Predictive ability of four auroral precipitation models as evaluated using Polar UVI global images

Abstract

[1] Auroral precipitation models have been valuable tools for several decades, but it has been difficult to estimate their objective accuracy. The use of global UV imagers, which make relatively instantaneous estimates of hemispheric auroral power, provides one approach forward. We present the first such validation and quantitative comparison of auroral precipitation models. Specifically, we correlated Polar UVI images with the predictions of four precipitation models. These are the Hardy Kp model, the Brautigam IMF-based model, the Evans nowcast model currently used at NOAA, and OVATION Prime, recently introduced by Newell and colleagues. Because calibration uncertainties exist for all particle detectors, and for imagers as well, we focus on correlation coefficients rather than the absolute magnitudes. To minimize dayglow, the nightside precipitating power (1800–0600 MLT) is considered and only for cases where that entire region is within the Polar UVI field of view. Also, only instances where each model has a prediction are considered (i.e., there must be IMF data, and there must be a NOAA satellite pass within the last 1 h). Altogether, 27,613 1 min (“instantaneous”) images satisfied these criteria from 1996 to 1997. The four models investigated predict roughly half the variance in auroral power. From least to best at predicting instantaneous auroral power, the results are Brautigam IMF model (r = 0.68, r2 = 46%); Evans nowcast model (r = 0.70, r2 = 49%); Hardy Kp model (r = 0.72, r2 = 52%); and OVATION Prime IMF (r = 0.75, r2 = 56%). We also considered 1 h averages of UVI images. All four models improved, but the nowcast jumped from third to first: Brautigam IMF (r = 0.69, r2 = 48%); Hardy Kp (r = 0.74, r2 = 55%); OVATION Prime IMF (r = 0.76, r2 = 58%); Evans nowcast (r = 0.77, r2 = 59%). The nowcast approach benefits most from hourly averaging because at times more than one satellite pass is available. In principle, with enough satellites, the nowcast approach would be ideal. The advantage of OVATION Prime IMF over Brautigam IMF mirrors their respective organizing solar wind coupling functions. The reasonable performance of the Hardy Kp model, despite its 3 h cadence and limited maps, suggests that an updated model based on Kp interpolated to 1 h cadence would be competitive.