Skip to content

Improve 4D-Var demo: sparse observations, longer window, accurate claims - #57

Open
havogt wants to merge 2 commits into
swm_2026_halofrom
claude/review-4dvar-demo-qm0S7
Open

havogt wants to merge 2 commits into
swm_2026_halofrom
claude/review-4dvar-demo-qm0S7

Conversation

@havogt

@havogt havogt commented Mar 4, 2026

Copy link
Copy Markdown
Owner

No description provided.

claude added 2 commits March 4, 2026 11:23
- Make observations sparse (50% grid coverage) instead of 100% dense,
  demonstrating 4D-Var's ability to propagate information via dynamics
- Extend assimilation window from 20 steps (30min) to 100 steps (2.5h)
- Add observation mask visualization
- Increase optimizer maxiter from 200 to 500 for better convergence
- Update cost function with proper sparse observation operator H
- Tone down ECMWF comparison in summary; add honest discussion of
  simplifications (diagonal B, small grid, single outer loop)
- Clear stale outputs

https://claude.ai/code/session_015hk7HYttY5wQ7zvkqix1Fy
The pressure field has a ~50000 mean, making spatial variations (~1-10)
invisible when plotting absolute values on a shared colorscale. Now plots
forecast errors (forecast - truth) which clearly show where and how much
the analysis forecast improves over the background forecast.

Also extends forecast to 2x the assimilation window to test whether the
improved IC retains its advantage beyond the observation period.

https://claude.ai/code/session_015hk7HYttY5wQ7zvkqix1Fy
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants