Multimodal Feasibility
The core project hypothesis is that VLF radio data may add useful predictive signal when combined with seismic and astronomical context.
Early Questions
- Which VLF datasets are available for Italy-relevant stations or reception paths?
- Do VLF records have timestamps precise enough for seismic window alignment?
- What licensing permits reuse and derived features?
- Which astronomical and geomagnetic indexes are available at useful cadence?
- Can all sources be aligned to UTC windows without leaking future information?
First Multimodal Window
Use the seismic smoke window as the first alignment target:
- region: Italy, with Central Italy subset
- window: 7 days
- seismic target: event with magnitude
>= 3.0 - VLF features: Cumiana live image availability first, then image-derived signal summaries
- astronomical features: USNO lunar phase, NOAA solar and geomagnetic indexes
Evaluation Rule
Build seismic-only baselines first, then compare:
- seismic-only
- seismic plus VLF
- seismic plus astronomical
- seismic plus VLF plus astronomical
Only treat multimodal features as useful if they improve held-out performance and calibration after ablation testing.