Radar DA case study — Macaé INEA, 30 November 2023 Afternoon weak convection · DA 12–15 Z · forecast 15 Z → 12 Z+1 · draft for WRF User Meeting 2026

CONVECT / Project Sharktopus · run 2026-06-11 · cross-case synthesis

The event. A weak afternoon convective episode on 30 November 2023, active 16–21 Z (area-mean max ~0.7 mm/h). The DA window ran 12–15 Z — 4 hourly 3D-Var cycles assimilating INEA Macaé radial velocity and reflectivity. The free forecast ran from 15 Z for 21 h (to 12 Z+1) on d01 → d02 → d03 via ndown.

Both branches (control and radar-cycled) use the same GFS 12Z pre-forecast as background — fair same-lineage baseline (no cold start, no spin-up mixing with the DA signal).

The nest picker placed d03 to the NE of the radar (centre −21.99°, −41.31°), climo <z> +0.32σ (indicative — DJF climatology applied to 30 Nov). The d03 box landed in a gauge-sparse region (10 stations), so the d03 station scores rest on a handful of events and are indicative only.

21-h accumulation scatter d02 — 21-h accumulated rainfall scatter — d02. MAE 20.5→19.9 mm.

case date

2023-11-30

afternoon event, 16–21 Z; weak (max ~0.7 mm/h area-mean)

climo <z>

+0.32σ

indicative — DJF climo applied to 30 Nov

DA cycles

12, 13, 14, 15 Z — hourly 3D-Var, rv + rf

headline

MODE d02 55.9→10.9 km

3 matched hours; d03 stations POD 0→0.33 at ≥2.5 mm/h (1 hit of 3)

Honest reading. Hourly point scores at d02 improve modestly in the afternoon window (POD ≥1 mm/h 0.301→0.372, ≥2.5 mm/h 0.250→0.304) while the frequency bias, already high, rises further (3.52→4.14) — DA injects moisture that over-forecasts light rain. The 21-h accumulation on d02 is essentially unchanged (MAE 20.5→19.9 mm). The object-based metric is where the gain sits: MODE centroid distance on d02 falls from 55.9 to 10.9 km over 3 matched hours. On d03 the radar run scores its only hits (POD 0→0.33 at ≥2.5 mm/h, 1 hit of 3 events) but the MODE centroid there worsens on a single matched hour. The d03 box fell in a gauge-poor region (10 stations) — lesson: in weak cases the picker should consider observation density when choosing the fine-domain centre.

Navigate the story

01 · placement

Where did the nest go?

Interactive picker — composite anomaly, per-variable convective index frames, placed d02/d03 boxes. Climo <z> +0.32σ.

Open the picker

02 · innovation

Did the DA move the state?

4 × 3D-Var cycles, cost function, OmB→OmA fit, and analysis increments of U, V, T, Q_vapor around the radar.

Open the innovation report

03 · verification

Did the free forecast benefit?

Control vs radar-cycled WRF at d01/d02/d03 — station POD/CSI/BIAS, 21-h accum, MODE, HiRA, and CREF viewer (obs × control × radar).

Open verification