Fire Emissions AI is an artificial neural network designed to predict fire emissions information based on previously collected data for an area. Fire Emmissions AI is trained using freely available data provided by the NASA EarthData Global Fire Emissions Database (GFED) V4.11.
It is advised that you run all commands from within the root directory of the project. First you will need a .csv file with lines in the following format. Each value should be extracted directly from the GFED to ensure the correct format of individual values.
[Year],[Month],[Latitude],[Longitude],[Region ID],[BB],[NPP],[Rh],[C],[DM],[Burned Area]
- You will need Python 3.5 or greater installed.
- Install Pipenv, if you do not already have it, with
$ pip install pipenv. - Install dependencies for the project using
$ pipenv install - Run the unit tests and linters to ensure correct behavior
$ pipenv run python setup.py test. - Run the predict.py with
$ pipenv run python fireemissionsai/predict.py [path to .csv].
The repository contains two main scripts; fireemissionsai/predict.py and fireemissionsai/preprocess.py. The former contains the code relating to the neural network directly (including training, testing, and predicting), the latter contains the code for a utility used to convert the Global Fire Emissions Database files into training examples for the predictor. The preprocess.py outputs .csv files that are used for training the predictor.
To train the predictor on your own data simply do the following.
- Run the preprocess.py with
$ pipenv run python fireemissionsai/preprocess.py [GFED hdfs directory]. - Run the predict.py with
$ pipenv run python fireemissionsai/predict.py [path to .csv] --retrain --persist.
Using the --persist flag will save your newly trained model and weights over the existing model. To train a model for one-time-use simply omit the --persist flag.
- Randerson, J.T., G.R. van der Werf, L. Giglio, G.J. Collatz, and P.S. Kasibhatla. 2017. Global Fire Emissions Database, Version 4.1 (GFEDv4). ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1293