Empirical Modeling and Stochastic Simulation of Sea Level Pressure Variability

The scope of this work is stochastic emulation of sea level pressure (SLP) for use in error estimation and statistical prediction studies. The input SLP dataset whose statistics are to be emulated was taken from the 1979-2013 ERA-Interim dataset at full 6-hourly temporal and 0.758 spatial resolutions over the Northern Hemisphere. Upon subtracting the monthly climatological mean value and mean diurnal cycle, the SLP anomalies (SLPA) were projected onto the subspace of 1000 leading empirical orthogonal functions of the daily-mean SLPA, which account for the vast majority (>99%) of the full 6-hourly fields' variance for each season. The main step of this method is the estimation of a linear autoregressive moving-average empirical model for the daily SLPA principal components (PCs) via regularized multiple linear regression; this model was driven, at the stage of simulation, by state-dependent (multiplicative) noise. Last, a diagnostic statistical scheme has been developed and implemented for accurate interpolation of simulated daily SLPA to 6-hourly temporal resolution. Upon transforming the simulated 6-hourly SLPA PCs into the physical space and adding a seasonal climatological mean and mean diurnal cycle, the resulting SLP variability was compared with the actual variability in the ERA-Interim dataset. It is shown that this empirical model produces independent realizations of SLP variability that are nearly indistinguishable from the observed variability over a wide range of statistical measures; these measures include, among others, spatial patterns of bandpass- and low-pass-filtered variability, as well as diverse characteristics of midlatitude cyclone tracks.