http://mtc-m16c.sid.inpe.br/col/sid.inpe.br/mtc-m16c/2020/12.15.14.49/doc/p14.

pdf

Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

Spatial-temporal Analysis of active fire classified by INPE's

Fire Risk Model in Brazil using Python language
Gabriel Máximo da Silva1, Bruno Vargas Adorno1, Gilberto Ribeiro Queiroz1, Thales
Sehn Körting1, Fabiano Morelli1, Silvana Amaral1, Yosio Edemir Shimabukuro1
1
Divisão de Observação da Terra e Geoinformática - Instituto Nacional de Pesquisas
Espaciais (INPE)
12.227-010 – São José dos Campos – SP – Brazil
{gabriel.maximo,bruno.adorno,gilberto.queiroz,thales.korting,
fabiano.morelli,silvana.amaral,yosio.shimabukuro}@inpe.br

Abstract. The use of programming languages such as Python to analyse

geospatial data has simplified the analysis of remote sensing data. In this study,
we evaluated the spatial-temporal distribution of active fires detected by
MODIS in Aqua Satellite by classes of fire risk, as defined by INPE Fire Risk
model. The study area was the federative units, geographical regions, and
biomes in Brazil. The temporal assessment comprised different time seasons and
the years from 2015 to 2019. Active fires were higher during winter and spring
seasons when also more Critical and High risks were noticed. The same risk
was prevalent in Northeastern and Southeastern Brazil, as well as in the
Caatinga and Cerrado biomes.

1. I d c
Fire in natural environments is the result of a combination of factors, such as vegetation
type and climate combined with human actions or natural causes [Fearnside 2006, Phillips
et al. 2009]. Therefore, it requires a complex territory-level network to be properly
managed. For such endeavors, large databases are necessary to gather information aiming
the support of fire monitoring as a way of preventing serious forest fires, especially in
fire-prone biomes, such as the Brazilian Cerrado [Schmidt et al. 2016, Tedim et al. 2016].
Fire use can be convenient for anthropological purposes but its abusive use can
worsen health problems, due to smoke-carrying combustion products [Souza et al. 2012].
Also, it features a positive feedback relationship with climate change, which can cause
harm to fire-prone ecosystems. Lately, these ecosystems have had more contact with fire,
especially in the Amazon [Aragão et al. 2018]. Monitoring fire risk is, therefore,
necessary for land management over time and within the same year, according to the
region under analysis. Considering a continental country like Brazil, with such
heterogeneity of vegetation cover, time seasons, land use, and occupation, a complex
regional monitoring system is needed [Nogueira et al. 2017].
As consequence, another typical challenge on this issue is dealing with large
volumes of data. In Brazil, the National Institute for Space Research (INPE) gathers a
database with more than 10 satellites, capable of detecting the occurrence of active fire,
generally indicating fire occurrences. Besides, Weather Forecasting and Climate Studies
Center CPTEC/INPE, has implemented fire risk observation and forecasting models,
including as variables: vegetation cover, accumulation of days without rain, air
temperature, altitude and latitude, and the active fire observation itself [Setzer et al. 2019].

153
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

The use of computational languages, such as Python allows the creation and
documentation of geospatial data processing architectures. By doing so, the scripts can
either be improved in different versions afterward or be easily adapted to other regions of
interest [Teodoro and Duarte 2013]. In this sense, such routines related to fire monitoring
can be adjusted to allow more input data or even improve time series to perform different
analyses and their new results are easily compared to previous ones [Gomes et al. 2017].
Based on the preceding, this work aimed to evaluate active fire occurrence relative
to different risk classes, according to the Fire Risk M de de e ed b INPE Wildfire
Monitoring Program. For this purpose, the analysis was performed across all Brazilian
regions, states, and biomes, also considering different seasonality, using programming
languages to provide adaptive and replicable processing routines.

2. Ma e a a d Me d
The following sections give details on the study area, data, and processing routines chosen
for this study. In summary, active fire classified by fire risk were spatially distributed,
considering seasonal assessments from 2015 to 2019. Python libraries inserted in the
programming routine, which was developed and documented in a Github platform using
Jupyter Notebook application, supported all the processing and plotting of main results.

