Tracking turbulence for wind-driven halo removal using MagAO-X pyramid wavefront sensor images
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.
Vera: We're looking at the paper: "Tracking turbulence for wind-driven halo removal using MagAO-X pyramid wavefront sensor images".
Jocelyn: Here's the gist of the paper. The wind-driven halo (WDH) is a persistent, low spatial frequency noise artifact that arises due to the servo-lag error inherent to all adaptive optics (AO) instruments.
Vera: Using archival WFS telemetry from MagAO-X, which is the “extreme” AO instrument for the 6.5-meter Magellan-Clay telescope, we demonstrate a novel workflow for WDH reconstruction and removal in individual coronagraphic science images.
Jocelyn: The wind-driven halo (WDH) manifests as a butterfly-shaped artifact spanning across this region of the image, arising due to the servo-lag error inherent to all AO instruments.
Vera: Spatial filtering is detrimental to diffuse, often faint structures associated with circumstellar disks.
Paper summary: Vera: So, to summarize this paper, the main thesis is that we can use archival wavefront sensor telemetry from MagAO-X to track atmospheric turbulence and reconstruct the wind-driven halo artifact.
Jocelyn: The core claim is that spatial filtering isn't a good option because it hurts diffuse structures, so they instead use the closed-loop wavefront data to measure wind vectors layer by layer.
Subrahmanyan: The paper claims their pipeline, windsoCC, successfully cross-correlates this telemetry to measure the wind vectors of several turbulent layers above Las Campanas Observatory.
Vera: And from those measured wind parameters, they use a parametric model called WINDFM to build a full basis of WDH models that sum up to the best-fit total solution.
Jocelyn: This basis is then used in a non-negative least-squares fitting routine to subtract the WDH artifacts from each science frame. That’s how they remove the noise from individual images.
Subrahmanyan: The paper emphasizes that this approach provides a dramatic improvement in object recovery when applied to on-sky MagAO-X images of the disk around HR 4796A at visible wavelengths <ref:2611.06873#pg1>.
Vera: It seems the methodology is centered on using WFS data to model the noise structure rather than just trying to filter it out blindly.
Jocelyn: And they are showing that this technique helps image an entire disk without needing aggressive PSF subtraction, which is a major point because PSF subtraction is so tricky for extended objects.
Subrahmanyan: So, the implication here is that we can get cleaner data on circumstellar disks by accounting for the wind structure affecting our adaptive optics system.
Conclusion: Vera: Looking at the title, "Tracking turbulence for wind-driven halo removal using MagAO-X pyramid wavefront sensor images," it really tells you exactly what this work is doing: they are tracking atmospheric turbulence to remove that specific noise artifact using a specific instrument's data.
Jocelyn: The authors, Kuenya and collaborators, focused on showing how the closed-loop wavefront data from MagAO-X can be used for this kind of postprocessing reconstruction.
Subrahmanyan: The implication is that we gain a much more reliable way to study faint, extended structures near stars because we can account for the dynamic noise component from the atmosphere itself.
Vera: It moves us away from just trying to fix the AO loop speed or gain—the traditional ways—to actually using what's already being measured in real time, like those wind vectors.
Jocelyn: So, ultimately, this work suggests that leveraging WFS telemetry for noise removal is a powerful tool for high-contrast imaging science.
Subrahmanyan: It confirms that modeling the WDH as a combination of multiple layers from different turbulent regions is a necessary step if we want to accurately map those environments around stars.
Jay K. Kuenya, Jared R. Malesa, Alycia J. Weinbergere, Laird M. Closea, Joseph D Longd, Joshua Libermana, Sebastiaan Haffertf, aSteward Observatory, University of Arizona, Center for Computational Astrophysics, Flatiron Institute, Earth and Planets Laboratory at Carnegie Institution for Science, Leiden Observatory at Leiden University, Subaru Telescope at National Observatory of Japan, Department of Astronomy at University of Michigan, Olivier Guyong, Logan Pearceh, Parker T Johnsona, Katie Twitchella, Jialin Lia, Alex Hedgleni Avalon Gowerj Warren Fostera Jhen Lumbresk Lauren Schatzl
Steward Observatory, University of Arizona · James C Wyant College of Optical Sciences, University of Arizona · Center for Computational Astrophysics, Flatiron Institute · Earth and Planets Laboratory at Carnegie Institution for Science · Leiden Observatory at Leiden University · Subaru Telescope at National Observatory of Japan · Department of Astronomy at University of Michigan
astro-ph.IM
Submitted: 2026-11-04
Updated: 2026-11-04
Comments: 59 pages, 11 figures, 6 tables; Accepted for publication in SPIE JATIS
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 83/100
The gist: The wind-driven halo (WDH) is a persistent, low spatial frequency noise artifact that arises due to the servo-lag error inherent to all adaptive optics (AO) instruments.
Key concepts
- Wind-Driven Halo (WDH)
- A persistent, low spatial frequency noise artifact caused by servo-lag errors in adaptive optics systems. It appears as a butterfly shape and is difficult to filter out because it changes intensity and position over time.
- windsoCC Pipeline
- A data processing workflow that cross-correlates the recorded closed-loop wavefront data from the pyramid WFS. This process measures the apparent motion of residual wavefront structure across the telescope's pupil, allowing researchers to map out wind vectors in different atmospheric layers.
- Forward Modeling (WINDFM)
- A technique where a parametric model, based on Gaussian envelopes and 'pinch' parameters defining wing opening angles, is used to simulate the WDH. This simulated model is then used in a least-squares fitting routine to subtract the noise from science images.
- Pyramid Wavefront Sensor (WFS)
- A specific type of WFS on MagAO-X capable of recording high-cadence wavefront telemetry. This high-resolution data stream is crucial because it provides the necessary input for windsoCC to measure the turbulent layers above Las Campanas Observatory.
Summary
The wind-driven halo (WDH) is a persistent, low spatial frequency noise artifact that arises due to the servo-lag error inherent to all adaptive optics (AO) instruments. p. 0 , Tucson, AZ 85721,
USA
cNational Science Foundation Graduate Research Fellow
dCenter for Computational Astrophysics, Flatiron Institute, 162 5th Ave, New York, NY
eEarth and Planets Laboratory, Carnegie Institution for Science, 5241 Broad Branch Road NW, Washington, DC
20015-1305
fLeiden Observatory, Leiden University, PO Box 9513, 2300 RA Leiden, The Netherlands
gSubaru Telescope, National Observatory of Japan, Hilo, HI
hDepartment of Astronomy, University of Michigan, Ann Arbor, MI
iNorthrop Grumman Corporation, 600 South Hicks Rd, Rolling Meadows, IL
jDraper Laboratory, 555 Technology Square, Cambridge, MA
kNorthrop Grumman, Pasadena, CA
lStarfire Optical Range, Kirtland Air Force Base, Albuquerque, NM
Abstract. The wind-driven halo (WDH) is a persistent, low spatial frequency noise artifact that arises due to the
servo-lag error inherent to all adaptive optics (AO) instruments. Since wavefront sensors (WFSs) continuously
track the closed-loop residual wavefront error, WFS telemetry presents the ideal resource for combating this type of
noise artifact through postprocessing. Using archival WFS telemetry from MagAO-X, which is the “extreme” AO
instrument for the 6. Using archival WFS telemetry from MagAO-X, which is the “extreme” AO
instrument for the 6. 5-meter Magellan-Clay telescope, we demonstrate a novel workflow for WDH reconstruction
and removal in individual coronagraphic science images.Where the paper says it
Where the paper says it
Where the paper says it
Introduction to the Problem
The wind-driven halo (WDH) manifests as a butterfly-shaped artifact spanning across this region of the image, arising due to the servo-lag error inherent to all AO instruments. Spatial filtering is detrimental to diffuse, often faint structures associated with circumstellar disks. p. 2 When imaging point sources,
high-pass filtering or aggressive PSF subtraction may perhaps be employed with limited efficacy. However, spatial filtering is detrimental to diffuse, often faint structures associated with circumstellar disks ([11], [12]). There is value in mitigation
strategies for the WDH, as simulations have shown improvements in raw contrasts on the order of
∼ 1 magnitude in the areas closest to the star ([10]; [13]). Subtracting the point-spread function (PSF) is a challenging aspect of high-contrast imaging
post-processing especially for resolved extended objects like circumstellar disks.Where the paper says it
Where the paper says it
Wind Vector Retrieval Pipeline
The pipeline, windsoCC, cross-correlates the recorded closed-loop wavefront to measure the wind vectors of several turbulent layers of the atmosphere above Las Campanas Observatory. The workflow involves several discrete stages:
-
Partitioning raw telemetry into short, time-ordered segments to treat turbulence as approximately stationary. p. 7
Where the paper says it
For characterizing the wind, these data must be grouped on timescales short enough that the behavior of the atmospheric turbulence can be treated as approximately stationary, while also spanning sufficient integration time to minimize the noise floor. We therefore partition the raw WFS telemetry 8
-
Reducing WFS image preparation and calibration to remove quasi-static detector and illumination structure. p. 8
Where the paper says it
2 Reduce: WFS image preparation and calibration After partitioning the raw telemetry into shorter temporal segments, each set of pyramid WFS images is calibrated and reformatted for cross-correlation analysis. The purpose of this reduction step is to remove quasi-static detector and illumination structure from the raw WFS frames while preserving just the residual wavefront structure.
