High-Resolution Stratospheric Tracer Fields Reconstructed with Lagrangian Techniques: A Comparative Analysis of Predictive Skill

Abstract

Numerical experiments and statistical analyses are conducted to determine the skill of different Lagrangian techniques for the construction of tracer distributions. High-resolution potential vorticity (PV) maps are calculated from simulations of the 1996/97 arctic winter stratospheric dynamics using two different numerical schemes?reverse domain filling trajectories (RDF) and contour advection with surgery (CAS)?and data from three meteorological agencies (NCEP, the Met Office, and ECMWF). The PV values are then converted into ozone (O3) concentrations and statistically compared to in situ O3 data measured by the electro chemical ozone cell (ECOC) instrument during the Airborne Polar Experiment (APE) using cross correlation, rms differences, and the Kolmogorov?Smirnov (KS) test. Results indicate that while Lagrangian techniques are successful in increasing the presence of lower-scale tracer structures with respect to the plain meteorological analyses, they significantly improve the statistical agreement between the simulated and the measured tracer profiles only when there is clear evidence of filaments in the measured data. This better fit is most clearly seen by using the KS test, rather than cross correlation. It is argued that this difference in the performance of Lagrangian techniques can be partly related to the treatment of mixing processes in the framework of the Lagrangian schemes. Statistical analyses also show that the temporal rather than the spatial resolution of the input meteorological fields, used to advect tracers, enhances the predictive skill of the Lagrangian products. The best overall performance is obtained with the Lagrangian product (not gridded) based on high-resolution reverse trajectories calculated along a flight track, in particular when the simulation is initialized with ECMWF data. Other products, such as CAS initialized with ECMWF and 3D-gridded RDF initialized with the Met Office data, show fairly good performances, thus with lower statistical confidence.