…And We’re Back!! Sunspotter Round 2

I suppose Friday the 13th is as good a day as any to launch a Citizen Science project.

For those of you who helped us classify the previous data set: Welcome back!
And hello to all of our new Sunspotters!

avatar_sunspotter

If you haven’t a notion of what Sunspotter is all about, check out our science section. In short, our goal is to determine the complexity of sunspot groups. It is well known (to solar physicists) that more complicated looking sunspot groups produce more solar flares than simple looking ones. But so far, scientists have not found a good way to quantify sunspot group complexity. This is not a task easily accomplished by a computer. Humans, on the other hand, can easily point to the more complex in a pair of objects, ideas, images, and so on.

I’m pretty sure you have an idea of which is the more complex: a graduate text on quantum mechanics, or an Italian cookbook?
On the other hand, it would not be straight-forward for a computer to make that choice. The same is true with sunspot groups.

In round one (lasting only a month!), ~1,600 volunteers helped us to rank ~13,000 images of sunspot groups by choosing the more complex one in ~300,000 pairs of images. This has allowed us to quantify ‘true’ sunspot group complexity for the first time! Now that we have a handle on how to give complexity a number, we want to determine how the complexity of a sunspot group changes over time.

A less complex sunspot group.

A less complex sunspot group.

To do this we have automatically detected thousands of sunspot groups and tracked them over time. Each sunspot group has been detected about 15 times per  day. Some of these sunspot groups were included in the previous dataset, but now they are being detected in a different way. That means that you will see a number of similar-looking images- but don’t worry if you can’t tell which sunspot group is more complex, just do your best! When graphing the complexity of a sunspot group over time, we are hoping to see clear jumps in the data when the sunspot group became more complex and we expect this to be followed by the occurrence of solar flares.

A more complex sunspot group

A more complex sunspot group

There are a few ‘biases’ that we could not easily correct for with the previous dataset, that we hope to get a handle on this time. For instance, depending on a sunspot group’s position, it will look more squished as it nears the edge of the Sun. As mentioned in an earlier post, we are now using a projection technique to ‘de-squish’ the sunspot groups. Also, it is likely that the most complex sunspot groups are always the largest. However, humans might also be biased toward thinking bigger things are more complex, even when they are not. So, to help reduce this bias, we have ‘de-scaled’ the images so that all of the sunspot groups, big or small, will appear roughly the same size on your screen.

We think you will find it much easier to focus on complexity with this new data set.

Thanks for listening, and happy classifying!!

Advertisements

About Dr. Paul A. Higgins

I am a postdoctoral research fellow in the Astrophysics Research Group at Trinity College Dublin in Ireland. Currently I am a visiting researcher at LMSAL in Palo Alto, CA. I am investigating the causes of solar eruptions. To do this I use image processing and data mining techniques to study the evolution of sunspot groups as they are born, cross the solar disk, producing flares and coronal mass ejections, and then quietly decay and fade away.

3 responses to “…And We’re Back!! Sunspotter Round 2”

  1. Art says :

    This round is definitely tougher and it stakes me longer to classify things.

    Will these classifications help you to write a program which determines the complexity of even larger amount of sunspots much easier?

    • Dr. Paul A. Higgins says :

      Hi Art,

      Well the main science goal here is to determine the complexity evolution of sunspot groups that produce large eruptions, compared to those that do not.

      However, we also intend to test a machine learning algorithm on this dataset to see if we can get a reliable automated complexity measure, as you alluded to. But, at this point the results of that test would be a proof of concept, since we would have to retrain the algorithm again in the third phase of Sunspotter (requiring a new set of people-powered classifications), as the new data coming from Solar Dynamics Observatory/Helioseismic Imager are at a much higher resolution than the current data from Solar and Heliospheric Observatory/Michelson Doppler Imager.

      In other words we will need to train the automated algorithm on similar data to that being used, going forward.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: