Skip to content

GeoDevelopment Observatory of Bolivia / Indicators

Satellite embeddings

What 333 municipalities look like from space, each compressed by a deep-learning model into 64 numbers: features for machine learning.

Sixty-four satellite embedding dimensions (A00–A63) for 333 of Bolivia’s 343 municipalities in 2017, averaged from the annual 10 m Google Satellite Embedding dataset (AlphaEarth Foundations). The model summarises Sentinel-2, Landsat and other imagery into abstract features that capture land cover, texture, roads and buildings. Two versions are published: emb, the area-weighted mean over each municipality, and embpw, weighted by population so that it describes where people live. The dimensions are not interpretable one by one; they are inputs to models, for example to predict poverty. The values were computed on DS4Bolivia’s polygons of the 2015 municipal map, so they are published for the 333 municipalities whose territory did not change; the six that lost territory to the four units created since (Colquechaca, Mizque, Porco, Tomave, San Ignacio and Santa Ana) and the four new units have no data.

Variables of emb

Sources

Variables of embpw

Sources

Back to the GeoDevelopment Observatory of Bolivia