2.1. Study Area

Three Brazilian territory delimitations were considered as spatial analysis unit: Federative
Units, or States, Geographical Regions, and Biomes (Figure 1). Either considering the
regions or states, the analysis intends to show important results for administrative
purposes [Nogueira et al. 2017]. Moreover, important insights can be realized regarding
the active fires on the biome delimitation, according to its specific vegetation and other
environmental characteristics.

Figure 1. Study area and spatial analysis units: A) Brazilian Biomes B)

Geographical Regions and Federative Units (states).

Besides, active fires were also evaluated for different seasons (summer, fall,
winter, and spring) from 2015 to 2019. Fundamentally, time assessment may stress the
possibilities of different patterns in the monitored events during a defined period.

154
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

Identifying these spatial and temporal fire risk patterns are relevant considering the
changes in climate the world is facing lately or even the planning of public policies
directed to land and environmental administration [Fonseca et al. 2019].

2.2. Database
The database for this study were mainly active fire points (shapefiles) from January 1st to
December 31st between 2015 and 2019, available at INPE's Wildfire Monitoring website.
Only data from the reference satellite (Aqua afternoon) were selected for this study. The
active fire data already bring attributes on date/time, state, biome, and fire risk when
registered. Only geographical region delimitation was complemented with data from the
Brazilian Institute of Statistics and Geography (IBGE). Season information derived from
date/time attribute in terms of days of the year. Summer was defined from December 21st
to March 20th; Fall, from March 21st to June 20th; Winter, from June 21st to September
20th, and Spring, from September 21st to December 20th.

2.3. Data processing and analysis

For data processing and analysis, e ed he f i g P h ib a ie : Pa da ,
Geopandas, Numpy, and Matplotlib. The entire sequence of commands was organized
into two Jupyter Notebooks and a module, named as riscofogo.py1. Some active fires,
whose fire risk values were null, were excluded from the analysis. A random 5% sample
was selected from active fire data in each a series of routines were applied to consolidate
all other necessary information (time seasons and regions) in the same Geodataframe (i.e
Ge a da da a ga i ed in table structure including the object geometry as an
attribute).
Numpy, Pandas, Geopandas were libraries applied for organizing the data and
operating on them. Thus, logic operators to derive time season attributes from date/time
and spatial join function were used to include geographical region classes in active fire
Geodataframe. Another fundamental process was to define classes for fire risk attributes
using logical operators, according to Setzer et al. (2019) classification (Table 1).
Table1. Risk classes according to Fire Risk Values registered in active fire database

Risk classes Fire Risk Values (RF)

Minimum RF < 0,15
Low 0,15 < RF 0,40
Medium 0,40 < RF 0,70
High 0,70 < RF 0,95
Critical RF > 0,95
Adapted from Setzer et al. (2019)

Pandas' Pivot Table function was applied to summarize the results per season,
geographical regions, biomes, and federative units, which were then represented in stack
horizontal bars, showing the relative occurrence of active fire in different risk classes.
Figures and maps showing active fire distribution in the study area were plotted using
Matplotlib. Other details can be found in the Jupyter Notebooks developed for this work.

1
https://github.com/ser-347/risco-de-fogo
155
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

3. Re
We observed more occurrences of active fire in the winter and spring seasons. Notably,
there is a pattern that persists for the entire time studied. The Critical risk class showed
more active fires in the Midwest and Southeast during winter and the Northeast during
spring (Figure 2). There are also considerable active fires in the Minimum risk class,
mainly in the Amazon and Pampa biomes. Likewise, in the summer, states such as Pará
(PA), Mato Grosso (MT), Rondônia (RO), and Tocantins (TO), which are in the Arc of
Deforestation, presented a substantial number of active fires with Minimum risk.

