Clustering the PlanetFour results
Our beloved PlanetFour citizen scientists have created a wealth of data that we are currently digging through. Each PlanetFour image tile is currently being retired after 30 randomly selected citizens pressed the ‘Submit’ button on it. Now, we obviously have to create software to analyze the millions of responses we have collected from the citizen scientists, and sometimes objects in the image are close to each other, just like in the lower right corner of Figure 1.
Figure 1: Original HiRISE cutout tile that is being shown to 30 random PlanetFour citizen scientists.
And, naturally, everybody’s response to what can be seen in this HiRISE image is slightly different, but fret not: this is what we want! Because the “wisdom of the crowd effect” entails that the mean value of many answers are very very close to the real answer. See Figure 2 below for an example of the markings we have received.
Figure 2: Original markings of P4 Citizen scientists.
Note the amount of markings in the lower right, covering both individual fans that are visible in Figure 1. It is understandable that the software analyzing these markings needs to be able to distinguish what a marking was for, what visual object in the image was meant to be marked by the individual Citizen scientists. And I admit, looking at this kind of overwhelming data, I was a bit skeptical that it can be done. Which would still be fine, because one of our main goals is wind directions to be determined and as long as every subframe results in the indication of a wind direction, we have learned A LOT! But if we can disentangle these markings to show us individual fans, we could even learn more: We can count the amount of activity per image more precisely, to learn how ‘active’ this area is. And we even can learn about changes of wind direction happening, if at the same source of activity two different wind directions can be distinguished. For that, we need to be able to separate these markings as good as possible.
And we are very glad to tell you that that indeed seems possible, using modern data analysis techniques called “clustering” that looks at relationships between data points and how they can be combined into more meaningful statements. Specifically, we are using the so called “DBSCAN” clustering algorithm (LINK), which allows us to choose the number of markings required to be defined a cluster family and the maximum of distance allowed for a different marking before being ‘rejected’ from that cluster family. Once the cluster members have been determined, simple mean values of all marking parameters are taken to determine the resulting marking, and Figure 3 shows the results of that.
Figure 3: Clustered markings for P4 tile ZG7
Just look at how beautifully the clustering has merged all the markings into results have combined all the markings into data that very precisely resembles what can be seen in the original data! The two fans in the lower right have been identified with stunning precision!
For an even more impressive display of this, have a look at the animated GIF below that allows you to track the visible fans, how they are being marked and how these markings are combined in a very precise representation of the object on the ground. It’s marvelous and I’m simply blown away by the quality of the data that we have received and how well this works!
Figure 4: Animated GIF for the clustering of the markings of APF0000zg7
This is not meant to say though that all is peachy and we can sit back and push some buttons to get these nice results. Sometimes they don’t look as nice as these, and we need to carefully balance the amount of work we invest into fixing those because we need to get the publication out into the world, so that all the Citizen scientists can see the fruit of their labor! And sometimes it’s not even clear to us if what we see is a fan or a blotch, but that distinction is of course only a mental help for the fact if there was wind blowing at the time of a CO2 gas eruption or not. So we have some ideas how to deal with those situations and that is one of the final things we are working on before submitting the paper. We are very close so please stay tuned and keep submitting these kind of stunningly precise markings!
For your viewing pleasure I finish with another example of how nicely the clustering algorithm works to create final markings for a PlanetFour image:
Figure 5: Animation for the clustered markings process of P4 image ZMJ
2 Years On from BBC Stargazing Live
In the UK, tonight starts the latest installment of BBC Stargazing Live. Three nights of live astronomy television hosted by Professor Brian Cox and Dara Ó Briain. Just over two years ago, we were preparing for the launch of the Planet Four live on television as part of Stargazing Live. Professor Chris Lintott from the BBC’s Sky at Night and PI of the Zooniverse went out on the program broadcast live from Jodrell Bank and introduced to the world Planet Four, asking for viewers help to map the seasonal fan and blotches visible in images of the Martian South Pole taken by the HiRISE camera.
For the past 9 years, the HiRISE camera aboard the Mars Reconnaissance Orbiter has been capturing stunning and dynamic images of the defrosting South Pole. During this time, carbon dioxide geysers loft dust and dirt through cracks in a thawing carbon dioxide ice sheet to the surface where it is believed that surface winds subsequently sculpt the material into dark fans observed from orbit. 30% of Mars’ atmosphere condenses out to form this ice sheet. Understanding the direction, frequency, and appearance of these fans (a proxy for the geysers) and how these properties are impacted by varying factors we can better understand the Martian climate and how it differs from Earth.
This is a project that we truly couldn’t do with out the help of citizen scientists and BBC Stargazing Live. Hundreds of thousands of fans are visible in HiRISE observations, but for years this rich dataset was not tapped to its full potential. Automated computer algorithms have not been able to accurately identify and outline individual fans in the HiRISE images, but a human being intuitively can distinguish and outline these features. And thus Planet Four was born.
