def derobertis(Sv, iax, jax, m, n, r, alpha, bgnmax=-125):
"""
Estimate background noise as in:
De Robertis and Higginbottom (2007) ‘A post-processing technique to
estimate the signal-to-noise ratio and remove echosounder background
noise’, ICES Journal of Marine Science, 64: 1282–1291.
Args:
Sv (float) : 2D array with Sv data (dB)
iax (int/float): 1D array with i axis (nsamples or metres)
jax (int/float): 1D array with j axis (npings, seconds, metres, etc)
m (int/float): i resampling length (nsamples or metres)
n (int/float): j resampling length (npings, seconds, metres, etc)
r (float) : 1D array with range data (metres)
alpha (float) : absorption coefficient value (dB m-1)
bgnmax (int/float): maximum background noise estimation (dB)
Returns:
float: 2D numpy array with background noise estimation (dB)
bool : 2D array with mask indicating valid noise estimation.
"""
# calculate TVG
r_ = r.copy()
r_[r<=0] = np.nan
TVG = 20*np.log10(r_) + 2*alpha*r_
# subtract TVG from Sv
Sv_noTVG = Sv - np.vstack(TVG)
# get resampled i/j axes
iaxrs = np.arange(iax[0], iax[-1], m)
jaxrs = np.arange(jax[0], jax[-1], n)
# proceed if length of resampled axes is greater than 1
if (len(iaxrs)>1) & (len(jaxrs)>1):
# resample Sv_noTVG into m by n bins
Sv_noTVGrs = rs.twod(Sv_noTVG, iax, jax, iaxrs, jaxrs, log=True)[0]
# compute background noise as the minimun value per interval in Sv_noTVGrs
jbool = np.isnan(Sv_noTVGrs).all(axis=0)
Sv_noTVGrs [:,jbool] = 0
bgn_noTVGrs = np.nanmin(Sv_noTVGrs, 0)
bgn_noTVGrs[ jbool] = np.nan
bgn_noTVGrs = np.tile(bgn_noTVGrs, [len(Sv_noTVGrs), 1])
# Prevent to exceed the maximum background noise expected
mask = np.ma.masked_greater(bgn_noTVGrs, bgnmax).mask
bgn_noTVGrs[mask] = bgnmax
# resample background noise to previous Sv resolution, and add TVG
bgn_noTVG, mask_ = rs.full(bgn_noTVGrs, iaxrs, jaxrs, iax, jax)
bgn = bgn_noTVG + np.vstack(TVG)
# return background noise as NAN values otherwise
else:
bgn = np.ones_like(Sv)*np.nan
mask_ = np.ones_like(Sv, dtype=bool)
warnings.warn("unable to estimate background noise, incorrect resampling axes", RuntimeWarning)
return bgn, mask_
def ccamlr(raw, prepro=None, jdx=[0, 0]):
"""
CCAMLR processing routine.
Process EK60 raw data and returns its variables in a dictionary array.
"""
#--------------------------------------------------------------------------
# check for appropiate inputs
if (isinstance(prepro, dict)) & (jdx[0] >= 0):
raise Exception('Preceeding raw data needs appropiate j indexes')
#--------------------------------------------------------------------------
# Load variables
rawfiles = raw['rawfiles']
transect = raw['transect']
alpha120 = raw['alpha']
r120 = raw['r']
t120 = raw['t']
lon120 = raw['lon']
lat120 = raw['lat']
nm120 = raw['nm']
km120 = raw['km']
knt120 = raw['knt']
kph120 = raw['kph']
pitchmax120 = raw['pitchmax']
rollmax120 = raw['rollmax']
heavemax120 = raw['heavemax']
Sv120 = raw['Sv']
theta120 = raw['theta']
phi120 = raw['phi']
#--------------------------------------------------------------------------
# join preceeding raw data, if there is continuity in the transect
if prepro is not None:
if prepro['transect'] == raw['transect']:
t120 = np.r_[prepro['t'][jdx[0]:], t120]
lon120 = np.r_[prepro['lon'][jdx[0]:], lon120]
lat120 = np.r_[prepro['lat'][jdx[0]:], lat120]
nm120 = np.r_[prepro['nm'][jdx[0]:], nm120]
km120 = np.r_[prepro['km'][jdx[0]:], km120]
knt120 = np.r_[prepro['knt'][jdx[0]:], knt120]
kph120 = np.r_[prepro['kph'][jdx[0]:], kph120]
Sv120 = np.c_[prepro['Sv'][:, jdx[0]:], Sv120]
theta120 = np.c_[prepro['theta'][:, jdx[0]:], theta120]
phi120 = np.c_[prepro['phi'][:, jdx[0]:], phi120]
else:
jdx[1] = 0
else:
jdx[1] = 0
#--------------------------------------------------------------------------
# report about the transects being processed
trsct = np.arange(jdx[1], nm120[-1], 1)
logger.info('Processing transect %03d : %2.2f - %2.2f nmi...' %
