02 · Innovation Eight 3D-Var radar cycles — did the DA actually move the model state?

Eight cycles, summed up. Once the nest is placed, the pipeline runs 8 WRFDA 3D-Var cycles on the d01 9 km domain — 4 for the morning window (00, 01, 02, 03 Z) and 4 for the afternoon (06, 07, 08, 09 Z). Each cycle assimilates radial velocity (rv) and reflectivity (rf) from the INEA Macaé volume scans, using the WRFDA radar forward operator with the same physics_suite="CONUS" the be.dat background-error covariance was built with.

The two windows are independent: the afternoon chain restarts from the GFS 20/12Z pre-forecast at 06 Z (it does not continue the morning analyses). Each cycle runs max_ext_its=2 outer loops; "OmA" below is the final analysis fit (audit revision 2026-06 — the original page showed the loop-1 fit). Every cycle produces a positive impact on the radial-velocity RMSE — from +16% in the quietest cycle to +61% in the most active one. The reflectivity impact is smaller but consistently positive (+10% to +38.5%).

RMSE OmB to OmA summary across 8 cycles
OmB → OmA RMSE per cycle. Lower bars after DA = analysis fitted the obs.

Open the interactive innovation report

8 cycles · ~200 figures · dropdowns for cycle / variable / level · cost function, OmB→OmA scatter, increment maps

Open report

What each piece of the report shows