← Flood monitoring

13  Was the event caught?

Cell-level scores say how well a product maps a flood. Many uses only need to know whether it noticed the flood at all, and how early. These rules *were fixed on 14 October, before any re**sult existed (ADR 0007)*:

measure	rule	why
caught	on some day from 3 days before the event’s first map to 3 days after its last, the product flags at least 10% of the cells flooded on any map day of the window, and at least 5 cells	a product need not match the outline to have seen the flood; 10% with a floor of 5 cells is the rule the first run used
days to detection	the first day the “caught” rule holds, counted from the first gold map	early warning value
false event	the same rule applied to the same cells on ordinary days: days 30 to 10 before the first map, which leaves a buffer in case flooding began before mapping	without it, a product that always says flood would catch everything
false events away from floods	the same rule on the global random sample blocks (no gold, ordinary days). Deferred: a sample block has no gold footprint for the rule to count against, so it joins the ordinary-day precision work (Next steps, 6)	catches products that fire everywhere, not just near floods

Reported per product and for fusion (P ≥ 0.5), per continent and per land type: the share of events caught, the false-event rate, and the median days to detection.