WSR88D Data Abstraction
- Radar
- The Radar data abstraction sits at the top of the data hierarchy. It contains all of data for one radar cycle and is composed of one or more Volumes.
- Volumes
- A Volume describes one moment in three dimensional space. Level II WSR-88D data contains three moments: Reflectivity, Radial Velocity, and Spectrum Width. A Volume is composed of Sweeps at different elevation angles.
- Sweeps
- Sweeps represent a two dimensional cut of the volume space. Cuts occur at different elevation angles, which typically includes .44, 1.36, 2.42, 3.34, 4.31, 5.93, 9.84, 14.68, and 19.51 degrees. Sweeps are composed of radial sections of data called Rays and typically contain enough rays to cover the entire 360 degree area of the sweep.
- Rays
- The Ray represents radar data collected along the same radial path. That is to say that rays extend from radar much like spokes on a bicycle wheel. Rays are composed of Pulse Volumes. All Pulse Volumes in a given Ray have the same azimuth, but are located a different distance from the radar station.
- Pulse Volumes
- Pulse Volumes contain the actual radar data that is discerned from received echoes. The pulse volume typically has a width of 1km and a depth ranging from 250 - 1000m. The value of the pulse volume is more or less an average from the entire area spanned by that pulse volume.
Moments
- Reflectivity
- Reflectivity is the standard echo received by the radar. It is described by signal intensity on a decibel scale. The raw value is typically referred to as Z (which becomes dbZ on a decibel scale).
- Radial Velocity
- The addition of Doppler technology to weather radars allowed them to use the Doppler shift in the received echo to calculate the velocity of the object reflecting the radar signal. The Doppler radar can only detect the portion of movement along the same line as the radar beam. Therefore an object moving perpendicular to the radar beam would appear to have a 0 velocity. Velocity is typically measured in m/s.
- Spectrum Width
- Spectrum Width is a measure of the variance of a signal within a given pulse volume.
- Variance
- This is the standard statistical form of variance.
- Kurtosis
- Kurtosis is a measure of peakedness in a distribution. It can indicate whether the variance in a distribution is due to infrequent extreme deviations or if it is due to mild often occurring deviation.
- Skewness
- As the name implies, Skewness is a measure of skew or lopsided-ness in a distribution.
Classification Methodology
Out approach to volume classification begins with training a classifier on individual
pulse volumes spanning a number of training sweeps. A classification for an
unknown volume is chosen by first classifying the individual pulse volumes
and then aggregating these lower level classifications in some meaningful
way. Possible aggregation techniques include using a 'Majority Wins' or
'Threshold' rule.
Classifier
Definitions
- Entropy
- Although there are many ways to define entropy depending on the use (e.g. thermodynamics, statistics, etc.), we are concerned with the definition that uses entropy to describe the randomness in a set of nominal values. Entropy is typically defined as a real number between 0.0 - 1.0. If every value in the set was the same, the set would have 0 entropy. An equally distributed set would have an entropy of 1.0.
- Information Gain
- Information Gain measures possible entropy reduction. It can be used to determine the most important attributes to use for classification.