Figure 2. Spatial distribution of active fire in Brazil per fire risk classes from 2015
to 2019 and in different seasons.
Considering the absolute number of active fires, the Amazon biome exhibited
more active fires in the Critical than in the Minimum class between 2015 and 2017, but
this pattern reversed from 2018 onwards. In 2019, there was a peak of 1,900 active fires
in the Minimum risk class for the Amazon, reaching about twice the amount of Critical

156
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

risk active fires for the same year in the same biome. In contrast, although active fires
were classified as Minimum risk increased, Critical risk class was dominant in the
Cerrado (Figure 3).

Minimum - Amazon Critical - Amazon

Minimum - Cerrado Critical - Cerrado
3500
3000
Number of active fires

2500
2000
1500
1000
500
0
2015 2016 2017 2018 2019

Figure 3. Number of active fires classified as Minimum and Critical fire risk for
the Amazônia and Cerrado biomes between 2015 and 2019.
The mean relative occurrence of active fire by seasons, regions, biomes, and states
are shown in Figure 4. The model was more accurate during the winter and spring seasons,
followed by fall and summer when more active fires were detected at Minimum risk
locations (Figure 4.A). Regarding Brazilian administrative regions (Figure 4.B),
Northeast and Southeast presented about 80% of active fire classified as Critical risk,
whereas North and South less than 40%. Midwest showed almost 60% of active fire in
the Critical risk region.
Given the Brazilian biomes (Figure 4.C), the patterns are somewhat related to
those observed among geographical regions. The Caatinga biome (located mainly in
northeastern Brazil) presented a greater relative occurrence of active fire in Critical risk
areas (about 90%), followed by the Cerrado (about 80%). The opposite took place in the
Pampa biome, which is found in southern Brazil, followed by the Amazon, showing less
than 40% of active fires classified as Critical risk.
Finally, with concerns to the Brazilian States (Figure 4.D), they are somehow
linked to the results presented in the geographic regions (Figure 4.B). For instance, states
like the Rio Grande do Norte (RN), Sergipe (SE), and Piauí (PI), which are located in the
Northeast, and São Paulo (SP), Rio de Janeiro (RJ), and Minas Gerais (MG), in the
Southeast, were among those with the greater relative occurrence of active fire classified
as a Critical risk. Conversely, Acre (AC), Amazonas (AM) from the North or the Rio
Grande do Sul (RS) and Santa Catarina (SC) from the South showed the greater
occurrence of active fire classified as Minimum risk.

157
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

Figure 4. Mean relative occurrence of active fire by fire risk classes in different
seasons (A), regions (B), biomes (C), and states (D). The bars were arranged in
descending Critical risk order.

4. D c
Python libraries used in this study enabled active fire analysis in different states, biomes,
regions, and seasons in Brazil from 2015 to 2020. By documenting and making the
routines available in an open source platform, however, one could easily reproduce them
in other analysis units (e.g. municipal level) and/or time span.
A possible limitation to consider before replicating this analysis is the spatial
resolution of the model. Its input variables are typically in kilometric-scale, which may
restrict local studies depending on the size of the study area. However, taking advantage
of current open source culture and the popularization of geoinformatics, new modeling of
fire risk with local variables (generated from finer spatial resolution) is encouraged. Once
developed, its architecture could be shared, so other researchers could either test it, by
applying variables from their research area or contribute to its processing efficiency
[Teodoro and Duarte 2013].

