When Context Compensates for Sparse Event History: AlphaEarth for Spatio-Temporal Point-Process Forecasting
2026-07-01 • Machine Learning
Machine Learning
AI summaryⓘ
The authors studied models that predict events happening over space and time, especially when there is little past data about events in a new area. They tested if adding external spatial information, called AlphaEarth embeddings, helps these models make better predictions for emergency medical calls in regions not seen before. Their findings show that adding this context improves prediction accuracy, especially when very little recent event history is available. The improvements are strongest for short histories and still noticeable when more history is available. This suggests that using external spatial data can help make better forecasts when local event data is sparse.
spatio-temporal point processeslog-Gaussian Cox processspatial contextAlphaEarth embeddingsevent historyemergency medical services forecastingspatial transferpredictive performance
Authors
Yahya Aalaila, Mouad Elhamdi, Gerrit Großmann, Daniel Jenson, Elizaveta Semenova, Sebastian Vollmer
Abstract
Spatio-temporal point-process models must often generalise across space when local event histories are sparse. We study whether exogenous spatial context can compensate in such regimes. Using a fixed log-Gaussian Cox process backbone, we compare an event-only model with the same model augmented by AlphaEarth embeddings as linear spatial context. We evaluate spatial transfer on emergency medical services (EMS) forecasting across eight held-out regions, fixed forecast anchors, and a sweep over history length $w$, using only AlphaEarth (AE) embeddings available strictly before each anchor. AE improves out-of-region predictive performance across all history regimes, with the largest gains under scarce histories: approximately $2$--$6\times$ multiplicative improvements at $1-2$ weeks, tapering to roughly $10$--$20\%$ at $w=20$--$104$ weeks. These results show that contextual information can substantially stabilise spatially transferred point-process forecasts when event history is limited.