Reading ADV Data with MHKiT

This example presents a simplified workflow for analyzing Acoustic Doppler Velocimetry (ADV) data using MHKiT. MHKiT incorporates the DOLfYN codebase as a module to handle ADV and Acoustic Doppler Current Profiler (ADCP) data.

A standard ADV data analysis workflow can be segmented into the following steps:

Raw Data Review: Evaluate the original data by verifying timestamps and assessing the quality of velocity data, specifically looking for any data spikes.
Data Cleaning: Identify and eliminate any spurious data points. If needed, bad data points can be replaced with interpolated values.
Data Rotation: Transform the data into the principal flow coordinates, which are the streamwise, cross-stream, and vertical directions.
Data Averaging: Aggregate the data into bins or ensembles, each of which spans a predefined time length, typically between 5 and 10 minutes.
Statistical Analysis: Compute turbulence statistics such as turbulence intensity, Turbulent Kinetic Energy (TKE), and Reynolds stresses for the observed flow field.

Start your analysis by importing the necessary tools:

[1]:

from mhkit import dolfyn
from mhkit.dolfyn.adv import api

Quality Control

ADV velocity data tends to have spikes due to Doppler noise, and the common way to “despike” the data is by using the phase-space algorithm by Goring and Nikora (2002). DOLfYN integrates this function using a 2-step approach: create a logical mask where True corresponds to a spike detection, and then utilize an interpolation function to replace the spikes.

[5]:

# Clean the file using the Goring+Nikora method:
mask = api.clean.GN2002(ds.vel, npt=5000)
# Replace bad datapoints via cubic spline interpolation
ds["vel"] = api.clean.clean_fill(ds["vel"], mask, npt=12, method="cubic", maxgap=None)

print("Percent of data containing spikes: {0:.2f}%".format(100 * mask.mean()))

# If interpolation isn't desired:
ds_nan = ds.copy(deep=True)
ds_nan.coords["mask"] = (("dir", "time"), ~mask)
ds_nan["vel"] = ds_nan["vel"].where(ds_nan["mask"])

Percent of data containing spikes: 0.73%

Coordinate Rotations

Now that the data has been cleaned, the next step is to rotate the velocity data into true East, North, Up (ENU) coordinates.

ADVs use an internal compass or magnetometer to determine magnetic ENU directions. The set_declination function takes the user supplied magnetic declination (which can be looked up online for specific coordinates) and adjusts the orientation matrix saved within the dataset.

Instruments save vector data in the coordinate system specified in the deployment configuration file. To make the data useful, it must be rotated through coordinate systems (“beam”<->”inst”<->”earth”<->”principal”), done through the rotate2 function. If the “earth” (ENU) coordinate system is specified, DOLfYN will automatically rotate the dataset through the necessary coordinate systems to get there. The inplace set as true will alter the input dataset “in place”, a.k.a. it not create a new dataset.

[6]:

# First set the magnetic declination
dolfyn.set_declination(
    ds, declin=10, inplace=True
)  # declination points 10 degrees East

# Rotate that data from the instrument to earth frame (ENU):
dolfyn.rotate2(ds, "earth", inplace=True)

Once in the true ENU frame of reference, we can calculate the principal flow direction for the velocity data and rotate it into the principal frame of reference (streamwise, cross-stream, vertical). Principal flow directions are aligned with and orthogonal to the flow streamlines at the measurement location.

First, the principal flow direction must be calculated through calc_principal_heading. As a standard for DOLfYN functions, those that begin with “calc_*” require the velocity data for input. This function is different from others in DOLfYN in that it requires place the output in an attribute called “principal_heading”, as shown below.

Again we use rotate2 to change coordinate systems.

[7]:

ds.attrs["principal_heading"] = dolfyn.calc_principal_heading(ds["vel"])
dolfyn.rotate2(ds, "principal", inplace=True)

Averaging Data

The next step in ADV analysis is to average the velocity data into time bins (ensembles) and calculate turbulence statistics. These averaged values are then used to calculate turbulence statistics. There are two distinct methods for performing this operation, both of which utilize the same variable inputs and produce identical datasets.

Object-Oriented Approach (standard): Define an ‘averaging object’, create a dataset binned in time, and calculate basic turbulence statistics. This is accomplished by initiating an object from the ADVBinner class and then feeding that object with our dataset.
Functional Approach (simple): The same operations can be performed using the functional counterpart of ADVBinner, turbulence_statistics.

Function inputs shown here are the dataset itself: - n_bin: the number of elements in each bin; - fs: the ADV’s sampling frequency in Hz; - n_fft: optional, the number of elements per FFT for spectral analysis; - freq_units: optional, either in Hz or rad/s, of the calculated spectral frequency vector.

