How mzhen77/neural-ca-wildfire works
Neural CA Wildfire trains a CNN-parameterized probabilistic cellular automaton to predict daily fire spread from LANDFIRE terrain/fuel, ERA5 wind, and observed fire histories. The main workflow loads and preprocesses configured events, initializes the model, simulates spread, evaluates losses, and saves checkpoints; included notebooks use a saved checkpoint for metric and spatial plots. The graph follows the sampled training path and README-documented data extraction and plotting capabilities; notebook internals were not sampled.
Main entry points: main.py Training entry point · training.py Optimizer and training loop · data_extractor.py HDF5 extractor · data_loader.py Event loader
Components 18
Data preparation
- HDF5 extractor
data/data_extractor.py - HDF5 dataset
data/hdf5/dataset.hdf5 - Event rasters
data/data_real - Event loader
src/data_loader.py - Feature preprocessing
src/preprocessing.py
Fire simulation
- Fuel embedding
src/model.py - Multi-scale CNN
src/model.py - Wildfire CA
src/model.py - Simulation loops
src/simulation.py
Training and evaluation
- Training entry point
src/main.py - Training configuration
src/config.yaml - Optimizer and training loop
src/training.py - Loss metrics
src/loss.py
Results and analysis
- Saved checkpoints
result_saved/best_model - Metric plots
src/paper_plots_1d.ipynb - Spatial plots
src/paper_plots_2d.ipynb - Generated figures
result_saved/paper_plots
Other components
- Researcher
Connections 22
- Researcher → HDF5 extractor: runs extraction
README.md - HDF5 extractor → HDF5 dataset: reads
README.md - HDF5 extractor → Event rasters: writes events
README.md - Researcher → Training entry point: starts training
README.md - Training entry point → Training configuration: loads config
src/main.py - Training entry point → Event loader: loads events
src/main.py - Event loader → Event rasters: reads rasters
README.md - Event loader → Feature preprocessing: preprocesses inputs
src/data_loader.py - Training entry point → Fuel embedding: initializes model
src/main.py - Fuel embedding → Multi-scale CNN: embeds fuel
src/model.py - Multi-scale CNN → Wildfire CA: generates parameters
src/model.py - Training entry point → Optimizer and training loop: starts training
src/main.py - Optimizer and training loop → Simulation loops: simulates spread
src/training.py - Simulation loops → Wildfire CA: steps CA
src/simulation.py - Optimizer and training loop → Loss metrics: computes loss
src/training.py - Optimizer and training loop → Saved checkpoints: saves model
- Researcher → Metric plots: runs notebook
README.md - Researcher → Spatial plots: runs notebook
README.md - Metric plots → Saved checkpoints: loads checkpoint
README.md - Spatial plots → Saved checkpoints: loads checkpoint
README.md - Metric plots → Generated figures: writes figures
README.md - Spatial plots → Generated figures: writes figures
README.md