Truth in the Data: a chat with Angèle Syty

Welcome to Growing Worlds, the podcast where we explore the science of exoplanets through conversations with early career researchers on the front line of discovery. I'm Lorenzo.

And I'm Jenny. In each episode, we meet someone who helps us understand planets, stars, and the tools we use to study them.

In today's episode we turn our attention to what goes behind the scene of exoplanet atmosphere observations, the data processing and instrumentation side of things, because achieving that exquisite precision needed for a mission like Ariel means taming systematics, designing pipelines and building algorithms before the first photon even arrives.

help us navigate that essential but often hidden side of exoplanet science, we're joined by Angèle Citi, a second year PhD student at the Institut d'Astrophysique de Paris. Angèle, thank you for joining us.

Thank you for the invitation.

Angele is working on the data processing pipelines for aerial missions. Her focus is on simulating how the aerial detectors will behave in real life, from jitter to bad pixels to detecting non-realities and ensuring that when we analyze a panoramic spectrum, we truly understand and correct for what the instrument, spacecraft, might have done to the data. Her work contributes directly to aerial to guarantee that the stability and the noise level meets the mission goals and requirements.

So, know, no pressure.

So, Angelle, let's start with your path. How did you get into working on the pipelines and the instrumentation for exoplanets, rather than say, know, like modeling the atmospheres? Was there some kind of experience you had? What drew you into this? Because it's kind of quite niche, really.

So I think it started when I did my first year of master internship. I went to Sapienza in Rome to work. Yes, it was a very

lucky!

nice experience also because of the city. But I started to work on data processing because I was working with a team of people doing data processing for the Arial space mission. And so... So, yes, this is the point where I started to get more involved in this.

Yeah. Were you, per chance, did you work with Enzo? Yes.

Yes, I worked with Enzo Vascare, yes.

Look at this. This is the thing that connects us all. We've all worked with Enzo at some point. So when I finished my masters and just before I started my PhD and then the start of my PhD, I worked with Enzo on exoplanet data, not aerial data, but trans and exoplanets.

Okay.

And that's how me and Enzo met.

Yeah, was starting my thesis when she was finishing her PhD basically and then I met Enso, I did my thesis with Enso and my PhD with Enso and I think that when I was finishing my PhD, you Angel end up with... in meeting Enso.

Yes, I think so, yes. And

Ugh.

Enzo is so impressive with data processing that it really motivated

them. Yeah.

me to

We will make sure that you listen to this episode.

The man is...yeah.

Yes, definitely.

So did you have like an interest in astronomy or exoplanets from childhood or is it something you kind of came into later in life?

Yeah, I started to get interested into astronomy basically by just looking at the stars. When I was a child I got a telescope and so yeah, I was in an association to look at the planets of the solar system and some

Mmm.

galaxies and things like that and so I started to be a bit passionate about it, to read some general science, watch some science content. And so I wanted to do my studies in this direction.

Yeah. Yeah. And you've made it happen, which is amazing.

Yeah, yes definitely.

So you made it. You are now in the in the XFID field and you contributing to a major space mission. Which, I mean, I think is something for the little kid that was looking at the sky through the telescope, isn't

Mm-hmm.

You

it? Can you bring us with you in the importance of your work? Because when we think about observing this...

planets, Like transmissive spectroscopy, we always think about light curves, spectrum from these atmosphere of planet, but what you are dealing with are the systematics of the instrument that actually happen even before all these astronomical signals are So can you walk us through what are all these systematics and all these things that you are defending us against?

Yes, so basically, Systematics are all the signals that do not come from the targets that we want to observe. So it's signals that are introducing systematically bias to the data. And it's very important to be able to correct it for different reasons. First, we are dealing with a with simulated data with Ariel because the mission will be launched in a few years. So with simulation data, we are able to know the ground truth because we put it as an input in our simulation. So basically we do our simulation, we extract the results from the simulation and we can compare it to the true values that we wanted to obtain.

