The effects of seasons and weather on sleep patterns measured through longitudinal multimodal sensing

The study examines whether and how seasons and weather (including day length and temperature) influence human sleep patterns—specifically sleep duration, bedtime, and wake time—in real-world industrial settings where artificial light and climate control may modulate circadian processes. Prior work shows sleep is governed by circadian mechanisms entrained by zeitgebers such as light and temperature, but findings about seasonal effects on sleep in humans are mixed across laboratory, observational, and preindustrial contexts. The authors aim to clarify these effects using objective, continuous, longitudinal sleep measures, while controlling for demographics and psychological traits, and by incorporating localized daily weather metrics.

Existing studies report inconsistent seasonal effects on sleep due to differences in setting (lab vs. in situ), measurement (wearables, diaries, apps), populations (age groups, industrial vs. preindustrial), and geography (latitude). Small laboratory studies found limited or no seasonal changes in sleep duration but shifts in sleep timing (e.g., earlier onset/offset in winter). Wearable-based studies with brief sampling windows often found ambiguous effects; some reported longer sleep in autumn or winter than spring/summer with minimal changes in timing. Preindustrial societies generally show longer sleep in winter and later wake times in summer, with seasonality attenuated by electricity. Large-scale self-report and app-based datasets reveal later wake times in winter and longer sleep in colder seasons, with stronger effects at higher latitudes, in children or the elderly, and among evening chronotypes; anomalously warm nights increase insufficient sleep in summer. Overall, when effects are present, sleep tends to be longer in winter and shorter in summer, potentially driven by longer photoperiods and higher temperatures, but results vary by subgroup and context. Prior work has rarely integrated detailed daily weather beyond temperature or used continuous, long-term objective measures across all seasons.

Design and ethics: Longitudinal observational study approved by University of Notre Dame IRB (17-05-3870) with written informed consent. Participants and period: Data collected across the U.S. from Feb 5, 2018 to Mar 15, 2019. Of 649 with year-long collection (757 started), 216 participants met completeness criteria after cleaning (≥50% of weekdays in each season; ≥33 weekdays/season), yielding 51,836 weekday observations (mean 239.9 days/participant). Sleep measurement: Participants wore Garmin Vivosmart 3 devices continuously (24/7). Sleep duration, bedtime, and wake time were obtained from the Garmin Health API. Accuracy was enhanced by fusing wearable data with smartphone usage to refine bed/wake detection. Location and travel: Two Gimbal Series 21 Bluetooth beacons placed at home and office enabled home/office detection; smartphone location and beacon sightings identified travel. Travel days (no beacon sightings and >300 miles from home or >200 miles average distance during day) were excluded (3,261 days total). Home latitude/longitude derived from beacon data. Weather and seasons: Daily localized weather retrieved from World Weather Online API using home ZIP codes: 14 numeric variables (e.g., sunrise/sunset, min/mean/max temperature, humidity, cloud cover, wind speed, visibility, pressure). Principal component analysis reduced weather to three components (temperature, wind, humidity/cloud cover) explaining 71.2% variance. Day length computed from sunrise/sunset. Astronomical season start dates: spring Mar 20, summer Jun 21, fall Sep 22, winter Dec 21. Weekdays only; the 5 weekdays after DST changes were excluded due to transient effects. Analysis: Mixed linear-effects models with participant random intercepts predicted each outcome separately (sleep duration, bedtime, wake time). Nested models: Baseline (no fixed effects); Model 1 adds demographics and traits (age, gender, affect balance, Big Five, organization, supervisory role, latitude, longitude, PSQI, chronotype/MEQ); Model 2 adds season and day length; Model 3 adds weather PCs (temperature, wind, humidity/cloud cover). For bedtime, same-day predictors used; for sleep duration and wake time, previous-day predictors used. Coefficients standardized via Gelman’s method (divide by two SDs). Parametric bootstrap provided p-values and 95% CIs. Multicollinearity assessed via VIF (max GVIF^(1/(2*Df))=2.80 for day length). Model fit assessed with marginal/conditional pseudo-R² and BIC for comparison. R language and packages (dplyr, tidyr, lubridate, lme4, car, jtools, ggplot2, cowplot, sf, sp) used.

Sample: 216 participants; 51,836 weekday observations; approximately uniform distribution across seasons. Sleep duration (Model 3 full):

