Spatial Prediction of Wildfire Susceptibility Using Global NASA MODIS Fire Products and Machine Learning Approaches

Omid Ghorbanzadeh, Khalil Valizadeh Kamran, Thomas Blaschke

Publikation: KonferenzbeitragPoster

Abstract

Although wildfires are recognized as a natural part of a forest ecosystem, increasing the frequency of the number of events, the areas damaged by the ignition, and the severity of wildfires present great challenges in forestry areas. Some environmental factors like droughts have great impacts on fire occurrence and spread, but in many cases, fires are caused by humans. Study area  The case study was the forestry area of Amol County in the Mazandaran province of northern Iran. This region is important for its natural forests and as a popular recreational centre of the country. Conclusions Although we got a different resulting spatial prediction of wildfire susceptibility maps and the highest accuracy from the RF approach, all revealed that the central, east and southern regions of the case study area are more susceptible to wildfire in the future.Our wildfire inventory maps included precise fire locations and extension which is greatly required for wildfire susceptibility assessments. We use global NASA Moderate Resolution Imaging Spectroradiometer (MODIS) fire products (from 2012 to 2017) as the wildfire inventory dataset along with sixteen conditioning factors to evaluate the potential of different machine-learning (ML) approaches to spatial prediction of wildfire susceptibility in Amol County, northern Iran. The applied ML approaches are Artificial neural network (ANN), support vector machines (SVM) and random forest (RF). The effectiveness of each ML approaches was specified by evaluation of the existing any uncertainty among the resulting wildfire susceptibility maps. In this regard, the receiver operating characteristics (ROC) curves were conducted for all resulting maps. The area under the curve (AUC) which is considered as the accuracy index was calculated for each resulting maps. Although we got a different resulting spatial prediction of wildfire susceptibility maps and the highest accuracy from the RF approach, all revealed that the central, east and southern regions of the case study area are more susceptible to wildfire in the future
OriginalspracheEnglisch
PublikationsstatusVeröffentlicht - 6 Jul 2019
VeranstaltungGI_Forum 2019: Symposium and Exhibit Geographic Information Science - University of Salzburg, Salzburg, Österreich
Dauer: 2 Jul 20195 Jul 2019
http://www.gi-forum.org
http://gi-forum.org/archive2019

Konferenz

KonferenzGI_Forum 2019
KurztitelGI_Forum
LandÖsterreich
OrtSalzburg
Zeitraum2/07/195/07/19
Internetadresse

Systematik der Wissenschaftszweige 2012

  • 105 Geowissenschaften

Zitieren