Planetary prediction engine: Automating global models via Earth AI

Addressing humanity's most pressing global challenges — from forecasting regional food security and environmental disaster risks to tracking real-time disease outbreaks and mapping socio-economic vulnerability — requires high-fidelity geospatial modeling. However, building these models is hindered by a fragmented data ecosystem that requires specialized teams to spend weeks on manual data curation, feature engineering, and specialized spatial validation. While existing AutoML and LLM-based agents effectively automate standard machine learning pipelines, they rely on pre-curated tabular data and lack the specialized capabilities needed to autonomously handle geospatial workflows. Consequently, planetary-scale analytics remains a significant bottleneck, a limitation that is especially severe when rapid response is critical during humanitarian crises.

Today we introduce the planetary prediction engine (PPE), the latest experimental research capability within our broader Google Earth AI initiative that turns planetary information into actionable insights. Previously, we demonstrated Earth AI’s reasoning capabilities in intelligently bringing together diverse geospatial assets. As an autonomous AI system, PPE executes the full geospatial prediction workflow, from planetary-scale data discovery and cleanup to training and evaluation, directly from natural-language queries. Given a geospatial predictive query, PPE will autonomously retrieve relevant data, engineer features, train and evaluate predictive models, and generate a comprehensive report. We demonstrate that by delivering improvements over diverse benchmarks without manual intervention, PPE effectively reduces the time needed to build complex planetary prediction models from weeks that include manual data engineering to mere minutes that result in autonomous insight.

How the planetary prediction engine works

The PPE decomposes the predictive workflow into three modular stages, each orchestrated by an LLM:

• Stage 1: Intelligent geospatial data selection. Given a natural-language query, the PPE translates the prompt into strict geographic constraints (spatial granularity, join-keys, temporal scope). It then conducts grounded signal discovery, formulating domain hypotheses and identifying both direct and causal proxy signals validated against published literature. PPE systematically retrieves covariates from established geospatial repositories including Data Commons and Google Earth Engine. For signals not available in established repositories, the system performs live open-web discovery, searching government portals and academic repositories at inference time.
• Stage 2: Multimodal dataset curation. The assembled covariates are fused with pre-trained geospatial foundation model embeddings, Population Dynamics Foundation Models (PDFM) for socio-demographic latent states and AlphaEarth for satellite imagery semantics. The system implements a strict “Feature Gate” that enforces automated target leakage mitigation, evaluating every candidate covariate against four anti-leakage criteria (filtering out mathematical sub-components, shared survey data, downstream causal effects, and future temporal data) to ensure the integrity of downstream evaluations.
• Stage 3: Automated model building and prediction. Here, PPE shifts from data curation to model optimization and evaluation. The system searches over multiple model families such as regularized linear models, gradient-boosted decision trees (GBDT), and multi-layer perceptrons, with custom logic to prevent overfitting (Overfitting Guard Protocol) that pre-assesses dataset risk and implements a self-correction loop to detect and recover from generalization failures.

Critically, each stage operates independently on well-defined inputs and outputs to prevent data bottlenecks. Data artifacts are passed between stages via opaque handles rather than serialized into LLM prompts, avoiding context-window limitations.

Improvements over diverse benchmarks across tasks

We evaluated the PPE across a multidimensional matrix of machine learning paradigms, geographies, and scientific domains.

Spatial regression: US public health and environment

Across 21 CDC health indicators, the PPE’s intelligent data selection and multimodal fusion achieve a mean R² of 76.8% vs 60.0%, compared to a manual expert pipeline. We observe similar gains for predicting FEMA national risk indicators (mean R² of 64.9% vs. 60.0% baseline) and Social Vulnerability Index (mean R² of 66.2% vs. 58.6% baseline).

Super-resolution downscaling: Nigeria food security

In data-scarce regions, coarse regional reporting often obscures local vulnerability. By autonomously integrating localized market shocks, food price anomalies, and microclimate indicators, PPE doubles baseline accuracy when downscaling food security from the provincial state (ADM1) level to the local government area (ADM2) level (R² 66.1% vs. 31.5%), providing humanitarian organizations with actionable, high-fidelity vulnerability maps.

Epidemiological nowcasting: DRC Ebola outbreak

For real-time prediction of new disease transmission hotspots during the 2026 Bundibugyo ebolavirus outbreak in the Democratic Republic of the Congo, the PPE achieves a Recall@10 of 83.3%, correctly identifying 15 of 18 newly invaded health zones across five sequential weekly forecasts. This represents a +10.3 percentage point absolute improvement over the published state-of-the-art Bayesian modeling baseline (~73%), driven by fusing epidemiological signals with PDFM embeddings and PPE-selected geospatial covariates.

The power of multimodal fusion with intelligent data selection

A key finding across all experiments is the synergistic value of combining structured statistical covariates with latent foundation model embeddings. Neither modality alone captures the full picture: statistical covariates provide explicit, interpretable signals, while Population Dynamics Embeddings and AlphaEarth Foundations embeddings encode complex non-linear patterns learned from large pre-training datasets. Our ablation studies consistently show that multimodal fusion along with intelligent data selection outperforms baseline approaches, confirming that these representations are complementary rather than redundant. In future work, we want to expand these capabilities to include more geospatial data sources, and foundation model embeddings such as Remote Sensing Foundations multimodal embeddings.

Conclusion

The PPE demonstrates that autonomous AI systems can match or improve performance across a wide range of geospatial prediction tasks: from estimating chronic disease prevalence in the United States, to improving resolution of food insecurity in Nigeria and nowcasting active viral outbreaks in the Democratic Republic of the Congo. By combining intelligent data discovery, multimodal foundation model fusion, and automated model optimization, the PPE lowers the technical barrier to planetary-scale analytics, enabling rapid deployment when time-sensitive decisions are critical.

We believe this experimental research represents a meaningful step toward democratizing geospatial prediction. By shifting the focus from manual data engineering to high-level hypothesis direction, the planetary prediction engine helps researchers, humanitarian organizations, and policymakers build models without needing specialized engineering teams. While PPE is an early-stage research project, we’re excited to explore more use cases and see how this approach can help organizations better anticipate and solve complex global challenges.

Acknowledgements

We thank and acknowledge the contributions from all of the co-authors of the paper. We are grateful to the UN World Food Programme (WFP) and Vulnerability Analysis and Mapping (VAM) team for the data and research support. We also thank the Institut National de Recherche Biomédicale (INRB) for their collaboration on the DRC Ebola nowcasting, and the teams behind Data Commons, Google Earth Engine, Population Dynamics Foundation Models, and AlphaEarth for providing the foundational data and model infrastructure that power the PPE.