A methodological framework for analysis of participatory mapping data in research, planning, and management

ABSTRACT Today, various methods are applied to analyze the data collected through participatory mapping, including public participation GIS (PPGIS), participatory GIS (PGIS), and collecting volunteered geographic information (VGI). However, these methods lack an organized framework to describe and guide their systematic applications. Majority of the published articles on participatory mapping apply a specific subset of analyses that fails to situate the methods within a broader, more holistic context of research and practice. Based on the expert workshops and a literature review, we synthesized the existing analysis methods applied to the data collected through participatory mapping approaches. In this article, we present a framework of methods categorized into three phases: Explore, Explain, and Predict/Model. Identified analysis methods have been highlighted with empirical examples. The article particularly focuses on the increasing applications of online PPGIS and web-based mapping surveys for data collection. We aim to guide both novice and experienced practitioners in the field of participatory mapping. In addition to providing a holistic framework for understanding data analysis possibilities, we also discuss potential directions for future developments in analysis of participatory mapping data.


Introduction
The last two decades of the Western and non-Western world have witnessed increasing interest in the participatory mapping approaches, applied in a variety of fields of research and practice (Brown et al. 2020). Various terms have been used to describe these approaches, the most prominent being public participation geographical information systems (PPGIS), participatory GIS (PGIS), and volunteered geographic information (VGI) (Verplanke et al. 2016). Today, participatory mapping approach has roused the interest of academics and a wide user community. This is evident from the increasing number of academic publications, conferences, workshops, and journal special issues pertaining to this field (see, e.g., Brown and Fagerholm 2015, Mukherjee 2015, Brown and Kyttä 2018. Furthermore, participatory mapping now has an international professional society, comprising scholars and practitioners, invested in its integrity, accuracy, data collection, and the equitable distribution of knowledge (International Society for Participatory Mapping 2020).
As noted by Brown and Kyttä (2014), PPGIS, PGIS, and VGI are related spatial terms with sufficient differences to warrant nuanced descriptions. PPGIS approaches promote the use of GIS and modern communication technologies to engage the general public and stakeholders to carry out informed participatory planning and decision-making, particularly in the context of urban and regional development (Sieber 2006). The term PGIS emphasizes empowerment and can be traced to the merger of Participatory Learning and Action methods with geographic information technologies in the Global South (Rambaldi et al. 2006). The term VGI, introduced by Goodchild (2007), describes a phenomenon where citizens voluntarily create, collect, validate, analyze, and disseminate geographic information. Collecting VGI is based more on contribution and communication of information, than on participation (Verplanke et al. 2016). Collecting VGI conceptually resembles PPGIS approaches owing to the use of typical online tools to harness spatial information (Hall et al. 2010). In this paper, we have adopted the use of the term PPGIS, although some of the empirical work we discuss could be described as PGIS or collecting VGI.
PPGIS approaches seek to understand location-specific human values, perceptions, behavior, and preferences for future land use and development. Methods for analyzing spatially referenced data, collated using PPGIS, have been developed in diverse directions. Among others, these include analyzing sampling effects and response bias (e.g., Brown et al. 2014a, Brown 2017, Munro et al. 2017, representing diversity, abundance, or rarity of value points (Bryan et al. 2011); examining the level of overlap in values across different stakeholder groups (Muñoz et al. 2019), identifying the potential for value or preference conflicts (e.g., Brown and Raymond 2014, Kahila-Tani et al. 2016, Plieninger et al. 2018, Wolf et al. 2018; assessing environmental justice issues (Raymond et al. 2016); and bridging the divide between experts and the public (Whitehead et al. 2014, Zolkafli et al. 2017a. Despite the plethora of analysis methods available to explore, explain, and predict spatial attributes collated using PPGIS, most published articles apply a specific subset of analyses that fails to situate the methods within a broader, more holistic context of research and practice. Hence, the field currently lacks a methodological framework, making it essential to synthesize the various existing analysis methods to guide their processes and applications. In this paper, we aim to produce a systematic framework of PPGIS data analysis methods supported by examples from published empirical studies. Our focus is mainly limited to the online PPGIS approaches, particularly the web-based mapping survey, the most common administrative technique for PPGIS data collection (Brown and Kyttä 2014). In order to provide context for our proposed framework, we first discuss spatial and nonspatial attributes in PPGIS surveys and implications of the combination of these on the quality of the PPGIS data produced and the methods for data analyses. This is followed by presentation of our views on the different phases in PPGIS data analysis (Explore, Explain, Predict/Model) and identifying the methods, purposes, and analytical approaches, highlighting each with example studies and application domains, based on expert workshops and extensive literature review of peer-reviewed articles. Finally, we recommend potential future development directions for PPGIS analysis methods. The methods of analyses described herein are relevant to different applications including conservation and natural resource planning and urban and regional development. We aim to guide researchers and practitioners, both new and experienced, interested in PPGIS approaches, to address academic, or applied questions relevant to exploration, explanation, and prediction.