But with real data, we cannot do this because we do not know the ground truth. So it's very important to trust our methods. And to do so, we need to be able to understand the different systematics that can be included in our observations by, for example, the instrument, the detector, or other astrophysical reasons.

because if we do not understand those systematics, we cannot trust the data we are getting.

Can you tell us a bit more about what those systematics are and like where they come from? Is it like the materials that the telescope is made of or the particular sensors or the electronics or is it what sort of things are they?

So there are a lot of different kinds of systematics. For example, in my work, I work with a jitter. So it's the jitter of the line of sight of the telescope. Basically, the telescope is not always still during the observation. You look at the target, but the telescope will be moving a little bit.

even in space.

Yes, even in space. For example, actually.

Why is that?

So.

There are, in the IELTS spacecraft, there are some reaction wheels, for example, that are supposed to maintain the telescope still during the observation. And it works with wheels that are spinning, and sometimes they reach resonance frequency. And so when they reach this frequency, the telescope can start to vibrate, and this vibration will distort the light that is reaching the focal plane of the telescope. And so it will introduce...

And guess there's no getting around that, right? Because you have to make it out of all these, like you have to have the reaction wheels and they will

Yes.

have a given resonant frequency no matter what you make them out of. Every material have some kind of frequency. So it's kind of, you just can't get away from it.

Yes, so basically there is of course some people that are focusing on the instrument design that make sure that the effect is as small as possible but the effect cannot be totally removed so we have to

Yeah.

to post-process it.

Can you explain

and

us how this vibration causes systematic signal?

Yes, so basically you are looking at a star in the sky and the photons of the star are reaching the focal plane of your detector. So the photons are reaching the pixels of your detector and when the spacecraft is jittering, the photons will reach different parts of your detectors.

So it can be different pixels or even in a single pixel, it will reach a different part of the pixel. And since the detector is not perfectly homogeneous, it will introduce photometric variations.

and yeah, when you're looking for tiny signals, those variations, I guess, can actually be pretty significant.

Yes, actually when I started to work on JITR in my first year of master internship, I was thinking why are we doing this because it's so tiny, but actually we are looking at exoplanet atmosphere spectrum, so it's a tiny variation of an already very faint signal, so we have to make sure that the systematics are well controlled.

Yeah. And so aside from the jitter, are there any others which are quite important?

So I'm especially looking to how different systematics interact with jitter, for example, bad pixels or

Okay.

the pixels response nonlinearities, because in the end, all these effects on pixels behavior are introducing more inhomogeneities to the detector. So they will increase the effect of of the photometric variations introduced by Jeter.

pixels always interests me because it's a very strange concept that we make these detectors, right, with a bunch of pixels but not all of them work.

Yep. Yep.

But why? Why, why don't like, because why can't we just like replace them? Like why are they, I mean you understand that once it's in space and it's being bombarded by like cosmic rays and things like that it's gonna interfere with the electronics and they, the bad pixels will appear but even like your launch spacecraft and they have bad pixels. Like how how do we allow this or is it just an inevitability?

So basically we are still trying before launch to choose the area of the detector where there are almost no bad pixels or not on the important part of the detector. But yes, it's inevitable because it's very complex electronics and the pixels have to be mounted on the detector and linked to the electronics and in the process a lot of different things can happen.

it's very difficult to produce all these kind of detectors, right?

Yeah.

I Know this is the thing is like I'm criticizing it and have I ever tried to make a detector? No, am I ever gonna try and make one? No, like I

Yeah.

you

am NOT the person like I should have no opinion on this, but it it fascinates me

I always remember a lecture that I, when I was a student, I was an undergrad and it was a class about electronic lab. And I learned a lesson, I think it's the only thing I remember out of that class. I hope the professor is not listening to me now. Yes, yeah, I remember something.

Hey look, at least you remember something. That's the thing. It was many years ago when you remember something.