-
Cross-correlating reduced WFS pupil cubes using a segmented, overlapping CC method to measure apparent motion of residual wavefront structure across the pupil. p. 9
Where the paper says it
, 8 WFS frames at a 2 kHz AO loop speed) and thus would be easily trackable in our data. 3 Cross-correlate: retrieving the turbulence layers The reduced WFS pupil cubes are analyzed using the CC to measure the apparent motion of residual wavefront structure across the pupil.
-
Distilling the CC response maps by combining measurements from the four pyramid WFS pupil images and applying unsharp masking and matched-filter (MF) templates. p. 7
Where the paper says it
P artitio n Re d u c e Cro s s- c orrelate Distill Profile Crop pupils Average CC response cubes across 4 quadrants Derotate by parallactic angle CC across range of delays Subtract image reference Read in pupil quadrant set Concatenate time series Assemble CC response cube Divide by noise Chunk temporally into mini datasets Saved image data Download images MagAO-X observations Allocate WFS data Image coadding Save pupils separately by quadrant Apply unsharp masking Make matched-filter (MF) template Make MF response cube Extract wind CC peaks Match extractions with tracks Predict next-frame locations Repeat for remaining pupil sets Repeat for remaining frames in cube Reject non-physical wind tracks Save outputs to disk Fig 1 Flowchart illustrating the various operations performed in the windsocc pipeline in left to right progression. This workflow is explained in detail in Section 3.
-
Profiling wind layer attributes by detecting coherent peaks using Source Extractor facilitated by the SEP Python library and matching them to a wind track. p. 4
Where the paper says it
Thus, profiling the turbulence above a site using instruments inherent to an AO system continues to be an active area of research. Recent work by [33, 34] showed that cross-matching the wind vectors measured from the WFS telemetry with weather data led to increased knowledge on the turbulence above Mauna Kea.
Wind Layer Characterization
Wind layer attributes are measured from the MF response cubes produced by the distillation stage, where a wind layer appears as a compact peak whose displacement from the origin increases linearly and with constant velocity with temporal delay. The measurement stage detects and tracks coherent peaks due to individual wind layers as they move through the correlation space. p. 7 P
artitio
n
Re
d
u
c
e
Cro
s
s- c
orrelate
Distill Profile
Crop pupils
Average CC
response
cubes across
4 quadrants
Derotate by
parallactic
angle
CC across
range of
delays
Subtract
image
reference
Read in pupil
quadrant set
Concatenate
time series
Assemble
CC response
cube
Divide by
noise
Chunk
temporally into
mini datasets
Saved
image
data
Download
images
MagAO-X
observations
Allocate
WFS data
Image
coadding
Save pupils
separately by
quadrant
Apply
unsharp
masking
Make
matched-filter
(MF) template
Make MF
response
cube
Extract
wind CC
peaks
Match
extractions
with
tracks
Predict
next-frame
locations
Repeat for
remaining pupil sets
Repeat for
remaining
frames in cube
Reject
non-physical
wind tracks
Save
outputs to
disk
Fig 1 Flowchart illustrating the various operations performed in the windsocc pipeline in left to right progression. This workflow is explained in detail in Section 3.Where the paper says it
WDH Forward Modeling and Removal
The WDH is modeled using a parametric model assuming a Gaussian envelope along the direction of the wind and a “pinch” parameter that dictates the opening angle of the wings. This model is then forward-modeled through KLIP in lieu of a traditional disk model, referred to as WINDFM. p. 3 Going further, KLIP forward modeling (FM) methods are being developed
