Everyday movement through nature linked to nighttime cardiac regulation

Chronic stress is a highly and increasingly prevalent issue in urbanised societies^1,2. It can lead to dysregulation of the autonomic nervous system (ANS), which has been associated with several common non-communicable diseases, including cardiovascular disease and depression^3,4,5. The ANS plays a central role in the body’s stress response system, regulating respiratory and cardiovascular function in the face of environmental, physical or psychological demands^6,7, making stress an important contributor to autonomic regulation⁵. Experimental studies show that exposure to natural settings can both improve urban dwellers’ mental health outcomes⁸ and support recovery from stress^9,10,11, suggesting that stress reduction may help explain its mental health benefits. However, experimental studies face a challenge of replicating real-world nature interactions, often relying on virtual environments and/or short exposure periods^12,13. Meanwhile, observational studies relating residential greenspace to reduced stress¹⁴ may misrepresent actual time spent in nature as well as introduce confounding by personal and lifestyle factors^15,16. Together, these limitations highlight the need to examine how real-world day-to-day changes in people’s exposure to natural and urban settings (i.e., within-person differences) are associated with changes in autonomic regulation.

One promising avenue lies in the use of non-invasive indicators of autonomic regulation. The influence of the ANS on the heart, known as cardiac autonomic regulation, can be assessed through various heart rate (HR) indices. These include resting heart rate (RHR), the number of heartbeats per minute at rest, which is a risk factor in cardiovascular disease¹⁷. They also include heart rate variability (HRV), the variation in time between consecutive heartbeats, which reflects interactions between the heart and the ANS¹⁸, with higher HRV generally indicating a healthier nervous system capable of adequately and quickly adapting to changes in physiological demands.

Wearable technologies now enable long-term continuous real-world monitoring of HR indices. Recent observational studies in real-world conditions have used wearables for within-person analyses linking subjective stress with higher RHR and lower HRV^19,20. Separately, dynamic exposure assessment based on Global Position Systems (GPS) has been linked to momentary health outcomes such as depressive symptoms²¹. However, to our knowledge, no study has yet combined these approaches to examine within-person associations between nature exposure and autonomic regulation in everyday life. Such an approach opens the door to addressing several knowledge gaps.

First, the health benefits of nature exposure might depend on the timescale over which exposure is measured. For example, one study suggested that spending at least 30 minutes in nature per week could reduce the population prevalence of depression by 7%²². However, studies focusing specifically on autonomic regulation tend to rely on short-term single-bout nature exposure in experimental settings^12,13, limiting insight into cumulative effects in everyday life.

Second, beyond total exposure, how individuals encounter nature may shape its restorative effects. Growing evidence suggests that active movement (i.e., walking or cycling, also known as active travel) through nature can confer larger benefits than either active movement in non-natural environments (e.g., urban settings) or stationary time in nature^23,24,25. Such benefits may reflect nature exposure, physical activity, or both^23,24, and could also involve processes specific to movement in nature, such as reduced rumination via effortless attentional engagement with absorbing stimuli²⁵. This highlights the importance of investigating both stationary and active nature exposure while accounting for physical activity, using real-world data.

Third, while sex differences in autonomic regulation and their links to health outcomes are well established^26,27,28,29, it remains unclear how nature exposure influences autonomic regulation across sexes in everyday life. A review of studies using residential greenspace as a proxy for nature exposure found that a lack of greenspace appeared more clearly linked to chronic stress among females than males³⁰. This highlights the need to investigate sex as a potential moderator of the relationship between dynamic, real-world nature exposure and autonomic regulation.

