Enhanced detection limits in the SHINE F150 survey through the Regime Switching Model Optimizing thresholds and investigating environmental noise

TL;DR

This work revisits the SHINE F150 high-contrast imaging dataset with the Regime Switching Model (RSM) to push detection limits under varying environmental noise. By clustering observations into six environmental-condition groups and optimizing RSM parameters per cluster, the study shows detection thresholds vary significantly with observing conditions and PSF-subtraction configurations, with log-normal and F1-score thresholds offering complementary advantages. RSM, especially when leveraging multiple PSF subtraction techniques, yields notable gains in sensitivity—up to fivefold at inner working angles and about twofold at larger separations—leading to 87 detected signals, including 38 new detections and one promising planet candidate requiring follow-up. The results demonstrate the value of environment-aware noise modeling for robust companion detection and pave the way for applying this framework to the full SHINE survey and other ground-based high-contrast imaging programs, aided by publicly available code and data access.

Abstract

In high-contrast imaging, a novel detection algorithm for angular differential imaging (ADI) sequences has recently been introduced: the Regime Switching Model (RSM). In this study, we apply the RSM algorithm to analyze the F150 sample from the SHINE high-contrast imaging survey carried out with VLT/SPHERE, aiming to enhance detection limits and identify new exoplanet candidates. Additionally, we investigate how environmental conditions influence post-processed noise distributions and detection thresholds. We generate detection maps and contrast curves for 213 observations in the F150 SHINE sample using the RSM algorithm. A clustering approach based on environmental parameters is used to group observations with similar noise characteristics. We propose two methods for defining radial detection thresholds in the RSM maps: fitting a log-normal distribution to the post-processed noise and maximizing the F1 score. We also assess the performance of various combinations of post-processing techniques within the RSM framework to identify optimal configurations. This study demonstrates the utility of clustering based on observational parameters, effectively distinguishing features like wind-driven halos and low-wind effects. Detection thresholds vary significantly across clusters, differing by up to a factor of 10, highlighting the importance of considering observational environments. Log-normal thresholds provide conservative, noise-aware limits, while F1 score-based thresholds offer observation-specific results, both showing compatibility overall. RSM improves detection limits by an average factor of two at 1arcsec and five at inner working angles compared to standard PCA processing. This study reports more than 30 newly detected signals, including one promising candidate awaiting second-epoch confirmation.

Figures (26)

• Figure 1: Image representation of single frames from various observations across different clusters, labeled (a) to (f) corresponding to clusters 1 to 6 presented in log scale.
• Figure 2: Projection of cluster distributions across parameters: seeing, coherence time, wind speed, raw contrast, Strehl ratio, total parallactic angles, and number of frames.
• Figure 3: Quantile-Quantile Plot of the RSM noise in Cluster 1 at $0\farcs25$ (left) and $1\arcsec$ (right) using APCA. The plot compares the RSM noise distribution to the Lognormal Distribution, excluding values above the $3 \times 10^{-7}$ False Alarm Probability of the fit. The Coefficient of Determination ($R^2$) is also displayed, indicating a good fit.
• Figure 4: Top. Log-normal distributions obtained by fitting RSM noise histograms at $0\farcs25$ for various observing conditions (clusters) using RSM with APCA. Bottom. Detection thresholds derived from the lognormal distributions at $3\times 10^{-7}$ FAP, as a function of angular separation and of observing conditions.
• Figure 5: Variation of the true positive rate (TPR), false positive rate (FPR), and F1 score for the cluster 1 center at $0\farcs7$. The black dashed line marks the threshold that maximizes the fitted F1 score curve (grey curve), balancing high true positive (blue curve) recovery with minimal false positives (red curve).