There is emerging evidence of the influence of visual exposure to harmful or helpful health-related behaviours, such as smoking or engagement in physical activity, in the public health literature. These studies tend to focus on the social contagion or normalisation of behaviours through their observed presence in particular places. For example, studies on physical activity indicate that observing others engaging in physical activity in one’s neighbourhood was associated with increased walking and a lack of observing others engaging in physical activity was associated with a negative perception of exercise for individuals [1]. Likewise, research on observed smoking indicates an association between the frequency of observed smoking in some locations and the perception that smoking is socially acceptable, particularly for youth [2, 3]. The normality of smoking can reduce the likelihood of quit attempts and successful quitting [4–6]. Social ambience, including smoking visibility, may influence smoking relapse, particularly at bars and cafés [7]. Policies aimed at smoking prevalence reduction increasingly involve the de-normalisation of smoking via smokefree policies in public outdoor places [8].

While there has been a range of studies that have measured the extent of smoking outdoors for particular types of venues such as parks or bus stops [9], there is a lack of research presenting a systematic and comprehensive spatial approach to quantifying visual exposure to health-related behaviours over a relatively large geographic area. Over the past 15–20 years, advances in geographic techniques for automating the measurement and display of viewsheds (visible areas) have facilitated their use in landscape ecology [10], archeology [11], and site selection for minimal visual impact [12]. In viewshed analyses, one or more observation points are used to determine which cells in an input raster (a digital matrix of cells or pixels) can be seen from each location. All cells are assumed visible unless there is an intervening feature between itself and the observer point. Typically, viewshed analyses rely on topography for determining visibility of various locations (e.g., from a hilltop). However, in small scale urban settings, particularly those that are relatively flat, buildings may be more important obstructions to visibility. In some research observational or photographic data are combined with elevation or visual obstruction data to quantify and represent visibility of objects in viewshed analyses. In the case of smoking behaviours, quantification of visibility allows for measurement at various locations throughout the study area and for highlighting differential exposure, including hotspots of harmful visual exposure. These spatial data can be used in tandem with data, say, on distributions of populations (whether resident or frequent visitors), whose health could be influenced or harmed by the behaviour.

As there has been a lack of mapping of outdoor smoking or its visibility over large areas, we aimed to use viewshed analysis to understand and quantify visual exposure to this harmful behaviour. This quantification is useful for highlighting areas of high exposure and may aid in expansion of smokefree public spaces, as reductions in outdoor visibility of smoking may reduce cues to smoke.

Over the two-week observation period, a total of 411 smokers were recorded during a total of 28 hours observation time at 14 different pavement leases of bars/cafés/restaurants in the Wellington CBD. The observations were in a range of weather conditions, from rain and wind to warm sunshine. We observed a smoking point prevalence of 16%, with a maximum number of 44 lit cigarettes observed at one venue in one 15-minute observation period and 14 lit cigarettes in a 30-second scan. In the evenings, bars experienced the highest number of observed smokers, with an average of 10 lit cigarettes every 15 minutes, while restaurants and cafés had an average of three. At midday, all three venue types had similar levels of observed smoking. However, levels were all considerably lower than during evening observations, with an overall average of one cigarette per 15-minute period.

Visibility analyses indicated that estimated visible smoking occurring at pavement leases was highest in the evenings (7-8 pm), where values ranged from zero up to 92 visible smokers (for a particular viewpoint) when averaging Wednesday and Friday evening estimates (average = 1.44) (Table 1). In contrast, estimated visible smoking at midday ranged from zero to 13 across the CBD public spaces (average = 0.27). Figure 2 shows the high levels of estimated visual exposure on evenings (max = an average 92 smokers across Wednesday and Friday). Estimated visible smoking values were also higher in the evening at the end of the week compared with the middle of the week. Figure 3 indicates that estimates for Wednesday midday include larger concentrated areas of higher smoking visibility values than for Friday, particularly in the northern quadrant. This is also reflected by the higher average on Wednesday midday for visible smoking across the CBD (Table 1). Maps also indicate that streets with high levels of retail shops and hospitality venues had high values of estimated visible smokers, particularly in the evening where numbers were consistently above 50. Specifically, the hospitality and retail areas marked in Figure 4 had an average smoking visibility value of 35 across all observation periods.

Time of day or week	Estimated visible smoking numbers, for any given public place (1 metre x 1 metre units) in Wellington CBD
	Range	Median for all cells with smokers visible	Average
Wednesday midday	0 – 21	3	0.46
Wednesday evening	0 – 76	8	1.26
Friday midday	0 – 20	2	0.34
Friday evening	0 – 116	13	1.70
Midday	0 – 13	2	0.27
Evening	0 – 92	10	1.44

We found the highest estimated visible smoking levels in the evenings, where values reached as high as 92 visible smokers (averaged across Wednesday and Friday). Values were also higher at the end of the week compared with midweek in the evening. During 12-1 pm, when children are more likely to be downtown in public outdoor spaces, our estimated visible smoking levels ranged from zero to 13. Given New Zealand’s smoking prevalence of about 19% [14], countries with higher smoking prevalences, including the UK, the USA and parts of Europe [15], may find that levels of exposure to visible smoking may be even higher. In addition, in places with strong pedestrian cultures, such as parts of Europe, the potential of pedestrians to be visually exposed could be elevated. In addition to the utility of this method to understand smoking behaviours, this method could be applied to measure other health-related behaviours including physical activity or consumption of unhealthy foods.

