FloodScope

What this is

A platform that converts water levels into statements about specific public assets: flooded road length, affected critical facilities, isolated assets, park and open-space flooding, and a preparedness priority score — rolled up to the scale an official actually manages. A tract for outreach, a municipality for operations, a county for coordination.

It is the third of a trilogy. FloodOps asks what floods at stage X? for one town. FloodSentinel asks is stage X coming, and who should know? FloodScope is the platform a county emergency-management office would actually commission.

The load-bearing decision

R owns batch geoprocessing; Python owns everything at runtime. Intersecting every scenario polygon with every road, asset, park, and tract happens once, locally, in R’s sf, producing exposure tables. The six-hourly job is then pure Python: fetch forecasts, snap each to the nearest scenario at or below it, roll up precomputed numbers, score, publish. No R in scheduled jobs at all.

That split is what makes the refresh fast, cheap, and dependency-light — and it is why runtime needs no geometry. Snapping down to the nearest library scenario is also conservative by construction for flood extent, which is the right direction to be wrong in.

The two halves communicate only through file formats — GeoPackage, COG, Parquet, DuckDB. Neither side reads the other’s internals. Every metric surface will carry its source, vintage, and uncertainty range, and every page will carry the demonstration disclaimer: this is not an operational warning system.