Seasons, weather, and device-measured movement behaviors: a scoping review from 2006 to 2020

Background This scoping review summarized research on (a) seasonal differences in physical activity and sedentary behavior, and (b) specific weather indices associated with those behaviors. Methods PubMed, CINAHL, and SPORTDiscus were searched to identify relevant studies. After identifying and screening 1459 articles, data were extracted from 110 articles with 118,189 participants from 30 countries (almost exclusively high-income countries) on five continents. Results Both physical activity volume and moderate-to-vigorous physical activity (MVPA) were greater in summer than winter. Sedentary behavior was greater in winter than either spring or summer, and insufficient evidence existed to draw conclusions about seasonal differences in light physical activity. Physical activity volume and MVPA duration were positively associated with both the photoperiod and temperature, and negatively associated with precipitation. Sedentary behavior was negatively associated with photoperiod and positively associated with precipitation. Insufficient evidence existed to draw conclusions about light physical activity and specific weather indices. Many weather indices have been neglected in this literature (e.g., air quality, barometric pressure, cloud coverage, humidity, snow, visibility, windchill). Conclusions The natural environment can influence health by facilitating or inhibiting physical activity. Behavioral interventions should be sensitive to potential weather impacts. Extreme weather conditions brought about by climate change may compromise health-enhancing physical activity in the short term and, over longer periods of time, stimulate human migration in search of more suitable environmental niches. Supplementary Information The online version contains supplementary material available at 10.1186/s12966-021-01091-1.

Two seminal reviews of research on seasonality, weather, and physical activity across the lifespan were published over a decade ago [10,11]. Physical activity was typically greatest during spring and summer and lowest during winter, but regions with more extreme weather conditions sometimes yielded different conclusions. For example, a study conducted in Galveston, Texas, where the average temperature during summer months is over 28°C (82°F), revealed lower levels of physical activity in the summer than in winter [12]. However, the general pattern of seasonal trends demonstrated that people accumulated greater levels of physical activity in warmer and more arid conditions. Weather conditions were also found to have differing impacts on physical activity between sub-groups in the population [10,13,14]. For instance, physical activity was not associated with wind gusts for most individuals, but individuals with lower body mass were less active in the presence of stronger wind gusts than individuals with higher body mass [15].
The social context for research on weather and physical activity changed in three significant ways around the time of the seminal Tucker and Gilliland review [10]. First, public interest in weather and health increased following the 2007 Nobel Peace Prize that was awarded jointly to the Intergovernmental Panel on Climate Change and former US Vice President Al Gore [16]. As Earth's surface temperature rises, extreme weather conditions will increase air pollution and ultraviolet radiation exposure, increasing risk for cardiovascular and lung diseases as well as cancer [17]. Second, the mobile and wearable technology industries experienced major disruptions in 2007 due to the launch of the Apple iPhone and the founding of Fitbit [18,19]. The widespread adoption of mobile technologies such as smartphones and wearable activity monitors enabled researchers to monitor both physical activity and location-specific weather indices in real-time. Third, sedentary behaviorwaking activity conducted in a seated or reclined posture involving low energy expenditurehas emerged as a distinct behavior that has important health consequences independent of physical activity levels [20,21].
Despite summarizing over 60 studies with over 300, 000 participants from approximately 18 countries, these seminal reviews possess several limitations. First, although they both examined seasonal differences in physical activity, most findings at the time focused on a relatively narrow range of specific weather indices with temperature and precipitation being most common. Understanding associations between additional weather indices and physical activity could both explain seasonal differences and facilitate the development of just-in-time interventions that leverage information from short-term weather forecasts. Second, weather and physical activity data were aggregated into person-level summary measures, yet both weather and physical activity are dynamic. Understanding the timing of and changes in weather conditions is essential because current weather conditions are likely to have a more immediate influence on physical activity than average weather conditions. Third, both reviews focused on a general physical activity outcome. Physical activity can be quantified as total volume (to represent energy expended) or the duration of activities completed at specific intensities (to represent time allocated to different effort levels). Sedentary behavior may also be impacted as weather conditions alter people's activity choices. Additional reviews published since these seminal reviews have been limited by incomplete search strategies and a focus on narrow segments of the population [14,22]. In light of these limitations and the broader context described earlier (increasing public interest in climate, advances in mobile technology, the emergence of sedentary behavior), an updated review of weather and movement behavior, including both physical activity and sedentary behavior, would be a valuable contribution.
A scoping review was conducted to examine associations between device-based measures of physical activity, sedentary behavior and a broad array of weather-related phenomena at different levels of specificity, ranging from seasons (e.g., spring, summer, fall, winter) to specific weather indices (e.g., humidity, precipitation, temperature). A scoping review was selected over a systematic review or meta-analysis based on the breadth of weather indices available and the need to both analyze the available evidence and identify existing knowledge gaps [23].

