Crop Map Generation

End-to-end workflow for generating high resolution cropland maps.

Contents

• Creating a crop map
• Training a new model
• Setting up a local environment
• Adding new labeled data
• Tests
• Previously generated crop maps
• Acknowledgments
• Reference

Creating a crop map

To create a crop map run the following colab notebook (or use it as a guide):

Training a new model

To train a new model run the following colab notebook (or use it as a guide):

Two models are trained - a multi-headed pixel wise classifier to classify pixels as containing crop or not, and a multi-spectral satellite image forecaster which forecasts a 12 month timeseries given a partial input:

Setting up a local environment

Ensure you have anaconda installed.

1. For development

Ensure you have gcloud installed.

conda install mamba -n base -c conda-forge  # Install mamba
mamba env create -f environment-dev.yml     # Create environment with mamba (faster)
conda activate landcover-mapping            # Activate environment
gcloud auth application-default login       # Authenticates with Google Cloud

2. For shapefile notebook

conda env create -f environment-lite.yml    # Create environment
conda activate landcover-lite               # Activate environment
jupyter notebook

Adding new labeled data

Ensure the local environment is setup.

1. Add the csv or shape file for new labels into data/raw
2. In dataset_labeled.py add a new LabeledDataset object into the labeled_datasets list and specify the required parameters.

# Activate environment setup in: Setting up a local environment
conda activate landcover-mapping 

dvc pull                                    # Get latest data from dvc
earthengine authenticate                    # Authenticates Earth Engine 
python -c "import ee; ee.Initialize()"      # Will raise error if not setup

# Pull in latest EarthEngine tifs (you may need to rerun this command)
gsutil -m cp -n -r gs://crop-mask-tifs2/tifs data/

# Create features (you may need to rerun this command)
python scripts/create_features.py

dvc commit                                  # Save new features to repository
dvc push                                    # Push features to remote storage

# Push changes to github
git checkout -b'new-Ethiopia-Tigray-data'
git add .
git commit -m 'Added mew Ethiopia Tigray data for 2020'
git push

Tests

The following tests can be run against the pipeline:

flake8 . # code formatting
mypy .  # type checking
python -m unittest # unit tests

# Integration tests
python -m unittest test/integration_test_labeled.py
python -m unittest test/integration_test_model_bbox.py
python -m unittest test/integration_test_model_evaluation.py

Previously generated crop maps

Google Earth Engine:

• Kenya (post season)
• Busia (in season)

Zenodo

• Kenya (post season) and Busia (in season).

Reference

If you find this code useful, please cite the following paper:

Gabriel Tseng, Hannah Kerner, Catherine Nakalembe and Inbal Becker-Reshef. 2020. Annual and in-season mapping of cropland at field scale with sparse labels. Tackling Climate Change with Machine Learning workshop at NeurIPS ’20: December 11th, 2020