We addressed the above-mentioned gaps by leveraging wearable and geospatial data to investigate nature exposure and autonomic regulation in everyday life. We analysed 3224 person-days of GPS and HR data collected over 10 months (February–November 2022) from 45 participants in Gävle, Sweden (Fig. 1a), using a smartphone app paired with Garmin vívosmart 4 smartwatches³¹. We measured three HR indices: RHR and two HRV measurements—the standard deviation of interbeat intervals (SDNN) and the root mean square of successive differences between normal heartbeats (RMSSD). These were quantified during nighttime (01:00–04:00) to reflect resting autonomic regulation. Applying classification and geoprocessing techniques to GPS and land cover data (Fig. 1b, c), we quantified nature exposure and different types of mobility (stationary time versus active movement versus passive movement) in minutes between 08:00 and 22:00 across the preceding day, week and month of each person-day with HR values (Fig. 1d). By examining within-person daily deviations in nature exposure, our analysis isolates the temporal association with resting autonomic regulation, reducing bias relative to designs comparing individuals. We aimed to examine within-person associations between HR indices and daily (1) time in nature, (2) active and passive movement, and (3) stationary time or active movement in nature vs. non-natural environments. We also aimed to explore how associations for nature exposure vary when averaging across different exposure windows (day, week, and month). In all analyses, we evaluated moderation by sex. We hypothesised that nature exposure and active movement on their own and in combination would be negatively associated with RHR and positively with HRV, and that these associations would be more pronounced for females compared with males.

Table 1 presents characteristics of the 45 participants (18 males and 27 females). In total, we analysed 3224 person-days, with a median (quartile 1-quartile 3) of 55 (27–115) nights per participant (Fig. 1e). We obtained relatively more data during spring, with declining numbers over summer and autumn (Fig. 1f, g). All HR indices varied more between than within individuals, with intraclass correlations of 0.76–0.80 (Table 1). Comparisons by sex showed no substantial differences in mean HR indices or median daily nature exposure. However, females engaged in more active movement than males, a difference that was nearly statistically significant (Wilcoxon rank sum: W = 326, p = 0.055).

Time in nature and nighttime heart measures

Across the whole sample, within-person variation in time spent in nature was not associated with night-time HR indices (Fig. 2a, Tables S4 and S5). However, there was evidence of moderation by sex of associations for RHR (F(1, 9.6) = 6.79, p_raw = 0.027, p_adjusted = 0.049) and RMSSD (F(1, 9.6) = 6.20, p_raw = 0.033, p_adjusted = 0.049). Among females, an additional 10 minutes in nature during the day was associated with a 0.09 BPM lower-than-usual RHR (95% CI [−0.16, −0.02]) and higher-than-usual RMSSD (β = 0.16, 95% CI [0.02, 0.30]). The association for SDNN was in the positive direction (β = 0.18, 95% CI [-0.02, 0.38]), but evidence for moderation by sex was weaker than for the other outcomes (F(1, 9.6) = 3.58, p_raw = 0.089, p_adjusted = 0.089). Among males, estimated associations were close to zero.

Active and passive movement and nighttime heart measures

Next, we analysed how active and passive movement in any environment was associated with HR indices (Fig. 2b, Tables S4 and S6). We found no evidence of associations for the full sample, and a possible indication of moderation by sex for all three outcomes (RHR: F(2, 16.7) = 3.25, p_raw = 0.064, p_adjusted = 0.096; SDNN: F(2, 16.7) = 2.46, p_raw = 0.116, p_adjusted = 0.116; RMSSD: F(2, 16.7) = 3.52, p_raw = 0.053, p_adjusted = 0.096). Among females, point estimates were negative for RHR and positive for HRV, while among males, estimates were reversed in sign relative to females.

Active movement in nature and nighttime heart measures

We then analysed how combinations of nature exposure and movement behaviour were associated with HR indices (Fig. 2c, Tables S4 and S7). In the full sample, these combinations were associated with lower-than-usual RHR (F(3, 13.8) = 5.31, p_raw = 0.012, p_adjusted = 0.028) and higher-than-usual RMSSD (F(3, 13.8) = 4.69, p_raw = 0.018, p_adjusted = 0.028). An additional 10 minutes of active movement in nature during the day was associated with a 0.27 BPM lower-than-usual RHR (95% CI [−0.41, −0.12]) and higher-than-usual RMSSD (β = 0.58, 95% CI [0.20, 0.96]). The overall test for associations between exposure-movement combinations and SDNN was weaker than for the other outcomes (F(3, 13.8) = 3.01, p_raw = 0.066, p_adjusted = 0.066), with the coefficient for active movement in nature estimated to β = 0.58 ms (95% CI [0.16, 1.00]). Across all three outcomes, we found no evidence of moderation by sex. Estimated associations for active movement in nature among females mirrored those in the whole sample in magnitude, while estimates for males pointed in the same direction but were closer to zero.

