What an embedding adds

An embedding is a set of numbers that compresses patterns from complex data so another model can use them. Instead of handing a model every raw satellite detail, a geospatial foundation model can provide a learned representation of what tends to appear across space and time.

PPE combines named variables with embeddings from Population Dynamics Foundation Models and AlphaEarth satellite representations. Named variables are easier to explain. Embeddings can capture patterns we did not think to name. Neither one replaces local measurements or tells us why a place is at risk.

The best use is combination: explicit signals for accountability, learned representations for additional pattern-finding, and field checks for reality.