Lead: Shuyi Chen (UW Atmospheric Sciences), whose work on the connection between the Madden–Julian Oscillation and atmospheric rivers anchors this thread.
Why this sits in a geohazards project¶
The most damaging events in our use cases do not begin in the ground. An atmospheric river makes landfall, days of rain saturate a hillslope whose strength was already set by its wetting history, and the failure that follows is recorded as a landslide or a flood. Treating the atmospheric forcing as an external boundary condition — something handed to us by a weather product — throws away most of the predictability.
The claim this page tests is narrower and more useful: if the ocean sets up the atmospheric river days to weeks in advance, then the lead time on a geohazard forecast is not limited by the hillslope. It is limited by how far upstream in the coupled system we are willing to look.
The MJO → atmospheric river link (lead thread)¶
The Madden–Julian Oscillation is the dominant mode of intraseasonal tropical variability, and its eastward-propagating convective envelope modulates where and when atmospheric rivers form and make landfall on the west coast of North America. That modulation operates on the two-to-six-week horizon — the gap between weather forecasting and seasonal prediction, and precisely the horizon on which emergency managers can still act.
(outline — to be developed with Shuyi Chen: MJO phase compositing against AR landfall frequency and intensity; which phases favour Pacific Northwest versus California landfall; how far the skill extends; what the coupled ocean state contributes beyond the atmospheric signal alone.)
Atmospheric rivers and extreme precipitation (outline)¶
AR dynamics, detection and tracking; integrated vapour transport as the working variable; landfall geometry against terrain. Links to the weather products catalogued in ModelHub — AR index, ACE2, Clima-X.
Teleconnections and climate modes (outline)¶
SST anomalies, ENSO and other modes; what each contributes to seasonal predictability of extreme precipitation, and where those signals are and are not separable from the MJO.
Handoff to the hazard pillars¶
This page produces one thing the rest of the project consumes: precipitation forcing with an honest uncertainty and a stated lead time. Pillar 1 uses it to drive soil-moisture reanalysis; Pillar 3 uses it to extend the forecast horizon beyond what hillslope state alone supports. The interface, not the meteorology, is what has to be specified first.
(outline — the forcing contract: variables, resolution, ensemble treatment, and how lead time is reported alongside skill.)
Data and models (outline)¶
See DataHub for the observational and reanalysis holdings, and ModelHub for the AI weather models under evaluation.
Evaluation (outline)¶
Skill against persistence and climatology at each lead time, following the metric families in HazEvalHub. A forecast that beats climatology at two weeks is worth more to this project than one that beats it at two days.
Open questions¶
How much of the AR landfall signal is recoverable from the MJO phase alone, and how much requires the coupled ocean state?
Do AI weather models inherit MJO skill, or do they degrade at exactly the intraseasonal range where this thread is useful?
What is the shortest defensible lead time at which a coupled ocean–atmosphere signal changes a hillslope forecast?
References¶
(to be added)