Residual-Guided Multi-Resolution Refinement of Foundation Models: A Case Study in Drought Forecasting

2026-07-19

AI Summary

This work addresses the limitations of existing time series foundation models in regional climate forecasting, which typically rely on single-pass inference and thus struggle to capture multi-scale temporal dynamics or incorporate iterative refinement. To overcome this, we propose the Residual-Guided Multi-Resolution (RGMR) framework—the first approach to integrate climatological principles of multi-scale analysis and error diagnosis directly into the inference phase of foundation models. RGMR employs a coarse-to-fine, residual-guided strategy across multiple resolutions, enabling plug-and-play performance gains without modifying the backbone model’s parameters. The method is compatible with prominent models such as TimesFM, TimeGPT, and TabPFN. Evaluated on one-month-ahead SPEI prediction at three sites in South Australia, RGMR reduces the average MSE by 18.7% when applied to TimesFM and demonstrates consistent improvements across three additional external regions.

Abstract

Regional climate prediction presents unique challenges for time series foundation models, which typically process temporal patterns through single-pass inference. Expert climatologists, in contrast, employ multi-scale temporal analysis and iterative refinement based on systematic error diagnosis. We present RGMR (Residual-Guided Multi-Resolution Refinement), an inference-time framework that adapts pre-trained foundation models to perform structured coarse-to-fine refinement for climate forecasting without updating backbone parameters. Applied to drought forecasting using the Standardized Precipitation Evapotranspiration Index (SPEI), RGMR is architecture-agnostic across the three TSFM backbones evaluated per site (TimesFM, TimeGPT, TabPFN) and consistently lowers test-set MSE on three South Australian sites and three additional regions outside South Australia. Applied to TimesFM, the wrapper reduces one-month-ahead SPEI MSE by up to 18.9\% across the three South Australian sites (mean reduction $\approx$18.7\%). Overall, RGMR provides a practical route for deploying frozen TSFMs in regional climate forecasting workflows.

Problem

• drought forecasting
• regional climate prediction
• time series foundation models
• multi-resolution analysis
• SPEI

Innovation

• Residual-Guided Refinement
• Multi-Resolution Forecasting
• Foundation Models
• Drought Prediction
• Inference-Time Adaptation