A Multi-billion Parcel Atmospheric Trajectory Model

We present a new parallel implementation of an atmospheric trajectory modelling framework which provides improved numerical accuracy, greater flexibility for specifying experiments, and sufficient raw performance to simultaneously simulate billions of parcel trajectories on suitable computing platforms. The application is parallelized using the Message Passing Interface (MPI) library and can scale efficiently on a wide variety of modern computing platforms. The ability to treat such large numbers of parcels is expected to enable a new generation of experiments to explore questions related to global stratosphere-troposphere exchange, age-of-air spectra, and transport of trace gases and aerosols. The modelling framework is written in C++ for easy integration with other computing technologies. It also provides a great deal of flexibility by allowing users to select from (or add to) alternative subclasses for vertical coordinates (pressure, potential temperature), integration schemes (Runge-Kutta, Euler), meteorological data sources (NCEP/NCAR Reanalsyis, MERRA), data interpolation methods (linear, log-linear, splines), and output (parcel histories, summary statistics, min/max quantities encountered). Significantly improved numerical accuracy, especially near the poles, is provided by expressing integration in terms of purely geometric constructs which avoid various complications associated with spherical coordinates near the poles. The entire package has been rigorously developed using Test-Driven Development (TDD) which both provides confidence in the implementation and should also assist other developers that wish to extend the framework. Several tests are performed to demonstrate the fourth-order Runge-Kutta integration scheme with our spherical geometric constructs. Tilted solid body rotation provides a baseline synthetic wind field for assessing model performance, and a time-varying case is used to examine the errors introduced by interpolating linearly in time. We perform full model tests with NCAR/NCEP reanalysis wind fields to computer forward and backwards trajectories of thousands of parcels in kinematic and isentropic mode. Accuracy and performance statistics from these tests show that the model efficiently generates highly reproducible trajectories.