From AI Weather Prediction to Infrastructure Resilience: A Real-Time Correction-Downscaling Framework for Tropical Cyclone Impact Forecasting

Abstract

This paper addresses a missing capability in infrastructure resilience: turning fast, global AI weather forecasts into asset-scale, actionable risk intelligence. We introduce the AI-based Correction-Downscaling Framework (ACDF), which combines real-time bias correction, terrain-informed downscaling, and fragility-based power transmission system risk assessment for tropical cyclone impacts. ACDF separates storm-scale bias correction from terrain-aware refinement, mitigating error propagation while restoring the sub-kilometer wind variability that governs structural loading. Tested on 11 typhoons affecting Zhejiang, China under leave-one-storm-out evaluation, ACDF produces 500 m wind fields over a province-scale domain, reduces station-scale wind-speed MAE by 38.8% relative to Pangu-Weather, and runs in approximately 25 s per 12-h cycle on a single GPU. In the Typhoon Hagupit case, ACDF reproduced observed high-wind tails, identified a coastal high-risk corridor, and flagged the transmission line that subsequently failed, demonstrating actionable guidance at tower and line scales. ACDF provides an end-to-end pathway from global AI weather forecasts to operational, impact-based early warning for critical infrastructure.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.