Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
AlphaEarth Embeddings Predict Deforestation
Demonstration of using AlphaEarth 64-dimensional satellite embeddings and similarity-weighted k-NN with WRI driver data to predict forest loss probabilities with confidence intervals.
A deforestation risk prediction system that analyzes potential forest loss using Google’s new AlphaEarth embeddings dataset. Essentially, we find similar forests and see what happened. Instead of traditional ML approaches, we use similarity search to find forest patches that looked identical in 2017, then analyze what happened to them by 2024 to predict deforestation risks. The system combines 64-dimensional satellite embeddings with WRI’s forest loss driver data using similarity-weighted k-nearest neighbors. I’ll demo live ecosystem fingerprinting, show how cosine similarity reveals forest twins, and walk through the statistical approach that generates risk probabilities with confidence intervals.
Similarity-weighted K-NN uses AlphaEarth embeddings predicting specific deforestation risk drivers.
Conservation technology using geospatial intelligence, machine learning, and custom software solutions.