There were no indications of improved HR indices from spending more stationary time in nature or from active movement in non-natural environments.

Cumulative weekly and monthly nature exposure

When averaging over the past week or month rather than the previous day, there was no clear evidence of associations or sex moderation effects after adjustment for multiple testing, either for total nature exposure (Table S8) or for active movement in nature (Table S10). Point estimates for total nature exposure and active movement in nature were generally larger in magnitude at longer timescales, particularly among females. We report coefficient estimates in the supplementary information (Tables S9 and S11), although these should be interpreted cautiously and not as evidence of effects.

Sensitivity analyses

To confirm the robustness of our results we performed four sensitivity analyses (SI section 4): 1) We only included participants with at least 10 measurement days rather than at least five, 2) we only included days where the duration of GPS points summed to at least one hour, 3) we included GPS points closer than 50 m to the home coordinate, and 4) we removed adjustment for daily physical activity. We found the results of these sensitivity analyses to align with the main analyses on the whole. The second analysis provided stronger evidence of moderation by sex for associations between active and passive movement and RHR and RMSSD, respectively (Table S13). The fourth sensitivity analysis without adjustment for daily physical activity yielded similar patterns of association as the main analysis, though Wald test statistics were consistently smaller and no associations were supported after adjustment for multiple testing (Table S15).

In this study, we leveraged long-term wearable and geospatial data to examine within-person associations between night-time HR indices and previous-day daily (1) time in nature, (2) active and passive movement, and (3) stationary time or active movement in nature vs. non-natural environments. We further evaluated sex differences and estimated sex-specific associations, as well as how the associations varied for cumulative exposures across the daily, weekly and monthly timescales. We documented a within-person association between more-than-usual active movement in nature on the one hand and lower-than-usual RHR and higher-than-usual HRV on the other. This implies enhanced parasympathetic activity, extending prior experimental evidence¹⁰ to an everyday context. Analyses of moderation by sex suggested these associations may be more pronounced among females, though the smaller male subsample limits conclusions about sex differences.

In the full sample, daily nature exposure or active movement on their own were not associated with HR indices. However, spending an extra 10 minutes on active movement in nature in a day was associated with a 0.27 BPM lower-than-usual RHR and 0.58 ms higher-than-usual SDNN and RMSSD the following night. For perspective, this indicates that a 30-minute nature walk was followed by a reduction in nighttime RHR similar in magnitude to next-day increases observed after intense workouts³² or wheezing episodes among adults with allergic rhinitis³³. Our observed HRV associations with active movement in nature were relatively small: substantially larger increases in SDNN and RMSSD have been reported after exercise interventions spanning several weeks³⁴, or in the short term immediately after practicing paced breathing³⁵. Our results are consistent with a meta-analysis of nature walk experiments¹⁰. Direct comparison of effect sizes is limited as such experiments typically measure HRV directly after a 10–30-minute predefined walk, whereas our study relied on naturally occurring behaviour with HRV assessed the following night, several hours after the most recent bout of active movement in nature. Despite these differences, the similarities in direction of associations could indicate that the effects observed in experimental studies persist beyond the immediate exposure and into the nighttime, though attenuated in magnitude.

Among females, but not males, both nature exposure (irrespective of movement) and active movement (irrespective of nature exposure) were associated with lower-than-usual RHR and higher-than-usual HRV. These associations appeared driven by active movement in nature, with no indications of females benefitting from stationary time in nature or from active movement in non-natural environments. One possible explanation for observing associations in females but not males is that more females (n = 27) than males (n = 18) participated, providing greater statistical power for detecting associations among females. The smaller number of males and the particular lack of young and low-income males likely contributed to wide confidence intervals for this group. We therefore cannot conclude that males do not benefit from active movement in nature. The marked income imbalance between sexes (41% of females vs 5.6% of males in the lowest income category) also means that we cannot fully disentangle sex differences from income-related differences in stress exposure, nature access, or health behaviours. However, the observed results align with prior evidence that stress reduction effects in nature may be stronger among females. For example, walking in nature was found to reduce amygdala activity, which is linked to activation of the stress system, in females but not males³⁶, and a lack of residential greenspace was observed to be associated with chronic stress among females to a larger extent than among males³⁰. Like in other studies³⁷, females in our sample engaged more in active movement than males, and in a nature-abundant city-region, this might have translated into greater opportunities for active movement in nature. The social context and activities associated with nature exposure may also have differed between sexes in ways we could not capture. Notably, unlike in prior studies²⁸, females did not exhibit lower HRV than males. This is noteworthy, as females tend to report higher stress levels^38,39 and our sample featured more low-income earning females than males, with financial strain being a well-established source of chronic stress. That we did not observe lower HRV among females despite this income imbalance raises the possibility that active movement in nature played a role in autonomic regulation. This aligns with prior work showing that women report a stronger psychological connection to nature, which is associated with higher probabilities of experiencing restoration in nature⁴⁰. Despite limits to drawing firm conclusions regarding males’ autonomic regulation, this study adds to evidence on nature’s role in female autonomic regulation, although small effect sizes may imply limited practical relevance for health.

