Current paradigms for understanding the projected changes in extreme rainfall intensities indicate that extremes are affected to changes in the atmospheric thermodynamic state, convective mass flux, and vertical displacements of the mass flux profile. These findings are based upon diagnostic scaling relationships that relate the extreme precipitation rates to a few key predictive dynamical and thermodynamic fields, or predictor variables.

This work instead investigates the predictive power of these relationships: their ability to approximate extreme rain rates based on local values of the predictor variables. We analyze simulations of the superparameterized CESM on spatial scales of hundreds of kilometers and larger, from hourly to subseasonal time scales. First, we quantify the accuracy of the scalings across these ranges of scales. Then, we derive local estimates P_sc from these scalings and compare them to the modeled precipitation values P , in two ways: (1) in the largest percentile bins of the P distribution, the bias and the variance in Psc are decomposed into a sum of variance and covariance terms in the predictor variables; and (2) we quantify the variance explained and the colocation between extremes in P and P_sc, in order to explore the relationship between the tail distribution of P and the full probability distributions of predictor variables. The analysis is first implemented on a simple scaling where predictors are mid-tropospheric vertical velocity and saturation specific humidity at the surface, and then extended to a scaling expression that includes their full vertical structures.

These results provide insights on the amount of information missing to properly constrain the dynamics of extreme rainfall. We conclude by discussing the importance of other potentially contributing elements.