Google Unveils Planetary Prediction Engine for Automated Geospatial Workflows

Google Research has unveiled the Planetary Prediction Engine (PPE), an experimental tool designed to automate complex geospatial workflows from natural-language queries. PPE is part of the Google Earth AI program, aimed at bypassing bottlenecks in spatial data analysis by handling tasks from data discovery to model evaluation.

Geospatial analysts and researchers have long faced challenges collating fragmented data ecosystems that require weeks of manual retrieval, cleaning, and integration. PPE seeks to solve these problems using off-the-shelf Large Language Models (LLMs) as orchestrators in three modular stages: intelligent data selection, multi-model dataset curation, and automated model building.

The system uses natural-language prompts to translate geographic constraints, spatial granularities, and temporal scopes into specific queries. It then fuses results with trained geospatial foundation models, such as Population Dynamics Foundation Models for socio-economic work and AlphaEarth for satellite imagery. The third stage involves optimisation, testing multiple model families while employing an 'overfitting guard protocol' to secure model validity.

Initial testing indicates that the automated system can match or improve upon manual baseline performances across various benchmarks, including public health indicators, food security, and environmental risk assessment. Google hopes PPE will shift operational focus toward higher-level analysis by reducing the time needed to build complex models from weeks to minutes.