Search papers, labs, and topics across Lattice.
This study investigates the role of exogenous spatial context in enhancing spatio-temporal point-process models, particularly when local event histories are sparse. By employing a fixed log-Gaussian Cox process backbone, the authors compare the performance of an event-only model against one augmented with AlphaEarth embeddings as linear spatial context. The findings reveal that incorporating AlphaEarth significantly boosts out-of-region predictive performance, especially under conditions of limited historical data, yielding improvements of 2-6 times at 1-2 weeks and 10-20% at longer histories.
Contextual embeddings can dramatically enhance predictive accuracy in spatio-temporal forecasting, especially when historical data is sparse.
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.
