Back to Search
Start Over
Spatial pattern assessment of tropical forest fire danger at Thuan Chau area (Vietnam) using GIS-based advanced machine learning algorithms: A comparative study.
- Source :
- Ecological Informatics; Jul2018, Vol. 46, p74-85, 12p
- Publication Year :
- 2018
-
Abstract
- Thuan Chau is a serious district affected by forest fire in Vietnam, especially in 2016; however, no forest fire prediction research has been conducted for this region. Thus, knowledge of spatial patterns of fire danger of the district plays a key role in forest succession and ecological implications. This study's aim was to analyze the spatial pattern of fire danger for the tropical forest of Thuan Chau district using advanced machine learning algorithms, Support Vector Machine classifier (SVMC), Random Forests (RF), and Multilayer Perceptron Neural Network (MLP-Net). For this purpose, a GIS database for the study area was established with 564 forest fire locations and ten forest fire variables. Then, Pearson correlation method was used to assess the correlation of the variables with the forest fire. In the next step, three forest fire danger models, SVMC, RF, and MLP-Net, were trained and validated. Finally, global performance of these models was assessed using the classification accuracy (ACC), Kappa statistics (KS), Area under the curve (AUC). In addition, Wilcoxon signed-rank test was employed to check the prediction performance of these models. The result shows the three models performed well; however, the MLP-Net model has the highest prediction performance (ACC = 81.7, KS = 0.633, and AUC = 0.894), followed by the RF model (ACC = 81.1, KS = 0.621, and AUC = 0.883), and the SVMC model (ACC = 80.2, KS = 0.604, and AUC = 0.867). The result in this study is useful for the local authority and forest manager in forest management and fire suppression. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 15749541
- Volume :
- 46
- Database :
- Supplemental Index
- Journal :
- Ecological Informatics
- Publication Type :
- Academic Journal
- Accession number :
- 130876690
- Full Text :
- https://doi.org/10.1016/j.ecoinf.2018.05.009