Data reduction method for OPTICAM multiband time series of transiting exoplanets

This paper presents a data reduction methodology for the OPTICAM instrument that utilizes a 3×3-pixel median filter to effectively mitigate unpredictable warm pixels in sCMOS detectors, thereby improving photometric precision and Bayesian evidence for multiband transiting exoplanet light curves.

The Gist

Here is an explanation of the paper, translated into everyday language with some creative analogies.

The Big Picture: Catching a Cosmic Blink

Imagine you are trying to watch a tiny firefly (an exoplanet) fly in front of a giant, bright flashlight (a star). When the firefly passes in front, it dims the light just a tiny bit. By measuring that dimming, astronomers can learn about the firefly's size, speed, and atmosphere.

This paper is about a specific camera called OPTICAM, mounted on a telescope in Mexico, and a specific problem it has: it gets "hot" and starts glitching.

The Problem: The "Hot Pixel" Ghosts

The camera uses a special type of sensor (sCMOS) that is great at taking pictures very quickly. However, when the camera takes a picture for more than 10 seconds, some pixels (the tiny dots that make up the image) get "overheated" or "glitchy."

Think of these warm pixels like bad apples in a basket of fruit.

• Normal pixels are good apples; they represent the actual star light.
• Warm pixels are rotten apples. They suddenly light up with bright, random static.
• The Glitch: Unlike a normal bad apple that stays bad, these warm pixels are unpredictable. One second they are bright, the next they are dark. They jump around like ghosts.

Because they jump around, you can't just take a "before" picture of the bad apples and subtract them out later. They change too fast. This creates a "fuzzy" or "noisy" light curve, making it hard to see the tiny dimming caused by the planet.

The Experiment: Trying to Clean the Mess

The researchers wanted to find the best way to clean up these "rotten apples" without accidentally throwing away the "good apples" (the star's light). They tested six different cleaning methods on data from a planet called TOI-7149 b.

Here are the methods they tried, using a kitchen analogy:

1. The Standard Wash (Standard Reduction): Just washing the fruit with water (subtracting dark frames).
   • Result: The bad apples are still there. The noise remains high.
2. The Blender (Gaussian Convolution): Smoothing out the image by blending neighboring pixels together, like putting the fruit in a blender.
   • Result: It gets rid of the bright spots, but it also mushes up the star's shape. It's like turning a crisp apple slice into applesauce. You lose the detail you need.
3. The Sieve (Median Filters): This was the winner. Imagine looking at a 3x3 grid of pixels (a small window). If one pixel is a "rotten apple" (a huge outlier), you ignore it and replace it with the middle value of the other 8 pixels in the grid.
   • Result: The bad apple is removed, but the good apple stays crisp.

The Winner: The 3x3 Sieve

The researchers found that a 3x3 Median Filter was the perfect tool.

• It's small enough that it doesn't blur the star.
• It's big enough to catch the glitchy pixels.
• It acts like a smart bouncer at a club: "Hey, you're acting weird (too bright or too dark compared to your neighbors), so you're out. We'll replace you with the average behavior of the group."

When they used this method, the "noise" in the data dropped significantly, and the signal of the planet became much clearer.

The Solution: A New Recipe Book

The authors didn't just find a solution; they built a pipeline (a step-by-step recipe) so other astronomers can use it too.

• They created a free computer program called PROFE (Python modules).
• This program automatically applies the "3x3 Sieve" to the raw photos before they are analyzed.
• It then passes the clean data to a popular tool called AstroImageJ to do the final math.

Why Does This Matter?

Before this paper, astronomers using this specific camera might have thrown away data or gotten inaccurate results because of the "hot pixels." Now, they have a reliable way to fix the data.

In short: The camera has a glitchy sensor that creates random static. The authors figured out that using a specific mathematical "sieve" (a 3x3 median filter) cleans up the static without ruining the picture, allowing them to see exoplanets much more clearly. They packaged this fix into a free tool so everyone can use it.

Technical Summary

Here is a detailed technical summary of the paper "Data reduction method for OPTICAM multiband time series of transiting exoplanets."

1. Problem Statement

The paper addresses a specific instrumental challenge affecting high-cadence photometric observations using the OPTICAM instrument mounted on the 2.1m telescope at the Observatorio Astronómico Nacional en la Sierra de San Pedro Mártir (OAN-SPM).

• The Issue: OPTICAM utilizes three Andor Zyla 4.3-Plus sCMOS detectors. These detectors generate a significant number of "warm pixels" (pixels with elevated signal or high variance) during exposures longer than 10 seconds.
• Why Standard Methods Fail: Unlike standard hot pixels, these warm pixels exhibit unpredictable, frame-to-frame variability (changing mean counts and standard deviation). Consequently, they cannot be removed by standard calibration techniques like dark-frame subtraction or flat-field division, as the dark frames do not perfectly match the warm pixel behavior in science frames.
• Impact: These warm pixels introduce red noise (time-correlated noise) and white noise into light curves, degrading photometric precision and potentially biasing the retrieval of exoplanet parameters (e.g., transit depth, radius ratio).

