Connection to the thesis
A candidate way to initialise or refine local fields; high-resolution output must be validated at the support of independent observations.
Available inputs and prework
The Physics-AI repository and prior D-LEGOS-PISR scaffold are identified. A completed Caribbean training run, retained holdout outputs and an acquired pressure-level data manifest have not been verified here.
D-LEGOS-PISR protocol; earlier ERA5 acquisition work; observation-scale and downscaling manuscripts.
What the next checkpoint needs
Inspect the current acquisition workflow and retained inputs, freeze spatial/event splits, and compare interpolation and constrained allocation before a neural super-resolution experiment.
Provenance and limits
Prior research context and identified Physics-AI repository; source contents require a separate experiment audit.
This is an inventory note, not a new empirical paper. Third-party research is kept in the reference library and is not attributed to Adrian Dunkley.