All of the variables in the returned dataset have been bin-averaged, where each average is computed using the number of elements specified in n_bins. Additional variables in this dataset include the turbulent kinetic energy (TKE) vector (“ds_binned.tke_vec”), the Reynold’s stresses (“ds_binned.stress”), and the power spectral densities (“ds_binned.psd”), calculated for each bin.

[8]:

# Option 1 (standard)
binner = api.ADVBinner(n_bin=ds.fs * 600, fs=ds.fs, n_fft=1024)
ds_binned = binner.bin_average(ds)

# Option 2 (simple)
# ds_binned = api.calc_turbulence(ds, n_bin=ds.fs*600, fs=ds.fs, n_fft=1024, freq_units="Hz")

The benefit to using ADVBinner is that one has access to all of the velocity and turbulence analysis functions that DOLfYN contains. If basic analysis will suffice, the turbulence_statistics function is the most convienent. Either option can still utilize DOLfYN’s shortcuts.

See the DOLfYN API for the full list of functions and shortcuts. A few examples are shown below.

Some things to know: - All functions operate bin-by-bin. - Some functions will fail if there are NaN’s in the data stream (Notably the PSD functions) - “Shorcuts”, as referred to in DOLfYN, are functions accessible by the xarray accessor velds, as shown below. The list of “shorcuts” available through velds are listed here. Some shorcut variables require the raw dataset, some an averaged dataset.

For instance, - bin_variance calculates the binned-variance of each variable in the raw dataset, the complementary to bin_average. Variables returned by this function contain a “_var” suffix to their name. - turbulence_intensity is calculated from the ratio of the standard deviation of the horizontal velocity magnitude (equivalent to the RMS of turbulent velocity fluctuations) to the mean of the horizontal velocity magnitude - power_spectral_density calculates the power spectral density (velocity spectra) of the velocity vector - cross_spectral_density calculates the cross spectral density between each direction of the supplied DataArray. Note that inputs specified in creating the ADVBinner object can be overridden or additionally specified for a particular function call. - dissipation_rate_LT83 uses the Lumley and Terray 1983 algorithm to estimate the TKE dissipation rate from the isoropic turbulence cascade seen in the spectral. This requires the frequency range of the cascade as input. - turbulent_kinetic_energy calculates the TKE (Reynolds normal stress) components - reynolds_stress calculates the Reynolds shear stress components

[9]:

# Calculate the variance of each variable in the dataset and add to the averaged dataset
ds_binned = binner.bin_variance(ds, out_ds=ds_binned)

# Calculate the turbulence intensity
ds_binned["TI"] = binner.turbulence_intensity(ds.velds.U_mag)

# Calculate the power spectral density
ds_binned["auto_spectra"] = binner.power_spectral_density(ds["vel"], freq_units="Hz")

# Calculate the cross power spectral density
ds_binned["cross_spectra"] = binner.cross_spectral_density(
    ds["vel"], freq_units="Hz", n_fft_coh=512
)

# Calculate dissipation rate from isotropic turbulence cascade
ds_binned["dissipation"] = binner.dissipation_rate_LT83(
    ds_binned["auto_spectra"], ds_binned.velds.U_mag, freq_range=[0.5, 1]
)

# Calculate the Reynolds stresses
ds_binned['tke_vec'] = binner.turbulent_kinetic_energy(ds["vel"])
ds_binned['stress_vec'] = binner.reynolds_stress(ds["vel"])

Plotting can be performed using matplotlib. As an example, the mean spectrum in the streamwise direction is plotted here. This spectrum shows the mean energy density in the flow at a particular flow frequency.

[10]:

import matplotlib.pyplot as plt

%matplotlib inline

plt.figure()
plt.loglog(ds_binned["freq"], ds_binned["auto_spectra"].sel(S="Sxx").mean(dim="time"))
plt.xlabel("Frequency [Hz]")
plt.ylabel("Energy Density $\mathrm{[m^2/s^s/Hz]}$")
plt.title("Streamwise Direction")

[10]:

Text(0.5, 1.0, 'Streamwise Direction')

Saving and Loading DOLfYN datasets

Datasets can be saved and reloaded using the save and load functions. Xarray is saved natively in netCDF format, hence the “.nc” extension.

Note: DOLfYN datasets cannot be saved using xarray’s native ds.to_netcdf; however, DOLfYN datasets can be opened using xarray.open_dataset.

[11]:

# Uncomment these lines to save and load to your current working directory
# dolfyn.save(ds, 'your_data.nc')
# ds_saved = dolfyn.load('your_data.nc')

Reading ADV Data with MHKiT

Read Raw Instrument Data

Quality Control

Coordinate Rotations

Averaging Data

Saving and Loading DOLfYN datasets