Efficient Data Representation for Wildfire Predictions

Wildfires continue to pose severe threats to ecological systems, communities, and economies worldwide. Early and accurate prediction of wildfire occurrences is crucial for effective preparedness and response strategies. This study investigates the application of machine learning methods to predict global wildfire events, utilizing the SeasFire Cubes dataset — a scientific datacube designed explicitly for seasonal fire forecasting on a global scale. Data cubes, representing three-dimensional data (time, latitude, and longitude) with 54 features, present a holistic view of the Earth’s climate variables.
The outcomes of this research contribute valuable insights into the potential of machine learning for wildfire predictions. By evaluating three different approaches on the SeasFire Cubes dataset, we offer a thorough and meaningful comparison of their performance. This comparative analysis assists wildfire management and emergency response agencies in making informed decisions regarding the adoption of machine learning methodologies based on data availability and prediction requirements. The direct comparison of three diverse approaches using the same data facilitates the extraction of best practices from each method, potentially leading to the development of new machine learning models based on these insights.

Faculty Supervisor:

Steve Easterbrook

Student:

Partner:

Lviv Polytechnic National University

Discipline:

Computer science

Sector:

Artificial Intelligence; Environmental Science and Technology; Forestry

University:

University of Toronto

Program: