Identifying and Measuring Satellite Streaks in DECam Images

This paper presents a proof-of-concept workflow using the Hough Transform and SatChecker to detect, identify, and measure the brightness of satellite streaks in archival DECam images, demonstrating the feasibility of characterizing orbital debris impacts on astronomical surveys while highlighting challenges in detecting faint and transient glints.

The Gist

The Sky's New Traffic Jam: How Astronomers Are Learning to Count Satellite Streaks

Imagine you are a photographer trying to take a perfect, long-exposure picture of a starry night. You set your camera up, wait for the perfect moment, and press the shutter. But just as you're capturing the beauty of the universe, a bright, glowing line zips across your photo, ruining the shot.

That's exactly what is happening to modern astronomy. Thanks to companies like SpaceX launching thousands of satellites (like the Starlink internet constellation), the night sky is becoming crowded with "traffic." These satellites move so fast that, in a long-exposure photo, they look like bright streaks or scratches across the image.

This paper is about a team of scientists who decided to stop just complaining about the streaks and start measuring them. They wanted to answer a simple question: How bright are these space cars, and how often do they crash into our photos?

Here is how they did it, broken down into simple steps:

1. The Detective Work: Finding the Streaks

The team used a giant camera called DECam (Dark Energy Camera) sitting on a mountain in Chile. This camera has taken millions of photos of the sky over the last decade.

• The Problem: Looking through millions of photos to find a few satellite streaks is like trying to find a specific needle in a haystack the size of a city.
• The Solution: They used a digital tool called the Hough Transform. Think of this as a super-smart "highlighter" for computers. Instead of looking at the whole picture, the computer scans for straight lines. If it sees a straight line that isn't a star or a galaxy, it flags it as a "suspect" satellite streak.

2. The ID Check: Who is Driving?

Once they found a streak, they needed to know which satellite made it. Was it a Starlink satellite? An old rocket? A navigation satellite?

• The Tool: They used a program called SatChecker.
• The Analogy: Imagine you hear a siren in the distance. You look at a map of all the police cars, ambulances, and fire trucks in the city and their current locations. By matching the time and direction of the siren with the map, you can guess exactly which vehicle passed by.
• How it worked: The scientists fed the time and location of the photo into SatChecker. The program calculated where every known satellite should have been at that exact second. If a predicted satellite path overlapped perfectly with the streak in the photo, they knew, "Aha! That streak is Starlink Satellite #2559!"

3. The Speedometer: Measuring the Brightness

Now that they knew what the streak was, they needed to know how bright it was. This is crucial because brighter streaks ruin more astronomical data.

• The Method: They used two different mathematical recipes to measure the light:
  1. The "Bucket" Method (Aperture Photometry): Imagine pouring the light from the streak into a bucket and weighing it. They counted every pixel of light in the streak and subtracted the background "noise" (the faint glow of the sky).
  2. The "Model" Method (PSF Fitting): This is more like a tailor measuring a person. They created a perfect mathematical model of what a moving satellite should look like (a blurry line) and tried to fit that model over the actual photo to see how well it matched.

What Did They Find?

They tested this workflow on just nine different streaks. Here are the big takeaways:

• It Works: They successfully identified a mix of objects: active internet satellites (Starlink), a GPS satellite, an old science satellite that was turned off, and even a piece of a rocket body.
• Brightness Varies Wildly: Some satellites were blindingly bright (like a car with high beams on), while others were much dimmer. This depends on the satellite's design, how it's angled toward the sun, and what kind of satellite it is.
• The "Glint" Problem: While they could measure the steady streaks, they noted that glints (sudden, brief flashes of light when a satellite's solar panel hits the sun just right) are much harder to catch. These are like a camera flash going off for a split second—very hard to predict and measure.

Why Does This Matter?

Think of this paper as the foundation for a new traffic report.

Before this, astronomers knew satellites were a problem, but they didn't have a systematic way to count them or measure their brightness across different types of satellites. This project proved that we can build a "database" of satellite impacts.

In the future, this workflow can be automated to scan millions of images. This will help astronomers:

1. Predict when the sky will be too crowded to take good photos.
2. Advocate for satellite companies to design dimmer satellites.
3. Clean up the data they already have by mathematically removing the streaks.

In short: The night sky is getting busy, but this team built a toolkit to count the cars, measure their headlights, and figure out how to keep the view of the stars clear for everyone.

Technical Summary

Here is a detailed technical summary of the paper "Identifying and Measuring Satellite Streaks in DECam Images":

1. Problem Statement

The rapid proliferation of Low Earth Orbit (LEO) satellite constellations, particularly SpaceX's Starlink, is increasingly degrading ground-based astronomical observations. These satellites create bright linear streaks in long-exposure images, contaminating data, obscuring scientific targets, and reducing the effective survey area. While the presence of these trails is well-documented, there is a lack of systematic, cataloged data linking specific observed trails to their satellite identities (NORAD ID) and precise brightness measurements. This gap hinders the ability to model the impact of satellite populations on astronomical surveys and to develop effective mitigation strategies.

