A Novel Statistical Framework for Recovering Radial Velocities from Line-by-Line Spectral Data

Detecting low-amplitude radial velocity (RV) signals is challenging because stellar variability can mimic or obscure the Doppler shifts caused by low-mass planets. Changes in the shapes of spectral lines provide valuable information for disentangling stellar variability from true Doppler shifts. In this work we introduce a novel framework for analyzing spectroscopic time-series data. Our model jointly estimates a shared temporal RV component and line-specific regression coefficients linking RV deviations to multiple shape descriptors extracted from hundreds of spectral lines. Though the proposed statistical model is classical, the novelty lies in the exploitation of estimated line-by-line shape information from the spectra, producing significant reductions in RV root-mean-square errors. Applied to a large spectroscopic dataset with known ground-truth RV signals, the model reduces the root-mean-square error by approximately 76% relative to uncorrected line-by-line Doppler shifts.

astrostatistics

exoplanet detection methods

astronomy data analysis

time series analysis