Predicting wildfire vulnerability using logistic regression and artificial neural networks: a case study in Brazil’s Federal District
Pablo Pozzobon de Bem A , Osmar Abílio de Carvalho Júnior![https://orcid.org/0000-0002-0346-1684](/media/client/orcid_16x16.png)
A Department of Geography, University of Brasília, Campus Universitário Darcy Ribeiro, 70904-970, Brazil.
B Department of Forest Engineering, University of Brasília, Campus Universitário Darcy Ribeiro, 70910-900, Brazil.
C Corresponding author. Email: osmarjr@unb.br
International Journal of Wildland Fire 28(1) 35-45 https://doi.org/10.1071/WF18018
Submitted: 6 February 2018 Accepted: 5 November 2018 Published: 23 November 2018
Abstract
Predicting the spatial distribution of wildfires is an important step towards proper wildfire management. In this work, we applied two data-mining models commonly used to predict fire occurrence – logistic regression (LR) and an artificial neural network (ANN) – to Brazil’s Federal District, located inside the Brazilian Cerrado. We used Landsat-based burned area products to generate the dependent variable, and nine different anthropogenic and environmental factors as explanatory variables. The models were optimised via feature selection for best area under receiver operating characteristic curve (AUC) and then validated with real burn area data. The models had similar performance, but the ANN model showed better AUC (0.77) and accuracy values when evaluating exclusively non-burned areas (73.39%), whereas it had worse accuracy overall (66.55%) when classifying burned areas, in which LR performed better (65.24%). Moreover, we compared the contribution of each variable to the models, adding some insight into the main causes of wildfires in the region. The main driving aspects of the burned area distribution were land-use type and elevation. The results showed good performance for both models tested. These studies are still scarce despite the importance of the Brazilian savanna.
Additional keywords: data mining classifiers, machine learning, remote sensing, risk, predictive models.
References
Adab H (2017) Landfire hazard assessment in the Caspian Hyrcanian forest ecoregion with the long-term MODIS active fire data. Natural Hazards 87, 1807–1825.| Landfire hazard assessment in the Caspian Hyrcanian forest ecoregion with the long-term MODIS active fire data.Crossref | GoogleScholarGoogle Scholar |
Adab H, Kanniah KD, Solaimani K (2013) Modeling forest fire risk in the north-east of Iran using remote sensing and GIS techniques. Natural Hazards 65, 1723–1743.
| Modeling forest fire risk in the north-east of Iran using remote sensing and GIS techniques.Crossref | GoogleScholarGoogle Scholar |
Ajin RS, Loghin A, Vinod PG, Jacob MK (2016) RS and GIS-based forest fire risk zone mapping in the Periyar Tiger Reserve, Kerala, India. Journal of Wetlands Biodiversity 6, 139–148.
Allgöwer B, Carlson JD, Van Wagendonk JW (2003) Introduction to fire danger rating and remote sensing – will remote sensing enhance wildland fire danger rating? In ‘Wildland fire danger estimation and mapping’. (Ed. E Chuvieco) pp. 1–20. (World Scientific Publishing: Singapore)
Alonso-Betanzos A, Fontenla-Romero O, Guijarro-Berdiñas B, Hernández-Pereira E, Canda J, Jimenez E, Luis Legido J, Muñiz S, Paz-Andrade C, Inmaculada Paz-Andrade M (2002) A neural network approach for forestal fire risk estimation. In ‘The 15th European conference on artificial intelligence, ECAI 2002’, 21–26 July, Lyon, France. (Ed. V Harmelen) pp. 643–647. (IOS Press: Amsterdam, the Netherlands)
Amatulli G, Camia A, San-Miguel-Ayanz J (2013) Estimating future burned areas under changing climate in the EU-Mediterranean countries. The Science of the Total Environment 450–451, 209–222.
| Estimating future burned areas under changing climate in the EU-Mediterranean countries.Crossref | GoogleScholarGoogle Scholar |
Bar Massada A, Syphard AD, Stewart SI, Radeloff VC (2012) Wildfire ignition-distribution modelling: a comparative study in the Huron–Manistee National Forest, Michigan, USA. International Journal of Wildland Fire 22, 174–183.
| Wildfire ignition-distribution modelling: a comparative study in the Huron–Manistee National Forest, Michigan, USA.Crossref | GoogleScholarGoogle Scholar |
Bisquert MM, Sánchez JM, Caselles V (2011) Fire danger estimation from MODIS Enhanced Vegetation Index data: application to Galicia region (north-west Spain). International Journal of Wildland Fire 20, 465–473.
| Fire danger estimation from MODIS Enhanced Vegetation Index data: application to Galicia region (north-west Spain).Crossref | GoogleScholarGoogle Scholar |
Bisquert M, Caselles E, Snchez JM, Caselles V (2012) Application of artificial neural networks and logistic regression to the prediction of forest fire danger in Galicia using MODIS data. International Journal of Wildland Fire 21, 1025–1029.
| Application of artificial neural networks and logistic regression to the prediction of forest fire danger in Galicia using MODIS data.Crossref | GoogleScholarGoogle Scholar |
Bivand R, Keitt T, Rowlingson B (2016) rgdal: bindings for the geospatial data abstraction library. R package version 1.1–10. Available at: https://cran.r-project.org/package=rgdal [Verified 30 September 2017].
