• Andrea Sulova
Recent studies suggest that due to climate change the number of wildfires across the globe will be increasing. Recently, massive wildfires hit Australia during the 2019-2020 summer season where 46 million acres of land burnt. This fire disaster is raising questions to what extent the risk of wildfires can be linked to various climate, environmental, topographical, and social factors and how to predict fire occurrence to take preventive measures. This study investigates the Australian wildfires-based on free remotely sensed data from Earth observation to uncover the general insights. In the last few years, machine learning (ML) has demonstrated to be successful in many domains due to its capability of learning from obvious but also hidden relationships. One of the aims of this study is to create an automatized process of creating a fire training dataset at a continental level with an efficient computational expense for the ML algorithms. These results of fire occurrence and no-fire occurrence locations are mapped alongside with fire causal factors. The training dataset is applied to different ML algorithms, such as Random Forest (RF), Naïve Bayes (NB), and Classification and Regression Tree (CART). The ML algorithm with the best performance, the RF model, is used to identify the driving factors using variable importance analysis. Typically, a model can learn certain properties from a training dataset to make predictions. Thus, the overall objective of this study is to disclose the fire occurrence probability across Australia as well as identify the driving factors of wildfires applying the fire occurrence dataset from the 2019-2020 summer season. Improved preventive measures can be implemented in the fire-prone areas to reduce the risk of wildfires in Australia by considering the identified factors.
Publication date4 Jun 2020
Number of pages69
ID: 333534467