Search strategy
PubMed, the Cumulative Index of Nursing and Allied Health Literature (CINAHL), and SPORTDiscus electronic databases were searched from January 1,2006 to October 31, 2020. This date range was chosen to capture all research since the end of the search period used by Tucker and Gilliland [10]. Three databases were selected based on their likelihood of including both environmental and health behavior data. Three main subject categories were included in our searches: movement-related behavior, movementrelated behavior measurement (i.e., intensity, volume), and weather. Movement-related behavior search terms related to physical activity or sedentary behavior were combined with "or" statements. These terms mirrored the search terms used to compile the literature for the 2018 Physical Activity Guidelines Advisory Committee [24]. Measurement search terms related to technologies commonly used to measure physical activity were combined with "or" statements (e.g., accelerometer or pedometer). Weather search terms related to seasonality or specific weather indices were combined with "or" statements (e.g., seasons or temperature or humidity or precipitation). The search strategies for each database used are available in Appendices 1, 2, and 3. The searches were restricted to articles that (a) were written in English, (b) examined human subjects only, and (c) included empirical studies only (i.e., no review papers). This search strategy was constructed in consultation with a trained reference librarian and search specialist at The Pennsylvania State University. The review protocol was not preregistered.

Selection process
Articles were included if (a) physical activity or sedentary behavior was an outcome variable of interest, (b) physical activity data were collected using device-based measurements (e.g., accelerometer, pedometer), and (c) results involved associations or differences between either movement-related behavior and either seasonality or weather indices. Articles were excluded if (a) studies were not published in English, (b) samples included non-human subjects, (c) results were limited to prevalence rates or other descriptive data, or (d) physical activity data were collected using self-report measures exclusively.
Titles and abstracts were independently reviewed in a blinded manner by coders trained by the first author using the eligibility criteria described above. The first author exported citations for each article found in the PubMed, CINAHL, and SPORTDiscus searches, and uploaded these citations into Rayyan. Rayyan is a web application that is used to facilitate collaborative screening of titles and abstracts for reviews [25,26]. Each coder accessed their assigned articles via the Rayyan web interface, reviewed titles and abstracts, and recorded decisions to include or exclude each article. Figure 1 summarizes study selection [27].

Data extraction
Articles containing studies that met inclusion criteria during full-text review were advanced for data extraction. Prior to data extraction, the first author and four trained coders used a standardized coding guide to code three identical papers that were eligible for data extraction. The coding guide was developed to ensure that sufficient descriptive information was collected to properly characterize associations of interest.
The coders then met to compare codes and discuss disagreements to promote consistency for final extraction. After this calibration exercise, each coder independently extracted data from the remaining articles using the standardized coding guide. Prior to analysis, the first author reviewed each article to ensure that the necessary data was extracted in a manner consistent with the coding guide.
A copy of the coding guide is available in supplementary online files. Extracted sample characteristics included age, sex, education, race, and the country where the data were collected. If countries included multiple diverse climate zones, the region of the country was also collected. Physical activity, sedentary behavior, weather indices (temperature, precipitation, wind speed, photoperiod, snow, cloud coverage, humidity, visibility, barometric pressure, windchill, and air quality; see Appendix 4 for definitions), and seasonality were all characterized by assessment timeframe and method of measurement. Research designs were classified as cross-sectional or longitudinal. The following statistics were extracted when available: t-scores, F-scores, correlation coefficients, β coefficients, p-values, effect sizes, and odds ratios. Significance thresholds were kept consistent with the author's prespecified level of significance, and all results were coded having a positive or negative association, or failing to reach significance.

