Rapid-Scan Radar Data from Supercells and Tornadoes: How Do We Get Science from the Movie?

Over the last 5 years, a number of new mobile Doppler radars have been used by various severe storm research groups to obtain radial velocity, reflectivity, and even dual-polarization variables from violent tornadoes with spatial resolution ~O(10 m) - O(100 m) and temporal sampling as frequent ~O(10 s).  These data are unprecedented in their spatial and temporal resolution and as such could provide valuable new insights into supercell and tornado dynamics.  Ideally, one would like to be able to initialize a numerical model from these data such that the complete dynamic and thermodynamic evolution of the storm and its vortices could be studied in detail.  The spatial and temporal resolution of the information present unique challenges for conventional data assimilation techniques (e.g., the ensemble Kalman filter).  For example, it is well known that current non-hydrostatic model grids cannot resolve features having spatial scales much less than 6Δ.   It is even less clear what temporal frequencies these models can effectively resolve.  These spatial and temporal scales, inherent within any numerical model, are almost always larger than the available radar data observations.  A further complication is that the mis-match between the observations and the model background arising from both sampling and model error will generate imbalances in the posterior analysis.  These imbalances will generate unphysical high-frequency oscillations in the solution which need to be propagated away or dissipated from the region of interest so that the solution can then evolve in a more balanced way.  This additional (and longer) time scale limits the assimilation’s cycling frequency, waiting for the model solutions to “settle-down” prior to the next analysis update.  

For the highest resolution data sets, current computational limitations prevent employing numerical model grids finer than the radar resolution.  An optimal data assimilation system therefore requires that the radar observations be thinned in both in space and time.  For the rapidly evolving and turbulent flows within mesocyclones and tornadoes, this thinning presents a unique challenge.  Currently, observation thinning for dense data sets are done using mostly heuristic rules developed from experience for coarser resolution observing systems (e.g., the WSR-88D, satellite data) and where the atmospheric flow is less turbulent and evolving on time scales of ~O(5 min).

Using an OSSE study, we will describe a new approach to appropriately thin the radar observations to optimally produce an analysis for a given model grid with and without model error present, and potentially show an example using real rapid-scan radar data.