Disentangle country effects from resolution and unit effects in transferred calibration

Determine whether the variation in transferred calibration coverage is attributable to the country, the spatial resolution, or the type of spatial unit, rather than solely to the resolution and unit differences identified in the transfer experiments.

Background

Puffin calibrates its uncertainty intervals using a global factor $\kappa$ fitted from fine-resolution validation labels. The paper tests whether this factor transfers across settings that differ in country, spatial resolution, and unit type. Although self-calibration achieves coverage near the nominal 90% level, transferred calibration produces coverage ranging from 0.32 to 1.00.

The authors conclude that fine-scale residual scales differ across the tested settings, but the experimental design changes country, resolution, and unit type simultaneously. Consequently, it does not identify the independent contribution of country, leaving unresolved whether country-specific effects explain any part of the transfer failure.