Boubeta M, Lombardía MJ, Marey-Pérez MF, Morales D (2015) Prediction of forest fires occurrences with area-level Poisson mixed models. Journal of Environmental Management 154, 151–158.
| Prediction of forest fires occurrences with area-level Poisson mixed models.Crossref | GoogleScholarGoogle Scholar |
Brazil Ministry of the Environment (2016) i3GEO. Available at: http://mapas.mma.gov.br/i3geo/ [Verified 30 September 2017].
Brazilian Institute of Geography and Statistics (IBGE) (2017) Geoscience downloads. Available at: http://downloads.ibge.gov.br/downloads_geociencias.htm [Verified 30 September 2017].
Cardoso MRD, Marcuzzo FFN, Barros JR (2014) Classificação climática de Köppen-Geiger para o estado de Goiás e o Distrito Federal. Acta Geographica 8, 40–55.
| Classificação climática de Köppen-Geiger para o estado de Goiás e o Distrito Federal.Crossref | GoogleScholarGoogle Scholar |
Chang Y, Zhu Z, Bu R, Chen H, Feng Y, Li Y, Hu Y, Wang Z (2013) Predicting fire occurrence patterns with logistic regression in Heilongjiang Province, China. Landscape Ecology 28, 1989–2004.
| Predicting fire occurrence patterns with logistic regression in Heilongjiang Province, China.Crossref | GoogleScholarGoogle Scholar |
Chuvieco E, Allgöwer B, Salas FJ (2003) Integration of physical and human factors in fire danger assessment. In ‘Wildland fire danger estimation and mapping’. (Ed. E Chuvieco) pp. 197–218. (World Scientific Publishing: Singapore)
Chuvieco E, Aguado I, Jurdao S, Pettinari ML, Yebra M, Salas FJ, Hantson S, de la Riva J, Ibarra P, Rodrigues MMT, Echeverría M, Azqueta D, Román MV, Bastarrika A, Martínez S, Recondo C, Zapico E, Martínez-Vega FJ (2014) Integrating geospatial information into fire risk assessment. International Journal of Wildland Fire 23, 606–619.
| Integrating geospatial information into fire risk assessment.Crossref | GoogleScholarGoogle Scholar |
Cipriani HN, Pereira JAA, Silva RA, De Freitas SG, De Oliveira LT (2011) Fire risk map for the Serra de São Domingos municipal park, Poços de Caldas, MG. Cerne 17, 77–83.
| Fire risk map for the Serra de São Domingos municipal park, Poços de Caldas, MG.Crossref | GoogleScholarGoogle Scholar |
Conceição AA, Alencar TG, Souza JM, Moura ADC, Silva GA (2013) Massive post-fire flowering events in a tropical mountain region of Brazil: high episodic supply of floral resources. Acta Botanica Brasílica 27, 847–850.
| Massive post-fire flowering events in a tropical mountain region of Brazil: high episodic supply of floral resources.Crossref | GoogleScholarGoogle Scholar |
Costafreda-Aumedes S, Comas C, Vega-Garcia C (2017) Human-caused fire occurrence modelling in perspective: a review. International Journal of Wildland Fire 26, 983–998.
| Human-caused fire occurrence modelling in perspective: a review.Crossref | GoogleScholarGoogle Scholar |
Coutinho LM (1990) Fire in the ecology of the Brazilian Cerrado In ‘Fire in the Tropical Biota: Ecosystem Processes and Global Challenges’. (Ed. JG Goldammer) pp. 82–105. (Springer, Berlin Heidelberg, Germany)
Eskandari S, Chuvieco E (2015) Fire danger assessment in Iran based on geospatial information. International Journal of Applied Earth Observation and Geoinformation 42, 57–64.
| Fire danger assessment in Iran based on geospatial information.Crossref | GoogleScholarGoogle Scholar |
Eugenio FC, dos Santos AR, Fiedler NC, Ribeiro GA, da Silva AG, dos Santos ÁB, Paneto GG, Schettino VR (2016) Applying GIS to develop a model for forest fire risk: a case study in Espírito Santo, Brazil. Journal of Environmental Management 173, 65–71.