Evidence grading
All of the available evidence from independent samples was observational so all studies were deemed to have a high risk of bias. Instead of rating study-level bias, evidence was graded based on the quantity and consistency of findings when five of more studies were available for a specific comparison or association. A "strong" grade was assigned when a conclusion was based on highly consistent findings related to the direction of a difference or association (or moderately consistent findings when a large number of studies was available). A "moderate" grade was assigned when a conclusion was based on mixed findings but the preponderance of evidence pointed to a consistent direction of a difference or association. A "limited" grade was assigned when a conclusion could not be drawn because of equivocal evidence related to the direction of the difference or association. When fewer than five studies were available, we noted that a grade was not assignable. This grading system was drawn from criteria used by the 2018 Physical Activity Guidelines Advisory Committee and adapted to match the state of this literature [24].

Results
A total of 1459 unique articles were identified during the initial search. Following title and abstract screening, 1208 articles were excluded. Full-text review of the remaining 251 articles led to an additional 141 exclusions. A total of 110 articles reporting 144 studies of independent samples were identified as eligible for inclusion in this review. Among those 110 articles, 26 reported two independent samples, one reported three independent samples, and two reported four independent samples to examine physical activity behaviors between sex, age group, or region, or examined both seasonality and weather [28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44]. Table 1 summarizes participant characteristics and study designs for all included articles. In total, articles included 118,189 participants (62.0% female participants, median N = 272, IQR = 85-722) from 30 unique countries on five continents. Figure 2 summarizes the sampling density across the globe. Most data represented western countries such as the United Kingdom, United States, Norway, Australia, Denmark, and Canada. Studies were primarily conducted in countries that the World Bank currently (July 2020) classifies as high income (27/ 30, 90.0%) and three were upper-middle income (3/30, 10.0%) [45]. None were from low-or middle-income countries.
Approximately equal numbers of studies examined weather (77/144, 53.5%) and seasonality (       . Although most studies with accelerometer-based measurements used wearable devices, one study measured physical activity using the accelerometer contained within a smartphone [46]. Physical activity measures included volume (i.e., step counts, total accelerometer counts) and intensity-specific durations. Volume measures represent the total amount of energy expended in physical activity whereas intensity-specific durations represent how individuals allocate their time to more and less effortful forms of physical activity. Figure 3 summarizes the frequency of studies using volume and intensity-specific duration measures in samples of youth, adults, and older adults. Seasonal differences in physical activity were more frequently estimated in youth than adults or older adults. Weather indices associated with physical activity were studied most frequently in youth, and approximately equally in adults and older adults.

