Plain-English search over AlphaEarth satellite embeddings, OpenStreetMap and NASA POWER climate,
in four geographical study areas.
Just a quick demo to explore using satellite embeddings with climate data for natural language
queries. Be gentle, it's not production-grade and will make mistakes, it needs further work for wider coverage of vocabulary and land types. 🙂
How it works, and how well: see the blog post .
Method
Method 1: Aligning text and satellite embeddings (AETHER's recipe)
Method 2: Aligning text and satellite embeddings, with NASA POWER for climate
Method 3: Using satellite embeddings to build a classifier with AI Planner
Examples: click one to run it
How many places to show (most similar)
Top 0.5% of the study area
Top 1%
Top 2%
Top 5%
Data: AlphaEarth Foundations Satellite Embedding (Google and Google DeepMind), USDA NASS Cropland Data Layer,
USFWS National Wetlands Inventory, USGS/LBNL US Large-Scale Solar PV Database, NASA POWER (NASA Langley
Research Center), © OpenStreetMap contributors. Imagery: USGS The National Map. Language model:
openai/gpt-oss-120b (open weights) via Hugging Face. Answers come from automated maps and can be wrong.