Environment setup for Python 3.11

Create a new virtual environment with

python3 -m venv env

activate it

source env/bin/activate

install requirements

pip3 install -r requirements.txt 

change directory

cd src/

install package

python setup.py install

Data

The notebooks directory contains examples for yield and crop types data. The script 'install_cropdata.sh' in scripts/data/ installs several data sources for crop types from different climate regions for domain adaptation experiments. Nevertheless, the dataset was not expanded due to emerging data sources. For future experiments with crop types, there is now very interesting benchmark data such as EuroCrops [6].

Paper self-supervised learning

The paper used crop type data (time series) from Sentinel-2 for Bavaria to train a model using data from 2016 and 2017 and apply it to 2018. Experiments were also run with 5 and 10 percent data from 2018. 2018 has deviating climate conditions compared to 2016 and 2017. Two example python files for reproducing the results can be found under src/experiments/paper.

Yields

The yield data at the field level were part of the publication 'Prediction of multi-year winter wheat yields...' and were documented and collected in detailed investigations [4]. The example 'notebooks/demo_yields_data.ipynb' shows how to load yield data. It includes data from a combine harvester as well as time series for weather and Sentinel-2 data. The notebook 'yield_pred_pixel.ipynb' predicts yields for each pixel. In addition to the combine harvester data, high-precision weighed yield data is also available for each field to investigate predictions at the field level or to compare with combine harvester yields. Mainly winter wheat was considered, but some winter barley yields are also available.

Citation

These data sets include yields and crop types and were introduced in following publications:

@article{Data1,
title = {Prediction of multi-year winter wheat yields at the field level with satellite and climatological data},
journal = {Computers and Electronics in Agriculture},
volume = {194},
pages = {106777},
year = {2022},
issn = {0168-1699},
doi = {https://doi.org/10.1016/j.compag.2022.106777},
url = {https://www.sciencedirect.com/science/article/pii/S0168169922000941},
author = {Michael Marszalek and Marco Körner and Urs Schmidhalter},
}

@article{Data2,
author = {Marszalek, Michael and Saux, B. and Mathieu, P.-P and Nowakowski, Artur and Springer, Daniel},
year = {2022},
month = {05},
pages = {1327-1333},
title = {SELF-SUPERVISED LEARNING – A WAY TO MINIMIZE TIME AND EFFORT FOR PRECISION AGRICULTURE?},
volume = {XLIII-B3-2022},
journal = {ISPRS - International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences},
doi = {10.5194/isprs-archives-XLIII-B3-2022-1327-2022}
}

References

[1] Breizhcrops, https://breizhcrops.org
[2] Radiant Earth Foundation, https://www.radiant.earth/
[3] Remelgado, R., Zaitov, S., Kenjabaev, S. et al. A crop type dataset for consistent land cover classification in Central Asia. Sci Data 7, 250 (2020). https://doi.org/10.1038/s41597-020-00591-2
[4] Chair of Plant Nutrition, TUM, https://www.pe.wzw.tum.de/
[5] Lightly, https://www.lightly.ai/
[6] EuroCrops, https://www.eurocrops.tum.de/index.html