Our results suggest that moving through nature is associated with resting autonomic regulation independently of physical activity. While we did not measure the intensity of the active movement itself, we controlled for daily average physical activity in our main analyses and conducted a sensitivity analysis removing this adjustment. If associations with active movement in nature operated primarily through increased physical activity, removing this adjustment should have strengthened the associations. Instead, associations were attenuated throughout (Table S15), suggesting that physical activity acts primarily as a confounder rather than a mediator in our data. This finding strengthens confidence that the observed associations are unlikely to be explained by physical activity during active movement and may instead reflect other processes. Reduced rumination during movement through natural settings is one possibility²⁵. Interestingly, an experiment found that walking in nature helped coping with real-life stress, whereas merely viewing nature was only effective for an induced stressor unconnected to rumination²³. Research on green exercise also suggests that physical activity in natural settings yields greater psychological benefits than comparable activity in non-green environments⁴¹, while evidence for greater physiological benefits remains inconclusive⁴². As our sensitivity analysis provided evidence against physical activity accounting for the associations, coping with stress through active movement in nature remains as a possible contributing factor to our results, although we could not test this pathway directly.

When averaging exposure over the preceding week or month instead of the past day, point estimates were generally larger in magnitude, particularly among females, but there was no clear evidence of associations after adjustment for multiple testing. While the roughly two- to threefold larger point estimates could indicate that effects accumulate over days, substantially wider confidence intervals, likely reflecting the influence of data gaps, uncertainty about the timing of exposure, and limited sample size, preclude firm conclusions. While inconclusive, the pattern warrants investigation in future studies with greater statistical power.

A limitation of the study, in addition to those mentioned above, concerns data quality challenges inherent in smartphone-based exposure assessment. Smartphone-based adaptive sampling of GPS data, while necessary for long-term monitoring, entails practical issues such as interruptions in background tracking, battery saving features, and signal loss, leading to data gaps and difficulties distinguishing true stationary activity from missing data. These data gaps likely result in underestimation of nature exposure and may have attenuated associations for active movement in nature, as well as obscured effects of stationary time in nature, where prolonged lack of movement can resemble missing data and may partly explain the absence of associations for this variable. We view this as an inherent trade-off in long-term smartphone-based research: reduced measurement precision on any given day is a notable limitation, whereas the large number of repeated measures within individuals enables within-person analysis that would not be possible with shorter, more precise monitoring. Furthermore, to maximise participation, individuals who already owned a compatible Garmin device were allowed to connect their own device, a pragmatic choice that may have introduced variability in measurement accuracy across devices, although our focus on within-person deviations makes this less of a concern. Finally, the association between active movement in nature and autonomic regulation the following night may be explained by various (interrelated) mediators, including reduced stress, improved sleep, or more efficient recovery processes. However, direct measurement of these variables was outside the scope of the study, and the potential mechanisms involving them should be further investigated. Similarly, it was outside the scope of the study to measure potentially confounding factors that might vary day-to-day (such as illness or daily obligations) or over intermediate timescales (such as fitness level, BMI, smoking or medication use), entailing a possibility of residual confounding due to non-measured factors.

