This project focuses on predicting various features of wildfires, such as fire area, brightness, radiative power, and count of fires per day, using weather and climate statistics. The study is crucial in understanding the impact of climate change on the increasing frequency and severity of wildfires.
- Riley Neher
- Akshay Srinivasan
- Amando Xu
The increasing pervasiveness of wildfires, particularly in Australia, has been linked to climate change. This project aims to utilize machine learning models to predict wildfire characteristics based on accessible weather statistics, providing valuable insights for fire management and prevention strategies.
The project involves:
- Data cleaning and preparation from various sources including IBM’s Spot Challenge for Wildfires.
- Principal Component Analysis to address multicollinearity in the dataset.
- Development of various machine learning models (Linear Regression, Ridge Regression, Lasso Regression, Fully Connected Neural Network, LSTM, XGBoost, LightGBM, Random Forest Regressor, and Extra Trees Regressor).
- Model performance assessment using RMSE and R2 metrics.
- The study presents the model performance for different wildfire features.
- Discussions on the model's suitability for various outcome variables and their implications.
The findings provide insights into the relationship between climate factors and wildfire characteristics. It also highlights the potential of using machine learning for predicting wildfires, aiding in risk management and mitigation strategies.
- The research is specific to Australian wildfires, which may not generalize to other regions.
- The study can be extended to incorporate more diverse factors affecting wildfires.
- Riley Neher ([email protected])
- Akshay Srinivasan ([email protected])
- Amando Xu ([email protected])
Special thanks to IBM and NASA for providing the data and resources necessary for this research.