I can remember launch day like it was yesterday, waiting on the Talk Discussion tool for the flood of volunteers to start posting questions and sharing their thoughts and ideas about the images they were seeing. I and the rest of the Planet Four team anxiously waiting at our keyboards could tell immediately when the Planet Four segment aired. The response from Stargazing was incredible and overwhelming. Each night, the Zooniverse servers struggled to keep up serving images of Mars as the number of people on the site continued to rise. Thanks to the Stargazing Live viewers we were able to complete nearly all of the Season 2 and Season 3 HiIRSE monitoring campaign images.
So where are we now? Thanks to help of Planet Four volunteers including Stargazing Live viewers, we’ve made great progress since January 8, 2013. Over 4.6 million blotches and 3.8 million fans have been drawn to date (the great majority of these markings were made during BBC Stargazing Live). In the past two years, Planet Four has captured the equivalent of a full year of non-stop human attention (a single person working non-stop/no breaks for an entire year!). The science team has been working to create a software pipeline to combine the multiple classifications to identify fans and blotches. We have also been working to create an expert dataset classified by the science team for a very small subset of Planet Four images to compare to the volunteer classifications to show that Planet Four citizen scientists are very efficient and effective at detecting the seasonal fans and blotches in the HiRISE images.
I’m pleased to say the science team is very close to submitting the project’s first science paper to a journal before the end of the year (we’re aiming for end of Spring/Summer). We have more than half of the paper draft currently written. One of the last lines of the paper is: ‘We thank all those involved in BBC Stargazing Live 2013.’ This is just the beginning. With this paper, we’ll be able to eventually produce the largest areal coverage wind measurement of the Martian surface to date spanning two Martian years. These maps will reveal how the fan properties and numbers change from Martian year to year and location to location on the South Pole. We also have 3 more Martian seasons of HiRISE data that we’ve just barely scratched the surface of. The majority of these images have yet to be classified, including right before a Martian dust storm, so we can see how the dust storm has impacted the Martian climate and how long its effects last in the atmosphere and the ice sheet by looking at the fans and geysers that are created in the seasons before and after the storm
This year the Zooniverse has something new up their sleeve that will be revealed during the broadcast, but while you’re waiting for the return of BBC Stargazing tonight, if you can spare a minute or two , we could use your continued help mapping the seasonal fans visible in the HiRISE images. There is so much of the South Pole (and 3 additional years of data to get through) that we have yet to study and explore! Classify a HiRISE tile or two at http://www.planetfour.org
Geometry of HiRISE observations is on Talk!
Good news: our wonderful development team has added new feature that many of our volunteers have asked for! Now you can see north azimuth, sub-solar azimuth, phase angle, and emission angle on the Talk pages directly. You can see an example here. These angles give you information about how HiRISE took the image and where the Sun was at that moment.
To understand what those angles are, here is an illustration for you:
You see how the MRO spacecraft flies over the surface while HiRISE makes an image. The Sun illuminates the surface .
Consider a point on the martian surface P.
Emission angle: HiRISE does not necessarily look at point P straight down, i.e. the line connecting point P and HiRISE has some deviation from vertical line – it is noted as angle e on the sketch. This is emission angle. It tells you much we tilted spacecraft to the side to make the image.
Phase angle: Because all the images you see in our project is from polar areas, the Sun is often low in the sky when HiRISE observes. To get an idea on how low, we use phase angle – it is the angle between the line from Sun to the point P and line from point P to the HiRISE. It is noted φ in the sketch. The larger phase angle is, the lower the Sun in the sky, the longer are the shadows on the surface.
Sub-solar azimuth: To understand what is the direction towards the Sun in the frame of HiRISE image, we use sub-solar azimuth. In any frame that you see on our project it is an angle between horizontal line from the center of the frame towards right and the Sun direction. It is counted clock-wise. The notation for it in the sketch is a.
North azimuth: The orbit of MRO spacecraft defines orientations of HiRISE images. North azimuth tells us direction to the Martian north pole. In the frame of an image it’s counted same as sub-solar azimuth, i.e. from the horizontal line connecting center of the frame and its right edge in the clock-wise direction.
I hope this helps you enjoy exploring Mars with HiRISE!
Anya
NASA funding for utilizing Planet Four markings
It is really-really tough to get funding to do research. You have to have an idea to do something really new and important, something that will be interesting and useful. You need to gather a team that can do it. Then you have to write a proposal to explain your idea and to convince other scientists that this project is worth pursuing. And you’ll be competing with other projects for the limited budget pot. It was even tougher than usual this year for planetary research at NASA: only 14% of all submitted projects got funding.
But we did!
A little project that utilizes the data from Planet Four will be funded by NASA so that we can compare directions of winds mapped by our citizen scientists (via fans, of course!) to the prediction of martian climate models!
This is so very exciting!
We have a great team here and I am convinced this project will be a great success! Thanks NASA and thanks all of our helpers on planet4!
Anya
Planet Four Blog 