def unfold_phi(phidp, rho, width=5, copy=False):
"""Unfolds differential phase by adjusting values that exceeded maximum \
ambiguous range.
Accepts arbitrarily dimensioned arrays, but THE LAST DIMENSION MUST BE
THE RANGE.
This is the fast Fortran-based implementation (RECOMMENDED).
The algorithm is based on the paper of :cite:`Wang2009`.
Parameters
----------
phidp : :class:`numpy:numpy.ndarray`
array of shape (...,nr) with nr being the number of range bins
rho : :class:`numpy:numpy.ndarray`
array of same shape as ``phidp``
width : int
Width of the analysis window
copy : bool
Leaves original ``phidp`` array unchanged if set to True
(default: False)
"""
# Check whether fast Fortran implementation is available
speedup = util.import_optional("wradlib.speedup")
shape = phidp.shape
assert rho.shape == shape, "rho and phidp must have the same shape."
phidp = phidp.reshape((-1, shape[-1]))
if copy:
phidp = phidp.copy()
rho = rho.reshape((-1, shape[-1]))
gradphi = util.gradient_from_smoothed(phidp)
beams, rs = phidp.shape
# Compute the standard deviation within windows of 9 range bins
stdarr = np.zeros(phidp.shape, dtype=np.float32)
for r in range(rs - 9):
stdarr[..., r] = np.std(phidp[..., r:r + 9], -1)
phidp = speedup.f_unfold_phi(
phidp=phidp.astype("f4"),
rho=rho.astype("f4"),
gradphi=gradphi.astype("f4"),
stdarr=stdarr.astype("f4"),
beams=beams,
rs=rs,
w=width,
)
return phidp.reshape(shape)
def unfold_phi_naive(phidp, rho, width=5, copy=False):
"""Unfolds differential phase by adjusting values that exceeded maximum \
ambiguous range.
Accepts arbitrarily dimensioned arrays, but THE LAST DIMENSION MUST BE
THE RANGE.
This is the slow Python-based implementation (NOT RECOMMENDED).
The algorithm is based on the paper of :cite:`Wang2009`.
Parameters
----------
phidp : :class:`numpy:numpy.ndarray`
array of shape (...,nr) with nr being the number of range bins
rho : :class:`numpy:numpy.ndarray`
array of same shape as ``phidp``
width : int
Width of the analysis window
copy : bool
Leaves original ``phidp`` array unchanged if set to True
(default: False)
"""
shape = phidp.shape
assert rho.shape == shape, "rho and phidp must have the same shape."
phidp = phidp.reshape((-1, shape[-1]))
if copy:
phidp = phidp.copy()
rho = rho.reshape((-1, shape[-1]))
gradphi = util.gradient_from_smoothed(phidp)
beams, rs = phidp.shape
# Compute the standard deviation within windows of 9 range bins
stdarr = np.zeros(phidp.shape, dtype=np.float32)
for r in range(rs - 9):
stdarr[..., r] = np.std(phidp[..., r:r + 9], -1)
# phi_corr = np.zeros(phidp.shape)
for beam in range(beams):
if np.all(phidp[beam] == 0):
continue
# step 1: determine location where meaningful PhiDP profile begins
for j in range(0, rs - width):
if (np.sum(stdarr[beam, j:j + width] < 5) == width) and \
(np.sum(rho[beam, j:j + 5] > 0.9) == width):
break
ref = np.mean(phidp[beam, j:j + width])
for k in range(j + width, rs):
if np.sum(stdarr[beam, k - width:k] < 5) and \
np.logical_and(gradphi[beam, k] > -5,
gradphi[beam, k] < 20):
ref += gradphi[beam, k] * 0.5
if phidp[beam, k] - ref < -80:
if phidp[beam, k] < 0:
phidp[beam, k] += 360
elif phidp[beam, k] - ref < -80:
if phidp[beam, k] < 0:
phidp[beam, k] += 360
return phidp
def test_gradient_from_smoothed(self):
x = np.arange(10).reshape((2, 5)).astype("f4")**2
result = util.gradient_from_smoothed(x)
shouldbe = np.array([[1.0, 2.0, 1.5, 0.0, 0.0],
[11.0, 12.0, 6.5, 0.0, 0.0]])
np.testing.assert_allclose(result, shouldbe)
def test_gradient_from_smoothed(self):
x = np.arange(10).reshape((2, 5)).astype("f4") ** 2
result = util.gradient_from_smoothed(x)
shouldbe = np.array([[1., 11., 2., 1.5, 0., 0.],
[1., 11., 12., 6.5, 0., 0.]])
np.testing.assert_allclose(result, shouldbe)