The key strength of this study is that, to our knowledge, it is the first to integrate smartphone-based GPS data with smartwatch-derived HR measures for tracking everyday mobility, environmental exposure and autonomic regulation. Furthermore, the high resolution and extensive duration of our data allowed a solid within-person day-to-day analysis, providing more realistic estimations of associations over an extended period and reducing the bias of physiological and lifestyle factors, which are important in studies of determinants of HRV⁴³. Finally, we applied classification and geoprocessing techniques to the GPS data in combination with land cover data to obtain an ecologically valid account of everyday active movement, passive movement and stationary time, in nature and non-natural environments. This approach not only ensured realistic exposure estimates but also allowed us to disentangle the roles of stationary time and active movement in nature.

This study adds to the growing evidence base supporting urban designs that enable urban dwellers to move through and be immersed in natural surroundings, providing to our knowledge the first within-person real-world evidence linking such movement to autonomic regulation in daily life. We documented a within-person association between more-than-usual active movement in nature and lower-than-usual night-time RHR as well as higher-than-usual HRV, both indicating a better autonomic regulation. Neither time in nature nor active movement on their own were associated with night-time HR indices in the full sample. At the weekly and monthly timescales, point estimates were larger than at the daily timescale but there was no clear evidence of associations after adjustment for multiple testing. Disaggregated by sex, associations for total nature exposure, active movement, and active movement in nature were observed among females but not males. While this suggests that active movement in nature might support autonomic regulation in everyday life among females, it does not establish that sex differences exist (given the modest sample size, particularly for males), and potential such differences warrant further investigation. Active movement in nature constitutes a low-cost and—where the environmental context permits—generally accessible behaviour that is easy to integrate into routine. As all participants lived in a small nature-rich city-region, further larger studies in more diverse populations and settings are needed to understand the generalisability of these findings, whether small daily associations accumulate into practically meaningful effects over time and whether sex differences exist in these associations. The methods we applied for linking long-term, real-world repeated measures of location and HR data can be used to address this question in future research.

Study population and setting

The study was conducted over 15 months, from 1 September 2021 to 30 November 2022, in Gävle municipality, Sweden (population: 104,000 in 2025). The municipality spans roughly 20 km east-west and 80 km north-south, with over 95% of its area non-built-up and mostly forested. The city of Gävle is situated on the Baltic Sea, about 150 km north of Stockholm, covering about 3% of the municipality’s area while containing about 85% of its population. The city area within which participants generally spent most of their time includes a city centre with mostly 3–5 story buildings surrounded by modernist apartment suburbs from the 1960s–1970s, intermixed with low-density detached housing. About 45% of the city area is constituted by natural land covers. Gävle has a humid continental climate, with winter daily maxima often below freezing, summer maxima ~20–25 °C, and moderate precipitation all year round.

Participants were recruited among inhabitants of Gävle in the summer and autumn of 2021 through convenience sampling, with information about the study disseminated through the University of Gävle’s web page and social media channels. Inclusion criteria were: aged 15–79 years, residence in Gävle municipality, ability to speak Swedish or English, absence of dementia and cardiovascular disease, and no mobility limitations preventing outdoor activity during daily life. Prospective participants provided informed consent in compliance with the study protocols approved by the Swedish Ethical Review Authority (application number 2021–02212).

GPS and smartwatch data

The MyGävle smartphone application used for data collection has been extensively described in a previous paper²⁷. Prospective participants downloaded the app and provided in-app informed consent before registering for the study. MyGävle continuously collected GPS data from 1 September 2021 to 30 November 2022. To facilitate long-term tracking, the application used the phone’s accelerometer for adaptive sampling, recording GPS locations every three seconds when in motion, but stopped recording when the phone transitioned to being stationary.

Participants could opt in to contribute HR data, either by purchasing a Garmin vìvosmart 4 smartwatch offered through the study at a discounted price or by connecting their own Garmin smartwatch (any model) to MyGävle. Even entry-level Garmin smartwatches have demonstrated acceptable accuracy for measuring heart rate during sedentary activity⁴⁴. MyGävle collected interbeat interval (IBI) data in milliseconds using a module developed with the Garmin Companion Software Development Kit (SDK) that was added in a version update of the app on February 10th 2022. Additional variables compiled by the smartwatch were accessed through Garmin’s Health Application Programming Interface (API).

