← Map / Eval dashboard Crop-sequence Sankey ↗ Full report (PDF) ↗
Field-level analysis of 9.16 M Romanian fields × 9 years (model MLP v2). Mapped WOSR area runs ~1.2× the official survey (0.86–1.33 by year, 2017–2024 vs INSSE AGR108A) — year-to-year pattern validated, r = 0.95. The earlier “~2×, commission bias” reading was mostly a pixel-area unit error, corrected 2026-09-03; correlation is scale-invariant and is unchanged. The “failed establishment” panel is an annual-maps proxy reporting upper bounds. Cross-border bbox fields excluded.

Where & when Romania sows winter rape

WOSR share of arable by county (NUTS3). Drag the year — a field mapped as rape in year Y was drilled the previous autumn.

2022 drilled aut. 2021
0%
14%+ of arable
–
mapped WOSR area (kha)
–
share of arable
–
Eurostat area (kha)
–
top county this year

National WOSR area vs Eurostat

Mapped (bars) vs official Eurostat (line). Same 2019 collapse, 2023/25 peaks — r = 0.964.

What comes before & after WOSR

% of WOSR-field area. Cereal break-crop slot: 91% cereal before, 60% back to wheat after.

Failed autumn establishment → spring replacement (proxy)

On rape-prone fields: spring-crop share (bars) vs WOSR (line). 2019 spikes above median = autumn-2018 drought re-sow. Upper bounds.