with the goal to recovering the astrophysical signal lost to PSF subtraction (e. , [18]; [19]; [20])
by simulating the instrument response and PSF subtraction artifacts on a model object. We then make use
of the wind parameters learned through windsoCC to reconstruct the WDH footprint by leveraging a parametric
model. Notably, we demonstrate a dramatic improvement in object recovery using on-sky MagAO-X images of the
disk around HR 4796A at visible wavelengths. , [18]; [19]; [20])
by simulating the instrument response and PSF subtraction artifacts on a model object. Reference
Differential Imaging (RDI; [21], [22]) is an alternative method to ADI where images of another
star that does not harbor an astrophysical object is used to subtract the PSF from the science target
images.Where the paper says it
Where the paper says it
Where the paper says it
Key Findings and Implications
The study demonstrates dramatic improvement in object recovery using on-sky MagAO-X images of the disk around HR 4796A at visible wavelengths. The results show that the WDH affecting MagAO-X images is an agglomeration of multiple WDHs stemming from the turbulent layers above LCO at the time of observations. p. 6 3 windsoCC pipeline description
In order to gauge how different layers of atmospheric turbulence affect MagAO-X’s performance,
we implemented a pyramid WFS [40] image processing pipeline, windsoCC, based on CC. The
basic premise is that under the assumption of frozen-flow, turbulent layers will appear as 2D static
phase screens undergoing translation across the telescope pupil images. MagAO-X is equipped with a pyramid WFS capable of
recording wavefront telemetry at a high-cadence which is saved during data acquisition. Given this, we detail how
our WFS data processing pipeline, windsoCC, cross-correlates the recorded closed-loop wavefront to measure the
wind vectors of several turbulent layers of the atmosphere above Las Campanas Observatory. There are strategies that may be employed to combat the WDH. When imaging point sources,
high-pass filtering or aggressive PSF subtraction may perhaps be employed with limited efficacy. Using archival WFS telemetry from MagAO-X, which is the “extreme” AO
instrument for the 6. 5-meter Magellan-Clay telescope, we demonstrate a novel workflow for WDH reconstruction
and removal in individual coronagraphic science images.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Future Directions
Going forward, the authors hypothesize that forward modeling a WDH for all wind layers detected using closed-loop WFS images will lead to a better understanding of how the current atmospheric profile will affect observations. One planned upgrade for this project is computing the forward modeling procedure within the framework of JAX [56] which is a Python machine learning library that facilitates autodiff and GPU-accelerated array computations. The ultimate goal for this project is to remove the WDH artifacts in the science focal plane images using solely the WFS telemetry. p. 6 Since the CC peaks encode the residual error that propagates to the science focal
plane, we can then leverage the windsoCC output to inform our WDH model parameter priors
when reconstructing the WDHs. Our pipeline follows a number of discrete stages that we illustrate
using a flowchart in Figure 1.Where the paper says it
Disclosures
A portion of the codebase used to facilitate the analyses presented in this manuscript were developed with the assistance of artificial intelligence (AI). The authors have personally verified that all of the code in the latest version generates output that is correct to the best of the authors’ knowledge and abilities.