While almost all large scale visibility analysis scenarios involve the challenge of incorporating complex terrain and vegetation, a host of new problems come to light when studying a built-up environment. Man-made structures, specifically buildings, are difficult to accurately incorporate into terrain models as they create new obstacles and sometimes interesting visual interactions (e.g., reflective glass). However, we included building height and footprint information in our analyses. Dynamic obtrusions such as vehicles and pedestrians also provide a unique challenge, especially in areas of cities with high vehicle and pedestrian volumes which limit the accuracy of visibility analyses. Fortunately, characteristics of the Wellington CBD simplified analyses. We were able to mitigate many of these issues. Firstly, due to the flat terrain of all the streets with pavement leases, we were able to manually generate a high resolution raster to represent the ground surface, which meant terrain had a negligible influence and the process of overlaying accurate building footprints was simplified. Secondly, the heavily built-up nature of the CBD diminished the influence of vegetation which often proves difficult in visibility analyses, due to its varying transparency and height. As such, the application of this methodology in other locations would require consideration of these and other place-specific challenges.

A number of limitations to this research are important to note. Observations were limited to pavement lease areas, and were for only four days in two weeks in April. While this period gave a range of weather conditions, a longer period of observations may provide more representative smoking data. This study did not incorporate smoking at building entrances, or in transit (walking), or that could be seen through windows or on balconies. Studies indicate that entrance areas are important sites of smoking [16]. As such, the visibility of smoking found will be an underestimate of total visible smoking in the CBD. Unrecorded smoking would have occurred with people standing outside hospitality venues without pavement leases, with people walking or standing in other areas of pavement, and in private vehicles. In addition, similar research could develop a more sophisticated method for extrapolation of observational data to locations for which data were not available. For example, land use regression or a similar methodology may produce more reliable, place-specific estimates for smoking based on small area characteristics which promote or hinder outdoor smoking.

A number of assumptions were made to conduct the visibility analyses. Firstly, no distance decay function was used, implying that smokers could be seen regardless of the distance from the viewpoint if a building did not obstruct the view. However, due to the relatively small area of the Wellington CBD (about 1.4 km²) and short viewing distances, this was not considered to be a major issue in this study. Still, future analyses could benefit from utilisation of a distance decay function to terminate visibility at a specified distance, to increase realism. Furthermore, analyses were conducted under the assumption that perfect visibility conditions held constant. Visibility can vary under different climatic and diurnal conditions and future research could produce estimates using place-specific climatic data (e.g., the chance of fog, mist and rain) and produce estimates based on realistic night versus daytime visibility. An important strength of this study was the novel application of a geographic tool in quantifying visibility of smoking across a relatively large area.

At present, in Wellington, in other New Zealand cities, and in many other jurisdictions, pavements in downtown shopping areas are not currently smokefree. Those using the pavements are exposed to SHS, and to the normalising of smoking. There are a number of policy options to mitigate this situation. These include smokefree policies for a distance from buildings used by the public (eg., 5 m, which would cover most pavements); smokefree areas near doors and windows of buildings; smokefree spaces where food and/or alcohol is served; smokefree policies for the areas with a number of premises with alcohol and food licenses; and policies for whole smokefree streets or larger areas. Successful examples of policies designated smokefree spaces include legislated requirements for smokefree areas 10 feet from doors, windows and ventilation intakes in all public places and places of employment in Oregon [17], 25 feet in Washington State [18], within 3–4 metres of building entrances in parts of Australia [19–25] and 3–5 metres around workplaces in certain provinces in Canada [26].

The resulting visual data of this method (similar to the Figures presented in this paper) would likely serve as important mobilisers for public policy action and change, particularly at a local level. This is because they provide systematic evidence of visual exposure to smoking at particular places that can otherwise only be conveyed by anecdotal statements or by film/video. These viewshed maps are compelling and can be easily absorbed and interpreted by lay audiences. Health agencies may find this methodological approach of particular interest in conveying a type of health risk to the public and to urban or national policymakers.

In conclusion, the method highlighted in this research is a useful approach for measuring the extent of visual exposure to smoking behaviours across relatively large areas. Applying this method in other locations would require consideration of place-specific characteristics which impact on visibility and could be improved through more sophisticated extrapolation of observational data across the study area. The findings of this and similar research could ultimately support the expansion of smokefree public spaces.