# Variance Ellipses¶

In this lab, you'll be working with current meter data, that is, observations of the horizontal ocean currents made from a fixed point. The main analysis tool we will learn are variance ellipses, a fundamental tool for analyzing velocity datasets or any kind of bivariate (or complex-valued) data. You'll see their relationship to the two-dimensional histogram, and will learn some associated analysis and visualization ideas, in particular the value of finding a suitable coordinate rotaiton.

For this assignment, we are going to use a mooring from the Labrador Current on the west side of the Labrador Sea known as the ‘m1244’ mooring. Please download my version of it here. (This is included in the full distribution of the course materials.)

Many thanks to Jan-Adrian Kallmyr for helping with translating the Matlab tutorial into Python.

# A Quick Look at the Data¶

Now let's load and examine the data.

The variables are as follows:

depths --- Depths of 4 different currents meters, in meters
num -- date in Matlab's "datenum" format
p --- Pressure in decibar
cv --- Complex velocity u+iv in cm/s, u = eastward, v = northward

Before we proceed let's find out the sampling interval and duration of the time series.

So we have a roughly 1.7 year record of the currents sampled every two hours.

Let's take a look at the currents the deepest depth.

In what follows, we're going to take advantage of Python's very useful ability to define functions on the fly. This will let us make efficiently re-use code in later plots.

Here a mean value is readily apparent in both the eastward and northward components. In addition, we see a lot of small-scale variability, possibly with a greater amplitude in the eastward component. Hints of multiple timescales of variability are present, with a fine noise-like variability superposed on somewhat longer timescales.

It is a little difficult to make sense of this plot, however, because it is not rotated into the most useful coordinate system. Let's figure out a sensible rotation angle.

# Choosing a Coordinate Rotation¶

To do this, we will look at the progressive vector diagram. For a current measurement at a point, the progressive vector diagram is defined as a plot of its time integral. In other words, the progressive vector diagram shows the displacement that would occur if a particle were advected with the given currents.

We see a strong mean flow at all depths directed to the east-southeast. This matches what we saw above with the line plot, where the positive mean of the eastward currents is roughly twice as large as the negative mean of the southward currents.

Note there is an important line in the "provec" function that sets the aspect ratio. If we had omitted it,we could have gotten something tlat looks like this:

Here, the currents appear to be directed due southeast. However, there is a problem with this plot: the x and y axes, which are the same physical quantity of displacement, correspond to different physical lengths! If we re-scaled the figure window, the direction of the mean flow would appear to change. Thus our perspective on the comparison between the eastward and northward currents is distorted.

This problem is fixed by setting the data aspect ratio. Not setting the data aspect ratio correctly is actually a common problem, even, regrettably, in many published papers.

# Rotating by the Mean Flow Direction¶

It's natural to rotate our current meter data so that the east-southeastward direction of the mean flow corresponds to the first velocity component, and the direction normal to that to the second velocity component. In other words, we will rotate our $(u,v)$ data to become $)\tilde u,\tilde v)$ with $\tilde u$ corresponding to the ‘downstream’ direction and $\tilde v$ corresponds to the ‘cross-stream’ direction. We will take a little time to do this so we understand how rotations work.

Firstly we find the direction of the mean flow at the deepest current meter, where the flow is strongest.

So the angle is about 30 degrees clockwise from due east. That looks about right!

Rotation of a complex-valued number $z=u+iv$ are straightforward. To rotate $z=|z|e^{i\varphi}$ through some angle $\phi$, we simply multiply by $e^{i\phi}$. This leads to a new version of $z$ , denoted $\tilde z =\tilde u+ i\tilde v=|z|e^{i(\varphi+\phi)}$.

In this case, we want to choose the rotation angle $\phi$ as the negative of the angle of the mean flow. Let's look at the mean real and imaginary components of the velocity before and after this rotation.

This shows that after the rotation, the mean of the cross-stream velocity (the imaginary part) is zero, as expected. Now let's re-plot the progressive vector diagram.

At first glance, this plot appears rather boring. But actually, boring can be a good sign because it means we've found a way to look at our dataset in such a way that it simplifies!

Let's return to the line plot we made earlier, but now make it for the rotated velocity data.

The downstream and cross-stream components are seen to be distinctly different. The cross-stream flow oscillates about a mean of zero, which is by construction, and also presents higher-frequency variability than does the downstream flow. Or, perhaps more accuately, it appears that the cross-stream flow is lacking an intermediate-timescale component that is present in the downstream flow.

Let's take a closer look at the timescales present in this dataset with a simple lowpass filter. We'll use a 24-point filter, which corresponds to a 2-day running mean since the sampling interval is 2 hours.