://doi.org/10.1109/LGRS.2005.858485
  » https://doi.org/10.1109/LGRS.2005.858485
• SANTOS, S. M. B.; BENTO-GONÇALVES, A.; FRANCA-ROCHA, W.; BAPTISTA, G. Assessment of Burned Forest Area Severity and Postfire Regrowth in Chapada Diamantina National Park (Bahia, Brazil) Using dNBR and RdNBR Spectral Indices. Geosciences, v. 10, n. 106, 2020. https://doi.org/10.3390/geosciences10030106
  » https://doi.org/10.3390/geosciences10030106
• SANTOS, S. M. B.; DUBERGER, S. G.; BENTO-GONÇALVES, A.; FRANCA-ROCHA, W.; VIEIRA, A.; TEIXEIRA, G. Remote Sensing Applications for Mapping Large Wildfires Based on Machine Learning and Time Series in Northwestern Portugal. Fire, v. 6, n. 43, 2023. https://doi.org/10.3390/fire6020043
  » https://doi.org/10.3390/fire6020043
• SILVA, S. L.; BAPTISTA, G. M. M. Análise do grau de severidade de áreas queimadas na estação ecológica de águas emendadas por meio de dados do landsat 8. Revista Brasileira de Geografia Física, v. 8, n. 2, p. 431-438, 2015. https://doi.org/10.26848/rbgf.v8.2.p431-438
  » https://doi.org/10.26848/rbgf.v8.2.p431-438
• SRIWONGSITANON, N.; GAO, H.; SAVENIJE, H. H. G.; MAEKAN, E.; SAENGSAWANG, S.; THIANPOPIRUNG, S. Comparing the normalized difference infrared index (ndii) with root zone storage in a lumped conceptual model. Hydrology and Earth System Sciences, v. 20, p. 3361-3377, 2016. https://doi.org/10.5194/hess-20-3361-2016
  » https://doi.org/10.5194/hess-20-3361-2016
• SRIWONGSITANON, N.; GAO, H.; SAVENIJE, H. H. G.; MAEKAN, E.; SAENGSAWANG, S.; THIANPOPIRUNG, S. The Normalized Difference Infrared Index (NDII) as a proxy for soil moisture storage in hydrological modelling. Hydrology and Earth System Sciences, v. 19, 2015. https://doi.org/10.5194/hessd-12-8419-2015
  » https://doi.org/10.5194/hessd-12-8419-2015
• VERNOOIJ, R.; GIONGO, M.; BORGES, M. A.; COSTA, M. M.; BARADAS, A. C. S.; WERF, G. R. V. D. Intraseasonal variability of greenhouse gas emission factors from biomass burning in the brazilian cerrado. Biogeosciences, v. 1, p. 1375-1393, 2021. https://doi.org/10.5194/bg-18-1375-2021.
  » https://doi.org/10.5194/bg-18-1375-2021

Data availability

Data citations

EARTH RESOURCES OBSERVATION AND SCIENCE (EROS) (USA). Washington, United States of America: 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997 . Washington: EarthExplorer, 2023. Disponível em: https://earthexplorer.usgs.gov/. Acesso em: 05 nov. 2020

EUROPEAN SPACE AGENCE (ESA) (EUROPE). Europe: Copernicus Open Access Hub, s.d., 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997. Disponível em: https://scihub.copernicus.eu/dhus/#/home Acesso em: 05 nov. 2022

GEOPORTAL (Brasil). Brasília, Distrito Federal. Secretaria de Estado de Desenvolvimento Urbano e Habitação, Brasília: dados vetoriais de Cidades Administrativas, Vias Rodoviárias e Unidades de Conservação, 2022. Disponível em: https://www.geoportal.seduh.df.gov.br/geoportal/. Acesso em: 02 fev. 2022

GOOGLE (USA) . Google Earth Engine, 2023. 12 imagens de satélite, color. Satélite Sentinel-2, instrumento MSI. Intervalo de tempo: de 05 mai 2016 a 02 mai. 2022. Lat. -15.6669, Long. -47.9997. Disponível em: https://code.earthengine.google.com/. Acesso em: 30 jan. 2023

IBGE (Brasil). Brasília, Distrito Federal. Instituto Brasileiro de Geografia e Estatística, Brasília: dados vetoriais de biomas brasileiros. Disponível em: https://www.geoportal.seduh.df.gov.br/geoportal//. Acesso em: 02 fev. 2022

Publication Dates

• Publication in this collection
  18 Sept 2023
• Date of issue
  2023

History

• Received
  07 Nov 2022
• Accepted
  10 Apr 2023
• Published
  14 June 2023