Radar data assimilation for WRF Macaé INEA case studies · 3 cases · draft for WRF User Meeting 2026

CONVECT / Project Sharktopus · last updated 2026-06-12 · mirrored from GitHub (migration in progress)

Three end-to-end WRF radar-DA experiments over the INEA Macaé S-band radar. Each case uses objective nest placement (composite convective-index anomaly, swept against a 3-month WRF climatology), hourly 3D-Var cycling of radial velocity and reflectivity, and a free forecast out to 21 h on a 3-domain 9/3/1 km grid via ndown. Every case uses a fair same-lineage control — no cold-start mixing with the DA signal.

The experiment spans two calendar years (2023–2024), covers three convective regimes (organised nocturnal flooding, scattered afternoon, and strong nocturnal line), and includes a full multi-agent audit of the flagship case (C2).

Nest decision overview — Macaé 2024-01-21
C2 morning nest decision — d01 9 km composite anomaly with placed d02/d03/d04 boxes. Macaé radar marked in black.
cases
3

2023-11-30 · 2024-01-21 (2 windows) · 2024-12-16

radar
INEA Macaé

S-band, −22.406°, −41.860°; rv + rf assimilated

DA method
3D-Var

WRFDA hourly cycling, 4 cycles per case, same-lineage control

highlight
C3 d03

control missed the event entirely; DA recovered it (POD 0→0.385)

Cases and synthesis

C2 morning climo composite
C2 · 2024-01-21

The audited flagship case

Two windows (morning 00–03 Z / afternoon 06–09 Z). Organised nocturnal flooding event. Full placement + innovation + verification site, with 104-finding multi-agent audit. Morning window shows clear, robust DA benefit.

Open case 2
C1 nest decision
C1 · 2023-11-30

Afternoon weak convection

DA cycles 12–15 Z. Weak event; d03 box in a gauge-sparse region. Accum MAE d02 −29%; MODE d03 centroid 51.6→10.6 km (1 hr matched). A design lesson in observation density.

Open case 1
C3 d03 POD — DA recovers missed event
C3 · 2024-12-16

The model was blind; DA recovered the event

DA cycles 14–17 Z. Nocturnal convective line, Arraial do Cabo, 64 dBZ. Control: 0 hits at every threshold. Radar DA: POD ≥1 mm/h 0.00→0.385 on d03. The clearest operational use-case in the set.

Open case 3
Cross-case synthesis
synthesis

Three cases, one fair baseline

Cross-case tables, key numbers control→radar for all 4 experiment windows, honest reading (5 bullets), and caveats on comparability.

Open synthesis
Macaé radar beam-blockage map
beam blockage

Where the radar is blind

Does it rain in directions the radar never sees? Found two ways — the radar's own dead sectors and a terrain beam-blockage simulation. ~19% of d03 is blind (W/NW behind the Serra do Mar). Map + WRFDA mask file.

Open beam-blockage study
Precipitable water — C2 morning radar DA
synoptic explorer

Browse the synoptic situation

Pre-rendered d01 (9 km) frames for 8 synoptic fields (MSLP, T2, q850, Z500, jet, PW, CAPE/shear, cref) across 2 runs — radar DA morning and the fair control. Use the run/field/hour selectors.

Open explorer