158
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

Our results indicated to be coherent to other studies [Nogueira et al. 2017]. More
than 80% relative occurrence of fire in Critical or High-risk areas from Cerrado and
Caatinga, for instance, maybe explained by many days of drought and available biomass
as fire fuel in those biomes, which are important parameters for the fire risk model [Setzer
et al. 2019]. Fire is commonly used as a technique for deforestation and pasture
management in the Midwest [Schmidt and Eloy 2020] and in the North and Northeast
regions, to where agricultural frontiers are expanding [Aragão 2018, Silva 2020].
On the other hand, in the Amazon or Pampa active fires were majorly classified
as Minimum to Medium risk. According to the model's parameters, Lower risks may be
associated to regions with dense forest vegetation and frequent rainfall as commonly seen
in the Amazon, or Lower mean temperatures, as typically noticed in the Pampa [Setzer et
al. 2019]. Perhaps, considering the current shift in political scenario [Fonseca et al. 2019,
Amigo 2020] and agricultural expansion in the Amazon biome, other parameters should
be considered for future modeling, such as areas threatened by deforestation, as they have
been increasingly associated with fire events [Silvério et al. 2019].
Be ee 2015 a d 2019, INPE da aba e ec ded e ha 19,000 active fires
spread in the Amazon. Even though 2017 had the largest absolute number (5,336 active
fires), 2019 had the highest percentage of active fires from deforested areas (~34%)
[Silvério et al. 2019]. This pattern has changed over the past years and mainly after the
c e B a i ia g e e a ie i e a age da [E c ba 2019]. The ef e,
despite important Brazilian monitoring programs such as the Real-Time Deforestation
Detection System (DETER) and the Program for Monitoring Deforestation of the
Amazon by Satellite (PRODES), and the Wildfire Monitoring program itself, more
stakeholders, including politicians, are needed to mitigate this problem.
Finally, even though uncertainties may be expected in fire risk modeling, the
ongoing active fire monitoring is necessary to support the environmental agenda in Brazil.
Since 2004, Brazilian policies have played an important role to reduce deforestation and
carbon emissions and thus helping improve air quality and human health [Reddington et
al. 2015, Wiedinmyer 2015, Oliveira 2020]. However, as stated before, this scenario has
changed over the past four years, when decision-makers have been favoring agricultural
expansion over natural areas, besides dismantling environmental laws [Fonseca et al.
2019, Take action to stop the Amazon burning 2019, Amigo 2020]. Consequently,
respiratory problems may be aggravated by the constant burning of vegetation at large
distances due to particulate transport through the atmosphere [Aragão et al. 2020].

5. C c
In this study we demonstrated how using programming language, like Python, may be
useful for visualizing and analyzing geospatial data in time and space. Our focus was on
active fire and risk distribution exploratory analysis throughout Brazilian states, regions,
and biomes from 2015 to 2019 and in different seasons. Thus, it could be noted how fire
risk and possible occurrence behave differently across the study area and time considered
in the assessment.
Programming language contributed to the spatial-temporal analysis of active fire
by risk classes and allowed documentation of every processing routine to adapt or update
it in the future, whenever needed. Python's libraries like Pandas, Geopandas, Numpy, and
Matplotlib were useful for processing and analyzing data, as well as presenting the results.

159
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

In addition, using GitHub as an open source development platform to host codes like the
ones developed on these analyses may allow developers to share and adapt different
versions of processing routines.
Active fires classified as High and Critical fire risk is more frequent during winter
and spring. Regarding space, the same was verified in the Northeast and Southeast
regions, as well as in the Caatinga and Cerrado biomes. In the Amazon we identified an
interesting pattern of a greater relative register of active fires classified as Critical fire risk
until 2017, which reversed from 2018 onwards, indicating a greater likelihood of fire
occurrence in less fire-prone regions in the last years.
Lastly, although this study used different geographical limits in the analysis, fire
negative effects are usually boundaryless. This implies that, in addition to all the
monitoring work being carried out through satellite products, more actions should be
taken by the government and other agencies to curb fire and activities associated with it,
which might be challenging by itself, considering the climate change scenario.

