Incidence-angle dependence of commission error in SWOT KaRIn water masks

The Ka-band Radar Interferometer (KaRIn) on the Surface Water and Ocean Topography (SWOT) mission gives the first wide-swath radar view of inland water, but over tropical deltas, it reports water over dry ground. Prior work blames both the land surface and nearness to the ground track. We separate the two. We measured the false-positive rate: dry cells that SWOT labels water. Across 133 region-dates on 13 Mekong Delta tiles (2024-2025), we binned 14.6 million dry cells by SWOT incidence angle (the beam's angle from vertical), controlling for season, land cover and preprocessing. Dry cells came from a Modified Normalized Difference Water Index (MNDWI) on Harmonised Landsat and Sentinel-2 imagery. To rule out location, we fit the incidence slope within each of 151 regions and split one scene at the ground track. We repeated it in the Red River Delta and Australia's Lake Eyre basin (no rice cycle). The Mekong false-positive rate falls from 0.803 below 1 degree to 0.137 beyond 3.5 degrees, a factor of 5.9 (95 percent bootstrap interval 5.1 to 6.9), while recall falls by a factor of 1.48. Season adds almost nothing (McFadden pseudo-R-squared gain 0.0026). The decline holds in 132 of 151 regions and on both sides of the ground track, which no landscape gradient can produce and survives recommended filtering. Red River (factor 6.0) and Lake Eyre (factor 21.9) match; a desert basin rules out a static prior water map. Commission error over land rises steeply as the beam nears straight down. A study sampling only one region, season or crop stage can mistake this geometry effect for a surface effect because the two cannot be told apart there. Discarding the near-nadir swath (an incidence-angle screen) raises the Mekong F1 score from 0.495 to 0.536, at a recall cost matching the swath discarded.