def compute_spectra_cfs(raw, array):
raw['dtime'] = ('time',
np.array((raw.time - raw.time.isel(time=0)).values * 1e-9,
dtype=float))
raw = raw.swap_dims({'time': 'dtime'})
return dsp.spectrogram(raw[array],
dim='dtime',
fs=86400 / dsp.get_sampling_step(raw, dim='dtime'),
nperseg=128)
def bandpass_velocity(raw, low_f, high_f):
import xrscipy.signal as dsp
# TODO: fix naming of variables
# convert datetime to seconds since t=0
raw['dtime'] = ('time',
np.array((raw.time - raw.time.isel(time=0)).values * 1e-9,
dtype=float))
# raw['dtime'] = ('time', np.array( (raw.time - raw.time.isel(time=0)).values, dtype=float))
# make dtime a dimension
raw = raw.swap_dims({'time': 'dtime'})
# filtering proceduce
# determine sampling timestep and Nyquist frequency
T = (dsp.get_sampling_step(raw, dim='dtime'))
# T = 5000
fs = 1 / T
ny = 0.5 * fs
# limits for bandpass
# low_f = gsw.f(lat_mean)*low # in 1/s
# high_f = gsw.f(lat_mean)*high # in 1/s
eps = 0 # how to fill nans
# # pick an order?
# print(low_f/ny)
# print(high_f/ny)
ulow = dsp.bandpass(raw.u.fillna(eps),
low_f / ny,
high_f / ny,
dim='dtime',
in_nyq=True,
order=4)
vlow = dsp.bandpass(raw.v.fillna(eps),
low_f / ny,
high_f / ny,
dim='dtime',
in_nyq=True,
order=4)
# ulow = dsp.bandpass(raw.u.fillna(eps), low_f, high_f, dim='dtime', in_nyq=False, order=4)
# vlow = dsp.bandpass(raw.v.fillna(eps), low_f, high_f, dim='dtime', in_nyq=False, order=4)
# swap dims back
ulow = ulow.swap_dims({'dtime': 'time'})
vlow = vlow.swap_dims({'dtime': 'time'})
raw = raw.swap_dims({'dtime': 'time'})
# remove time and space means?
ulow = ulow #- ulow.mean(dim='z') - ulow.mean(dim='time')
vlow = vlow #- vlow.mean(dim='z') - ulow.mean(dim='time')
mask = ~np.isnan(raw.u) & ~np.isnan(raw.v)
raw['uNI'] = ulow.where(mask)
raw['vNI'] = vlow.where(mask)
return raw
def bandstop_variable(raw, array, low_f, high_f):
import xrscipy.signal as dsp
# TODO: fix naming of variables
# convert datetime to seconds since t=0
raw['dtime'] = ('time',
np.array((raw.time - raw.time.isel(time=0)).values * 1e-9,
dtype=float))
# raw['dtime'] = ('time', np.array( (raw.time - raw.time.isel(time=0)).values, dtype=float))
# make dtime a dimension
raw = raw.swap_dims({'time': 'dtime'})
# filtering proceduce
# determine sampling timestep and Nyquist frequency
T = (dsp.get_sampling_step(raw, dim='dtime'))
fs = 1 / T
ny = 0.5 * fs
# limits for bandpass
eps = 0 # how to fill nans
# # pick an order?
# print(low_f/ny)
# print(high_f/ny)
ulow = dsp.bandstop(raw[array].fillna(eps),
low_f / ny,
high_f / ny,
dim='dtime',
in_nyq=True,
order=4)
# swap dims back
ulow = ulow.swap_dims({'dtime': 'time'})
raw = raw.swap_dims({'dtime': 'time'})
# remove time and space means?
ulow = ulow #- ulow.mean(dim='z') - ulow.mean(dim='time')
mask = ~np.isnan(raw[array])
varname = array + 'NI'
raw[varname] = ulow.where(mask)
return raw