Monday, March 12, 2012

The sound of a solar storm, and a tour of "audification"

A composer and NASA fellow at Michigan Engineering is turning satellite measurements into sound as a new data mining approach. Here's a demonstration.



It does sound like stardust. That, with maybe a little electronic whale song mixed in.

This "sonification" of some of the most recent solar storm activity illustrates a unique new approach to data mining in the University of Michigan's Solar and Heliospheric Research Group. Its creator is Robert Alexander, a design science Ph.D. student on a NASA fellowship to explore how turning data into sound could help scientists hear patterns or anomalies that their eyes might miss.

For this project, Alexander started with 90 hours worth of raw information from two NASA spacecraft, MESSENGER at planet Mercury and the Solar and Heliospheric Observatory near Earth. (The instrument he used from MESSENGER is the Fast Imaging Plasma Spectrometer, or FIPS, built right here at Michigan. It's said to be the first to register the latest storm.)

In the particular type of sonification he employs, called audification, each data sample becomes a single audio sample. That means 44,100 pieces of information play back as one second of sound at the common sampling rate of 44,100 hertz. So to make sense of the audio and extract anything meaningful from it, Alexander has to stretch it and in this case run it through additional algorithms.

It's a groundbreaking process that he and his colleagues say is giving rise to a new research tool.

"I can listen to a million data points in approximately 22 seconds," Alexander said.

Last year, this technique led to a new discovery: It turns out that a particular ratio of carbon atoms that scientists had not previously keyed in to can reveal more about the source of the solar wind than the ratios of elements they had currently relied on. He's the second author on the paper published in December in Astrophysical Journal.

Alexander, the Solar and Heliospheric Research Group's data sonification specialist, emailed me a detailed explanation of this work, complete with audio snippets. I think it's all worth sharing.
Before you proceed, pull out a nice pair of headphones and set your volume to a very low level.
For starters, have a listen to the heart-beat of the sun, a sound file which I generated by translating 47 years worth of solar proton speed data directly to audio. What we're listening to here is a flow of charged particles called the solar wind.


If you listen closely you can hear a low hum rise and fall in amplitude as the sun oscillates between solar minimum and solar maximum (order and chaos). This rise and fall occurs in an 11 year cycle. That low hum is generated by the rotation of the sun. We're listening to features on the surface of the sun (coronal holes) that swing around to hit the satellite every 26.4 days. The first time I heard this hum in all my data I thought there was an error in my algorithm, until I did the math.

When we speed up the data and listen closely, you can hear harmonics, which relate directly back to the spherical harmonics of the sun's magnetic field. No worries if you don't hear the harmonics immediately... the ear is a powerful pattern detecting machine, and it will slowly tease out even the subtlest of details from the noisiest of data. If you listen long enough, certain features will suddenly jump out of nowhere....

Here is that same file run through a single filter to boost some areas of interest. If you listen even closer you'll start to hear all sorts of things happening. There is a set of pulses near the middle of the audio file that is caused by a phenomenon known as a "Mid-Term Quasi Periodicity," which tells us something about how one of the sub-surface layers of the sun is interacting with the larger magnetic field. Anything that you hear that isn't pure noise could potentially be a new scientific discovery (though it's more common that the phenomenon has already been documented)

If we apply another filter, and yet another... these underlying periodicities become undeniably clear. Listening again to the original file, it's much easier to separate these patterns from the noise. This gives you an idea for what the process for auditory data analysis is like.

I became very interested in the underlying solar harmonics. In this audio file I've strongly accentuated the third harmonic, such that you can get a sense of exactly what you should be listening for. In musical terms, you're listening to a frequency that's an octave and a fifth above the fundamental. You should notice that it gets louder right towards the end of the file.

I began listening to these harmonics as they appeared in different types of data, including oxygen and carbon ratios as recorded by the NASA ACE satellite. These ratios are used as indicators of temperature very close to the sun, due to the fact that particles all "freeze" into a certain temperature before they move very far away from the sun. The ratio of oxygen 7 to oxygen 6, or O7/O6 was the leading indicator that we had for determining where energy was originating from on the sun, that is to say, the source regions of the solar wind.

However... when I listened to the data, I found that the harmonics were not quite as strong as in the original helium velocity data. Then came the breakthrough.

I started listening to carbon data (C4, C5 and C6), and I found the harmonics were much more prominent in C4 and C6, and noticeably absent in C5. This led me to ask the group if any research had been conducted into the utilization of carbon charge state 4. Here is a page that will allow you to compare the sounds for carbon against the sounds for helium velocity. Do a bit of close listening and see what you determine...

I noticed that C5 was quite "noisy" in comparison to C4 and C6. The research group used traditional data analysis methods to investigate the C6/C4 ratio, and it turns out that it is indeed a much better indicator of the temperature of the solar wind source regions. The paper "Carbon ionization states as a diagnostic of the solar wind" provided deeper insight into this finding.
Alexander says his process is reminiscent of "old-school recording," when audio data was stored on magnetic tape as variations in magnetic intensity. Magnetometers on satellites similarly pick up changes in magnetic intensity, though they don't use tape.

"If we listen to the data stream," Alexander said, "we're essentially able to turn these satellites into fancy recording studios, and the sun becomes our performer, with it's magnetic field lines swirling through the solar system."

No comments:

Post a Comment