Lagrangian Trajectory Analysis of Severe Convective Storms Using Variable Lossy Compression

Lagrangian trajectories are an important part of the severe storms analysis toolkit, but the large volumes of data are prohibitive enough that such analysis comes in two flavors: online trajectories, which are integrated by the model itself during its run time, or offline trajectories that are integrated from snapshots of saved data at a coarser temporal resolution. It is often prohibitive to save numerical simulations at every model time step to perfectly recreate them offline, while the online trajectories mean running the model multiple times, which depending on the size of the simulation, can also be prohibitive.

In this presentation we will explore the effect of lossy compression on Lagrangian parcel analysis. Using CM1 model data saved in LOFS format, a filesystem comprised of HDF5 files, we explore trajectory positions and interpolated values along these paths as a function of compression level as determined by the ZFP accuracy parameter (Lindstrom 2014). The accuracy parameter is a value set at model runtime that represents the largest amount of absolute error tolerated for a given model variable. We save model data at the model time step which allows us to validate our offline approach by first comparing it to online CM1 trajectories, the "gold standard". Each simulation will be saved with several different accuracy parameter choices, allowing direct comparison to online CM1 trajectories. The impact of lossy compression on Lagrangian trajectories in terms of location accuracy and budget accuracy will be explored by examining various compression configurations compared to uncompressed data at coarse (250m), medium (100m), and high (30m) resolution numerical simulations of supercell thunderstorms.