Aggregated Gaussian Processes with Multiresolution Earth Observation Covariates

Authors: Harrison Zhu*, Adam Howes, Owen van Eer, Maxime Rischard, Yingzhen Li, Dino Sejdinovic, Seth Flaxman

*Corresponding author: harrison.zhu15@imperial.ac.uk

Paper

https://arxiv.org/pdf/2105.01460.pdf

Data

Download scripts available in data

Requirements

tensorflow
gpflow
matplotlib
scipy
numpy
pyjson
contextily
geopandas
pandas
sklearn
tqdm
earthengine-api
shapely

Simply run python setup.py install .

Experiments

Data Processing

Obtain access to Google Earth Engine Python API first https://earthengine.google.com/new_signup/.

Run cd data/crops/; python download_yields.py; python processed_yields.py; cd ../../. You may need to authenticate once.

Model Training

Go to the respective files in notebooks. One caveat is that to create the inducing points (saved as a csv), you need to run cd notebooks/usa_crops; python MVBAgg.py first.

Results are saved in results.

Visualisations

The visualisations are in the visualisations.ipynb notebooks for each experiment.

License

License available at License

harrisonzhu508 / MVBAgg