| Applying GIS to develop a model for forest fire risk: a case study in Espírito Santo, Brazil.Crossref | GoogleScholarGoogle Scholar |
Giglio L, Randerson JT, Van Der Werf GR (2013) Analysis of daily, monthly, and annual burned area using the fourth-generation global fire emissions database (GFED4). Journal of Geophysical Research. Biogeosciences 118, 317–328.
| Analysis of daily, monthly, and annual burned area using the fourth-generation global fire emissions database (GFED4).Crossref | GoogleScholarGoogle Scholar |
Goldberg DE (1989) ‘Genetic algorithms in search, optimization and machine learning.’ (Addison-Wesley Professional, Reading, Massachusetts)
Grubbs FE (1950) Sample Criteria for Testing Outlying Observations. The Annals of Mathematical Statistics 21, 27–58.
| Sample Criteria for Testing Outlying Observations.Crossref | GoogleScholarGoogle Scholar |
Guo F, Su Z, Wang G, Sun L, Lin F, Liu A (2016) Wildfire ignition in the forests of south-east China: identifying drivers and spatial distribution to predict wildfire likelihood. Applied Geography 66, 12–21.
| Wildfire ignition in the forests of south-east China: identifying drivers and spatial distribution to predict wildfire likelihood.Crossref | GoogleScholarGoogle Scholar |
Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH (2009) The Weka data mining software. SIGKDD Explorations 11, 10–18.
| The Weka data mining software.Crossref | GoogleScholarGoogle Scholar |
Haykin SO (2008) ‘Neural networks and learning machines.’ (Pearson: Upper Saddle River, NJ, USA)
Hijmans RJ (2016) raster: geographic data analysis and modeling. R package version 2.5–8. Available at: https://cran.r-project.org/package=raster [Verified 30 September 2017].
Hornik K, Buchta C, Zeileis A (2009) Open-source machine learning: R meets Weka. Computational Statistics 24, 225–232.
| Open-source machine learning: R meets Weka.Crossref | GoogleScholarGoogle Scholar |
Instituto Nacional de Pesquisas Espaciais (INPE) (2016) Portal do monitoramento de queimadas e incêndios. Available at http://www.inpe.br/queimadas/ [Verified 30 September 2017].
Jafari Goldarag Y, Mohammadzadeh A, Ardakani AS (2016) Fire risk assessment using neural network and logistic regression. Photonirvachak 44, 1–10.
| Fire risk assessment using neural network and logistic regression.Crossref | GoogleScholarGoogle Scholar |
Jung J, Changjae K, Jayakumar S, Seongsam K, Han S, Dong HK, Heo J (2013) Forest fire risk mapping of Kolli Hills, India, considering subjectivity and inconsistency issues. Natural Hazards 65, 2129–2146.
| Forest fire risk mapping of Kolli Hills, India, considering subjectivity and inconsistency issues.Crossref | GoogleScholarGoogle Scholar |
Köppen W (1936) Das geographische System der Klimate [The geographic system of climates]. In ‘Handbuch der Klimatologie’. (Eds W Köppen, R Geiger) pp. 1–44. (Gebrüder Bornträger: Berlin, Germany)
Kuhlmann M, Ribeiro JF (2016) Evolution of seed dispersal in the Cerrado biome: ecological and phylogenetic considerations. Acta Botanica Brasílica 30, 271–282.
| Evolution of seed dispersal in the Cerrado biome: ecological and phylogenetic considerations.Crossref | GoogleScholarGoogle Scholar |
Mahdavi A, Shamsi SRF, Nazari R (2012) Forests and rangelands’ wildfire risk zoning using GIS and AHP techniques. Caspian Journal of Environmental Sciences 10, 43–52.
Marsett RRCR, Qi J, Heilman P, Biedenbender SH, Watson MC, Am S, Weltz M, Goodrich D, Marsett RRCR (2006) Remote sensing for grassland management in the arid Southwest. Rangeland Ecology and Management 59, 530–540.
| Remote sensing for grassland management in the arid Southwest.Crossref | GoogleScholarGoogle Scholar |
Melchiori AE, Setzer AW, Morelli F, Libonati R, Cândido P de A, de Jesús SC (2014) A Landsat-TM/OLI algorithm for burned areas in the Brazilian Cerrado: preliminary results. In ‘Advances in forest fire research’. (Ed. DX Viegas.) pp. 1302–1311. (Imprensa da Universidade de Coimbra: Coimbra, Portugal)
Mohammadi F, Bavaghar MP, Shabanian N (2014) Forest fire risk zone modeling using logistic regression and GIS: an Iranian case study. Small-scale Forestry 13, 117–125.
