# Grid-scale noise: native vs HRRR, and the initial condition

- noise_zoom_{precip,OLR}_wyoming.png: x2 and x3 at 18:55 UTC over Wyoming,
  native pg2 output (left) vs HRRR-remapped output (right). The squares are
  already in the native output, so the online remap is not the cause. They
  sit in the same places in x2 and x3.
- ic_x3_era5_blocks.png: the x3 HICCUP initial condition over NE Wyoming.
  T, qv, U and qc are piecewise constant on 0.25 degree squares (ERA5 grid),
  i.e. HICCUP's horizontal remap from ERA5 was first order. The model
  inherits these jumps.

Made with analysis/noise_zoom.py and analysis/plot_ic_blocks.py, e.g.

    python analysis/noise_zoom.py precip 55 out.png \
        --run "x2=<x2 run dir>=files_map/map_conus1024x2v1pg2_to_hrrr_traave.20260804.nc" \
        --run "x3=<x3 run dir>=files_map/map_2026-incite-conus-1024x3-pg2_to_hrrr_traave_rank1.20261014.nc"
    python analysis/plot_ic_blocks.py files_init/HICCUP.atm_era5.2011-05-19.18.1024x3.L128v4.20260918.nc out.png

(The x3 pg2 SCRIP file in files_grid is truncated, so noise_zoom.py takes
the pg2 cell centres from the pg2-to-HRRR map file instead.)
