Texture Dependence of Advection Skill Relative to Persistence for Hourly Satellite-Derived Solar Irradiance over China

The benefit of motion-based solar irradiance forecasts can vary with spatial texture even when their pooled skill is favorable. We compare a fixed history-only Farnebäck extrapolation with solar-geometry-adjusted persistence on the hourly Tang/TPDC product. A pilot-fixed texture split was applied to 21 geographically selected cores in 2013–2014. At 2 h, the equal-year high-minus-low contrast in advection-minus-persistence mean absolute error was −7.99 W m−2 (exploratory 95% interval: −9.96 to −6.02). Relative error reduction was 5.43% in high texture and 1.45% in low texture; their difference was 3.98 percentage points (2.97–5.09). Thus, the contrast was not confined to absolute error units. A screened morning disadvantage recurred on 322 previously unscored 2012 dates at one selected core, and retrospective diagnostics found recurring spring losses. Fixed blending reduced the local loss without demonstrating superiority over persistence. SETORS comparisons remained exploratory because satellite pixel timing remains unresolved, station records are ten-minute means, QC coverage is incomplete, and the 2013–2014 pooled method-difference interval includes zero. These findings support reporting the evaluated operator’s skill by texture, location and origin time, with ground validation before operational switching. They do not establish a causal cloud-motion mechanism or general performance limits of advection.