A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran

Abolfazl Jaafari, Davood Mafi Gholami, Eric Zenner

Research output: Contribution to journalArticle

14 Citations (Scopus)

Abstract

The preparation of probability distribution maps is the first important step in risk assessment and wildfire management. Here we employed Weights-of-Evidence (WOE) Bayesian modeling to investigate the spatial relationship between historical fire events in the Chaharmahal-Bakhtiari Province of Iran, using a wide range of binary predictor variables (i.e., presence or absence of a variable characteristic or condition) that represent topography, climate, and human activities. Model results were used to produce distribution maps of wildfire probability. Our modeling approach is based on the assumption that the probabilities reflect the observed proportions of the total landscape area occupied by the corresponding events (i.e., fire incident or no fire) and conditions (i.e., classes) of predictor variables. To assess the effect of each predictor variable on model outputs, we excluded each variable in turn during calculations. The results were validated and compared by the receiver operating characteristic (ROC) using both success rate and prediction rate curves. Seventy percent of fire events were used for the former, while the remainder was used for the latter. The validation results showed that the area under the curves (AUC) for success and prediction rates of the model that included all thirteen predictor variables that represent topography, climate, and human influences were 84.6 and 80.4%, respectively. The highest AUC for success and prediction rates (86.8 and 84.6%) were achieved when the altitude variable was excluded from the analysis. We found slightly decreased AUC values when the slope-aspect and proximity to settlements variables were excluded. These findings clearly demonstrate that the probability of a fire is strongly dependent upon the topographic characteristics of landscapes and, perhaps more importantly, human infrastructure and associated human activities. The results from this study may be useful for land use planning, decision-making for wildfire management, and the allocation of fire resources prior to the start of the main fire season.

Original languageEnglish (US)
Pages (from-to)32-44
Number of pages13
JournalEcological Informatics
Volume39
DOIs
StatePublished - May 1 2017

Fingerprint

Bayesian Modeling
wildfires
wildfire
Iran
Fires
mountains
mountain
wildland fire management
modeling
Predictors
Curve
prediction
topography
Topography
Climate
weight-of-evidence
fire season
human activity
climate
Prediction

All Science Journal Classification (ASJC) codes

  • Ecology, Evolution, Behavior and Systematics
  • Ecology
  • Modeling and Simulation
  • Ecological Modeling
  • Computer Science Applications
  • Computational Theory and Mathematics
  • Applied Mathematics

Cite this

Jaafari, Abolfazl ; Gholami, Davood Mafi ; Zenner, Eric. / A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran. In: Ecological Informatics. 2017 ; Vol. 39. pp. 32-44.
@article{a86daf7b07bb4ea599d39fc8bde30f76,
title = "A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran",
abstract = "The preparation of probability distribution maps is the first important step in risk assessment and wildfire management. Here we employed Weights-of-Evidence (WOE) Bayesian modeling to investigate the spatial relationship between historical fire events in the Chaharmahal-Bakhtiari Province of Iran, using a wide range of binary predictor variables (i.e., presence or absence of a variable characteristic or condition) that represent topography, climate, and human activities. Model results were used to produce distribution maps of wildfire probability. Our modeling approach is based on the assumption that the probabilities reflect the observed proportions of the total landscape area occupied by the corresponding events (i.e., fire incident or no fire) and conditions (i.e., classes) of predictor variables. To assess the effect of each predictor variable on model outputs, we excluded each variable in turn during calculations. The results were validated and compared by the receiver operating characteristic (ROC) using both success rate and prediction rate curves. Seventy percent of fire events were used for the former, while the remainder was used for the latter. The validation results showed that the area under the curves (AUC) for success and prediction rates of the model that included all thirteen predictor variables that represent topography, climate, and human influences were 84.6 and 80.4{\%}, respectively. The highest AUC for success and prediction rates (86.8 and 84.6{\%}) were achieved when the altitude variable was excluded from the analysis. We found slightly decreased AUC values when the slope-aspect and proximity to settlements variables were excluded. These findings clearly demonstrate that the probability of a fire is strongly dependent upon the topographic characteristics of landscapes and, perhaps more importantly, human infrastructure and associated human activities. The results from this study may be useful for land use planning, decision-making for wildfire management, and the allocation of fire resources prior to the start of the main fire season.",
author = "Abolfazl Jaafari and Gholami, {Davood Mafi} and Eric Zenner",
year = "2017",
month = "5",
day = "1",
doi = "10.1016/j.ecoinf.2017.03.003",
language = "English (US)",
volume = "39",
pages = "32--44",
journal = "Ecological Informatics",
issn = "1574-9541",
publisher = "Elsevier",

}

A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran. / Jaafari, Abolfazl; Gholami, Davood Mafi; Zenner, Eric.

