“ We should support a software ecosystem in which centrally developed reference implementations may exist, but interfaces are simple and well documented so that alternative implementations can be swapped in — this will allow us to leverage innovation by the community.
We encourage support of International Virtual Observatory Alliance (IVOA) protocols, and specifically pyVO, for retrieval of test datasets (and possibly of satellite prediction data if appropriate protocols exist) . However, programs should always also allow import of datasets from a local disk. ”

“ We assume this is a small telescope, so the spatial extent of the satellite (defocus + resolved size) is not accounted for.
The effects of streaks on optical imaging data were discussed in the SATCON1 Report (Walker et al., 2020) . In this working group we also discussed the effects on other kinds of data, especially spectroscopy. Low-spectral-resolution fiber spectroscopy is especially vulnerable — the effect of a satellite streak is to add a solar spectrum to the target spectrum, and in the absence of any spatial information it may be hard to spot that your data has been affected. ”

“ In brief, the SATCON1 recommendations call for the ability to flag, mask and repair satellite trails affecting astronomical data (a software tool we call TrailMask) , to predict when satellite trails may or will affect specific observations (which we call PassPredict) , and to simulate the effects of satellite trails so that the community can assess the scientific impact of those effects on astronomical research.
Our main focus is on ground-based optical images of all kinds. However, we also considered space-based images and spectroscopy. ”

“ Keras, a deep learning API (applications programming interface) written in Python, runs on top of TensorFlow's machine learning platform, focusing on enabling fast experimentation and easy implementation. With Keras, a trained model can be used easily as Python code, standalone, or a part of a pipeline.
The deep models that are trained on a single instrument are likely not generically applicable but they can be used as pre-trained models for trail detection on new instruments. Training for the new instrument can then be achieved using a smaller training data set. ”

“ It is likely that many ground-based observatories around the world will be forced to make a significant investment in hardware such as auxiliary spotting cameras or other solutions to deal with the problem of satellite streaks. However, the effects of satellite constellations on a given observatory will depend on specifics such as aperture, etendue, pixel size, observing strategy, and other factors; for some the impact will be small and mitigations will not be required, while for others the impact will be serious. Each observatory will need to make its own assessment. ”

“ Red: visual magnitude at zenith of an example satellite as a function of orbit altitude. Blue to Green: observed magnitude of the same satellite accounting for trailing, for a series of increasing exposure times and assuming a 1-arcsecond resolution element. In a given telescope/instrument, as exposure time increases, the number of counts detected from a faint (e.g., 15th magnitude) star will increase, but the number of counts in the satellite trail will not, assuming that the satellite crosses the field of view in a time that is short compared to the exposure time. ”

“ Support development of a software application for observation planning available to the general astronomy community that predicts the time and projection of satellite transits through an image, given celestial position, time of night, exposure length, and field of view, based on the public database of ephemerides. Current simulation work provides a strong basis for the development of such an application.
A browser-based interface will enable astronomers to predict what satellites will intersect with any given single observation. ”

“ A satellite pass test suite is needed to test PassPredict. The ObsCore data model should also be used here. Each test case should include:
The observing conditions (telescope and camera parameters, pointing direction, date and time) A fixed test satellite database from which satellite ephemerides may be extracted, or (for some use cases) a single satellite ephemeris prediction (when you don’t want to test the database part) We should be able to (retro-) predict the passes of a few specific satellites at a specific epoch over a specific observatory and (if possible) predict their brightness. ”

“ Simulated satellite traces planted on real images and also the images without the traces as the training set for deep learning models Real satellite traces in images — coming from the test data suite and elsewhere. ”

“ We re-emphasize SATCON1 recommendations 1 to 3. New tools are critical to partially mitigate the constellation impacts on astronomy. The PassPredict software will allow astronomers to determine which observations may be affected and in conjunction with simulations may allow quantification of the degradation of science data expected in a particular situation. The TrailMask software will allow some science to be salvaged from some affected datasets and reduce the chance of spurious results being published. ”

“ Pretrained nets User supplied training set (one could call the simulation tools discussed below to generate simulated data from observation parameters) A deep learning generative model has been used by the CADC team to remove moving object trails; it uses the open source tensorflow library. However, designing the model is not easy. A deep learning model can be trained, with relevant data sets, to detect and model various objects in images, such as satellite traces in astronomical images. The models can then be used to remove the trails from the images using deep learning techniques. ”

“ The deep learning model that creates the latent representation is trained such that the compressed latent representation contains astrophysically meaningful information. Simultaneously, the latent vector outputs of deep models provide a homogenous input for deep downstream models trained to solve astrophysical problems. A set of deep probabilistic deep models can learn "what is in the images". ”

“ We are mostly concerned with the large low-Earth orbit (LEO) satellite constellations. However, spacecraft at near-lunar distance are regularly seen by asteroid surveys, so we should consider MEO and GEO (medium Earth orbit and geosynchronous Earth orbit) cases too. Note that the fainter magnitude of high-orbit satellites is offset by their lower apparent angular velocity, leading to larger effective exposure time on a streak pixel. ”

“ Note that covariance interpolation can lead to unphysical results (results which are non-positive-definite or have negative variance) . See the astroplan package (https://astronplan.readthedocs.io/en/latest/) for approaches to handling this. Calculate the difference vector between the observatory location and the satellite position. Transform this to a unit vector (the “line of sight”) in the observatory topocentric frame. Determine the telescope field of view at this time in the topocentric frame. There remain detailed issues to resolve: how to handle aberration and refraction ”

“ Flag noise from satellite trails to reduce corruption in science analysis. Reduce/remove the noise that satellite trails insert into images. Making images “look pretty” with various blending techniques is less important to the science mission. However, visual inspection of images is often important for analysis. Removal of satellite trails to allow deeper visual inspection (while ensuring the pixels impacted by the removal are flagged) is an important capability.
3.1.3. Relevant existing software
Rubin Observatory’s satellite trail finder maskStreaks.py [1] — all open software. ”

“ Operator orbit solutions for their own active satellites determined by active tracking of the satellite radio signal or derived from onboard GPS receivers Passive (radar or optical) tracking of satellites, including inactive satellites and debris. This is systematically done by US Space Force 18th Space Control Sqn (18SPCS) and the Russian Space Forces’ SKKP, and is now also being done by commercial companies LeoLabs and ExoAnalytics, and to some extent by ESA. For brighter satellites optical and radio-transmission tracking is also done by hobbyists. ”

“ We need to provide base images to add simulated trails to. One could argue that you can just use existing images and no basic image simulation tool is needed. However, it is useful to have the capability to create images with known (because simulated) source properties (e.g., faint sources with a variety of known magnitudes) to properly assess detection thresholds. ”

“ The output of this tool could then be passed to PassPredict or a similar algorithm to model the observability of the constellation at a given observatory and date. In an advanced mode, we should also predict the satellite brightness. Simulations should include cases for professional observations and also of the effect on the naked-eye sky. As to existing software, various unpublished research codes exist but would need to be modified to generate the ephemeris database output and improved to be robust enough for general use. ”

“ Higher resolution spectrographs have much shallower limiting magnitudes (15-20, even on large telescopes) , and will therefore be essentially immune to the contamination by all but the brightest satellites. Finally, although exoplanet transit spectroscopy already has the problem of subtraction at the limits of S/N, its high spectral resolution will prevent significant contamination problems.
1. General Software Considerations
The astronomy software community is investing heavily in Python and in particular in the astropy suite. ”

“ A large simulation and modeling effort will allow the community to assess impacts of current and future constellations on both ground- and spac-ebased observations and establish recommended constraints on the design of constellations. ”

“ However, depending on use case, some or several of these options might be superfluous, or even undesirable. For instance, in many science cases where the noise properties must be very well understood, it is preferable to flag and ignore affected pixels rather than trying to recover them. ”

“ ILRS Data Format Working Group (2006) . Consolidated Laser Ranging Prediction Format. Version 1.01, https://ilrs.gsfc.nasa.gov/docs/2006/cpf_1.01.pdf
International Business Machines. (n.d.) IBM/arcade. Github.com. ”