Capturing residents ' values for urban green space : Mapping , analysis and guidance for practice

Planning for green space is guided by standards and guidelines but there is currently little understanding of the variety of values people assign to green spaces or their determinants. Land use planners need to know what values are associated with different landscape characteristics and how value elicitation techniques can inform decisions. We designed a Public Participation GIS (PPGIS) study and surveyed residents of four urbanising suburbs in the Lower Hunter region of NSW, Australia. Participants assigned dots on maps to indicate places they associated with a typology of values (specific attributes or functions considered important) and negative qualities related to green spaces. The marker points were digitised and aggregated according to discrete park polygons for statistical analysis. People assigned a variety of values to green spaces (such as aesthetic value or social interaction value), which were related to landscape characteristics. Some variables (e.g. distance to water) were statistically associated with multiple open space values. Distance from place of residence however did not strongly influence value assignment after landscape configuration was accounted for. Value compatibility analysis revealed that some values co-occurred in park polygons more than others (e.g. nature value and health/therapeutic value). Results highlight the potential for PPGIS techniques to inform green space planning through the spatial representation of complex human-nature relationships. However, a number of potential pitfalls and challenges should be addressed. These include the non-random spatial arrangement of landscape features that can skew interpretation of results and the need to communicate clearly about theory that underpins results. Ives et al. (2017) Landscape and Urban Planning 161: 32-43


Introduction
In contrast to the study of the health and environmental benefits of green space, social values and 15 attitudes towards green spaces and the cultural services they offer have received less attention 16 (Hitchings, 2013). In their review of empirical research on urban ecosystem services, Luederitz to urban residents will not necessarily be evident from their use patterns alone (Ives & Kendal,24 2014; Swanwick, 2009). Indeed, Tyrväinen et al. (2007)

in their study of green space values in 25
Helsinki found open spaces that were identified by local residents to be their favourite were not 26 the most frequently used green spaces.  (Campbell, 2012). 36 planning process that would consider priorities for economic activities, urban growth and 114 conservation (see Raymond & Curtis, 2013 for details). The four suburbs selected were 115 Charlestown and Toronto (within the Lake Macquarie LGA), and Nelson Bay and Raymond 116 Terrace (within the Port Stephens LGA) (Fig. 1). These suburbs were chosen because they are 117 areas of current and future urban growth and contain a variety of green spaces. Population 118 statistics for the four suburbs were as follows (suburb initials used for brevity): Survey instruments were developed to ascertain the values that residents in the Lower Hunter 126 Valley assigned to the green spaces in their local area. Survey packets were mailed to a total of 127 1,000 residents from the four suburbs in July 2013. Survey recipients had expressed willingness 128 to participate via initial screening telephone calls from a larger database of residents phone 129 numbers. Recipients were asked to indicate their age to ensure that >20 % were 18-35 and >20 % 130 35-55 as a way of minimising the bias towards an older demographic which is typical in survey 131 respondents. 418 surveys were returned from a possible 972 (43%) (28 of the 1000 survey 132 packets were returned to sender  area, width etc.) were calculated using standard Spatial Analyst tools in ArcGIS. The 'near' tool 210 was used to calculate the distance of green spaces from water bodies (sea, lakes, rivers and 211 creeks) and resident's home addresses according to the closest point of approach between these 212 features. Finally, the management categories that green spaces were classified as were assessed. 213 Because the Local Environment Plans of the two LGAs contained different green space 214 management classes, consistency between the LGAs was maintained by assigning green space polygons to one of three management categories based upon the original plans (see Table 1   selected as a reasonable trade-off between explained variance and model parsimony) (Beckmw, 248 2013). The set of variables selected for further modelling were: percentage of vegetation cover, 249 distance from a significant water body, area, width, perimeter:area ratio, length:width ratio, and 250 the presence/absence of a walking path. 251 252 Quadratic terms of continuous predictor variables were also included to test for non-linear 253 relationships. Suburb was included and retained as a predictive factor in all the models to systematically account for any differences between the four study areas. The best models of 255 different green space values were determined through the following process: (1) a negative 256 binomial model was calculated using all predictors, (2) the variable with the highest P-value was 257 removed and the model recalculated, (3) the two models were compared using the "vuong" 258 function within the "pscl" R package, with the model with the lower AICc index retained, (4) 259 variables were sequentially dropped using this process until no further improvement in AICc was 260 found. We present only the model results for the positive counts because we are interested in 261 identifying the factors that influence the strength and type of values of green spaces that receive 262 marker dots, not the factors that determine whether or not green spaces receive marker dots at all. 263 Results of the final model were displayed by plotting predictor variable effects to allow visual 264 comparison of model differences. The influence of the green space management classification by 265 local councils (general, natural, sportsfield) on green space values was analysed in separate 266 models because it was not a physically observable variable associated with a green space. 267 Results of models with green space management classification were also displayed graphically, 268 with predicted means of value reported. 269

