African cropland dataset

[alanalutz]

Description

Brings in the African Annual Cropland Dynamics dataset (AFCD) to improve cropland detection accuracy for African PAs. This was specifically done to improve results for the Rob Walton Foundation's sites of interest in Africa. AFCD is an annual cropland detection dataset at 30m resolution from 2000-2022, matching well with the spatial and temporal resolution of the other land cover datasets.

When running results in the notebooks, there is now an option at the top to select whether or not AFCD should be used in the analysis. If use_africa_cropland=False, only the default datasets are used, and cropland is defined as the cropland classes in the Global Land Cover dataset. If use_africa_cropland=True, the AFCD dataset is used, and cropland is defined as cropland agreement between Global Land Cover and AFCD. In other words, cropland detected by Global Land Cover must be confirmed by AFCD to count.

AFCD data was downloaded from here and uploaded to EE assets.

Changes

• data_processor.py: functions to load and process AFCD data
• habitat_condition.py and habitat_loss.py: alternate functions for creating habitat rasters that detect cropland using AFCD
• run_absolute_effectiveness.py and run_relative_effectiveness.py: add a flag at the top to select whether or not to use AFCD data and call AFCD-specific functions if needed
• update map styling and add legends for visualizations to share with RWF

Unrelated changes:

• relative_scores.py: Fix computation timeout bug for large PAs by exporting per-cell scores to GCS