Estimating circadian rhythms often relies on intrusive and costly methods like melatonin sampling, misfitting in longitudinal field studies. Alternatively, ambulatory activity monitoring is frequently used, yet requires week-long data and signals fitting poorly with cosinor models. Ambulatory skin temperature monitoring is promising, but its feasibility, reliability, and robustness to homeostatic sleep pressure remain understudied. We collected sensor data in a randomized crossover field study manipulating sleep duration (8 h vs. 4 h time-in-bed, three consecutive nights) among 17 healthy participants (19-32 years old). Activity and skin temperatures (distal, proximal, distal-proximal gradient) were continuously monitored across conditions. Cosinor analysis estimated rhythm parameters (amplitude, mesor, acrophase) and goodness of fit. Non-parametric tests and linear mixed-effect models assessed difference and stability across sleep conditions. Cosinor models fitted better on skin temperature than activity-based data. Sleep restriction affected activity rhythms (reduced amplitude, delayed acrophase, increased mesor). Beside a reduced distal temperature amplitude in sleep restriction, skin temperature rhythm showed no statistically significant moderations by sleep. Rhythm parameters between data types did not significantly correlate. Skin temperature showed promising feasibility for in-field circadian rhythm estimation after only three days, potentially facilitating research on circadian rhythms and health. Validation against core body temperature and melatonin sampling is warranted.
