Researchers have developed an innovative artificial intelligence system that can automatically gather and analyze vast amounts of geospatial data to predict complex planetary-scale issues like food shortages, disaster risks, and disease spread. This new tool, called the Planetary Prediction Engine (PPE), aims to simplify and speed up the process of creating accurate, location-based forecasts—something that currently requires significant manual effort and expert knowledge.
Key Takeaways
- PPE autonomously collects and fuses diverse geospatial data from multiple online sources, including satellite observations and public data repositories.
- The system integrates cutting-edge “foundation model” embeddings, a type of AI representation that captures complex spatial patterns.
- Across various tests, PPE outperformed expert-crafted models in predicting health indicators, disaster risk, and social vulnerability in the US, as well as food security in Nigeria.
- In tracking the 2026 Ebola outbreak in the Democratic Republic of Congo, PPE improved prediction accuracy and successfully identified newly affected areas faster than existing models.
The Planetary Prediction Engine represents a significant step forward in geospatial modeling by automating a traditionally fragmented and labor-intensive process. Normally, building predictive models at this scale involves manually searching for relevant data sets, cleaning and combining information from different sources, and selecting the best statistical or machine learning methods to analyze the data. PPE streamlines all these steps into a single AI-driven workflow that starts with a simple natural-language query—meaning users can ask questions in plain English and receive ready-to-use predictions.
At the heart of PPE is its ability to synthesize “multimodal” data—information coming from different types of sources, such as satellite images, public health records, and social vulnerability indices. It taps into platforms like Google Earth Engine and Data Commons to dynamically retrieve the most relevant data based on location and time. The system then uses foundation model embeddings, which are advanced AI techniques that convert complex geospatial information into a format that the prediction models can easily interpret. This allows PPE to recognize subtle spatial patterns that might be missed by traditional approaches.
Another innovative aspect of PPE is its automated search for the best model architecture tailored to the specific prediction task. It tries out different types of models while guarding against overfitting—a common problem where a model performs well on training data but poorly on new data. This ensures that the predictions are both accurate and generalizable.
In practical tests, PPE showed remarkable improvements over existing expert-designed models. For example, when predicting 21 health indicators across the United States, it raised the average accuracy (measured by the statistical metric R²) from 60% to nearly 77%. For Nigeria’s food security, especially in regions with scarce data, PPE doubled the accuracy by incorporating localized proxy data. During the 2026 Ebola outbreak in the Democratic Republic of Congo, PPE’s forecasts identified most newly affected health zones ahead of time, outperforming publicly available models by over 10 percentage points in recall.
By automating the complex pipeline of data discovery, integration, and model optimization, the Planetary Prediction Engine lowers the technical barriers for planetary-scale analytics. This could empower researchers, policymakers, and humanitarian organizations to deploy expert-level predictive models quickly and customize them for diverse global challenges. While the system is still new, its promising results suggest a future where AI can help monitor and respond to critical environmental and public health issues with greater speed and precision.
Based on research published on arXiv by Evelyn Ma, Rama Kumar Pasumarthi, Kishwar Shafin et al..