Model predicts future Earth states from satellite data
SPEAR NeXT Causal Latent Forecasting Across Multiple Horizons for Spectral Temporal Earth Representation Learning
Computer Vision and Pattern Recognition
Summary
Earth observation data changes all the time, and understanding these changes can be tricky. The authors present SPEAR NeXT, a model that learns by predicting the Earth's future state based on past satellite and environmental data. It uses a clever encoding system to compactly represent current Earth conditions and a transformer to forecast how these conditions will evolve. This approach includes information about time order, seasons, and yearly patterns to improve its predictions.
What this means in practice
- •For environmental monitoring teams: Generate accurate multi-horizon forecasts of Earth conditions from past satellite and environmental data using compact model embeddings.
- •For climate data platform developers: Embed multimodal Earth observation data to enable efficient storage and future state prediction for climate monitoring services.
Authors
Rajiv Ranjan, Udaiveer Singh, Shashank Tamaskar, Dharmendra Saraswat
Abstract
Earth observation is inherently dynamic, yet temporal information in many foundation models is learned through reconstruction, invariance, or retrospective sequence summarization. SPEAR NeXT is introduced as a compact pixel-wise multimodal spectral temporal foundation model in which temporal self supervision is formulated as past only, multi horizon latent Earth state prediction. Instantaneous states are first encoded by the pretrained SPEAR model from optical, radar, and environmental observations into compact 32 dimensional embeddings. Their temporal evolution is then modeled by a causally masked Trans former that predicts multiple future latent states from pre ceding observations. Relative temporal order is represented using Rotary Position Embeddings, while month and year embeddings encode seasonal phase and interannual con text.