In: Ecological Informatics, Vol. 39, 01.05.2017, p. 32-44.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran

AU - Jaafari, Abolfazl

AU - Gholami, Davood Mafi

AU - Zenner, Eric

PY - 2017/5/1

Y1 - 2017/5/1

N2 - The preparation of probability distribution maps is the first important step in risk assessment and wildfire management. Here we employed Weights-of-Evidence (WOE) Bayesian modeling to investigate the spatial relationship between historical fire events in the Chaharmahal-Bakhtiari Province of Iran, using a wide range of binary predictor variables (i.e., presence or absence of a variable characteristic or condition) that represent topography, climate, and human activities. Model results were used to produce distribution maps of wildfire probability. Our modeling approach is based on the assumption that the probabilities reflect the observed proportions of the total landscape area occupied by the corresponding events (i.e., fire incident or no fire) and conditions (i.e., classes) of predictor variables. To assess the effect of each predictor variable on model outputs, we excluded each variable in turn during calculations. The results were validated and compared by the receiver operating characteristic (ROC) using both success rate and prediction rate curves. Seventy percent of fire events were used for the former, while the remainder was used for the latter. The validation results showed that the area under the curves (AUC) for success and prediction rates of the model that included all thirteen predictor variables that represent topography, climate, and human influences were 84.6 and 80.4%, respectively. The highest AUC for success and prediction rates (86.8 and 84.6%) were achieved when the altitude variable was excluded from the analysis. We found slightly decreased AUC values when the slope-aspect and proximity to settlements variables were excluded. These findings clearly demonstrate that the probability of a fire is strongly dependent upon the topographic characteristics of landscapes and, perhaps more importantly, human infrastructure and associated human activities. The results from this study may be useful for land use planning, decision-making for wildfire management, and the allocation of fire resources prior to the start of the main fire season.

AB - The preparation of probability distribution maps is the first important step in risk assessment and wildfire management. Here we employed Weights-of-Evidence (WOE) Bayesian modeling to investigate the spatial relationship between historical fire events in the Chaharmahal-Bakhtiari Province of Iran, using a wide range of binary predictor variables (i.e., presence or absence of a variable characteristic or condition) that represent topography, climate, and human activities. Model results were used to produce distribution maps of wildfire probability. Our modeling approach is based on the assumption that the probabilities reflect the observed proportions of the total landscape area occupied by the corresponding events (i.e., fire incident or no fire) and conditions (i.e., classes) of predictor variables. To assess the effect of each predictor variable on model outputs, we excluded each variable in turn during calculations. The results were validated and compared by the receiver operating characteristic (ROC) using both success rate and prediction rate curves. Seventy percent of fire events were used for the former, while the remainder was used for the latter. The validation results showed that the area under the curves (AUC) for success and prediction rates of the model that included all thirteen predictor variables that represent topography, climate, and human influences were 84.6 and 80.4%, respectively. The highest AUC for success and prediction rates (86.8 and 84.6%) were achieved when the altitude variable was excluded from the analysis. We found slightly decreased AUC values when the slope-aspect and proximity to settlements variables were excluded. These findings clearly demonstrate that the probability of a fire is strongly dependent upon the topographic characteristics of landscapes and, perhaps more importantly, human infrastructure and associated human activities. The results from this study may be useful for land use planning, decision-making for wildfire management, and the allocation of fire resources prior to the start of the main fire season.

UR - http://www.scopus.com/inward/record.url?scp=85016544653&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85016544653&partnerID=8YFLogxK

U2 - 10.1016/j.ecoinf.2017.03.003

DO - 10.1016/j.ecoinf.2017.03.003

M3 - Article

VL - 39

SP - 32

EP - 44

JO - Ecological Informatics

JF - Ecological Informatics

SN - 1574-9541

ER -