2. Methodology

The authors evaluated six different data reduction strategies using transit observations of the exoplanet TOI-7149 b (an M-dwarf host, $V \approx 18$ mag) observed in $g'$, $r'$, and $i'$ bands.

Pre-processing Approaches Tested:

1. Standard Reduction (st): Dark subtraction and flat-field division only (Baseline).
2. Gaussian Convolution (g1, g3): Smoothing raw images with Gaussian kernels ($\sigma=1$ and $\sigma=3$ pixels) before standard reduction.
3. Median Filtering (w3, w5, w7): Replacing pixel values with the median of a surrounding window ($3\times3$,$5\times5$, and$7\times7$ pixels) before standard reduction.

Analysis Workflow:

• Data Processing: All pre-processing was applied to both science and calibration frames.
• Photometry: Differential aperture photometry was performed using AstroImageJ (AIJ) with a consistent set of comparison stars across all methods to isolate the effect of pre-processing.
• Model Fitting: Light curves were fitted simultaneously across all three bands using the juliet package. The model included a transit component and Gaussian Processes (GPs) to detrend baseline variability and correlated noise.
• Metrics: Performance was evaluated using:
  • Bayesian Evidence ($\ln Z$): To determine the statistically preferred model.
  • RMS and RMS$_{10\text{min}}$: To measure total scatter and correlated (red) noise.
  • Radial Profiles: To assess signal degradation (stellar peak counts).

3. Key Contributions

• Characterization of sCMOS Warm Pixels: The authors provided a detailed behavioral analysis of OPTICAM warm pixels, categorizing them into four types (A: bias-dominated, B: stepwise jumps, C: salt events, D: pepper events). They quantified that warm pixels can affect up to 60% of the detector area in long exposures, rendering static pixel masks ineffective.
• Optimization of Pre-processing: The study systematically demonstrated that Median Filtering is superior to Gaussian smoothing for this specific noise profile. Gaussian kernels were found to smear warm pixel spikes into adjacent pixels, creating "fake" faint stars that contaminate photometric apertures.
• Development of a Hybrid Pipeline: The authors introduced PROFE, a new open-source Python module suite, designed to integrate with AstroImageJ. This pipeline automates the application of the optimal pre-processing method (3$\times$3 median filter) and generates diagnostic products.

4. Results

• Best Method: The 3$\times$3 pixel median filter (w3) applied to raw science and calibration frames prior to standard reduction yielded the best results.
  • Bayesian Evidence: The w3 method achieved the highest $\ln Z$ (approx. 5004.96 for $g'$), significantly outperforming the standard reduction ($\ln Z \approx 4464.34$) and Gaussian methods. A difference $\Delta \ln Z > 5$ indicates strong statistical preference; w3 exceeded this by a large margin.
  • Noise Reduction: The w3 method produced the lowest RMS$_{10\text{min}}$ (correlated noise) across all bands, bringing the time-averaging curves closest to the white-noise expectation.
  • Signal Integrity: Unlike the $\sigma=3$ Gaussian kernel, which reduced stellar peak counts by ~13%, the 3$\times$3 median filter only reduced peak counts by ~9% while effectively removing outliers without creating artifacts.
• Band Dependence: The improvement was most critical in the $g'$ band, where the target star's signal was closest to the dark current level (lower Signal-to-Noise Ratio). In redder bands ($r'$, $i'$), where stellar counts were higher, the standard reduction performed better, but the w3 method remained the most robust overall.
• Parameter Retrieval: The transit model fitted with w3-preprocessed data provided consistent planetary parameters ($R_p/R_*$, $a/R_*$, inclination) compared to previous studies, with tighter constraints on the scaled semi-major axis due to improved baseline normalization.

5. Significance

• Enabling Ground-Based Multiband Transit Science: This work validates the use of OAN-SPM 2.1m+OPTICAM for high-precision exoplanet characterization, a capability previously limited by detector noise.
• Generalizable Solution for sCMOS: The findings offer a critical solution for the growing field of sCMOS-based astronomy. As sCMOS detectors replace CCDs due to lower cost and faster readout, the "warm pixel" issue is becoming common. The paper demonstrates that local median filtering is a robust, non-destructive method to mitigate frame-to-frame variable noise without the artifacts introduced by Gaussian smoothing.
• Open-Source Tooling: The release of the PROFE pipeline provides the community with a standardized, reproducible workflow for reducing OPTICAM data, ensuring homogeneity in future exoplanet studies using this instrument.

In conclusion, the paper establishes that applying a 3$\times$3 median filter to raw frames is the optimal strategy for correcting warm-pixel noise in OPTICAM data, significantly improving photometric precision and model evidence for transiting exoplanet observations.