A Description and Evaluation of an Automated Approach for Feature-Based Tracking of Rossby Wave Packets

Abstract

This paper describes an automated approach to track Rossby wave packets (RWPs), and the sensitivity of various tracking parameters and methods used in filtering the raw data in the feature-based tracking. The NCEP-NCAR reanalysis meridional wind and geopotential height data at 300 hPa every 6 h were spectrally filtered using a Hilbert transform technique under the assumption that RWPs propagate along a waveguide defined by the 14-day running average of the 300-hPa wind. After some spatial and temporal smoothing, the local maxima in RWP amplitude (WPA) were tracked using two objective techniques: a point-based cost optimization routine and a hybrid approach using point identification and object-based tracking following rules. A variation of the total energy flux term of the eddy kinetic energy equation was used to subjectively verify RWP tracks in order to compare the performance of each tracking method. When tracking methods are verified over two winter seasons, the hybrid technique outperformed point-based tracking, particularly for track duration and propagation. Problems with tracking were found to be most common during periods when two RWPs merge, one RWP splits into multiple packets, or an RWP moves from one storm track to another. RWPs are found to move irregularly rather than linearly, with their motion and intensity best described as pulse like. The sensitivity to some of the parameters used in the tracking was also explored.

The Climatology and Characteristics of Rossby Wave Packets Using a Feature-Based Tracking Technique

“…A T21 Cholesky filter (Hodges 1999) was applied to the WPA data in order to filter out wavelengths too small to be associated with RWPs. A temporal smoothing (running mean) using a 24-h duration was also applied to the data in order to help track the progression of local maxima in RWPs (Souders et al 2014). Souders et al (2014) describe the details of the RWP tracking approach so only a brief overview is provided below.…”

“…A temporal smoothing (running mean) using a 24-h duration was also applied to the data in order to help track the progression of local maxima in RWPs (Souders et al 2014). Souders et al (2014) describe the details of the RWP tracking approach so only a brief overview is provided below. The method uses a hybrid approach to track RWPs, by combining a point-based object identification similar to the cyclone-tracking scheme of Charles and Colle (2009), and an feature-based tracking approach similar to methods used in the tracking of tropical convective clouds (Arnaud et al 1992).…”

“…A minimum WPA threshold (14 m s 21 ), a minimum object size (40 grid points from 2.58-resolution data), and a search range for track continuity (208N/S, 308W, and 908E) were chosen. Sensitivity of the tracking results to changes in these parameters is discussed in Souders et al (2014). The algorithm identifies unique RWPs by selecting significant local maxima in WPA (more detail on the meaning of significance follows), attributing the object space inside the minimum tracking threshold to each local maximum, and then tracking the objects by searching for significant overlap across a single time interval.…”

“…A minimum propagation distance (408 longitude) and duration (2 days) were regarded as significant RWPs (Souders et al 2014), and thus tracks not meeting these criteria were discarded. The end result was a database of RWP tracks and object parameters, such as RWP size (in square kilometers), zonal group velocity, maximum and areaaveraged RWP amplitude (meters per second), duration (6-hourly time steps), and total eastward propagation (degrees longitude).…”

“…Using Hovmöller diagrams to track RWPs can be problematic when more than one active waveguide exists along a single longitude, and there are sensitivities with the parameters used to define the objects. As a result, Souders et al (2014) developed a tracking algorithm for RWPs that includes both a pointbased object detection scheme and feature-based tracking rules. They verified RWPs manually over several months and found a probability of detection of ;93% and a false alarm probability of around 20%.…”