Heart rate characteristics

We quantified HR indices for the night period (01:00–04:00) when most participants were likely at rest, since HRV at rest reflects autonomic regulatory capacity⁵, and smartwatch HR measurements are most accurate during rest⁴⁴. To ensure that night-time waking activity did not influence the analysis, we removed nights with more than 100 steps recorded by the smartwatch between 01:00–04:00 (table S1).

To calculate HR indices, we used five-minute windows (36 per night) of IBI, a commonly applied period for short-time HRV measurements⁴³. We removed windows with <90% data completeness and then nights with <24 windows of data (i.e., two hours). For the remaining nights, RHR was quantified by first computing the mean HR in each five-minute window (({{HR}}_{{mean}}=60* 1000/{{IBI}}_{{mean}})), and then retaining the minimum such HR, as RHR should be measured over a short period while maximally at rest⁴⁵.

We calculated SDNN and RMSSD as our HRV variables, as these are two commonly applied complementary time-domain measures of HRV⁴³. SDNN is the standard deviation of all IBIs over a time interval (i.e., 5 min) after abnormal beats have been removed. In short-term recordings such as those used here, SDNN is predominantly influenced by parasympathetic (vagal) tone and breathing patterns¹⁸. RMSSD is obtained by calculating first the differences in milliseconds between successive IBIs over a time interval, then the mean of those differences, and lastly the square root of that mean. While correlated with SDNN, RMSSD is relatively more influenced by vagal activity and less by respiration¹⁸. Five-minute SDNN and RMSSD values were computed using the R package RHRV⁴⁶, before averaging across the night.

Movement and nature exposure

We quantified combinations of movement (stationary time vs. active movement vs. passive movement) and time in nature vs. non-natural environments in minutes per day between 08:00 and 22:00 (i.e., normal waking hours among Swedes⁴⁷), through a four-step process: 1) GPS data preprocessing, 2) activity classification, 3) nature classification, and 4) summing durations.

1) Each GPS point was assigned a start and an end time, set at the midpoints between the point’s timestamp and the timestamps of the preceding and following points, respectively. After this, we removed observations based on low accuracy (>150 m), long distance jumps (>1 km from the previous point) and unreasonable speed (>300 km/h). Nighttime points (i.e., starting after 22:00 and ending before 08:00) were removed. Points overlapping 08:00 or 22:00 were truncated to start at 08:00 or end at 22:00, respectively. Durations of the remaining points were then calculated by subtracting the start time from the end time.

2) GPS data were divided into stationary points (no spatial displacement), movement points and breaks (i.e., periods of low-quality/missing data). Due to the adaptive GPS sampling, we included in the stationary category both stops (i.e., sequences of active sampling with points spatially close to each other) and stays (i.e., periods without sampling temporally bounded by two spatially adjacent points). Candidate stops were identified through the DBSCAN-TE algorithm, which we applied in two stages, as this has been found to improve density-based clustering (see SI section 2 for details). In a post-processing step, we then filtered stop clusters based on directional changes and displacement distances (SI section 2). Stays and breaks were identified with a rule-based approach based on prior studies (see SI section 2 for details). We classified two points separated by <150 metres and >3 minutes as a stay, and those separated by >150 metres and >10 minutes as a break. Additionally, two points separated by more than 12 hours were, even if within 150 m of each other, classified as a break. Each remaining sequence of at least three points, all separated by <3 minutes were classified as a trip⁴⁸. We then identified active and passive trips following predefined rules informed by previous research (Table S2).

3) Nature areas were defined using a 10-metre resolution national land use land cover map from the Swedish Environmental Protection Agency (https://www.naturvardsverket.se/verktyg-och-tjanster/kartor-och-karttjanster/nationella-marktackedata/). To focus on routine daily behaviour, we cropped the map to the extent of Gävle municipality and re-coded its 25 land cover classes into nature (20 classes) vs. non-nature (5 classes) (SI section 2). As GPS points have a spatial margin of error, some previous studies have reduced exposure misclassification by smoothing underlying raster layers through kernel density estimation, thereby making exposure assignment less sensitive to positional error⁴⁹. As the margins of error of GPS points in our dataset were mostly between 5 and 25 metres⁵⁰, we applied a 25-m radius kernel density estimation to the nature map, with each 10-m cell obtaining a value 0–21 reflecting the number of nature cells within 25 metres. Cells with the value 21 were classified as nature, and the remaining cells as non-nature, with GPS points obtaining the class of their overlapping cell. Thus, exposure to nature required being surrounded by natural land covers within a 25-m radius, which is sufficiently spatially detailed to capture small urban parks but not individual green elements within the streetscape.

