We developed the following comprehensive methodology leveraging multiple Earth Observation datasets in order to elucidate changes in built-up area, urban dynamics, nightlight trends, and agricultural development in lowland Ethiopia.

Task 1. Mapping Built-up Changes

We applied two machine learning algorithms, a first unsupervised learning algorithm to reduce the search space for built-up area, and a second supervised classifier to predict built-up areas in 2000 and 2017. The workflow included the following procedures:

1. Landsat composition using Google Earth Engine
2. Creating masks with candidate built-up pixels using K-means clustering
3. Supervised classification within the built-up masks
4. Post processing

Task 2. Urbanization Patterns and Trends

Urban patterns and trends were generated using Landscan population distribution maps between 2000 and 2017.

Task 3. Nightlight trends

We used DMSP (2000 - 2013) and VIIRS (2012 - 2017) datasets to reconstruct historical nightlight trends for Ethiopia. Nightlights are well-established development indicator and have been used as a proxy for electrification, urbanization, and infrastructure changes.

Task 4. Mapping large commercial farms

Finally, we mapped commercial farms in the lowlands of Ethiopia using multiple satellite image sources, including Sentinel-2 and high-resolution satellite imagery available within the JOSM OpenStreetMap editor.