Bayesian Gaussian Methods for Robust Background Modeling in CALorimetric Electron Telescope (CALET) Gravitational-Wave Searches 2025
Bisweswar Sen · 2025
Read Bayesian Gaussian Methods for Robust Background Modeling in CALorimetric Electron Telescope (CALET) Gravitational-Wave Searches 2025 by Bisweswar Sen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-25893.txt.
Introduction
Bisweswar Sen improves how a space telescope spots gamma-ray bursts in a 2025 paper. The CALET instrument on the space station hunts for flashes that may pair with gravitational wave events. Telling a flash from background noise needs a good model of that background. Older runs used flat averages or simple curves, which miss linked noise and shifting levels, and they drop the doubt in the background from later math.
Sen's method uses Bayesian tools: a flexible curve model that tracks linked noise, plus a change-point model for sudden shifts. Doubt in the background flows into detection scores and upper limits. On stored data and fake injected flashes, the method caught weak short bursts far better than curve fits, up to tenfold.
Catalog record
| CREATOR | Bisweswar Sen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-25893
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian Gaussian Methods for Robust Background Modeling in CALorimetric Electron Telescope (CALET) Gravitational-Wave Searches 2025 Bisweswar Sen Abstract The search for gamma-ray counterparts to gravitational-wave events with the CALET Gamma-ray Burst Monitor (CGBM) requires accurate and robust background modeling. Previous CALET observing runs (O3 and O4) relied on averaged pre/post-event baselines or low-order polynomial fits, approaches that neglect correlated noise, temporal non-stationarity, and the propagation of background uncertainty into derived flux upper limits. These simplifications can lead to reduced sensitivity to faint or atypical transients. In this work, we present a novel Bayesian framework for background estimation based on Gaussian Process (GP) regression and change-point modeling. Our approach captures correlated structures in the detector background, quantifies predictive uncertainties, and propagates them into both detection statistics and Bayesian credible upper limits. We demonstrate, using archival CALET time-tagged event data and simulated signal injections, that our method improves sensitivity to weak short-duration bursts by up to an order of magnitude compared to traditional polynomial fits. This probabilistic background treatment enables a more physically robust interpretation of non-detections and offers a scalable, real-time compatible extension for future joint multi-messenger searches. All codes used in this paper are available at github.com/SMALLSCALEDEV 1 Introduction The detection of gravitational waves (GWs) from compact binary mergers has opened a new era of multi-messenger astrophysics ( Abbott et al. 2016 ; Abbott et al. 2017 ) . Joint electromagnetic (EM) and GW observations provide complementary insights into the physics of compact object formation, the mechanisms of relativistic outflows, and the environments of neutron star and black hole mergers. In particular, gamma-ray observations play a critical role in probing prompt emission associated with short gamma-ray bursts (sGRBs), which are widely believed to originate from binary neutron star or neutron star–black hole mergers ( Eichler et al. 1989 ; Nakar 2007 ; Berger 2014 ) . The CALET mission ( CALorimetric Electron Telescope ), mounted on the International Space Station, includes the CALET Gamma-ray Burst Monitor (CGBM), which is sensitive to hard X-ray and soft gamma-ray emission in the keV–MeV range ( Torii et al. 2015 ; Yamaoka et al. 2017 ) . Owing to its wide field of view and continuous all-sky monitoring, CGBM is well suited to rapid follow-up of GW triggers. During LIGO/Virgo observing runs O3 and O4, CALET participated in systematic searches for gamma-ray counterparts to GW events ( Adriani et al. 2021 ; Adriani et al. 2024 ) . Although no definitive detections were reported, these efforts yielded important upper limits on the prompt gamma-ray emission from compact binary coalescences. A key step in such searches is accurate background estimation. The detector background in low-Earth orbit is shaped by multiple factors, including charged particle fluxes in the South Atlantic Anomaly, geomagnetic rigidity variations, and variable contributions from the cosmic X-ray background ( Ajello et al. 2008 ; Tatischeff et al. 2019 ) . In previous CALET analyses, background rates were estimated either by averaging pre- and post-event count rates in fixed time windows (O3) or by fitting low-order polynomials across a larger window surrounding the trigger (O4) ( Adriani et al. 2021 ; Adriani et al. 2024 ) . While straightforward