Mean error as function of redshift, for 60 models (per column). The three rows show PACS 100 μm, SPIRE 250 μm and SPIRE 500 μm. The left column uses uncorrected fluxes, while the right column is K-corrected. The models were trained on redshifts below 0.1 (DustPedia + H-ATLAS). An ensemble of the 60 models is shown with the dark plum lines, with the full line being the mean, and the dashed lines being the 16th and 84th percentiles.