🔍 Objectives
- Define “catastrophe windows” when forcings align and stability margins shrink.
- Fuse multi-source signals into a probabilistic nowcast/forecast.
- Publish machine- and human-readable alerts for PSL ops & research.
🧠 Key Concepts
- Alignment Score: orbital bands × cryosphere state × node resonance.
- Load–Release Imbalance: accumulated stress vs. dissipation capacity.
- Hazard Lensing: node multipliers intensify local risk from global drivers.
- Horizon Bands: near-term (months), mid (years), deep (decades+).
📦 Data Inputs
- V9.2–V9.6 derived layers (pacing, torque ties, node ranks, coupling, sync scores).
- Real-time/semi-real-time feeds: GNSS velocity/strain, seismicity, GRACE/grav, altimetry/SLR.
- Seasonal to decadal climate indices (ENSO, PDO, SAM/NAO), cryosphere extent & melt proxies.
🧪 Methods
- Feature Stack: build node–time features (alignment, load, sync, anomalies).
- Windowing Engine: sliding time windows with Bayesian updating.
- Thresholding: stability metrics (gain/lag) → alert tiers (Advisory/Watch/Warning).
- Backtests: hindcast against known pulses (e.g., MWP events, major rifts, uplift bursts).
- Calibration: reliability diagrams & Brier/CRPS; tune by region/node class.
🗺️ Figures
- Global “window map” highlighting active nodes by alert tier.
- Timeline strips per node showing alignment vs. observed release.
- Reliability & ROC curves for hindcast performance.
✅ Outputs
- Windows Feed: CSV/JSON of node_id, start/end, tier, rationale, confidence.
- GeoLayers: GeoJSON/TopoJSON with current-tier styling for maps.
- Reports: human-readable bulletins with figure thumbnails and citations.
📚 References (to stage)
- GNSS & seismic feeds, GRACE/altimetry repositories, climate indices, cryosphere datasets.