Code and Data Availability
The code that we used in our analysis are available upon request from the corresponding author. p. 8 This pre-processing is critical since the subsequent
CC stage is sensitive to static aperture features and detector artifacts and makes the CC response
ambiguous. For each partitioned telemetry directory, windsoCC first identifies the available camwfs
FITS cubes and loads the corresponding reduction parameters from the pipeline configuration file. MagAO-X operates with the pupil fixed with respect to the sky to facilitate ADI. Additionally, we made use of MagAO-X’s small Lyot coronagraph (focal plane mask radius of
about 21 mas or ≲ 3λ/D) for all observations.Where the paper says it
Where the paper says it
Acknowledgments
J.K.K., J.R.M., Y.D.L., and A.J.W acknowledge support from the NSF, grant no AST-2307613. J.D.L was supported by the Flatiron Software Research Fellowship at the Flatiron Institute, a division of the Simons Foundation. p. 0 University Blvd. , Tucson, AZ 85721,
USA
cNational Science Foundation Graduate Research Fellow
dCenter for Computational Astrophysics, Flatiron Institute, 162 5th Ave, New York, NY
eEarth and Planets Laboratory, Carnegie Institution for Science, 5241 Broad Branch Road NW, Washington, DC
20015-1305
fLeiden Observatory, Leiden University, PO Box 9513, 2300 RA Leiden, The Netherlands
gSubaru Telescope, National Observatory of Japan, Hilo, HI
hDepartment of Astronomy, University of Michigan, Ann Arbor, MI
iNorthrop Grumman Corporation, 600 South Hicks Rd, Rolling Meadows, IL
jDraper Laboratory, 555 Technology Square, Cambridge, MA
kNorthrop Grumman, Pasadena, CA
lStarfire Optical Range, Kirtland Air Force Base, Albuquerque, NM
Abstract. We inset small thumbnail images illustrating data products at the
intermediate steps. The final outputs include a human-readable text file that contains the information on each detected
wind layer and the MF response cubes. We inset small thumbnail images illustrating data products at the
intermediate steps. The final outputs include a human-readable text file that contains the information on each detected
wind layer and the MF response cubes. MagAO-X operates with the pupil fixed with respect to the sky to facilitate ADI. Additionally, we made use of MagAO-X’s small Lyot coronagraph (focal plane mask radius of
about 21 mas or ≲ 3λ/D) for all observations.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
WindsoCC pipeline enables automated retrieval of wind vectors from archival pyramid WFS telemetry by cross-correlating pupil images at different times, allowing for a
real-time
description of optical turbulence within a line-of-sight without needing a separate instrument. -
WindFM forward modeling allows for the reconstruction and removal of the wind driven halo by using measured wind attributes to inform parametric model priors during Markov Chain Monte Carlo fitting.
-
The system can remove WDH artifacts from individual science images using WDH model components as part of a modal basis, which is then used in a non-negative least squares fitting routine against an estimate of the PSF.
Abstract
The wind-driven halo (WDH) is a persistent, low spatial frequency noise artifact that arises due to the servo-lag error inherent to all adaptive optics (AO) instruments. Spatial filtering may be employed to overcome this artifact, however, filtering out the WDH while simultaneously preserving signal from an extended astrophysical object of interest is exceptionally challenging. Additionally, since the WDH changes in intensity and position angle through an observation, data-driven algorithms (e.g., KLIP) that are commonly used to subtract the starlight need to be overly-aggressive to remove both the static and dynamic noise components. Since wavefront sensors (WFSs) continuously track the closed-loop residual wavefront error, WFS telemetry presents the ideal resource for combating this type of noise artifact through postprocessing. Using archival WFS telemetry from MagAO-X, which is the "extreme" AO instrument for the 6.5-meter Magellan-Clay telescope, we demonstrate a novel workflow for WDH reconstruction and removal in individual coronagraphic science images. MagAO-X is equipped with a pyramid WFS capable of recording wavefront telemetry at a high-cadence which is saved during data acquisition. Given this, we detail how our WFS data processing pipeline, windsoCC, cross-correlates the recorded closed-loop wavefront to measure the wind vectors of several turbulent layers of the atmosphere above Las Campanas Observatory. We then make use of the wind parameters learned through windsoCC to reconstruct the WDH footprint by leveraging a parametric model. Notably, we demonstrate a dramatic improvement in object recovery using on-sky MagAO-X images of the disk around HR 4796A at visible wavelengths.
Sources
- Multiwavelength optical properties of compact dust aggregates in protoplanetary disks
- PynPoint: An Image Processing Package for Finding Exoplanets
- Reference-star differential imaging on SPHERE/IRDIS
- Estimating Atmospheric Wind Speeds From Gemini Planet Imager AO Telemetry
- A search for a 5th planet around HR 8799 using the star-hopping RDI technique at VLT/SPHERE
- Astropy: A Community Python Package for Astronomy
Related papers
- On-sky demonstration of a vector Zernike wavefront sensor in a cascaded adaptive optics system
- Euclid Data Release 1 (DR1): Calibration strategy and observations
- The General Antiparticle Spectrometer (GAPS) Antarctic Balloon Payload
- Detection and excision of radio-frequency interference in radio arrays
- Unified matched-filter and excess-power pipeline for gravitational-wave detector characterization
- Ushering in sub-arcsecond angular resolution observatories in the far-infrared: The case for a double-Fourier interferometer