Next Day Wildfire Spread: A Machine Learning Data Set to Predict Wildfire Spreading from Remote-Sensing Data
This is the code accompanying the paper "Next Day Wildfire Spread: A Machine Learning Data Set to Predict Wildfire Spreading from Remote-Sensing Data".
The dataset can be found on Kaggle.
If you use this code or data, please cite:
F. Huot, R. L. Hu, N. Goyal, T. Sankar, M. Ihme, and Y.-F. Chen, “Next Day Wildfire Spread: A Machine Learning Data Set to Predict Wildfire Spreading from Remote-Sensing Data”, arXiv preprint, 2021.
dataset_demo.ipynb is a Jupyter Notebook demonstrating how to load and parse the data set.
To export the data yourself, ensure you have a Google Storage bucket. Here is an example command, substituting your own bucket.
python3 -m simulation_research.next_day_wildfire_spread.data_export.export_ee_training_data_main \
--bucket=${BUCKET} \
--start_date="2020-01-01" \
--end_date="2021-01-01"
With the output of the previous export job, you can extract ongoing fires (fires for which there is at least one positive fire label in the PrevFireMask and in the FireMask) and write them to TFRecords. Here is an example command, substituting the location of your data (copied from Google Cloud to a local directory).
python3 -m simulation_research.next_day_wildfire_spread.data_export.extract_ongoing_fires_main \
--file_pattern="${DATA_DIR}/*.tfrecord.gz" \
--out_file_prefix="${DATA_DIR}/"