270
To analyse the effect of distance from home residence on the assignment of value dots, it was 271 necessary to account for the configuration of green spaces in each suburb relative to the locations 272 of the respondents. For example, if most green spaces occurred close to respondents' home 273 addresses, the distance to green spaces for each respondent would tend to be small, potentially 274 indicating a strong effect of green space distance. But this may be spurious as even if their true 275 preference had no relationship to distance (or indeed their selection of value dots was completely 276 random), respondents would likely select more green spaces close by if these were the majority of green spaces to choose from. To this end, a null model of green space values was generated 278 for each suburb by randomly assigning 6 'dots' per respondent to green spaces in their suburb. 279 The distribution of the distances between these dots and their home addresses was then 280 calculated. The resulting output represented a distribution of green space distances that resulted 281 solely from the spatial locations of the respondents relative to the green spaces rather than any 282 sort of preference. This could then be compared to the real distribution from the mapped data, 283 with any difference representing the effect respondent's preferences as opposed the effect of the 284 geometry. To understand the difference between these two distributions, they were both plotted 285 as histograms.  The four suburbs contained a total of 318 distinct green spaces, and 9,186 points were assigned 299 to them by respondents out of a total of 9,691 points assigned to the maps. The most commonly assigned value marker type was "activity/physical exercise" (n = 1131) while "noisy" received 301 the fewest dots (n = 131) (see Fig. 2  markers nearer to place of residence for all values (particularly for distances <2 km), but this 359 pattern was relatively weak and more pronounced in some suburbs more than others (e.g. 360 Toronto) (see Fig. 6). Although some value attributes showed the strongest densities within 1 km 361 of respondents' place of residence (e.g. social interaction value), others (especially negative 362 qualities) displayed no relationship with distance from home (see supplementary material S5). 363  Table 2). These correlations are confirmed, with the first 380 factor receiving highest loadings of nature and culture values, the second health and activity 381 values, and the third negative values. Interestingly, the fact that some green spaces are 382 considered noisy does not seem to compromise their activity, social interaction or health values 383 (see factor 2). In contrast, the other negative qualities all loaded on a single factor, suggesting 384 that these rarely are found alongside other values in green spaces. 385 386

392
In this study we sought to understand how people in a rapidly urbanising region assign value to 393 green spaces and assess the influence of environmental variables on these values. These insights 394 are important for building the evidence base from PPGIS research methods that are increasing in 395 popularity. In particular, our study can provide guidance on how statistical methods can be 396 appropriately applied to PPGIS data. Further, given some continuing resistance to the use of 397 PPGIS methods by planning practitioners (Brown, 2015) a key research question of this study 398 was to explore useful insights into how PPGIS assessment of green spaces can be applied in 399 practice. These issues are discussed in turn below. 400 401