| Forest fire risk zone modeling using logistic regression and GIS: an Iranian case study.Crossref | GoogleScholarGoogle Scholar |
Myers N, Mittermeier RA, Mittermeier CG, da Fonseca GAB, Kent J (2000) Biodiversity hotspots for conservation priorities. Nature 403, 853–858.
| Biodiversity hotspots for conservation priorities.Crossref | GoogleScholarGoogle Scholar |
Oliveira S, Oehler F, San-Miguel-Ayanz J, Camia A, Pereira JMC (2012) Modeling spatial patterns of fire occurrence in Mediterranean Europe using multiple regression and random forest. Forest Ecology and Management 275, 117–129.
| Modeling spatial patterns of fire occurrence in Mediterranean Europe using multiple regression and random forest.Crossref | GoogleScholarGoogle Scholar |
Oliveira S, Pereira JMC, San-Miguel-Ayanz J, Lourenço L (2014) Exploring the spatial patterns of fire density in southern Europe using geographically weighted regression. Applied Geography 51, 143–157.
| Exploring the spatial patterns of fire density in southern Europe using geographically weighted regression.Crossref | GoogleScholarGoogle Scholar |
Pivello VR (2011) The use of fire in the cerrado and Amazonian rainforests of Brazil: past and present. Fire Ecology 7, 24–39.
| The use of fire in the cerrado and Amazonian rainforests of Brazil: past and present.Crossref | GoogleScholarGoogle Scholar |
QGIS Development Team (2017) QGIS geographic information system. Open Source Geospatial Foundation Project. Available at http://qgis.osgeo.org [Verified 30 September 2017]
R Development Core Team (2017) R: a language and environment for statistical computing. Available at https://www.r-project.org/ [Verified 30 September 2017]
Ribeiro JF, Walter BM (2008) As principais fitofisionomias do bioma Cerrado. In ‘Cerrado: ambiente e flora’. (Eds SM Sano, SP Almeida) pp. 89–166. (Embrapa Cerrados: Brasilia, Brazil)
Rodrigues M, de la Riva J (2014) An insight into machine-learning algorithms to model human-caused wildfire occurrence. Environmental Modelling & Software 57, 192–201.
| An insight into machine-learning algorithms to model human-caused wildfire occurrence.Crossref | GoogleScholarGoogle Scholar |
Rodrigues M, de la Riva J, Fotheringham S (2014) Modeling the spatial variation of the explanatory factors of human-caused wildfires in Spain using geographically weighted logistic regression. Applied Geography 48, 52–63.
| Modeling the spatial variation of the explanatory factors of human-caused wildfires in Spain using geographically weighted logistic regression.Crossref | GoogleScholarGoogle Scholar |
San-Miguel-Ayanz J, Carlson JD, Alexander M, Tolhurst K, Morgan G, Sneeuwjagt R, Dudley M (2003) Current methods to assess fire danger potential. In ‘Wildland fire danger estimation and mapping’. (Ed. E Chuvieco) pp. 20–61. (World Scientific Publishing: Singapore)
Segurado P, Araujo MB, Kunin WE (2006) Consequences of spatial autocorrelation for niche-based models. Journal of Applied Ecology 43, 433–444.
| Consequences of spatial autocorrelation for niche-based models.Crossref | GoogleScholarGoogle Scholar |
Sitanggang IS, Yaakob R, Mustapha N, Ainuddin AN (2013) Predictive models for hotspot occurrence using decision tree algorithms and logistic regression. Journal of Applied Sciences 13, 252–261.
| Predictive models for hotspot occurrence using decision tree algorithms and logistic regression.Crossref | GoogleScholarGoogle Scholar |
Tachikawa T, Hato M, Kaku M, Iwasaki A (2011) Characteristics of ASTER GDEM version 2. In ‘2011 IEEE international geoscience and remote sensing symposium’, 24–29 July 2011, Vancouver, British Columbia, Canada. pp. 3657–3660. (IEEE, Vancouver, Canada)
Vilar del Hoyo L, Isabel MPM, Vega FJM (2011) Logistic regression models for human-caused wildfire risk estimation: analysing the effect of the spatial accuracy in fire occurrence data. European Journal of Forest Research 130, 983–996.
| Logistic regression models for human-caused wildfire risk estimation: analysing the effect of the spatial accuracy in fire occurrence data.Crossref | GoogleScholarGoogle Scholar |
Zhang Y, Lim S, Sharples JJ (2016) Modelling spatial patterns of wildfire occurrence in south-eastern Australia. Geomatics, Natural Hazards & Risk 5705, 1–16.
| Modelling spatial patterns of wildfire occurrence in south-eastern Australia.Crossref | GoogleScholarGoogle Scholar |