Clutter#
The pylawr package comes with useful filters for adjusting radar data
according to effects of non-meteorological echoes (Clutter).
Clutter is characterised by high reflectivity values and is
caused by static objects, e.g. trees and houses, or moving objects, e.g. insects
and plains. It varies from time step to time step and from range gate to range
gate. For further general information, see [Lengfeld et al., 2014].
Concept#
ClutterFilter calculates a map based on the filter’s
mathematic attributes. For example, the TDBZFilter
computes the average of the squared logarithmic reflectivity difference between
adjacent range gates. Based on the computed map and filter specific thresholds,
a ClutterMap is created, which classifies if the radar
signal is clutter or not in the range gate. Back to the example of our
TDBZFilter, if the mean of squared reflectivity
difference within five consecutive range gates is higher than 3 dBZ, the range
gate is flagged as clutter. The derived clutter can be removed from the
reflectivity field with the method
transform(). of the class
ClutterMap
tdbz_filter = TDBZFilter(window_size=5, threshold=3.)
tdbz_clt = tdbz_filter.create_cluttermap(reflectivity)
refl_filtered = tdbz_filter.transform(reflectivity)
Despite the application of clutter filters, some clutter remains in the radar
reflectivity. Several algorithms are applied to identify clutter signals and
interferences. To combine the results of several
ClutterFilter, use the
ClutterMap. You can set a fuzzy threshold, which is a
relative value to the number of clutter maps, which indicate clutter, to fulfil
the clutter condition. Note, if you apply more than one
ClutterFilter of same type, add some name
extension with the parameter add_name, otherwise you would overwrite one
clutter map with the append function (see below).
tdbz_filter = TDBZFilter(window_size=5, threshold=3.)
tdbz_filter_two = TDBZFilter(window_size=11, threshold=30.)
spin_filter = SPINFilter(threshold=3., window_size=11, window_criterion=.1)
tdbz_clt = tdbz_filter.create_cluttermap(reflectivity)
tdbz_clt_two = tdbz_filter_two.create_cluttermap(reflectivity,
addname='_some_more_info')
spin_clt = spin_filter.create_cluttermap(reflectivity)
cluttermap = ClutterMap('ClutterMapName', fuzzy_threshold=0.5)
cluttermap.append(tdbz_clt)
cluttermap.append(tdbz_clt_two)
cluttermap.append(spin_clt)
refl_filtered = cluttermap.transform(reflectivity)
Filter algorithms#
In the following, you can find the implemented clutter filter algorithms.
TDBZ filter#
This filter calculates the texture of the logarithmic reflectivity (TDBZ) according to Hubbert et al. (2009) [Hubbert et al., 2009] modified to 1D computations following [Lengfeld et al., 2014]. The TDBZ field is computed as the mean of the squared logarithmic reflectivity difference between adjacent range gates:
where \(\mathrm{dBZ}\) is reflectivity and \(N\) is the number of range gates used. If the mean within five consecutive range gates exceeds \(3\,\mathrm{dBZ}\) (default parameters), the range gate is flagged as clutter [Lengfeld et al., 2014].
Calculate texture of reflectivity (TDBZ) for clutter detection according to Hubbert et al. (2009). |
SPIN filter#
This filter calculates SPIN change of the reflectivity according to Hubbert et al. (2009) [Hubbert et al., 2009] modified to 1D computations. The SPIN field is a measure of how often the reflectivity gradient changes sign along the radial direction, with the following conditions:
and
where \(X_{i+1}\), \(X_{i}\), \(X_{i-1}\) are three consecutive dBZ values along a radar radial. The number of sign changes is calculated within a window of 11 range gates around the centre range gate and the reflectivity threshold is \(5\,\mathrm{dBZ}\) (default parameters) [Lengfeld et al., 2014].
Calculate SPIN change of the reflectivity for clutter detection according to Hubbert et al. (2009). |
Spike filter#
The spike filter identifies clutter in the form of spikes by calculating the reflectivity gradients for consecutive radar beams. If the differences in radar reflectivity of adjacent radar beams to the beam of interest exceed a threshold, the gate satisfies the first condition for clutter detection. If this condition is fulfilled for a certain percentage of consecutive range gates within a window, the gate is identified as clutter. Summarized, the following conditions need to be fulfilled for a percentage of e.g. \(50\,\%\) within a window of e.g. eleven consecutive range gates, with a threshold of \(3\,\mathrm{dBZ}\) and with a spike width \(W\) of one:
and
The index is for different radar beams.
We recommend to use two spike filters with a spike with of one and two to identify spikes effectively. The spike and ring filters are quite similar, but are defined for different axes.
Calculates gradients of consecutive radar beams for clutter detection. |
Ring filter#
The ring filter identifies clutter in the form of rings by calculating the reflectivity gradients for consecutive range gates. The ring filter is similar to the spike filter, but is defined for range gates instead of beams. Summarized, the conditions of the Spike filter are with the index for different ranges and need to be fulfilled for a percentage of e.g. \(50\,\%\) within a window of e.g. eleven consecutive radar beams with a threshold of \(3\,\mathrm{dBZ}\) and with a ring width \(W\) of one.
We recommend to use two ring filters with a ring with of one and two to identify rings effectively.
Calculates gradients of consecutive range gates for clutter detection. |
Speckle filter#
The speckle filter assumes that rain pixels are connected and larger than a single pixel. Following this, the probability is high that single rain pixels are clutter. The clutter filter looks for a number of rain pixels within a two-dimensional neighborhood of size \(k x l\). If this number of rain pixels is lower than a given threshold \(t\), the center pixel is identified as clutter. This can be formulated in the following way, with \(X_{i,j}\) as reflectivity in dBZ at position \(i\) and \(j\):
The speckle filter checks the number of rain pixels in a neighborhood. |
Temporal filter#
The temporal filter assumes that rain is moving slowly compared of the radar image update frequency, while clutter normally “jumps” around. We therefore can compare the current rain image to a history of rain images to identify clutter. If the sum of rain pixels in the last \(n\) images of a given grid point \(p\) is lower than a given threshold \(t\) (normally \(n=t\)), then this rain pixel is identified as clutter:
The temporal clutter filter compares current radar image with n-saved radar images. |
Using external filters#
pylawr package is able to integrate external filters. For example, the
clutter filter by Gabella et. al (2002) [Gabella et al., 2002] is an integration
of wradlib. We use wradlib.clutter.filter_gabella() to detect clutter
signals, see example below. For further information read wradlib documentation.
gabella = wradlib.clutter.filter_gabella(reflectivity.values[0], wsize=5,
thrsnorain=0., tr1=6., n_p=8,
tr2=1.3, rm_nans=False,
radial=False,
cartesian=False)[None, ...]
gabella_clt = ClutterMap('GabellaFilter', gabella.astype(int))
refl_filtered = gabella_clt.transform(reflectivity)
Functional API#
For functional-api usage, please note the methods
remove_clutter_lawr() and
remove_clutter_dwd(). The clutter detection
methods are tuned for the different radar types.
Remove clutter from given reflectivity field of lawr. |
|
Remove clutter from given reflectivity field of DWD. |