Methods
To draft and develop the methodological framework, the authors conducted a one-day expert workshop in August 2018, at the Aalto University in Helsinki, Finland. This was the first of the two workshops held at Aalto University where we discussed possible ways to categorize analysis methods. The second workshop held in October 2018 saw further refinements to the framework along with drafting of the manuscript contents and planning the literature review. Moreover, in January 2019 we searched for peer-reviewed articles using the Scopus electronic database (document search: title, abstract, keywords, and publication year 2004 to 2019) to gather examples of empirical studies that applied a broad range of data analysis methods, and contained the keywords: participatory GIS, participatory mapping, public participation GIS, PPGIS, SoftGIS, and geo-questionnaire. We identified 279 such published papers. We reviewed these articles to describe the applied analysis methods. Based on expert judgment, we also included several published papers that did not appear in the search results. In the third and final workshop held in December 2018 at the University of Turku in Turku, Finland, we critically reflected on the manuscript content, results of the literature review, and discussed future directions in the field of participatory mapping.
Although the article focuses on the online PPGIS approaches, we acknowledge that the presented methods can also be applied to analyze data collected through analog approaches (e.g. interviews, workshops, and mail surveys). Hence, we have included some offline PPGIS approaches to cover the range of methods and highlighted important examples where specific methods were applied for the first time. Online PPGIS surveys also include data collected from individuals, which is then aggregated to the scale of the survey population (Brown et al. 2015a), as opposed to deliberative valuation where the emphasis is on group negotiation and compromise, including the mapping of shared and social values , Kenter 2016).

