This repository contains the code and notebooks for the technical validation part of the "SeasFire as a Multivariate Earth System Datacube for Wildfire Dynamics" paper. The technical validation is divided into three key components:
- Visual Inspection
Folder: /visual_inspection
Description: Coming soon! This section focuses on visually inspecting SeasFire data. Notebooks and scripts within this folder guide the user through visualization techniques, enabling a comprehensive understanding of the multivariate Earth system datacube for wildfire dynamics. The notebook is used to generate the results of figure 5.
- Causality
Folder: /causality
Description: Explore causality relationships within the SeasFire dataset. In this notebook, we utilize the PCMCI algorithm to identify potential causal factors influencing wildfire dynamics as well as their lagged effects. The notebook is used to generate the results of figure 6. The illustration of figure 6 is done by the Makie.jl library.
- Machine Learning Modeling
Folder: /ml_experiments
Description: Implement machine learning models to predict wildfire dynamics using SeasFire data. Code within this section covers the development, training, and evaluation of U-Net++ model for burned area pattern forecasting that can be defined as a segmentation task.