Fig. 2: Evaluation of model calibration.
From: Residual corrective diffusion modeling for km-scale atmospheric downscaling

Calibration is evaluated using spread skill ratios and rank histograms based on the same validation set used in Fig. 1 and Table 1. a, c, e, g show the ensemble standard deviation as a function of the RSME of mean prediction for 10-m eastward wind, radar reflectivity, 10-m northward wind and 2-m temperature, respectively. The standard deviation is adjusted with a factor √(1 + 1/n) (see Eq. 15 in ref. 55) so that a ratio of one represents a perfectly tuned model. b, d, f, h show the corresponding rank histograms for the same channels.