and computationally efficient, these methods assume that the background is stationary and smoothly varying, an assumption that is not always justified. Detector backgrounds often exhibit correlated noise, abrupt rate changes due to orbital effects, and complex time structures that cannot be captured by polynomial models. Moreover, both pre/post averaging and polynomial fitting treat the background estimate as exact, neglecting uncertainty in the fitted model. This omission has two important consequences. First, detection statistics such as the signal-to-noise ratio (SNR) are computed relative to a fixed background, potentially inflating false alarms or missing weak signals in the presence of underestimated variance. Second, flux upper limits derived from non-detections do not account for uncertainty in the background model, leading to limits that may be either overly optimistic or overly conservative ( Loredo 1992 ; Gregory 2005 ) . Bayesian statistical methods provide a natural framework to address these limitations. Gaussian Process (GP) regression ( Rasmussen & Williams 2006 ) is particularly well-suited to modelling astrophysical detector backgrounds. GPs are non-parametric, probabilistic models that can flexibly capture correlated structures on multiple timescales while returning predictive uncertainties. They have been applied successfully in diverse areas of astrophysics, from stellar light curve detrending ( Foreman-Mackey et al. 2017 ) to spectral energy distribution fitting ( Alsing et al. 2018 ) . By applying GP regression to CGBM light curves, we can construct background models that capture both the mean structure and the covariance of the counts, thereby propagating uncertainty into detection and upper limit calculations. An additional challenge arises from the fact that detector backgrounds can exhibit non-stationarity, including sudden changes in mean count rate. To handle such behaviour, change-point detection algorithms ( Killick, Fearnhead & Eckley 2012 ; Truong, Oudre & Vayatis 2020 ) can be integrated with GP regression, allowing different statistical models to be fit on either side of detected discontinuities. This hybrid approach offers both flexibility and robustness: GPs capture correlated fluctuations within stable segments, while change-point models account for abrupt transitions. In this paper, we introduce a Bayesian framework for background estimation in CALET GW follow-up searches, based on Gaussian Processes and change-point modeling. Our contributions are threefold: 1. We demonstrate that GP-based background modelling improves sensitivity to weak short-duration signals compared to traditional polynomial fits. 2. We develop a Bayesian procedure for propagating background uncertainty into detection statistics and credible upper limits, yielding more robust constraints on gamma-ray emission from compact binary mergers. 3. We validate our approach using archival CALET data and injection–recovery simulations, showing that it is computationally feasible for near real-time follow-up analyses. This work highlights the importance of principled, uncertainty-aware statistical methods in the search for faint electromagnetic counterparts to gravitational-wave sources, and provides a general framework that can be extended to other high-energy astrophysics instruments. 2 Model The central task in searching for gamma-ray counterparts to gravitational-wave events is to distinguish transient signals from the variable detector background. In this section, we present our Bayesian framework for background estimation, based on Gaussian Process (GP) regression and change-point modelling. We describe the statistical formulation of the problem, motivate the use of GPs over traditional approaches, and outline how uncertainties are propagated into detection statistics and upper limits. All codes and implementation details are publicly available at My GitHub Repository . 2.1 Problem Formulation The observed counts in a gamma-ray detector channel can be represented as d(t)=b(t)+s(t)+\epsilon(t), (1) where d(t) is the observed count rate at time t , b(t) is the true background contribution, s(t) is a possible astrophysical signal, and \epsilon(t) represents measurement noise, which is typically dominated by Poisson fluctuations. The goal of background estimation is to construct a probabilistic model p(b(t)\mid d(t)) that captures both the mean behaviour and the uncertainty in b(t) . Traditional approaches assume either: 1. a stationary background approximated by a constant mean from pre/post windows, or 2. a smoothly varying function, typically a low-order polynomial. These assumptions ignore correlated noise, non-stationarity, and the variance associated with the background model itself. Such oversimplifications can bias detection statistics and yield unreliable flux upper