Wrist Temperature-Based Sleep Detection for Maritime Environments

Abstract

This study investigates a physiology-based approach for sleep detection implemented through a lightweight, staged temperature-based algorithm. The method estimates daily sleep bouts by identifying sleep onset and offset from temperature trends. The approach is motivated by the limitations of conventional actigraphy-based methods, which rely on accelerometer data and may become unreliable in environments characterized by continuous external motion and polyphasic sleep settings, such as offshore sailing. Under these conditions, estimation of sleep parameters drops substantially. The proposed method was evaluated on data collected from 7 athletes over 25 nights, resulting in a total of 148 days analyzed. The Garmin Quantix 8 smartwatch has been worn during three on land nights and 22 offshore nights during a long-term regatta. The results were compared with open-source Python packages for accelerometer-based sleep detection and with sleep diaries that served as ground truth. The balanced accuracy of 67.1% and F1-score equal to 53.1% revealed a clear trend: while the accelerometry failed by underestimating or overestimating the sleep periods, our approach defined the best overall performance. These findings suggest that wrist temperature-based sleep detection can provide a robust and computationally efficient alternative or complement to actigraphy-based methods, particularly in environments where motion signals are unreliable.