• Significant predictors: longer day length associated with shorter sleep; each additional hour of day length decreased sleep duration by about 3.6 minutes (P<0.001). Spring associated with 12.6 minutes shorter sleep vs. winter (95% CI −16.8 to −8.4; P<0.001). Other seasons not significant. Weather PCs did not add significant explanatory power beyond seasons.
• Variance explained: marginal R²≈0.02; conditional R²≈0.16; ICC≈0.14. Seasonal Model 2 outperformed Model 1 (F=166.09, P<0.001); adding weather (Model 3) did not improve over Model 2 (F=1.22, P=0.30). Model 2 had lowest BIC. Bedtime (Model 3 full):
• Significant predictors: day length associated with slightly earlier bedtimes (≈1.8 minutes earlier per additional hour of day length; P<0.001), but seasons offset this with later bedtimes in spring (≈+4.2 minutes; P=0.018) and summer (≈+6.0 minutes; P=0.001) vs. winter. Higher temperature PC associated with slightly later bedtime (≈+0.6 minutes per unit; P<0.001). Age associated with earlier bedtime (≈−0.6 minutes/year; P=0.041). Higher Openness associated with later bedtime (≈+11.4 minutes/point; P=0.022). Higher chronotype (more morningness) associated with earlier bedtime (≈−3.0 minutes/point; P<0.001).
• Variance explained: marginal R²≈0.12; conditional R²≈0.30; ICC≈0.22. Seasonal Model 2 > Model 1 (F=12.04, P<0.001); adding weather provided small additional variance (F=4.66, P=0.003). Model 1 had lowest BIC. Wake time (Model 3 full):
• Significant predictors: day length associated with earlier wake times (≈5.4 minutes earlier per additional hour; P<0.001). Seasonal effects vs. winter: fall ≈1.8 minutes earlier (P=0.030), spring ≈8.4 minutes earlier (P<0.001), summer ≈7.8 minutes later (P<0.001). Higher chronotype associated with earlier wake (≈−2.4 minutes/point; P<0.001). Higher Openness associated with later wake (≈+11.4 minutes/point; P=0.004). Higher temperature PC associated with slightly later wake (≈+0.6 minutes/unit; P<0.001).
• Variance explained: marginal R²≈0.09; conditional R²≈0.20; ICC≈0.12. Model 2 > Model 1 (F=185.81, P<0.001). Adding weather (Model 3) improved over Model 2 (F=6.23, P<0.001). Model 2 had lowest BIC. Derived seasonal contrasts using observed day-length differences: spring vs. winter (≈+3.6 h day length) corresponded to ≈−25 min sleep duration, ≈−25 min earlier wake, and ≈−2 min earlier bedtime; summer vs. winter (≈+3.5 h) corresponded to ≈−12 min sleep duration, ≈−11 min earlier wake, and negligible bedtime change. Weather: Temperature had small but significant associations with slightly later bed and wake times; wind and humidity/cloud cover PCs were not significant for outcomes. Overall, effects were small in magnitude but statistically significant, with wake time showing the strongest seasonal influence.

Findings indicate modest but reliable seasonal and weather-related influences on sleep patterns in an industrialized adult population, even after controlling for demographics, location, personality, sleep quality, and chronotype. Sleep duration was shorter and wake times earlier with longer day lengths, particularly in spring; bedtime changes were smaller and partially offset seasonal effects. Temperature modestly delayed bed and wake times, whereas other weather components were not predictive. Results align with prior work showing longer sleep in winter and suggest that seasonal changes in sleep duration may be more strongly driven by wake time adjustments than by bedtime shifts. A plausible mechanism is seasonal modulation of melatonin via light exposure, though artificial lighting and climate control in modern environments may attenuate effect sizes. The study’s continuous, in-situ, year-long sensing and control of confounds help reconcile mixed prior findings, demonstrating that seasonal effects exist but are small in working adults with relatively stable schedules.

Continuous, objective, longitudinal sensing across all seasons shows that seasons—especially day length—and, to a lesser extent, temperature exert small but significant effects on sleep duration, bedtime, and wake time in working adults. The strongest seasonal impacts were observed for wake time and sleep duration (notably in spring). These findings clarify inconsistencies in prior literature by leveraging fine-grained temporal resolution and comprehensive covariate control. Future research should incorporate seasonal and daily environmental variables, examine weekend vs. weekday differences, and consider personal exposure to light and temperature. Applications include optimizing schedules for occupations with rigid hours, designing smart home systems that adapt lighting and temperature seasonally, and informing environmental controls in isolated or extreme settings (e.g., Arctic stations, submarines, spacecraft).

• Sample generalizability: predominantly U.S.-based, college-educated information workers; few participants from the U.S. West Coast; results may not generalize to other populations, cultures, or job types.
• Selection/compliance bias: exclusion criteria (≥50% weekday data per season) may favor participants with traits related to study compliance (e.g., conscientiousness); some missingness may reflect technical failures.
• Unmeasured confounders: exogenous events (e.g., major news, personal stressors, work pressures, dependents) were not controlled and could affect sleep.
• Schedule characteristics: participants likely had flexible schedules; effects may differ for hourly workers or populations with constrained schedules.
• Environmental exposure: weather and day length were estimated from location data; personal exposure to light and temperature was not directly measured; participants may mitigate environmental variation (e.g., HVAC, clothing).
• Weekdays only: weekend sleep (typically more variable and possibly more sensitive to seasonality) was excluded.
• DST handling: week after DST changes excluded, but broader interactions between DST and seasonal effects were not analyzed.
• Travel detection and weather localization, while robust, may still introduce measurement error.