def smoothed_linear_interpolation_between_tiepoints(l2, ssh_tiepoint_indices, filter_width):
"""
The main SLA computation method in this class
:param l2: Level-2 data container
:param ssh_tiepoint_indices:
:param filter_width:
:return: None
"""
# Step 1: Get the observed (noisy) sea surface elevations
sla_raw = np.full(l2.n_records, np.nan)
sla_raw[ssh_tiepoint_indices] = l2.elev[ssh_tiepoint_indices] - l2.mss[ssh_tiepoint_indices]
non_tiepoints = np.isnan(sla_raw)
# Step 2: Create a filtered version of the raw SLA, but don't interpolate yet
# Use python implementation of IDL SMOOTH:
# idl_smooth(x, w) equivalent to SMOOTH(x, w, /edge_truncate, /nan)
sla_filter1 = idl_smooth(sla_raw, filter_width)
sla_filter1[non_tiepoints] = np.nan
# Step 3: Fill nans with linear interpolation and constant values at borders
# python: fill_nan(x) = IDL: FILL_NAN(x, /NEIGHBOUR)
sla_filter2 = fill_nan(sla_filter1)
# Step 4: The sea level anomaly is the smoothed version
# of the gap filled sla
sla = idl_smooth(sla_filter2, filter_width)
# All done, return value
return sla
def _linear_smoothed_interpolation_between_tiepoints(self, l2):
""" Based in cs2awi code from Robert Ricker """
# Use ocean and lead elevations
self._ssh_tiepoints = l2.surface_type.lead.indices
if self._options.use_ocean_wfm:
self._ssh_tiepoints.append(l2.surface_type.ocean.indices)
self._ssh_tiepoints = np.sort(self._ssh_tiepoints)
# Get initial elevation at tiepoint locations
mss_frb = l2.elev - l2.mss
# Remove ssh tiepoints from the list if their elevation
# corrected by the median offset of all tiepoints from the mss
# exceeds a certain threshold
if self._options.pre_filtering:
# Startup
index_dict = np.arange(l2.surface_type.lead.num)
threshold = self._options.pre_filter_maximum_mss_median_offset
# Compute the mean distance to the mss
tiepoint_mss_distance = mss_frb[self._ssh_tiepoints]
valid_points = np.where(np.isfinite(tiepoint_mss_distance))[0]
tiepoint_mss_distance = tiepoint_mss_distance[valid_points]
median_mss_offset = np.median(tiepoint_mss_distance)
# Filter points for outliers
offset = np.abs(mss_frb[self._ssh_tiepoints] - median_mss_offset)
valid = np.where(offset < threshold)
self._ssh_tiepoints = self._ssh_tiepoints[index_dict[valid]]
self.ssa_raw = np.ndarray(shape=(l2.n_records)) * np.nan
self.ssa_raw[self._ssh_tiepoints] = mss_frb[self._ssh_tiepoints]
non_tiepoints = np.where(np.isnan(self.ssa_raw))
# Filtered raw values
# Use custom implementation of IDL SMOOTH:
# idl_smooth(x, w) equivalent to SMOOTH(x, w, /edge_truncate, /nan)
ssa_filter1 = idl_smooth(self.ssa_raw, self.filter_width)
# Leave only the original ssh tie points
ssa_filter1[non_tiepoints] = np.nan
# Fill nans with linear interpolation and contant values at borders
# python: fill_nan(x) = IDL: FILL_NAN(x, /NEIGHBOUR)
ssa_filter2 = fill_nan(ssa_filter1)
# Final smoothing
ssa = idl_smooth(ssa_filter2, self.filter_width)
self._value = ssa
def get_l2_track_vars(self, l2):
# Set the requested data
self.set_requested_date_from_l2(l2)