Data collection through PPGIS surveys and data quality
PPGIS data analysis methods are constrained by data quality. The PPGIS process includes the phases of survey/website design, participant recruitment, and data collection, followed by data analysis. All phases are important and should be carefully prepared. Survey design identifies the spatial and non-spatial information to be collected which influences the user experience (Swobodzinski and Jankowski 2014, Poplin 2015, Gottwald et al. 2016, participation rates, and ultimately, the quality of spatial data and the possibilities of analysis offered. There exists several recruitment methods for online PPGIS surveys, ranging from random samples drawn from a national population or household registers (Hausner et al. 2015, Kyttä et al. 2015, Laatikainen et al. 2019, purposive sampling (Garcia-Martin et al. 2017), and crowdsourced/volunteer sampling through traditional or social media (Kahila-Tani et al. 2016, Rall et al. 2017) to using internet survey panels , Munro et al. 2017. To improve the quality of spatial data generated, other participant recruitment strategies, such as collecting data in schools when studying children and young people (Kyttä et al. 2012(Kyttä et al. , 2018b or applying facilitated mapping processes where survey respondent receives assistance (Zolkafli et al. 2017b, Fagerholm et al. 2019a, can also be used. In a recent review, Kahila-Tani et al. (2019) found that the data collection strategy impacts sample representativeness; random sampling seems to promote good representativeness while crowdsourced/volunteer sampling poses a challenge to reaching a balanced respondent profile.
The quality of PPGIS data also depends on many other factors including mapping efforts, accuracy, and precision, type of spatial data collected, and data usability in terms of how it fits the purpose (Brown and Kyttä 2014, Brown and Fagerholm 2015, Jankowski et al. 2016, Kahila-Tani et al. 2019. There are always practical limitations to the time and efforts exerted by the respondents in a survey. A meta-analysis shows that household sampling groups always dedicate more mapping efforts as compared to volunteer groups (Brown 2017). Cognitive challenges may vary depending on the type of data being collected through mapping, that is, objective or subjective. For instance, place-related activities and experiences seem to be cognitively less challenging to map as compared to place-related values and concepts such as ecosystem services (Brown 2017).
PPGIS data collection through online surveys are often self-administered, where individuals map spatial attributes of importance without outside assistance. These attributes can relate to mapping of either points, lines, or polygons, with points being the most commonly used and simple geographic feature in PPGIS (Brown and Fagerholm 2015). The mapped spatial PPGIS data attributes can, for example, signify a respondent's: (1) Spatial values, perceptions, or attitudes, e.g., landscape values , perceived environmental quality factors , and ecosystem service benefits (Ridding et al. 2018, Fagerholm et al. 2019a, in addition to perceived problems or unpleasant experiences (Raymond et al. 2016); (2) Spatial behavior patterns, everyday practices, and activities, e.g., daily mobility patterns, and routes travelled (Laatikainen et al. 2017, Kajosaari et al. 2019, places visited (Sarjala et al. 2015), and their temporal characters, e.g., seasonality, length, or frequency of visitation (Bijker and Sijtsma 2017); (3) Spatially defined future preferences or visions, e.g. development preferences (Brown 2006, Jankowski et al. 2016, Kahila-Tani et al. 2016, Engen et al. 2018; and (4) Preferred place features referred to as 'geographic citizen science' (Haklay 2013), e.g., mapping road/trail networks (e.g., OpenStreetMap) and wildlife observations (Brown et al. 2018a). These spatial data can be used to augment and validate authoritative data.
In addition to mapping the spatial attributes, open or structured follow-up questions can be asked to describe the mapped attributes. These follow-up questions often appear in a pop-up window with relation to the mapped places in the survey. Videos, photos, and recorded stories can also be captured for the mapped places (e.g., Kahila-Tani et al. 2018). Along with their focus on mapping, spatial surveys often include non-spatial PPGIS data collected through traditional open or structured questions (Figure 1). Such non-spatial data may include, but are not limited to, questions addressing: (1) Socio-economic-demographic characteristics, e.g. age, gender, education, and income levels; (2) Personal general values, attitudes, and preferences, e.g., lifestyle preferences, environmental worldviews, beliefs, and norms; (3) Personal motivation and behavioral intentions, e.g., personal goals, and likelihood to engage in special behavior; (4) Personal well-being, happiness, health, and satisfaction, e.g., perceived health, perceived quality of life, and neighborhood satisfaction; and (5) Level of trust in planning and decision-making processes for land use.
An important trade-off in survey design relates to the abundance of spatial data versus descriptive depth of mapped places. When targeting a considerable number of mapped places by each survey respondent, the respondents are likely reluctant to spend significant time to describe the mapped places in-depth. Similarly, when aiming for Figure 1. PPGIS survey data structure with mapped spatial and non-spatial data, and links to other geospatial data. detailed descriptions of numerous-mapped places, it is likely that the respondent's efforts may not be sufficient. Hence, in terms of survey design, it is crucial to balance the quantity of mapped spatial data with the corresponding descriptions; in each case, it is important to critically reflect on the most essential information for the purpose of the study. Hence, data quality can be controlled when designing the survey and this affects analysis possibilities for the data.

Other geospatial data in PPGIS data analysis
PPGIS datasets are often combined with other geospatial data during analysis (Figure 1). Typical examples include land cover, land use, and road network data. CORINE land cover is an openly accessible European example of a land cover and land use dataset (https://land.copernicus.eu/pan-european/corine-land-cover/clc-2012). Although, road network datasets, comprising national road datasets, are available, the use of Open Street Map (www.openstreetmap.org), an open geospatial road dataset produced by a community of mappers, has become common. Furthermore, versatile geographically referenced statistical data pertaining to population, species, city plans, population and housing density, conservation areas, zoned land units, real estates, buildings, and service and company locations can be used concurrently with PPGIS data. Participatory mapping also provides an opportunity to conduct ex-post planning evaluation, to obtain feedback from inhabitants regarding the performance of neighborhood, city, or regional level planning solutions once they are accomplished. PPGIS datasets can therefore be analyzed along with urban or spatial plans . To understand the actual realization of these plans, this analysis can be complemented with expert audit data of the physical environment (2018a) and virtual audits, now possible with the help of Google Street View (Rzotkiewicz et al. 2018). Additional spatially referenced datasets include social media data (Toivonen et al. 2019) and data produced through GPS tracking (Wolf et al. 2015).