Seasonal differences in physical activity and movementrelated behaviors
Moderate evidence indicated mixed findings for physical activity volume between spring and autumn with most studies revealing either greater volume in spring than autumn (5/12, 41.7%) or no differences in volume (6/12, 50.0%). One study reported greater volume in autumn than spring (1/12, 8.3%). Moderate evidence indicated mixed findings for physical activity volume between autumn and winter with some studies indicating no difference (7/12, 58.3%) and others indicating great volume in autumn than winter (5/12, 41.7%). No studies found evidence of greater physical activity volume in winter than autumn. Limited evidence compared physical activity volume in spring and summer were equivocal: six studies indicated no difference in volume between summer and spring (6/ 10, 60%) but the remaining studies were evenly split between greater volume in summer (2/10, 20.0%) and spring (2/10, 20.0%). Limited evidence compared physical activity volume between autumn and summer. Findings generally indicated no difference (5/7, 71.4%) with the exception of one study indicating greater volume in autumn than summer (1/7, 14.3%) and another study indicating greater volume in summer than autumn (5/7, 14.3%).
Moderate evidence indicated mixed findings from comparisons of MVPA between spring and autumn, either indicating no differences (7/12, 58.3%) or greater MVPA in spring than autumn (5/12, 41.7%). No study demonstrated greater MVPA duration in autumn compared to spring. Based on limited evidence, LPA generally did not differ between spring and autumn (5/7, 71.4%); however, one study indicated greater LPA in spring than autumn (1/7, 14.3%) and another study indicated greater LPA in autumn than spring (1/7, 14.3%). Moderate evidence indicated that MVPA duration did not differ between summer and autumn in five studies (5/5, 100%). A grade was not assignable for LPA comparisons between summer and autumn.

Specific weather indices associated with movementrelated behaviors
Weather indices were typically collected from regional weather stations or national institutes (  Associations between temperature and physical activity volume were more often positive (19/38, 50.0%) than null (13/38, 34.2%) or negative (6/38, 15.8%). No studies explicitly tested for curvilinear relations, but four studies noted an inverted-U pattern in which volume was lower during normatively warmer or colder days [38,78,117,125]. Additional negative associations with volume were found when extreme temperatures were present [126,127]. This evidence was graded as strong in favor of an inverted-U relation between temperature and physical activity volume.
Limited evidence indicated that cloud coverage and physical activity volume exhibit a negative (2/5, 40.0%) or null associations (2/5, 40.0%). One study found a positive association between cloud coverage and physical activity volume (1/5, 40.0%). A grade was not assignable to evidence linking cloud coverage and durations of MVPA, LPA, or sedentary behavior.

