GIS-based forest fire susceptibility assessment by random forest artificial neural network and logistic regression methods

Eslami R., . and Azarnoush M., . and Kialashki A., . and Kazemzadeh F., . GIS-based forest fire susceptibility assessment by random forest artificial neural network and logistic regression methods. pp. 173-184. ISSN 0128-1283

The knowledge and prediction of spatial distribution of forest fire is essential for improving fire prevention strategies in forest areas. Forest fire susceptibility maps of the Babolrood Watershed in the Mazandaran Province of Iran were obtained from random forest artificial neural network and logistic regression models. The important factors identified to affect forest fires include first and secondary topography climate vegetation cover and related human activities. Forest fire susceptibility maps were prepared using three models and the accuracy of the results was evaluated using validation datasets kappa coefficient (K) and area under the receiver operating characteristic curve (AUC). All three methods produced forest fire susceptibility maps of reasonable accuracy; artificial neural network model with K 0.61 and AUC 0.88; random forest model with K 0.64 and AUC 0.93 and logistic regression model with K 0.52 and AUC 0.79. These results showed that the accuracy of forest fire susceptibility map obtained from the random forest method was slightly higher. According to the random forest results 6.18 and 16.08 of the study area had very high and high potential for fire occurrence respectively. In general the aforementioned methods can be applied for forest fire susceptibility mapping in forest areas with similar conditions.

Full text available from: Official URL: https://info.frim.gov.my/infocenter_applications/j...

Abstract

The knowledge and prediction of spatial distribution of forest fire is essential for improving fire prevention strategies in forest areas. Forest fire susceptibility maps of the Babolrood Watershed in the Mazandaran Province of Iran were obtained from random forest artificial neural network and logistic regression models. The important factors identified to affect forest fires include first and secondary topography climate vegetation cover and related human activities. Forest fire susceptibility maps were prepared using three models and the accuracy of the results was evaluated using validation datasets kappa coefficient (K) and area under the receiver operating characteristic curve (AUC). All three methods produced forest fire susceptibility maps of reasonable accuracy; artificial neural network model with K 0.61 and AUC 0.88; random forest model with K 0.64 and AUC 0.93 and logistic regression model with K 0.52 and AUC 0.79. These results showed that the accuracy of forest fire susceptibility map obtained from the random forest method was slightly higher. According to the random forest results 6.18 and 16.08 of the study area had very high and high potential for fire occurrence respectively. In general the aforementioned methods can be applied for forest fire susceptibility mapping in forest areas with similar conditions.

[error in script]

Item Type:	Article
AGROVOC Term:	Forest fires
AGROVOC Term:	Topography
AGROVOC Term:	Data collection
AGROVOC Term:	Climate
AGROVOC Term:	Geographical information systems
AGROVOC Term:	Ai (artificial intelligence)
AGROVOC Term:	Regression analysis
Depositing User:	Mr. AFANDI ABDUL MALEK
Last Modified:	24 Apr 2025 00:55
URI:	http://webagris.upm.edu.my/id/eprint/9832

Actions (login required)

View Item