(transect, trsct[0], trsct[-1]))
#--------------------------------------------------------------------------
# Clean impulse noise
Sv120in, m120in_ = mIN.wang(Sv120,
thr=(-70, -40),
erode=[(3, 3)],
dilate=[(7, 7)],
median=[(7, 7)])
#TODO: True is valid
# -------------------------------------------------------------------------
# estimate and correct background noise
p120 = np.arange(len(t120))
s120 = np.arange(len(r120))
bn120, m120bn_ = gBN.derobertis(Sv120, s120, p120, 5, 20, r120, alpha120)
Sv120clean = tf.log(tf.lin(Sv120in) - tf.lin(bn120))
#TODO: True is valid
# -------------------------------------------------------------------------
# mask low signal-to-noise
m120sn = mSN.derobertis(Sv120clean, bn120, thr=12)
Sv120clean[m120sn] = -999
# -------------------------------------------------------------------------
# get mask for near-surface and deep data
m120rg = mRG.outside(Sv120clean, r120, 19.9, 250)
# -------------------------------------------------------------------------
# get mask for seabed
m120sb = mSB.ariza(Sv120,
r120,
r0=20,
r1=1000,
roff=0,
thr=-38,
ec=1,
ek=(3, 3),
dc=10,
dk=(3, 7))
# -------------------------------------------------------------------------
# get seabed line
idx = np.argmax(m120sb, axis=0)
sbline = r120[idx]
sbline[idx == 0] = np.inf
sbline = sbline.reshape(1, -1)
sbline[sbline > 250] = np.nan
# -------------------------------------------------------------------------
# get mask for non-usable range
m120nu = mSN.fielding(bn120, -80)[0]
# -------------------------------------------------------------------------
# remove unwanted (near-surface & deep data, seabed & non-usable range)
m120uw = m120rg | m120sb | m120nu
Sv120clean[m120uw] = np.nan
# -------------------------------------------------------------------------
# get swarms mask
k = np.ones((3, 3)) / 3**2
Sv120cvv = tf.log(
convolve2d(tf.lin(Sv120clean), k, 'same', boundary='symm'))
m120sh, m120sh_ = mSH.echoview(Sv120cvv,
r120,
km120 * 1000,
thr=-70,
mincan=(3, 10),
maxlink=(3, 15),
minsho=(3, 15))
# -------------------------------------------------------------------------
# get Sv with only swarms
Sv120sw = Sv120clean.copy()
Sv120sw[~m120sh & ~m120uw] = -999
# -------------------------------------------------------------------------
# resample Sv from 20 to 250 m, and every 1nm
r120intervals = np.array([20, 250])
nm120intervals = np.arange(jdx[1], nm120[-1], 1)
Sv120swr, r120r, nm120r, pc120swr = rs.twod(Sv120sw,
r120,
nm120,
r120intervals,
nm120intervals,
log=True)
# -------------------------------------------------------------------------
# remove seabed from pc120swr calculation, only water column is considered
m120sb_ = m120sb * 1.0
m120sb_[m120sb_ == 1] = np.nan
pc120water = rs.twod(m120sb_, r120, nm120, r120intervals,
nm120intervals)[3]
pc120swr = pc120swr / pc120water * 100
# -------------------------------------------------------------------------
# resample seabed line every 1nm
sbliner = rs.oned(sbline, nm120, nm120intervals, 1)[0]
# -------------------------------------------------------------------------
# get time resampled, interpolated from distance resampled
epoch = np.datetime64('1970-01-01T00:00:00')
t120f = np.float64(t120 - epoch)
f = interp1d(nm120, t120f)
t120rf = f(nm120r)
t120r = np.array(t120rf, dtype='timedelta64[ms]') + epoch
t120intervalsf = f(nm120intervals)
t120intervals = np.array(t120intervalsf, dtype='timedelta64[ms]') + epoch
# -------------------------------------------------------------------------
# get latitude & longitude resampled, interpolated from time resampled
f = interp1d(t120f, lon120)
lon120r = f(t120rf)
f = interp1d(t120f, lat120)
lat120r = f(t120rf)
# -------------------------------------------------------------------------
# resample back to full resolution
Sv120swrf, m120swrf_ = rs.full(Sv120swr, r120intervals, nm120intervals,
r120, nm120)
#TODO: True is valid
