United Nations Sustainable Development Goal 3 highlights that ensuring healthy and happy lives is critical to all people (United Nations, 2021). However, physical inactivity hinders the achievement of this goal (Salvo et al., 2021). Approximately one-fourth of adults worldwide do not meet the physical activity (PA) level recommended by the WHO (Guthold et al., 2018). This leads to higher risks of both physical diseases (Sallis et al., 2021; Silveira et al., 2022) and mental illness (Schuch et al., 2018). Although people may engage in different types of PA (e.g., transport and structured exercise), the amount of leisure PA, a critical component of individuals' total PA, is often inadequate (Strain et al., 2020). In the USA, 74% of men and 81% of women lack sufficient leisure PA (Piercy et al., 2018). In China, 77% of adults fail to meet the leisure PA recommendation (Tian et al., 2016). Therefore, promoting leisure PA should be prioritized to ensure citizens’ health and happiness.

Besides sociodemographic attributes, the amount of leisure PA is associated with urban environment attributes (Gidlow et al., 2019; Kajosaari and Laatikainen, 2020). Therefore, improving the urban environment could be a promising strategy to promote leisure PA (Cerin et al., 2022). Although previous studies have examined the connection between the environment and leisure PA, a few questions have remained mostly unanswered in the literature.

First, which is more important to leisure PA, the built environment or the natural environment? The built environment refers to human-made spaces (Ewing, 2005). People residing in compact neighborhoods, characterized by high density, mixed land use, connected streets, and high accessibility to destinations and transit (Ewing and Cervero, 2010), tend to have more leisure PA than those living in sprawling neighborhoods (Chen et al., 2021). This is because a compact built environment provides diverse destinations (e.g., sports facilities) and walkable streets for leisure PA (Bibri et al., 2020; Stevenson et al., 2016). The natural environment refers to the urban environment that provides natural features, particularly blue and green spaces, and exposure to pollution (Bird et al., 2018; Frank and Wali, 2021; Hartig et al., 2014). People usually have a higher level of leisure PA in the environment with more green spaces (Jansen et al., 2017) and better air quality (Tainio et al., 2021).

Although both built and natural environments contribute to leisure PA (Bailey et al., 2018; Bird et al., 2018), they are often at odds with each other (Soga and Gaston, 2016; Xie et al., 2019). Neighborhoods that are closer to nature are often less compact, while compact neighborhoods often have fewer natural resources. Therefore, although a compact built environment may promote leisure PA by improving walkability and destination accessibility, it may hinder leisure PA due to the loss of contact with the natural environment, and vice versa. This dilemma makes urban planners wonder which intervention of the built or natural environment is more efficient in encouraging people to engage in leisure PA. However, there is no clear evidence on the relative importance of built and natural environments to leisure PA.

The second question is whether environmental attributes have nonlinear and threshold associations with leisure PA. Previous studies have often assumed that the environment is linearly related to leisure PA. Recently, there has been a growing interest in challenging this premise. Some scholars have explored nonlinear and threshold associations of environmental attributes with PA (Cerin et al., 2022; Wali et al., 2021). A nonlinear effect means that the association between an environmental attribute and PA is not constant but varies by the value of the attribute. For example, several studies have shown that population density has an inverse U-shaped association with walking (Cerin et al., 2022; Lu et al., 2019). A threshold effect, a type of nonlinear effects, denotes that the impact of an attribute on PA changes drastically after the attribute passes a certain value (Galster, 2018). For example, people may not engage in PA unless an adequate number of PA facilities are near their houses (the lower threshold) (Van Dyck et al., 2013), or people may stop participating in PA when air pollution exceeds a particular level (the upper threshold) (An et al., 2019). A study from Xi'an has shown that children's PA decreases as the distance to the nearest park increasing from 0 to 1.2 km, but it has no additional effect beyond the threshold (Huang et al., 2021). However, few studies have considered the nonlinear and threshold associations of the environment with leisure PA.

The third question is whether environmental attributes in different spaces have similar or differing effects on leisure PA. Previous studies have mostly focused on the association between the residential environment and leisure PA, but few studies have considered the environment in workplace neighborhoods that may also influence PA (Lin et al., 2020; Zhu et al., 2020). People are likely to engage in leisure PA in workplace neighborhoods during the lunch break and after work (Valerie and Daniel, 2016), given that they spend most of their waking hours in workplaces (Dannenberg et al., 2005; Sun et al., 2022a). Moreover, it remains unknown whether an environmental attribute that promotes leisure PA in residential neighborhoods facilitates or discourages leisure PA in workplace neighborhoods. On the one hand, the same environmental attribute in different spatial contexts may have similar associations with leisure PA (i.e., complementarity effects). This hypothesis is supported by the socioecological model, which stresses that a multipronged approach (i.e., a similar intervention in different contexts) is important to promote leisure PA (Heath et al., 2012; Sallis et al., 2015), implying that environment interventions in different places may have cumulative contributions to leisure PA. On the other hand, an environmental attribute in one place (e.g., residences) may have a differing association with leisure PA from the same attribute in another space (e.g., workplaces) (i.e., substitution effects). The ActivityStat hypothesis states that a person's total PA is stable, and with an increase in PA in one domain, PA in another domain decreases (Gomersall et al., 2013). Following this logic, people tend to have a fixed budget for leisure PA if transport PA is stable. Hence, when people engage in more leisure PA in one place, they may participate in less leisure PA in another place. For example, adolescents who engaged in more PA at school engaged in less PA at home and at other locations (Carlson et al., 2017). Because the workplace of some people is within others' residential neighborhoods, the opposite associations may lead to ineffective environmental interventions for the population as a whole. Therefore, it is interesting to compare and contrast the associations of the same environmental attributes with leisure PA in different spatial contexts so that planning scholars and practitioners can better understand the substitution and complementary effects and design context-specific interventions to promote leisure PA efficiently.

This study aims to determine the relative importance of the built and natural environment to leisure PA and investigate how these environments are associated with leisure PA in residential neighborhoods and workplace neighborhoods (called residences and workplaces, respectively, hereafter for simplicity). Using a sample of 1049 respondents collected in Shanghai in 2018 and 2019, we apply gradient boosting decision trees (GBDTs) to relax the linear assumption commonly adopted in the literature, and address three research questions: (1) Which environmental aspect (i.e., the natural or built environment) of residences and workplaces is more important to leisure PA? (2) Do environmental attributes have nonlinear and threshold associations with leisure PA? (3) Do environmental attributes of residences and workplaces have substitution or complementary effects on leisure PA?

This study contributes threefold to the literature. First, it identifies relative contributions of built and natural environments to leisure PA. In both residences and workplaces, the built environment is found to make a larger contribution to leisure PA than the natural environment. Second, considering nonlinear associations between the environment and leisure PA produces more accurate results, which inform planners of nuanced interventions to promote leisure PA. Most environment attributes are found to have nonlinear and threshold associations with leisure PA. Third, this study confirms the existence of both substitution and complementary effects of environmental attributes in different spatial contexts on leisure PA. In certain ranges, the land use mix and population density of residences and workplaces are shown to have differing effects on leisure PA, whereas the distance to the city center and the water area of residences and workplaces are associated with leisure PA in the same direction.