2. Methodology

The authors developed a proof-of-concept workflow to detect, identify, and measure the surface brightness of satellite streaks in archival images from the Dark Energy Camera (DECam) on the 4-meter Blanco Telescope. The workflow consists of four main stages:

• Data Selection & Retrieval:

  • Utilized the NOIRLab Astro Data Archive to access DECam images.
  • Instead of processing full mosaics, the team retrieved specific CCD extensions containing candidate linear features provided by a pre-existing catalog (Drlica-Wagner).
  • The sample included 9 streaks: 7 Starlink satellites, 1 GPS navigation satellite (NAVSTAR-70), 1 decommissioned science satellite (SORCE), and 1 rocket body (DELTA-2).

• Streak Detection and Alignment (Hough Transform):

  • Used the satmetrics package to apply the Hough Transform for detecting linear features.
  • Preprocessing: Images were thresholded and edge-detected (Canny).
  • Alignment: Detected trails were rotated to a horizontal orientation using an affine transformation (OpenCV) to standardize photometric analysis.
  • Parameter Tuning: While default parameters (threshold 0.075) worked for bright streaks, fainter trails required manual tuning of detection thresholds and blur kernel sizes per image.

• Satellite Identification (SatChecker):

  • Employed the SatChecker tool to propagate Two-Line Element (TLE) orbital data for cataloged objects.
  • Predicted satellite paths were overlaid on the DECam focal plane based on observation time and sky coordinates.
  • Visual matching was performed to correlate observed streaks with predicted passes, confirming the identity of the object (e.g., distinguishing between multiple Starlink satellites in the same shell).

• Photometric Analysis:
  Two distinct methods were used to measure surface brightness (SB) in units of counts/arcsec² and AB mag/arcsec²:

  1. Aperture Photometry: A non-parametric approach. The streak profile was modeled as a 1D Gaussian. Background was estimated from off-streak regions, and total flux was calculated by integrating the signal above the background within a defined aperture ($3\sigma$ width).
  2. PSF-Convolved Trail Fitting: A parametric approach modeling the streak as a moving point source convolved with a Gaussian Point Spread Function (PSF). This method used non-linear least-squares minimization (scipy.optimize.curve_fit) to fit the 2D trail model, accounting for the trail's length, orientation, and PSF width.

3. Key Contributions

• Systematic Workflow: Established a reproducible pipeline combining satmetrics (detection/rotation), SatChecker (identification), and custom photometry routines to process DECam data.
• Dual Photometric Validation: Validated two photometric methods (aperture vs. model fitting) against simulations using the GalSim library. Both methods recovered true surface brightness within ~4%, with the fitting method offering better robustness for faint trails and the aperture method offering faster computation.
• Cataloging: Successfully linked specific DECam exposures to specific satellite objects (including non-Starlink debris) and provided the first set of measured surface brightnesses for this specific sample.
• Open Source: Released the code (reca-streaks) and detection wrappers to facilitate future large-scale studies.

4. Results

• Detection Success: The workflow successfully detected and measured 9 streaks, including objects ranging from active constellations to space debris.
• Brightness Variability:
  • The measured surface brightness spanned over 6 magnitudes.
  • Brightest: NAVSTAR-70 (GPS) at $\sim 18.8$ mag/arcsec².
  • Faintest: Starlink-2559 at $\sim 24.8$ mag/arcsec².
  • This variation reflects differences in object size, altitude, attitude, and illumination geometry.
• Method Consistency: The two photometric methods yielded consistent results, typically agreeing within $\sim 0.2$ magnitudes.
• Limitations Observed:
  • Faint Streaks: Detection of faint trails required significant manual parameter tuning, highlighting a bottleneck for automation.
  • Glints: The current workflow focuses on continuous trails; brief, localized glints (specular reflections) remain difficult to detect and were not addressed.
  • Uncertainties: Formal statistical uncertainties were very small (e.g., $\pm 0.0002$ mag), but the authors note that systematic errors (flat-fielding, background gradients) likely dominate, making realistic uncertainties larger (few millimagnitudes).

5. Significance and Future Work

• Foundation for Large-Scale Studies: This work demonstrates that systematic cataloging of satellite trails with brightness and identity is feasible. It lays the groundwork for uploading this data to the SatHub Satellite Constellation Observation Repository (SCORE) to track population evolution.
• Mitigation Insights: By quantifying the brightness distribution of different object types, astronomers can better model the impact on survey efficiency and develop strategies to avoid or mitigate specific satellite populations.
• Future Directions:
  • Automation: Automating the Hough Transform parameter tuning and the visual matching of SatChecker predictions to enable processing of millions of images.
  • Glint Analysis: Developing specific algorithms to detect and characterize short-lived glints.
  • Apparent Magnitude Conversion: Converting surface brightness to apparent magnitude using TLE-derived exposure times to allow direct comparison with other brightness datasets.

In conclusion, the paper provides a critical technical step forward in quantifying the "satellite pollution" of the night sky, moving from qualitative observation to quantitative, object-specific measurement.