# -------------------------------------------------------------------------
# compute Sa and NASC from 20 to 250 m or down to the seabed depth
Sa120swr = np.zeros_like(Sv120swr) * np.nan
NASC120swr = np.zeros_like(Sv120swr) * np.nan
for i in range(len(Sv120swr[0])):
if (np.isnan(sbliner[0, i])) | (sbliner[0, i] > 250):
Sa120swr[0, i] = tf.log(tf.lin(Sv120swr[0, i]) * (250 - 20))
NASC120swr[0, i] = 4 * np.pi * 1852**2 * tf.lin(
Sv120swr[0, i]) * (250 - 20)
else:
Sa120swr[0, i] = tf.log(
tf.lin(Sv120swr[0, i]) * (sbliner[0, i] - 20))
NASC120swr[0, i] = 4 * np.pi * 1852**2 * tf.lin(
Sv120swr[0, i]) * (sbliner[0, i] - 20)
# -------------------------------------------------------------------------
# return processed data outputs
m120_ = m120in_ | m120bn_ | m120sh_ | m120swrf_
#TODO: True is valid
pro = {
'rawfiles': rawfiles, # list of rawfiles processed
'transect': transect, # transect number
'r120': r120, # range (m)
't120': t120, # time (numpy.datetime64)
'lon120': lon120, # longitude (deg)
'lat120': lat120, # latitude (deg)
'nm120': nm120, # distance (nmi)
'km120': km120, # distance (km)
'knt120': knt120, # speed (knots)
'kph120': kph120, # speed (km h-1)
'pitchmax120': pitchmax120, # max value in last pitching cycle (deg)
'rollmax120': rollmax120, # max value in last rolling cycle (deg)
'heavemax120': heavemax120, # max value in last heave cycle (deg)
'Sv120': Sv120, # Sv (dB)
'theta120': theta120, # Athwart-ship angle (deg)
'phi120': phi120, # Alon-ship angle (deg)
'bn120': bn120, # Background noise (dB)
'Sv120in': Sv120in, # Sv without impulse noise (dB)
'Sv120clean': Sv120clean, # Sv without background noise (dB)
'Sv120sw': Sv120sw, # Sv with only swarms (dB)
'nm120r': nm120r, # Distance resampled (nmi)
'r120intervals': r120intervals, # r resampling intervals
'nm120intervals': nm120intervals, # nmi resampling intervals
't120intervals': t120intervals, # t resampling intervals
'sbliner': sbliner, # Seabed resampled (m)
't120r': t120r, # Time resampled (numpy.datetime64)
'lon120r': lon120r, # Longitude resampled (deg)
'lat120r': lat120r, # Latitude resampled (deg)
'Sv120swr': Sv120swr, # Sv with only swarms resampled (dB)
'pc120swr': pc120swr, # Valid samples used to compute Sv120swr (%)
'Sa120swr': Sa120swr, # Sa from swarms, resampled (m2 m-2)
'NASC120swr': NASC120swr, # NASC from swarms, resampled (m2 nmi-2)
'Sv120swrf':
Sv120swrf, # Sv with only swarms, resampled, full resolution (dB)
'm120_': m120_
} # Sv mask indicating valid processed data (where all filters could be applied)
return pro
ek60 = EK60.EK60() ek60.read_raw(rawfile) #------------------------------------------------------------------------------ # get 120 kHz data raw120 = ek60.get_raw_data(channel_number=2) Sv120 = np.transpose(raw120.get_Sv().data) theta120 = np.transpose(raw120.angles_alongship_e) t120 = raw120.get_Sv().ping_time r120 = raw120.get_Sv().range #------------------------------------------------------------------------------ # Resample Sv in 2D, every 10 m in range and 100 seconds in time. r120rs = np.arange(r120[0], r120[-1], 10) t120rs = np.arange(t120[0], t120[-1], np.timedelta64(100, 's')) Sv120rs, Sv120rsper = rs.twod(Sv120, r120, t120, r120rs, t120rs, log=True) # By typing "log=True" the algorith knows that the variable to be resampled # is in logarithmic scale (Sv) and need to be converted into linear before # resampling and back to logarithmic after the resampling. # "Sv120rs" is the resampled array, and "Sv120per" is the percentage of valid # samples used in each resampling bin. It's a proxy of data quality. #------------------------------------------------------------------------------ # Resample theta in 1D, every 10 m in the vertical. r120rs = np.arange(r120[0], r120[-1], 10) theta120rs, theta120rsper = rs.oned(theta120, r120, r120rs, log=False) # This time we set "log=False" becaue theta is already at linear scale. # "log=False" is the default, so might skip this setting.