source: Google Research: Planetary prediction engine: Automating global models via Earth AI

level: research

google research unveiled the planetary prediction engine, an experimental ai system that automates the entire geospatial modeling workflow. given a natural language query, it discovers data, engineers features, trains models, and generates reports without human intervention. the system targets public health, food security, environmental risk, and socioeconomic prediction tasks. it is part of the broader google earth ai initiative, which aims to turn planetary information into actionable insights. the engine reduces model building time from weeks of manual data engineering to minutes of autonomous insight.

the engine uses three modular stages orchestrated by large language models. first, it selects relevant geospatial data from repositories like data commons and google earth engine, or discovers open web sources. second, it curates datasets by fusing statistical covariates with embeddings from population dynamics foundation models and alphaearth satellite imagery. a feature gate filters out target leakage. third, it searches over linear models, gradient-boosted trees, and neural networks with overfitting guards. in tests, it achieved mean r-squared of 76.8% versus 60.0% for cdc health indicators, and 66.1% versus 31.5% for downscaling food security in nigeria.

the system also showed strong results in epidemiological nowcasting, achieving recall@10 of 83.3% for predicting ebola spread in the democratic republic of the congo, a 10.3 percentage point improvement over a bayesian baseline. ablation studies confirmed that combining structured covariates with foundation model embeddings outperforms either alone. the engine is early-stage research, but google plans to expand data sources and embeddings. it could help humanitarian organizations and policymakers build models without specialized engineering teams, especially during crises when rapid response is critical.

why it matters: this system could let non-experts quickly build accurate geospatial models for disease outbreaks, food insecurity, and disaster risk, speeding up humanitarian response.

source: Google Research: Planetary prediction engine: Automating global models via Earth AI