Predicting locust breeding ground locations from satellite data.
Create a virtual environment and install requirements.
pip install -r requirements.txt
Run the notebooks with Google Colab or appropriate Docker container.
Notebooks | Link |
---|---|
Colab Intro (Python) | |
Pseudo-Absence Generation (R) | View |
Pseudo-Absence Generation Viz (R) | View |
Presence-Only (MaxEnt) Data Generation (R) | View |
Presence-Only (MaxEnt) Modelling (R) | View |
Add Environmental and Climate Data (Python) | View |
Model Training (Python) | View |
Model Interpretation (Python) | View |
Hypothesis Testing (R) | View |
Build the image running the following.
make build
Start a docker container in bash
make bash
To launch a notebook use make run_notebook
.
For the R Docker Container add version=r
to the build and run commands.
Download and extract the preprocessed data into data/
directory from here
To run the preprocessing workflow, the following datasets are required:
- FAO Locust Observation Data
- NASA GLDAS_NOAH025_3H Dataset
- ISRIC SoilGrids (Refer to this notebook on how to download SoilGrids data)
Run the following notebooks sequentially, to generate preprocessed data
- Pseudo-Absence Generation. You can run Pseudo-Absence Generation Viz for visualization.
- Add Environmental and Climate Data
If you find this project useful in your research please consider adding the following citation:
@proceedings{yusef2021locust,
title = {On pseudo-absence generation and machine learning for locust breeding ground prediction in Africa},
author = {Ibrahim Salihu Yusuf and
Kale-ab Tessera and
Thomas Tumiel and
Sella Nevo and
Arnu Pretorius},
journal = {Advances in Neural Information Processing Systems (NeurIPS) workshop, 2021, Sydney},
year = {2021},
url = {https://arxiv.org/pdf/2111.03904.pdf},
}
NeurIPS 2021 Workshops: