Wildfire Forecasting from Satellite Imagery
OxAI Labs Earth and Space Project
Initial setup
Conda environment
We recommend installing conda
to manage the python environment.
conda create --name <your-env-name> python=3.7
conda activate <your-env-name>
Install Dependencies
pip install -r requirements_minimum.txt
Set up Google Earth Engine
You need to sign up to use Google Earth Engine in order to use this free API. Please refer to the README for Google Earth Engine for more details.
Set up Sentinelhub
This is optional, and you may not require it if you are able to set up Google Earth Engine. Please refer to the README for Sentinel Hub for more details.
Datasets
FPA-FOD Dataset
Download FPA_FOD_20170508.sqlite and place it in the resources/fpa_fod/data_dir/
directory.
MODIS fire archive
Follow the instructions in the MODIS Fire directory.
GlobFire Dataset
Follow the instructions in the GlobFire directory.
Models
The code in the models
directory is at an experimental stage, and is not intended to be used at its current state.
We developed this code to test a simple CNN model to classify images that capture wildfire and those which don't.
While the result was promising, we were only able to demonstrate this for a small dataset with obvious wildfire images.
Please see here for more discussion on the challenges of approaches that use machine learning for wildfire identification.
Web app
There are optional setup instructions required only if you want to try out our Django webapp to visualise the map.
Errors
If you are encountering errors on Mac such as
Attempting to bind to HOST environment variable
this medium article might help you solve the problem. Set your HOST
variable to localhost
in bash.
You may encounter installation issues with GeoPandas on Windows. Try
conda install -c conda-forge geopandas
If that doesn't work, follow this instruction for Windows
and install GDAL, Fiona, pyproj, rtree, and shapely from Gohlke (unofficial repository for Windows binaries).