Spatiotemporal learning is challenging due to the intricate interplay between spatial and temporal dependencies, the high dimensionality of the data, and scalability constraints. These challenges are further amplified in scientific domains, where data is often irregularly distributed (e.g., missing values from sensor failures) and high-volume (e.g., high-fidelity simulations), posing additional computational and modeling difficulties. In this paper, we present SCENT, a novel framework for scalable and continuity-informed spatiotemporal representation learning. SCENT unifies interpolation, reconstruction, and forecasting within a single architecture. Built on a transformer-based encoder-processor-decoder backbone, SCENT introduces learnable queries to enhance generalization and a query-wise cross-attention mechanism to effectively capture multi-scale dependencies. To ensure scalability in both data size and model complexity, we incorporate a sparse attention mechanism, enabling flexible output representations and efficient evaluation at arbitrary resolutions. We validate SCENT through extensive simulations and real-world experiments, demonstrating state-of-the-art performance across multiple challenging tasks while achieving superior scalability.

Scientific data, like information from sensor networks or traffic patterns, is often complex and messy. It can have missing values due to sensor failures, be spread out irregularly, and come in huge volumes, making it difficult to analyze and predict future trends. To address this, we developed a new AI framework called SCENT. SCENT is designed to learn from this challenging spatiotemporal data (data that changes across space and time). It cleverly fills in missing information, reconstructs incomplete data, and forecasts future events, all within a single, unified system. We built SCENT to be scalable, meaning it can efficiently handle large amounts of data and complex analyses. SCENT can help scientists and engineers gain clearer insights and make more accurate predictions from their complex data. This has the potential to improve our understanding in important areas like traffic forecasting or analyzing data from moving sensors, ultimately aiding in scientific discovery and addressing real-world problems.