Resolution-Aware Retrieval Augmented Zero-Shot Forecasting

2025-10-18

AI Summary

Zero-shot forecasting faces the challenge of predicting microclimatic conditions at unseen locations lacking historical observations. This paper proposes a resolution-aware retrieval-augmented forecasting framework that decomposes spatiotemporal signals in the frequency domain to extract multi-scale features, and dynamically fuses global and local spatial contexts for adaptive microclimate prediction over unobserved regions. Its core contributions are: (1) coupling spatial dependency modeling with temporal frequency analysis; (2) designing a resolution-aware retrieval strategy that adaptively selects contextual granularity according to spectral bands; and (3) constructing an end-to-end retrieval-augmented forecasting architecture. Evaluated on the ERA5 dataset, the method reduces mean squared error by 71% and 34% compared to HRRR and Chronos, respectively—outperforming state-of-the-art time-series and physics-informed models.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Image and Video Retrieval

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Zero-shot forecasting aims to predict outcomes for previously unseen conditions without direct historical data, posing a significant challenge for traditional forecasting methods. We introduce a Resolution-Aware Retrieval-Augmented Forecasting model that enhances predictive accuracy by leveraging spatial correlations and temporal frequency characteristics. By decomposing signals into different frequency components, our model employs resolution-aware retrieval, where lower-frequency components rely on broader spatial context, while higher-frequency components focus on local influences. This allows the model to dynamically retrieve relevant data and adapt to new locations with minimal historical context. Applied to microclimate forecasting, our model significantly outperforms traditional forecasting methods, numerical weather prediction models, and modern foundation time series models, achieving 71% lower MSE than HRRR and 34% lower MSE than Chronos on the ERA5 dataset. Our results highlight the effectiveness of retrieval-augmented and resolution-aware strategies, offering a scalable and data-efficient solution for zero-shot forecasting in microclimate modeling and beyond.

Problem

• Enhancing zero-shot forecasting accuracy for unseen conditions without historical data
• Leveraging spatial correlations and temporal frequency to improve predictions
• Adapting to new locations dynamically with minimal historical context

Innovation

• Resolution-aware retrieval for dynamic data adaptation
• Frequency decomposition to separate spatial and local influences
• Spatial-temporal correlation leveraging for zero-shot forecasting