Refe e ce
A ig , I. 2020. The A a F agi e F e. Nature 578(February): 505 7.
Aragão, L. E. O. C.; et al. 21st Century drought-related fires counteract the decline of
Amazon deforestation carbon emissions. Nature Communications, v. 9, n. 1, pp. 1 12,
2018.
Aragão, L. E. O. C.; et al. O desafio do Brasil para conter o desmatamento e as queimadas
na Amazônia durante a pandemia por COVID-19 em 2020: implicações ambientais,
sociais e sua governança. São José dos Campos, 2020. 34p. SEI/INPE:
01340.004481/2020-96/5543324.
Escobar, H. B a i def e a i i e di g and 2020 will be worse. Science (80-. ).
10 (2019), doi:10.1126/science.aba3238.
Fearnside, P. M. Desmatamento na Amazônia: dinâmica, impactos e controle. Acta
Amazonica, v. 36, n. 3, p. 395 400, 2006.
Fonseca, M. G. et al. Effec f c i a e a d a d e cha ge ce a i fi e babi i
during the 21st century in the Brazilian Amazon. Global Change Biology, v. 25, n. 9,
p. 2931 2946, 15 set. 2019.
Gomes, V.; et al. Um ambiente para análise exploratória de grandes volumes de dados
geoespaciais: Explorando risco de fogo e focos de queimadas. Proceedings of the
Brazilian Symposium on GeoInformatics, pp. 301 309, 2017.
IBGE Instituto Brasileiro de Geografia e Estatística. Malha Municipal Digital da
Di i P ic Ad i istrativa Brasileira. Versão 2018. Rio de Janeiro: IBGE, 2018.
Retrieved from: https://www.ibge.gov.br/geociencias/organizacao-do-
territorio/malhas-territoriais/15774-malhas.html?=&t=sobre.
IBGE Instituto Brasileiro de Geografia e Estatística. Mapa de biomas e sistema costeiro-
marinho do Brasil. Rio de Janeiro: IBGE, 2019. 1 mapa. Escala 1:250 000. Projeção
policônica. Retrieved from: https://www.ibge.gov.br/geociencias/informacoes-
ambientais/estudos-ambientais/15842-biomas.html?edicao=25799&t=acesso-
aoproduto.

160
Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 153-161

Nogueira, J. M. P.; et al. Spatial pattern of the seasonal drought/burned area relationship
across Brazilian biomes: Sensitivity to drought metrics and global remote-sensing fire
products. Climate, v. 5, n. 42, pp. 1-21, 2017.
Oliveira, G. de.; et al. Smoke pollution's impacts in Amazonia. Science.
2020;369(6504):634-635. doi:10.1126/science.abd5942.
Phillips, O. L.; et al. Drought Sensitivity of the Amazon Rainforest. Science, v. 323, n.
5919, p. 1344 1347, 6 Mar. 2009.
Reddington, C. L.; et al. Air quality and human health improvements from reductions in
deforestation-related fire in Brazil. Nature Geoscience, v. 8, n. 10, p. 768 771, 16 out.
2015.
Schmidt, I. B.; et al. Experiências internacionais de manejo integrado do fogo em áreas
protegidas recomendações para implementação de manejo integrado de fogo no
Cerrado. Biodiversidade Brasileira, v. 6, n. 2, pp. 41 54, 2016.
Schmidt, I. B.; Eloy, L. Fire regime in the Brazilian Savanna: Recent changes, policy and
management. Flora: Morphology, Distribution, Functional Ecology of Plants, v. 268,
n. October 2019, p. 151613, 2020.
Setzer, A. W.; Sismanoglu, R.A.; Santos, J.G.M. Método do cálculo do risco de fogo do
programa do INPE - Versão 11, junho/2019. São José dos Campos: INPE, 2019. 27 p.
Silva, P. S.; et al. Drivers of Burned Area Patterns in Cerrado: The Case of Matopiba
Region. n. Idl, p. 542 547, 2020.
Silvério, D. V.; et al. Amazônia em chamas. Brasília, DF. IPAM, 2019.
Souza, L. S. de; et al. Air quality photochemical study over Amazonia Area, Brazil.
International Journal of Environment and Pollution, v. 48, n. 1 4, pp. 194 202, 2012.
Take Ac i S he A a B i g. 2019. Nature 573(7773): 163 163.
http://www.nature.com/articles/d41586-019-02615-3.
Tedim, F. et al.; A wildfire risk management concept based on a social-ecological
approach in the European Union: Fire Smart Territory. International Journal of
Disaster Risk Reduction, v. 18, pp. 138 153, 2016.
Teodoro, A. C.; Duarte, L. Forest fire risk maps: A GIS open source application - a case
study in Norwest of Portugal. International Journal of Geographical Information
Science, v. 27, n. 4, pp. 699 720, 2013.
Wiedinmyer, C. Breathing easier in the Amazon. Nature Geosci 8, 751 752 (2015).
https://doi.org/10.1038/ngeo2550.

161