Early results from the DEIMOS dataset: what your classifications are revealing

As we continue collecting measurements on new data, we wanted to share how your Zooniverse classifications are being used and show some early results from the first dataset. This is the same set of spectra used in the sandbox task, so many of you will recognize it (and you can still make more classifications on it)!

This first round of spectra was collected with an instrument called DEIMOS. We intentionally started with this dataset because it already has expert-determined redshifts, which lets us compare your results with known values and see how well we’re doing.

What are we doing with your classifications?

For an in-depth look at how we use a consensus method to average your classifications and use emission line positions to calculate redshift, check out this post on the talk boards diving deeper into our analysis process. In short: we take an average of all of the markings made on each spectrum, identify the features by matching known patterns of emission and absorption lines, and then determine the redshift based on how far the patterns have shifted. Your markings are what help us find those patterns in the first place.

What do our redshift measurements look like?

We first took a look at the results from our beta testing process, which was done with a representative subset of the DEIMOS data.

Each point on this plot is an object for which volunteers found enough spectral features for us to calculate a redshift. The y-axis shows the Redshift Wrangler result, the x-axis shows the expert value from the DEIMOS catalog, and the closer a point is to the diagonal orange line, the more those values match. The colors indicate which emission line was the brightest in that spectrum. The data suggests pretty good agreement for many of our measurements, with some scatter among a few outliers, though the limited number of data points makes it hard to be conclusive in this sample. Some of the questions we've already begun to look into include: How many objects are we missing, and why can’t we calculate redshifts for them? Is there a common cause for some of the outliers? Do classifiers improve over time, and has accuracy improved since beta testing?

When we look at the full DEIMOS dataset, we can tell more clearly how often Redshift Wrangler classifiers measured redshifts that match catalog values determined by experts. Once again, the diagonal line highlights where the Redshift Wrangler result and expert value from the DEIMOS catalog match.

The lower section of the plot shows that less than 8% of the objects differ from the expert value by more than 0.001 in redshift. In other words, over 92% of your measurements closely match the catalog redshifts.

So far, these results show a pretty close agreement between our redshifts and catalog values, at least for the objects that had multiple lines classified in their spectra. In both of these plots, we only show objects where classifiers identified multiple spectral lines. These are the objects where a reliable redshift can be calculated.

Why do we need multiple spectral lines to calculate redshift?

We calculate redshift (z) from emission line locations (λ) using this simple equation:

In order to calculate redshift, we need to identify which lines we’re looking at.

Going back to the review of what we do with your classifications, we match the features in a spectrum with known patterns of spectral lines. Spectral features always show up in the same characteristic patterns, like a fingerprint. Knowing what other features are around them makes a huge difference in identifying the line - if we only have one line, it could be any of them, and we have no way of gathering additional information from that singular datapoint. For example, a single bright line could be Hα, OII, or OIII, and we can’t tell which one it is without additional lines to make the characteristic pattern.

That’s why your classifications are so important and why it’s important to mark as many features in the data as you can find!

In order to match up a pattern of features, we need to map them onto multiple emission and/or absorption lines in a spectrum, and the more the better.

We’re still working on improving our results, and your careful line markings make a huge difference. Thank you so much for helping us explore this dataset, and keep an eye out for our next updates!

---

Redshift Wrangler in the News and media:

NASA Citizen Science: Look Back In Time To the Early Universe!

RIT News: RIT scientists unveil Citizen Science Project to search for distant galaxies

WHEC News10NBC: RIT researchers looking for public’s help to study galaxies

WROC Channel 8: Become a citizen scientist with RIT and their “Redshift Wrangler” project

NASA SEES: Redshift Wranglers: Learning About the History of the Universe

Everyday Spacer: Redshift Wrangler

SciStarter LIVE: Wrangling the Cosmos: Help Decode Galaxy Light with Redshift Wrangler