And it told us, we talk about amplifiers, right? And if think about it, a pixel is kind of an amplifier, you're saving a signal and it's producing something out of that signal, right? And humankind has been building amplifiers for ages. Let's just think about music, right? And it's impossible to produce two amplifiers absolutely identical. So the way I think about it is every time you listen to music, in a concert is always different because those amplifiers are all different but also

you

if you are observing through detectors, right? Every detector is always different so you never know what to get until you get it.

Yeah, and so while we're on the subject of not knowing what you get until you're gonna get it, I mean you try and circumvent that, right, by doing these simulations, trying to know what we're gonna get before we actually detect these signals. So how do you actually simulate things like the bad pixels, the jitter, like how do you know what the issues are gonna be before we know what the issues are gonna be? Because of course we don't know.

until it starts observing. So how do you simulate that?

So I think for this we have to benefit from the experience we get from previous space missions. For example, the Ariel Spectrometer Detector I'm working on is to some extent close to the detectors of James Webb Telescope. So sometimes it's possible to just look at what is happening on the James Webb detectors and say, okay, we can start by thinking that it would probably be the same for Ariel. And then there is also work on the ground before flight to do some calibration and to be able to characterize our detectors to do some tests.

Okay.

And these calibrations can be included in the simulations at a later step to make them more realistic.

Right, so I guess in the lab you can sort of identify the existing bad pixels, right? And you can like look at how pixels are responding to light and you can spin those reaction wheels and see when they start vibrating and things like that. So, okay, yeah, no, that makes sense, yeah.

And so you're working with this at simulation level and you said at a later stage we're going to have the calibration product as well. But why is it so important to do this job of checking all these systematics and preparing for them now, which is, as you said, a few years from the launch of the mission.

And so for Ariel, we plan to have an official pipeline that is currently being designed. So few years before the launch of the mission, we have to be ready before the mission is launched so that when it will be the case, we will be able to interpret data right from the beginning.

To do so, we work on simulated data because we know what we input in our simulation. So we can just process, do the data processing the same ways that we will do it with real data and then look at the results and compare it to what we put as an input of our simulations so that we know if...

if our data processing pipelines are biased or if they are giving a result that is resilient to all kinds of systematics.

Can you tell us what bias mean in this case?

Okay, so a bias is a...

Basically you are doing a measurement and you are systematically away from the true value. You are always shifted from the true value. For example, if you have a kitchen scale and you forget to do the zero on the kitchen scale, for example, your scale will always give you five grams too much.

for your ingredients. This is a bias, it says. It's a measurement that is always shifted from the true value.

And it is so important to make sure that you know about those before you start actually getting the data. There was a story a few months ago, I think it was about tests. So the Transcend exoplanet survey satellite, either stars or exoplanets. And there was like a consistent error that hadn't been accounted for in terms of masses. So they've had to like revise a bunch of stuff.

Ouch.

And the kind of outcome of it is that like some planets that we thought were say super earths are actually more like sub-Neptunes and it's just kind of like shifted all the boundaries. So it's definitely, you want those at the beginning, you don't have to go back and find them afterwards.

Yeah, yeah. Yeah, because

Yes.

other people are gonna grab your data after you analyze them and use them for their own research. Obviously, that is how science works. So if your data are biased, you're giving people, well, ouch, problems, not data.

Yeah, yeah. And we definitely don't want to have... do I want introduce more ouches? There's enough ouches in science in terms of trying to get funding and things like that so we don't need more ouches.

So your work is really kind of an interface between engineering and science, right?

Yes, I'm working actually a lot with engineers. For example, at CEA in France, it's a lab where they are doing some measurements on airs, so the spectrometer of Ariel, on detectors to characterize the pixels of the detector. And so it's very important to work with them to...