# CAVEAT: The method `update_external_data()` will fail
# if the requested date is February 29, since no
# corresponding source file exists. As a fix, the data in
# in this case is set back to the February 28.
if self.month == "02" and self.day == "29":
self.set_requested_date(int(self.year), int(self.month), 28)
# Update the external data
self.update_external_data()
# Check if error with file I/O
if self.error.status or self._data is None:
# This will return an empty container
snow = SnowParameterContainer()
snow.set_dummy(l2.n_records)
else:
# Extract along track snow depth and density
sd, sd_unc = self._get_snow_track(l2)
# Apply along-track smoothing if required
smooth_snowdepth = self.cfg.options.get("self.cfg.options", False)
if smooth_snowdepth:
filter_width = self.cfg.options.smooth_filter_width_m
# Convert filter width to index
filter_width /= l2.footprint_spacing
# Round to odd number
filter_width = np.floor(filter_width) // 2 * 2 + 1
sd = idl_smooth(sd, filter_width)
sd_unc = idl_smooth(sd_unc, filter_width)
# Collect Parameters and return
# (density and density uncertainty fixed from l2 settings)
snow = SnowParameterContainer()
snow.depth = sd
snow.depth_uncertainty = sd_unc
snow.density = np.full(sd.shape, self.cfg.options.snow_density)
snow.density_uncertainty = np.full(
sd.shape, self.cfg.options.snow_density_uncertainty)
# Register Variables
self.register_auxvar("sd", "snow_depth", snow.depth,
snow.depth_uncertainty)
self.register_auxvar("sdens", "snow_density", snow.density,
snow.density_uncertainty)
def get_l2_track_vars(self, l2):
# Set the requested data
self.set_requested_date_from_l2(l2)
# Update the external data
self.update_external_data()
# Check if error with file I/O
if self.error.status or self._data is None:
# This will return an empty container
snow = SnowParameterContainer()
snow.set_dummy(l2.n_records)
else:
# Extract along track snow depth and density
sd, sd_unc = self._get_snow_track(l2)
# Apply along-track smoothing if required
if self.cfg.options.smooth_snowdepth:
filter_width = self.cfg.options.smooth_filter_width_m
# Convert filter width to index
filter_width /= l2.footprint_spacing
# Round to odd number
filter_width = np.floor(filter_width) // 2 * 2 + 1
sd = idl_smooth(sd, filter_width)
sd_unc = idl_smooth(sd_unc, filter_width)
# Collect Parameters and return
# (density and density uncertainty fixed from l2 settings)
snow = SnowParameterContainer()
snow.depth = sd
snow.depth_uncertainty = sd_unc
snow.density = np.full(sd.shape, self.cfg.options.snow_density)
snow.density_uncertainty = np.full(
sd.shape, self.cfg.options.snow_density_uncertainty)
# Register Variables
self.register_auxvar("sd", "snow_depth", snow.depth,
snow.depth_uncertainty)
self.register_auxvar("sdens", "snow_density", snow.density,
snow.density_uncertainty)
def get_l2_track_vars(self, l2):
""" Get the snow depth, density and their uncertainties for the track in the l2 data object
including the potential modification of the original climatology and filters """
# Validate hemisphere
if l2.hemisphere == "south":
snow = SnowParameterContainer()
snow.set_dummy(l2.n_records)
msg = "Warren99 not valid for southern hemisphere, returning 0"
self.error.add_error("warren99-invalid-hemisphere", msg)
return snow
# Get the original warren climatology values
# NOTE: snow is a class with properties depth, depth_uncertainty, density & density_uncertainty
snow = self._get_warren99_fit_from_l2(l2)
# Filter invalid values
valid_min, valid_max = self.cfg.options.valid_snow_depth_range
invalid = np.logical_or(snow.depth < valid_min, snow.depth > valid_max)
invalid_records = np.where(invalid)[0]
snow.set_invalid(invalid_records)
# Apply ice_type (myi_fraction correction)
scale_factor = (1.0 -
l2.sitype) * self.cfg.options.fyi_correction_factor
# The scaling factor affects the snow depth ...
snow.depth = snow.depth - scale_factor * snow.depth
# ... and the uncertainty. Here it is assumed that the uncertainty
# is similar affected by the scaling factor.
snow.depth_uncertainty = snow.depth_uncertainty - scale_factor * snow.depth_uncertainty
# the uncertainty of the myi fraction is acknowledged by adding
# an additional term that depends on snow depth, the magnitude of
# scaling and the sea ice type uncertainty
scaling_uncertainty = snow.depth * scale_factor * l2.sitype.uncertainty
snow.depth_uncertainty = snow.depth_uncertainty + scaling_uncertainty
# Smooth snow depth (if applicable)
if self.cfg.options.smooth_snow_depth:
filter_width = self.cfg.options.smooth_filter_width_m
# Convert filter width to index
filter_width /= l2.footprint_spacing
# Round to odd number
filter_width = np.floor(filter_width) // 2 * 2 + 1
snow.depth = idl_smooth(snow.depth, filter_width)
# Register Variables
self.register_auxvar("sd", "snow_depth", snow.depth,
snow.depth_uncertainty)
self.register_auxvar("sdens", "snow_density", snow.density,
snow.density_uncertainty)
