AIFS-TCSAT: Causal Forecast Corrections from Intermittent Satellite Observations

Satellite products provide valuable information on tropical-cyclone (TC) inner-core structure, but their arrival times vary across observing systems. Retrospectively incorporating these observations can improve forecasts while inadvertently introducing information unavailable at the original issue time. To preserve this causal availability constraint, we introduce AIFS-TCSAT, an event-driven extension of the new AIFS-TC intensity correction that issues an initial forecast from the information available at the time and then produces immutable revisions as GEO, PMW, SAR, and operational-fix observations arrive. In a storm-disjoint retrospective evaluation of 30 North Atlantic and eastern North Pacific cyclones from 2023, the full multimodal correction reduces storm-balanced wind MAE by 6.3\% relative to the AIFS-TC baseline and 5.2\% relative to fix-only correction. These results show promise for correction mechanisms that incorporate newly available satellite observations to refine TC intensity forecasts in future forecast settings.