The impact of environmental variables on values for green spaces.
The values people assigned to green space were very positive overall, with comparatively few 403 marker dots assigned that denoted negative qualities. This was true regardless of the type of 404 management applied to the green spaces (Fig. 5) space networks for urban populations will require a 'portfolio of places' (Swanwick, 2009). 417 418 For many value attributes, green spaces closer to water bodies were valued more strongly than 419 those further away (see Fig. 4). This finding is consistent with most of the literature on public 420 preferences for landscapes (Swanwick, 2009), with people's affinity for water explained by the 421 theory that it enhances the perceived orderliness and naturalness of a scene (Kaplan & Kaplan, 422 1989), as well as adding to the coherence of a landscape (Litton, Tetlow, & Sorensen, 1974). 423 However, there is evidence that preferences for waterscapes can differ according to type and 424 context (Herzog, 1985). For example, a study in Victoria, Australia recently found that the public The proportion of vegetation present in a green space was related to the abundance of marker 436 dots for many value types (Fig. 4), yet the nature of its influence varied. For native plants and 437 animals, the relationship was a positive one, for social interaction values a negative relationship 438 was observed, while a quadratic relationship was found for aesthetic values (Fig. 4) Local governments in Australia regularly categorise green spaces according to their intended 460 purpose or use. Our study showed that in our case study areas, these categories had little to no 461 bearing on the abundance of value markers found in specific green spaces (Fig. 5). In particular, 462 we observed no statistical difference in the average abundance of marker dots for nature values 463 or native plants and animals values between green spaces designated as 'natural areas' and those 464 for 'general use' (Fig. 5). Our results suggest that formal categories may not have a strong 465 influence on the perceptions of local residents. This may either be because residents simply do 466 not strongly distinguish between these classes when valuing green spaces, or because residents 467 have little knowledge of the official designated purposes of the green spaces. Determining which 468 of these is the more accurate explanation is an area for future research. In terms of biodiversity 469 conservation, our findings present an opportunity for management agencies to maximise biodiversity across the whole landscape rather than focussing exclusively on formal nature 471 protection areas since residents value nature on all different kinds of green spaces. 472 473 Distance from place of residence did not have a clear relationship to the assignment of values to 474 green spaces, after accounting for landscape configuration (Fig. 6). Although distance from 475 home has been found to be an important factor influencing green space visitation (Neuvonen,476 Sievänen, Tönnes, & Koskela, 2007; Shanahan et al., 2015), it appears that landscape values, at 477 least in our case study, are quite different constructs and are less strongly influenced by spatial 478 proximity. The established theory of geographic or spatial discounting of values (Norton & 479 Hannon, 1997) supposes that PPGIS respondents will place disproportionately more markers 480 closer to their home than more distal locations, as has been empirically shown by Brown et al. 481 (2002). Although this pattern can be seen in the suburb of Toronto, it was not evident for the 482 other suburbs. Thus, our analysis highlights the importance of accounting for the spatial bias in 483 the locations of landscape features (for example via simulation) in order to further explore the 484 spatial discounting hypothesis in relation to PPGIS.

PPGIS in practice 501
In considering how the insights from this study should be applied to planning practice, it is 502 useful to recognise the different scales at which research and planning practice can be reconciled 503 as proposed by Lindholst et al. (2015). First we consider applying insights at the policy level (i.e. 504 deriving general principles for planning green space), and second at the applied level (by 505 providing guidance for practitioners considering using PPGIS in a local context). 506 seek to promote multiple values simultaneously in individual green spaces regardless of their 513 management category (Fig. 5). Based on the value compatibility assessment (Table 2), some 514 values may be promoted alongside one another more easily than others (e.g. health and social 515 interaction, or nature conservation, aesthetics and culture). Practitioners should therefore carefully plant and maintain vegetation in ways that are visually appealing and help to promote 517 biodiversity (Ives & Kelly, 2016). Of course, applying these general principles is only one 518 element of good planning practice; practitioners should also seek to engage the community and 519 encourage participation in the decision-making process, as difficult as this process can be 520 (Chiesura, 2004). Indeed, the effect of 'suburb' on some of our models of open space values 521 (namely activity value, nature value, health/therapeutic value, social interaction value, and noisy; 522 see Fig. 4) suggests that the valuation of green spaces may be influenced by unique demographic 523 and environmental characteristics of specific areas. It is imperative therefore that planners 524 supplement any general principles with knowledge of the needs specific to a region. 525 526