Discussion
The present review summarized 144 studies from 110 articles with 118,189 participants from 30 countries. It updates conclusions from two seminal reviews based on a total of over 60 studies on over 300,000 participants from approximately 18 countries [10,11]. In those reviews, studies on seasonal differences in physical activity greatly outnumbered studies on weather correlates. Over the past decade, attention has been divided more equally between seasonal differences in and weather correlates of movement behaviors. This review also extended the prior reviews by capturing the spectrum of movement behaviors ranging from physical activity volume to intensityspecific durations to sedentary behavior. Collectively, these reviews establish how key features of the natural environment are linked with a variety of movement behaviors.
Physical activity volume and MVPA duration were the most frequent measures of movement behavior. Consistent with prior work, winter and summer were marked by the lowest and greatest movement behavior, respectively [10,11]. The present review extended those conclusions by documenting a trend in favor of greater physical activity volume and MVPA in spring than autumn. Comparisons of spring vs summer and summer vs autumn largely revealed no differences. Overall, the pattern of seasonal differences in physical activity volume and MVPA duration resembled a sinusoidal pattern and corresponded with both fluctuations in temperature and the waxing and waning photoperiod across the calendar year.
One contribution of the present review was that it examined a broad spectrum of movement behaviors that contribute to physical activity volume. MVPA duration has enjoyed a privileged status in the scientific literature because it has the strongest connections with health benefits [24]. MVPA also exhibited the clearest pattern of relations with seasonality and specific weather indices, including temperature, precipitation, and photoperiod. Thus, MVPA may provide one pathway by which the natural environment gets "under the skin" to affect health [139,140].
Moving the needle on population-level MVPA has proven to be difficult [141,142]. Far more time is spent in LPA than MVPA and LPA is a greater contributor to physical activity volume for most people [143]. Recent work has established unique health benefits from LPA after adjusting for MVPA [144]. As a consequence, LPA is a desirable substitute for prolonged sedentary behavior when MVPA is not feasible. Although a trend for increased LPA in summer and spring compared to winter was observed in the literature, LPA in the spring did not differ from LPA in the summer or autumn. In the interest of understanding how the natural environment facilitates or inhibits movement, this common form of physical activity should be a priority measure in future research examining seasonal differences and weather correlates.
Sedentary behavior was consistently greater in winter than spring or summer. Weather has been cited as a barrier to physical activity that facilitates sedentary behavior [8,9]. This study extended previous work by reviewing how specific weather indices, as opposed to perceptions of the weather, are associated with sedentary behavior. People engage in more sedentary behavior on shorter days (photoperiod), when precipitation is greater, and when temperatures are lower. These findings revealed that weather is likely to be a third variable influencing the entire spectrum of movement-related behaviors.
These findings have two major implications. First, although the studies reviewed here were necessarily observational, they can inform behavioral interventions. Weather conditions that may serve as actual, as well as perceived, barriers to physical activity (e.g., too hot, cold, rainy or snowy, windy, or shorter days). Current or forecasted weather conditions may be useful for providing contextual information about opportunities for activity that could inform just-in-time interventions for movement behaviors. For example, users could be prompted to develop coping plans for exercise in adverse weather conditions and then reminded of those plans when adverse weather conditions were expected. Digital tools could also extend work on person-specific physical activity interventions by learning how to identify userspecific preferred weather patterns for movement behaviors and prompt users to ensure they capitalize on their preferred conditions to be active [145].
At a more general level, the impact of seasonal differences in movement behaviors on the implementation and evaluation of physical activity promotion programs should be considered when interpreting ambulatory behavior changes. Lifestyle physical activity intervention evaluations often last 1-6 months, and few last 12 months, so baseline and follow-up assessments are often conducted during different seasons. Seasonal influences on activity levels are typically allocated to the error term of statistical models but may be informative to include as a covariate or moderator of intervention effects. For example, including season as a moderator could reveal if interventions work better when days are lengthening (winter to summer) or when days are shortening (summer to winter).
Second, climate change is increasing the frequency of extreme weather conditions. Climatic zones that are currently favorable for movement behaviors may become inhospitable, inhibit physical activity, and contribute to health disparities. Some have speculated that climate change will increase migration as people seek to preserve an adaptive environmental niche [146]. Population-level data on movement behaviors could be investigated as a leading indicator of future health or migration due to climate change.
This review had limitations as well. The search was conducted using three databases and limited to English language publications. It is possible that studies were missed if they were published in journals not indexed by PubMed, CINAHL, or SPORTDiscus, or were written in languages other than English. All data were observational so strong causal inferences are not possible. Results were obtained from 30 countries on five continents but western and high-income countries were overrepresented in the data. Low-and middle-income countries are home to over 80% of the world's population and four times as many deaths are attributed to physical inactivity in those countries than in high-income countries [147,148]. Conclusions may not generalize equally well to all regions, climatic zones, or economic strata. Most studies did not report on race or ethnicity so it is unclear how physical, social or cultural differences influence seasonal differences or relations between weather and movement behaviors. Devices were used to obtain measures of physical activity volume and intensity-specific duration primarily MVPA -but some activity types do not lend themselves to accurate measurement with devices (e.g., cycling, swimming) or are sub-optimal with devices attached at the waist (e.g., sedentary behavior). Devices also provide no insights into the specific domains of physical activity (e.g., occupational, transport-related, occupational, domestic) or sedentary behavior (e.g., reading, screen time, socializing, eating). The available literature has focused almost exclusively on aggregated weather summaries during monitoring periods. Yet weather is dynamic so those summaries may not generalize to within-person change processes [149]. Additionally, physiological or psychological differences in environmental tolerances likely exist. Some people, for example, will be more heat tolerant or may simply enjoy running in the rain. Person-specific models of physical activity under different weather conditions could shed light on these dynamics [145]. Finally, the review focused on 11 common weather indices but other indices may also be relevant.