Planetary Prediction Engine: Autonomous Geospatial...

Planetary Prediction Engine: Autonomous Geospatial... | AI Research

Key Takeaways

The Planetary Prediction Engine (PPE) is an autonomous AI system designed to streamline the creation of high-fidelity geospatial models for global challenges...
Addressing critical global challenges, from food security and disaster risk to disease outbreaks and socio-economic vulnerability, demands high-fidelity geospatial modeling.
However, building predictive planetary models remains bottlenecked by a fragmented data ecosystem, requiring manual data retrieval, multimodal data curation and fusion along with iterative model selection.
We present the Planetary Prediction Engine (PPE), an autonomous AI system that executes this end-to-end workflow directly from natural-language queries.
Simultaneously, it searches over task-tailored model architecture families with automated overfitting guards.

Paper AbstractExpand

Addressing critical global challenges, from food security and disaster risk to disease outbreaks and socio-economic vulnerability, demands high-fidelity geospatial modeling. However, building predictive planetary models remains bottlenecked by a fragmented data ecosystem, requiring manual data retrieval, multimodal data curation and fusion along with iterative model selection. We present the Planetary Prediction Engine (PPE), an autonomous AI system that executes this end-to-end workflow directly from natural-language queries. PPE synthesizes multimodal datasets on the fly, retrieving spatiotemporally relevant covariates across open-web and Earth observation platforms (Data Commons, Google Earth Engine) and fusing them with geospatial foundation model embeddings (PDFM, AlphaEarth). Simultaneously, it searches over task-tailored model architecture families with automated overfitting guards. Across diverse tasks, geographies, and scientific domains, PPE consistently outperforms state-of-the-art or manually tuned expert baselines. For US spatial regression, PPE improves mean $R^2$ across 21 CDC health indicators (76.8% vs. 60.0%), FEMA national risk indices (64.9% vs. 60.0%), and the Social Vulnerability Index (66.2% vs. 58.6%). For spatial downscaling in data-scarce settings, PPE integrates localized proxies to double baseline accuracy in Nigerian food security indicators ($R^2$ of 66.1% vs. 31.5%). For epidemiological nowcasting of the 2026 DRC Bundibugyo Ebola outbreak, PPE achieves a Recall@10 of 83.3% (identifying 15 of 18 newly invaded health zones across five weekly forecasts), a +10.3 percentage-point improvement over the public state-of-the-art modeling (~73%). By combining autonomous multimodal planetary data discovery with targeted model optimization, PPE lowers the technical barrier to planetary-scale analytics, enabling rapid, customized, expert-level deployment.

The Planetary Prediction Engine (PPE) is an autonomous AI system designed to streamline the creation of high-fidelity geospatial models for global challenges like food insecurity, disease outbreaks, and socioeconomic vulnerability. Developed by researchers at Google Research and the Institut National de Recherche Biomédicale in the Democratic Republic of Congo, the system automates the labor-intensive process of data retrieval, multimodal curation, and model selection, allowing users to generate predictive models through natural-language queries.

Automating the Geospatial Workflow

Building planetary-scale models typically requires manual effort to navigate fragmented data sources, clean information, and design specific model architectures. The PPE automates this by decomposing the workflow into three modular stages: 1. Intelligent Data Selection: The system uses Large Language Models (LLMs) to parse user queries, identify the required predictive paradigm (such as spatial regression or epidemiological nowcasting), and retrieve relevant data from sources like Data Commons and Google Earth Engine. 2. Multimodal Dataset Curation: The engine fuses raw statistical covariates with high-dimensional embeddings from geospatial foundation models, specifically the Population Dynamics Foundation Model (PDFM) and AlphaEarth. 3. Automated Model Building: The system searches through various model architectures—including gradient-boosted trees and multi-layer perceptrons—while applying automated guards to prevent overfitting and target leakage.

Performance Across Global Challenges

The researchers evaluated the PPE across a matrix of tasks, geographies, and domains, finding that it consistently outperformed manually tuned expert baselines:

Epidemiological Nowcasting: During the 2026 DRC Bundibugyo Ebola outbreak, the PPE achieved a Recall@10 of 83.3% in identifying newly invaded health zones. This represents a 10.3 percentage-point improvement over the state-of-the-art Bayesian modeling baseline.
Food Security Downscaling: In Nigeria, the system successfully downscaled food security indicators from the state level (ADM1) to the Local Government Area level (ADM2). It achieved an $R^2$ of 66.1%, compared to 31.5% for the baseline.
Spatial Regression: For US-based tasks, the PPE improved mean $R^2$ scores across 21 CDC health indicators (76.8% vs. 60.0%) and the Social Vulnerability Index (66.2% vs. 58.6%) compared to expert-led pipelines.

System Capabilities and Design

The PPE is designed to lower the technical barrier for non-experts to deploy expert-level analytics. By abstracting the complexities of feature engineering and model optimization, the system enables rapid responses to time-sensitive humanitarian and policy needs. The architecture relies on LLMs as "black-box orchestrators" to manage tool calls and interpret user intent, ensuring that the chosen model is tailored to the specific constraints of the task, such as the need for mobility matrix construction in transmission modeling or demographic covariate retrieval in socioeconomic studies.

Considerations for Deployment

The effectiveness of the PPE is tied to its ability to integrate diverse data streams, including satellite imagery, climate data, and socioeconomic indicators. The researchers note that the system’s performance gains are driven by the fusion of these multimodal embeddings with intelligently selected covariates. While the system demonstrates significant improvements over traditional manual pipelines, it is intended to function as an autonomous agent that handles the end-to-end predictive pipeline, from initial query to final test-set prediction, without requiring manual intervention during the modeling process.