First, be able to help them with simulations because in calibration work they also need simulations. But also for simulation to use their knowledge on the detectors to produce more realistic simulations.

question for you. This is a question that kind of cropped up at the last meeting of the Astronomical Society that I'm president of and the question was, we had our amazing talk and one of the questions was kind of about whether engineers who sort of build the instruments or they build detectors, anything like that, do they actually then care and get excited about the science results that come out?

from their work or actually is their love and joy from the construction of the instrument and then once they've built it they're like, don't care, onto the next. Like what's your perspective? Because you're in with these engineers all the time.

Yeah, I think they are excited about the results, but they are also very, very much excited about the detectors and when I see

Mmm.

them in the white rooms doing some experiments, I see the joy in their eyes.

So what does a typical day in the life of a PhD student working on critical data pipelines actually look like? Because you've mentioned that you're in the engineering department, but you've got to be at computers at some point. Do you have days where you're with loads of people, days on your own? Are you around the world? What is it like?

Yeah, so it's a lot of working in front of computer, but I think for a lot of astrophysicists today, but I like it. So it's not a bad thing for me. But yes, basically I'm doing a lot of simulations on the cluster. And then I introduce different kinds of systematics in my simulations. And then I do another part of a pipeline process, so again more computer science to get the final results. But I think that working on a space mission, we are lucky to be able to travel quite a lot. So there are some consortium meetings to make sure that people are working together and to share our progress and to talk about what are the next steps of our work. So it's a lot of also discovering new people and traveling.

What's your favourite place that you've been to?

This is a hard one. think

You

Vancouver for the

Uhhhh

Europe conference on machine learning.

nice very nice i haven't made it

Would you tell us how do you get to Vancouver from machine learning if you working on a space mission?

Yes, so basically I was involved in the Arial data challenge. So in 2024 and 2025, it was two data challenges that were similar both years. So it's data challenges that are open to people that are not working in astrophysics, but are usually machine learning or data processing experts. and

so people listening to this could get involved if those ones run in the future, right?

Yes, definitely. Definitely it's open science. So it was very interesting actually to see how many people got really enthusiastic about this project. There were a lot of participants and so there were a conference to present the results in Vancouver.

Mmm. And you've got to go out there.

Yes.

That's fantastic, Anshel. And now before we close, have our signature question, which I promise is a fan house, at least for us.

Yeah, yeah, we seem to have a lot of fun with this question and we ask people and they get very stressed, but we it's honestly it's just for fun

So if you had to choose one celestial body, a star, a planet, a space mission or something else entirely, that you feel like it represents you or your work, what would it be and why?

Okay, so this was really a tricky question. Yeah, I spent a lot of time thinking about

Yeah!

it. so I thought about different exoplanets and then I was in Paris. I don't know if it's everywhere in Europe, but there is in France a science day and this day all the science laboratories are open to the general public. And so I was presenting the Solar System and Trappist One system in the street and people, families were coming to hear about it.

Nice.

I love that.

Yeah, it was really nice. And so each time I was presenting Trappist One E, people were really, really getting very enthusiastic about it because Trappist One E is a Earth-sized planet in the habitable zone of TRAPPIST-1. And I was thinking, I don't know if it's such an interesting planet, it's very likely to be a bare rock planet with no atmosphere. I thought that, I think as a scientist, I would like to be represented by TRAPPIST-1e.

Ooh.

Because I think it's important in the end to be able to convince people that exoplanetary science is important and it's cool. yeah, I think it just made me remember that we are doing it also for community and for the outreach of...

Yeah.

of the science so I'd like to be more like Trappist 1e to some extent.

I love that. That's a brilliant answer. I

That's really cool, yeah.

think that's, yeah, one of the best answers that we've had. Like, yeah, I love that. That's a great answer. So, Angèle,

Thank you.

thank you for illuminating the data processing foundations that allow us to trust what we see in exoplanet atmospheres.

And thank you everyone for listening. Remember that every extraplanar spectrum we interpret rests on a foundation of instrument calibration and pipeline work. So join us next time on Growing Worlds, how we explore another corner of the universe and the people making sense of it.