Assessment of Three Machine Learning Techniques with Open-Access Geographic Data for Forest Fire Susceptibility Monitoring—Evidence from Southern Ecuador

• Published in: Forests 2022-03, Vol.13 (3), p.474
• Main Authors: Reyes-Bueno, Fabián, Loján-Córdova, Julia
• Format: Article
• Language: English
• Subjects: Decision making; Decision trees; Elevation; Forest & brush fires; Forest fires; Humidity; Learning algorithms; logistic decision trees; logistic regression; Machine learning; Meteorological data; Moisture effects; Optimization; Rain; Seasons; spatial modeling; Strategic management; susceptibility; Topography; Variables; Vegetation
• ISSN: 1999-4907; 1999-4907
• DOI: 10.3390/f13030474

Forest fires have become a habitual threat in all types of ecosystems, which is the reason why it is necessary to improve management of the territories and optimization of prevention and means of extinction. This study compares three machine learning techniques: logistic regression, logistic decision tree, and multivariate adaptive regression spline to identify areas susceptible to forest fires in the Loja canton. In the training of the machine learning models, a multitemporal database with 1436 points was used, fed with the information from seven variables related to fuel moisture, proximity to anthropic activities, and ground elevation. After analyzing the performance of the three models, better results were observed with the LMT, thus offering application ease for local decision-makers. The results show that the technique used allowed generating a model with a good predictive capacity and that the maps resulting from the model can be updated in short periods of time. However, it is necessary to highlight the lack meteorological data availability at the local level and to encourage future researchers to implement improvements in this regard.