4) For each person-day and activity-exposure combination, we summed up the duration of GPS points further than 50 m from the home location. To reduce the bias of nature exposure estimations from indoor activities among those living surrounded by nature, points close to home were excluded in the main analysis but included in a sensitivity analysis.

Physical activity

We adjusted for daily physical activity to focus on the effects of active movement in nature on HR outcomes that are independent of physical activity. Decisions to engage in physical activity can shape mobility choices, while mobility choices can also influence activity levels, meaning that physical activity acts as both a confounder and a mediator. Adjusting for overall daily physical activity, therefore, helps rule out effects driven solely by decisions to engage in physical activity but may also attenuate effects of active movement that operate through physical activity. We used metabolic equivalents of task (MET) as a physical activity measurement. MET values, which Garmin estimates from user biometrics and HR data, and wear time for each 15-minute period between 08:00 and 22:00 were obtained through the Garmin Health API. Wear time was assessed across the full day, and only days with at least 10 hours of wear were included, consistent with physical activity research standards. MET values were then averaged, weighted by each 15-minute period’s wear time to obtain a daily mean MET.

Statistical modelling

We examined within-person associations between exposure and/or movement across a day and three HR indicators the following night, in three modelling steps: (1) time in nature, (2) active and passive movement mutually adjusted, and (3) six combinations of nature exposure (yes/no) and movement (stationary/active movement/passive movement) mutually adjusted. All predictors were measured in minutes per day and were rescaled so that one unit equalled 10 minutes. At each step, Wald tests were used to determine whether predictor variables were jointly associated with autonomic regulation, with false discovery rate (FDR) correction (Benjamini-Hochberg, q = 0.05) applied across the three outcomes. FDR correction addresses multiple comparison issues in the presence of dependence⁵¹, such as in our case when outcomes are correlated. We report unadjusted coefficient estimates to characterise the pattern of associations. This approach balances Type I error control at the model level while preserving power to interpret association patterns.

We used mixed-effects models with a random intercept for participants, isolating within-person associations by disaggregating the exposure, movement, and movement-exposure combination variables into person-mean values (that remove confounding from stable factors varying between individuals) and person-mean–centred daily deviations, representing each day’s difference from a participant’s own average. To enable detecting within-group effects, we only included participants with at least five observations.

At each step, in addition to associations for the full sample, we assessed effect modification by sex by fitting a model including an interaction term between sex and each exposure/movement variable.

In addition to the daily models, we estimated associations of variation in cumulative exposure-movement combinations over a week and a month, averaged over the number of days with measurement. To obtain weekly and monthly values, we calculated for each night the number of days with exposure-movement data going back 7 and 30 days in time, respectively. Among nights with at least 4 and 15 such observation days, respectively, we computed means across those observation days. We tested the joint contribution of these exposures using Wald tests, with FDR correction applied across the nine tests (3 outcomes × 3 exposure windows).

All models controlled for individual factors: sex (categorical: female/male), age (numeric) and monthly individual income (categorical: less than SEK 30,000/SEK 30,000–50,000/greater than SEK 50,000), as well as time-varying factors: time of year (numeric: days before/after December 21), weekend (binary: yes/no) and physical activity, i.e., mean MET the preceding day (continuous).

We screened our models for residual autocorrelation using the Durbin-Watson test and heteroscedasticity using the Breusch–Pagan test. As these tests confirmed the presence of both autocorrelation and heteroscedasticity (Table S3), we computed cluster-robust standard errors with CR2 correction, which have been shown to be robust to heteroscedasticity, autocorrelation, and non-standard error distributions⁵².

To confirm the robustness of our results we performed four sensitivity analyses: 1) We only included participants with at least 10 measurement days (rather than at least five), 2) we only included days where the duration of GPS points summed to at least one hour, 3) we included GPS points closer than 50 m to the home coordinate, and 4) we removed adjustment for daily mean MET.