Large artifacts in dealiasing

[SteepAtticStairs]

Hello there,

I've recently discovered this library, and I've been trying to use it with Py-ART. I've thrown together this code that I assume would be a correct implementation of UNRAVEL:

import matplotlib.pyplot as plt
import numpy as np
import pyart
from pyart.correct import dealias_region_based
from unravel import *


# read in file
radar = pyart.io.read_nexrad_archive('KTLX20130520_201643_V06.gz')

radar = radar.extract_sweeps([1]) # extract velocity sweep

corr_vel = unravel_3D_pyart(radar, velname='velocity', dbzname='reflectivity')
theCorrVel = dealias_region_based(radar)
theCorrVel['data'] = corr_vel # comment this line of code to plot Py-ART's dealiased data instead

radar.add_field('dealiased_velocity', theCorrVel, True)

rmin = radar.fields['dealiased_velocity']['valid_min']
rmax = radar.fields['dealiased_velocity']['valid_max']

fig = plt.figure(figsize=[10, 7])
display = pyart.graph.RadarDisplay(radar)
display.plot('dealiased_velocity', vmin=rmin, vmax=rmax, cmap='pyart_balance')
plt.show()

Here is a link to the NEXRAD radar file I'm using (KTLX20130520_201643_V06.gz - the 2013 Moore tornado). It's a direct download link to the file from NOAA's AWS explorer: https://noaa-nexrad-level2.s3.amazonaws.com/index.html

When I use Py-ART's region based algoritm (dealias_region_based), I get a correctly dealiased field:

However, when I use UNRAVEL's unravel_3D_pyart function, I get large artifacts:

Am I using or implementing UNRAVEL incorrectly? If so, could you correct my code? Or is this an issue with the algorithm itself, e.g. it doesn't do well for areas of rotation?

I would appreciate any information you have about this.

[vlouf]

Your implementation of Unravel is correct, it just seems to fail on that case (which happens some times). Try using with strategy='long_range', to see if it works better for that specific case:

unravel_3D_pyart(radar, velname='velocity', dbzname='reflectivity', strategy='long_range')

Probably on your case study the region_based technique from PyART is going to work better though. Hope it helps!