AODisaggregation: toward global aerosol vertical profiles

Getting started

Download curated dataset to run experiments here.

Run from root directory

Stochastic variational Gaussian process model to reconstruct extinction profiles

$ python run_svgp_vertical_profile.py --cfg=cfg/svgp.yaml --o=path/to/output/directory --plot

Exponential idealized baseline to reconstruct extinction profiles

$ python run_idea_vertical_profile.py --cfg=cfg/ideal.yaml --o=path/to/output/directory --plot

Example of vertical slices at latitude 51.29 of groundtruth ECHAM-HAM extinction profiles, predicted extinction posterior mean and 2.5%/97.5% quantiles of predicted extinction distribution

Reproduce experiments

SVGP Run the bash script to repeat the experiment with multiple seeds

$ bash repro/repro_svgp.sh --device=cpu

Aggregate results across seeds to recover results from the paper

$ python aggregate_results.py --i=experiments/data/outputs/svgp --o=experiments/data/outputs/svgp

Installation

Code implemented in Python 3.8.0

Setting up environment with `pyenv`

Create and activate environment

$ pyenv virtualenv 3.8.0 venv
$ pyenv activate venv
$ (venv)

Install dependencies

$ (venv) pip install -r requirements.txt

References

@article{bouabid2024aerosol,
  title={Aerosol optical depth disaggregation: toward global aerosol vertical profiles},
  author={Bouabid, Shahine and Watson-Parris, Duncan and Stefanovi{\'c}, Sofija and Nenes, Athanasios and Sejdinovic, Dino},
  journal={Environmental Data Science},
  volume={3},
  pages={e16},
  year={2024},
  publisher={Cambridge University Press}
}