Guidance for practitioners applying PPGIS 527
Many methods exist for public communication, consultation and participation, each with 528 strengths and weaknesses depending on the decision-making context (Reed, 2008). We consider 529 PPGIS to be a useful complement to existing methods for engaging communities in urban green 530 space planning. PPGIS is more participatory than approaches that emphasise information 531 dissemination such as town hall meetings or leaflets, more representative than charettes or 532 community planning forums, more spatially nuanced than public surveys, and more quantitative 533 than focus groups. Yet the mass collection of quantitative data can also mask certain issues and 534 subtle complexities that emerge through more deliberative, qualitative methods. PPGIS is 535 therefore likely to be a useful tool that builds upon existing understandings of the social-536 ecological landscape and feeds back into the planning process in order for a just and sustainable 537 outcome to be reached. 538 Our study identified a number of potential challenges and pitfalls that need to be considered by 540 urban landscape managers and planners seeking to apply PPGIS methods in a specific context. In 541 their study of participatory green space planning processes in Finland, Kahila-Tani et al. (2016)  542 noted that "though planners found the collected data and the analysis valuable, they still lacked 543 the skills and institutional motivation to use the data effectively" (p. 195). Below we provide 544 guidance along these lines that could assist urban planners in implementing PPGIS methods. 545 546

Evaluation of PPGIS design and analysis choices 547
If PPGIS data are used to inform decision-making, it is critical that they are accurate and reliable. 548 This study has identified a number of issues that need to be considered. First, it is important that 549 the sample frame is an accurate representation of the broader population's spatial, temporal and 550 socio-demographic variability. We strove to ensure a representative sample of participants, yet 551 even with appropriate survey design and administration measures taken we found some 552 demographic bias in our data. This has potential to overemphasise the importance of certain 553 values and places since different demographic groups interact with landscapes in different ways 554 (e.g. parents valuing safe areas for children to play). Any such bias should be recognised when 555 applying results to planning practice. Second, the spatial arrangement of respondents and 556 landscape features can impact results and their interpretation. By accounting for the relative 557 spatial distribution of green spaces to the respondents in our study areas, we found that the 558 distance of a green space from participants' place of residence did not have a strong effect on 559 marker abundance (Fig. 6). Failure to account for the relative locations of green spaces and 560 respondents could in many cases lead to inaccurate conclusions about how distance impacts 561 values, yet this kind of analysis is not a simple exercise for many management agencies. Another challenge in undertaking effective PPGIS research for green space planning is the 571 resources (time, money, expertise) it requires. Using physical paper maps is known to generate 572 higher response rates than online PPGIS methods (Pocewicz, Nielsen-Pincus, Brown,& 573 Schnitzer, 2012), yet printing and postal costs can be prohibitive for many small municipalities. 574 The substantial time taken to digitise markers and analyse responses may also be problematic if 575 it exceeds the personnel time allocated by management agencies for community engagement. A 576 related challenge is ensuring agencies have the appropriate expertise (particularly statistical) 577 required to appropriately analyse and interpret results. We encourage the continuing 578 development of new methods to engage citizens using new technologies (e.g. smartphone apps) 579 and assist practitioners in data analysis as a way of helping to meet these challenges. 580 Additionally, if limited analytical skills are available, it may be more appropriate to simply use 581 visualisations of mapped values to identify immediate management priorities or issues rather 582 than seeking to extrapolate results to more generalised principles. 583 Planning for green space is a complex process that brings together various social, environmental 586 and political considerations. Although the specifics of the planning process varies across 587 different places and times, Maruani and Amit-Cohen (2007) identified five general open space 588 planning models that have been applied in an urban context. In brief, these are (i) opportunistic 589 (random allocation of land for open space according to availability), (ii) space standards 590 (providing minimum area of open space for a given population), (iii) park systems (interrelated 591 parks and gardens), (iv) garden city (a comprehensive approach based on Ebenezer Howard's 592 principles), and (v) shape related models (such as green belts or green wedges). We suggest that 593 PPGIS can help transition urban green space planning from traditional standards-based or shape-594 based planning models to a participatory, 'needs-based' planning approach: one that accounts for 595 a population's "socio-demographic composition, their leisure and recreation preferences and types. Further, existing management categories were shown not to have a strong bearing on the 609 kinds of values people assign to green spaces. This research reveals a complex picture of how 610 different values are assigned to green spaces, and highlights the need for green space planners to 611 avoid the 'one size fits all' approach to the design of green space networks. We encourage 612 planners to pursue participatory techniques such as PPGIS as a means of ascertaining the values 613 and preferences of the urban public and planning for these accordingly. Yet we also emphasise 614 the need for careful consideration of the design and analysis of these methods to ensure that the 615 data used to inform decisions are accurate and reliable. 616 617