An optically independent administrative reference for validating built-surface products, and the tree-canopy bias it reveals

Percentage impervious surface from optical satellite products underpins stormwater regulation, land-change accounting and equity assessment, yet it is validated almost exclusively against other optical references. Any error mechanism that suppresses the built signal in the reference as well as in the product is therefore invisible by construction, and tree canopy is the obvious candidate. This paper contributes a validation reference that does not share that blind spot. Assessor floor area divided by storey count, summed over the parcels in a raster cell, gives a lower bound on the impervious fraction that cell holds. The bound is exact where the recorded roof falls wholly inside the cell, which is 35.2 percent of parcels, and holds in aggregate otherwise. It uses no optical input, so it cannot inherit optical occlusion, and it turns disagreement into a testable claim about named cells. The frame is 24,088 parcels under half an acre in Nevada County, California, each carrying a recorded dwelling, of which 21,931 predate the product epoch. Across four products, closed canopy is associated with 18 to 22 percentage points less measured National Land Cover Database (NLCD) impervious surface than open ground, and a 31 point higher probability of falling below the bound. The association holds within every settlement-density quartile of the frame, survives conditioning on the bound itself, and holds on the subset where the bound is exact. The reference is validated against an independent building-footprint product: where canopy is absent the median ratio of detected footprint to recorded roof is 1.01, and under closed canopy 0.54, so a different sensor and vendor reproduce the same gradient. Loss of built signal ranges from 98 percent for a hard land-cover class to 41 percent for a built-area estimate, though decision rule is confounded with resolution, sensor and epoch. Parcels built after the product epoch serve as a negative control, returning minus 2.45 against minus 21.90. The method needs only a parcel roll carrying floor area and storey count, which is a weaker requirement than imagery though not a universal one.