Today at Cloud Next, we're introducing the Preview of Population Dynamics Insights (PDI), a Google Earth AI dataset that is now available via Google Maps Platform. Powered by Google Research’s Population Dynamics Foundation Model (PDFM), PDI is a first-of-its-kind geospatial embeddings dataset designed to help organizations decode the complex relationship between human behavior and the physical world. By distilling aggregated trends from Google Search and Google Maps into rich 330-dimensional embeddings for locations across the globe, it enables a new era of spatial machine learning without the need for manual feature engineering.
Zero feature engineering: The power of ML-ready embeddings
Population Dynamics Insights enables a fundamental shift in how we understand our world—allowing data teams to move away from relying on static maps and decade-old census snapshots, and instead deploy highly-dimensional geospatial models. It eliminates the "data wrangling tax" by delivering analysis-ready embeddings directly into BigQuery.
PDI embeddings are updated monthly and distill millions of aggregated signals—including Google Search trends, Google Maps popular times, points of interest and environmental conditions like air quality and weather—into concise vectors. These embeddings are indexed at S2 cell level 12 granularity (roughly 3km² to 6km²), providing a standardized, high-resolution grid that captures hyper-local nuances while remaining computationally efficient for global-scale analysis.
PDI is designed to be an additional, drop-in ML model training dataset, so that your ML models gain deep human and environmental context instantly with zero manual feature engineering. We handle the complex AI and data aggregation so your data teams can instantly deploy a standardized, ML-ready feature layer with minimal architectural changes.
Supercharge spatial models with PDI
By providing a universal geospatial embedding layer, ML teams can overcome blind spots, outdated census data, and geographic gaps. Key applications include:
Similarity modeling for site selection: Find “sibling” regions with environmental and behavior attributes that mirror your most successful locations for optimized site selection.
Cold-start geographic analysis: Project known outcomes to new territories where ground-truth data is missing. By utilizing globally comparable embeddings, ML teams can train models in data-rich markets like the US and retain full predictive power when deploying in new territories like India or Brazil.
Contextual interpolation: Seamlessly fill in blind spots within your existing data by using the surrounding environmental and behavioral context to predict missing values from unsampled areas
Unlock super-resolution: Generate high resolution data from low resolution datasets, allowing you to zoom in from broad regional trends to granular insights for precise geotargeting. For example, health organizations in Australia are using PDI to identify hidden blind spots at the community level to prioritize life saving screenings in remote areas where care is needed most.
Precision forecasting: Ground time-series models in static spatial realities, correlating environmental and human characteristics with historical success to forecast product demand and market shifts.
Proven performance out of the box
PDI doesn’t just provide context about a location; it drives measurable accuracy improvements across diverse downstream ML tasks. Research and technical evaluations have demonstrated its impact across diverse sectors:
Global health: PDI has demonstrated value on multiple public health initiatives including an effort by researchers at Mount Sinai and Boston Children’s Hospital/Harvard to leverage PDI to super-resolve county-level vaccination data into aggregated, highly granular neighborhood-level insights. This allowed researchers to successfully uncover hidden clusters of undervaccination obscured by traditional surveillance that directly overlapped with active measles outbreaks and more broadly creates the potential for forecasting and anticipating public health needs.
Industry-leading global performance: PDI has achieved state-of-the-art accuracy across 29 distinct US target variables, outperforming traditional demographic approaches and industry-standard satellite models. Furthermore it has shown equivalent performance in 17 countries worldwide and improved predictions of critical socioeconomic and public health outcomes, even when applied across international borders.
Learn how Public Storage is using PDI
Public Storage is leveraging Google Earth AI’s Population Dynamics Insights to evolve how we model and respond to changing market dynamics.
As part of Public Storage’s data science strategy, we seized the opportunity to work directly with the Google team. Population Dynamics Insights provides massive aggregated behavioral, location, and environmental data that we integrate into our proprietary self-storage data and models—turning complex, real-world data into actionable analytics that helps us invest with superior confidence and precision.Chief Data and Analytics Officer, Public Storage
Get started today
PDI is now available across 17 countries—and with a roadmap for continuous global expansion, it provides a unified foundation for borderless spatial intelligence.
We’re excited about the possibilities PDI opens up for data scientists, business decision makers and AI/ML engineers across retail, logistics, financial services, healthcare, global health, and public sectors.
Ready to explore how PDI can help reveal the bigger picture for your organization? Sign up for access and visit our website to learn more.
