Students: Rahul Steiger, Flavia, Roman, Jan Wenkel
Mentor: Lionel Trebuchon
Hack4Good project responsible: Pepa Aran
IFPRI contact person: Koen Hufkens
Hack4Good is a 8-week long pro-bono program that matches data science talents from ETH Zurich with non-profit organisations that promote social causes. In close collaboration with such a non-profit, small student teams develop and implement data-driven solutions to increase these organizations' impact.
Paper from Koen Hufkens. https://www.sciencedirect.com/science/article/pii/S0168192318303484
Here are some resources to get and process the remote sensing and meteorological data.
- ERA5 reanalysis weather data: https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=form
- Sentinel-2 remote sensing data: https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-2/data-products
- Koen's repository containing code to get the data (for an old version of the extended images dataset): https://github.com/khufkens/eyes_on_the_ground
Project proposal (restricted access)
[More info](PDF here?)
-
Prediction of the type of damage done to a crop type
See definition by -
Prediction of the severity of the damage\
TODO
- Pure image prediction using the pre-trained convnext (newer resnet50) or similar model?
- Separate weather feature prediction
- Combining the two?
- Model (kedro + mlflow?)
- Report (who will read it?)
- Presentation Hack4Good
| Oct 6: Kick-off
| Oct 14: Hackday 1 - Project planning and Exploratory Data analysis
| Somewhen in October - ...: Feature engineering and Feature selection
| Somewhen - ...: Modeling
| Oct 29th - ...: Hackday2 - adjustments, alignment etc...
| End October ...: First "clean rerun" to check whether docu is sufficient and works
| November ...: Severity of damage?
| ...
| Nov 15: Final Presentations
| Nov 22: Final Event\
Project organization - Kedro Kedro How-to
Version control - Git and Gitlab Git How-to
Communication - Slack channel h4g-hs2023-ifpri