Analysis of PPGIS data: a framework with three phases
Analysis methods applied to PPGIS data can be represented as a framework of three analytical phases: Explore, Explain, and Predict/Model. These phases graduate from basic to advanced. The three phases of the framework relate to different types of knowledge claims, as an output of PPGIS data analysis. The Explore phase defines the exploratory and descriptive character of the analysis method ( Figure 2). Such methods are generally termed as exploratory spatial data analysis (De Smith et al. 2020). The analysis does not require high expertise and can be done by non-academics such as planners and stakeholders. The Explain phase aims to understand the relationship between PPGIS data and multiple other geospatial data sources. This phase demands expertise in analytical methods. The Predict/Model phase intends to generalize mapped attributes to other places and contexts, and to understand future realities. This phase typically requires advanced expertise to perform analyses that integrate multiple data sources to predict and model PPGIS data.
The presented framework suggests a logical progression in data analysis. However, in reality, analysis often proceeds iteratively going back and forth between the different Figure 2. Three analysis phases Explore, Explain, and Predict/Model for data gathered through PPGIS approaches and links in each phase between PPGIS data and other geospatial data.
phases. Moreover, it is not necessary to perform data analysis in all phases. For example, data analysis can simply focus on the first phase, Explore. Each phase of our framework and its related methods have been presented in the following sections. Since earlier literature have not given due justice to spatial PPGIS data analysis options, our paper specially focuses on these.

Explore
The first analytical phase, Explore, involves descriptive and univariate analysis of PPGIS data and generation of visual outputs. Spatial patterns are identified for one attribute at a time (univariate analysis) and compared across available attributes. Though the analyses in Explore phase focus on spatial and non-spatial PPGIS data, it incorporates other geospatial data merely as cartographic background information. The analyses are accomplished with basic GIS software or with the help of the interactive analysis tools provided by some online PPGIS services. An important part of Explore phase is the assessment of spatial data quality through validation. Before the data enter the exploration phase, PPGIS data need to be cleaned by detecting, correcting, or removing inaccurate spatial records, Table 1. PPGIS analysis methods, purposes, and example tools in Explore phase. Goals in Explore phase include assessment of spatial data quality and uncertainty, data exploration through simple univariate statistics, and visualization of spatial patterns. and organized for subsequent data analysis. Such data manipulation may include value (re)classification, data (re)ordering, data queries, and removal of outliers.

External and internal validation
External and internal assessment of spatial data quality and uncertainty is important (Lechner et al. 2014). External validation addresses assessment of sampling strategy, size, and representativeness (i.e., a comparison of sample characteristics within a wider population) (Table 1). Exploration can also include comparison of different sampling groups, e.g., comparison between random versus volunteer samples (Brown et al. 2014a) or different cohorts in the sample to assess sample representativeness (Munro et al. 2017). Testing whether the spatial results can be generalized to other locations, people, and situations is challenging because PPGIS studies are case studies with a unique mix of place-based contextual variables. External validity can be indirectly assessed by performing meta-analysis with multiple PPGIS studies to examine which spatial variables appear valid across different place settings. For instance, Brown and Hausner (2017) analyzed the distribution of mapped cultural ecosystem values in coastal areas of five countries and found that the mapped values were significantly more abundant in all coastal zones, regardless of ecosystem value category, country, population, or dominant land use. Internal data validation involves assessing validity of content, criterion, and construct (see Brown et al. (2017a) for application of data validity concepts to quantitative and qualitative PPGIS data). Measurement of positional accuracy, correctness, and completeness validity (Brown et al. 2015a, Jankowski et al. 2016, Rohrbach et al. 2016 is also a part of internal data validation. To explore global clustering, the data are commonly tested for spatial autocorrelation through nearest neighbor index (e.g., van Riper and Kyle 2014, Pietilä and Fagerholm 2016) and Moran's I (e.g., Garcia-Martin et al. 2017). These descriptive statistical methods help to understand spatial distribution of the mapped data.

Descriptive and visual analysis
Simple univariate descriptive analysis is applied in the Explore phase to study PPGIS data qualitatively and quantitatively (Table 1). Visual outputs, in form of thematic maps and charts, are often generated to examine the spatial patterns (e.g., Ramirez-Gomez et al. 2013, Samuelsson et al. 2018, 261, Kajosaari et al. 2019

Explain
The second phase, that is, Explain, aims to look more closely at observations than the Explore phase, in order to explain observations by further analysis. A wide variety of PPGIS data analysis methods are categorized within this phase. The Explain phase essentially combines spatial and non-spatial PPGIS data with other geospatial data. Thus, several methods including inferential and multivariate statistics are used in this phase; it also involves the use of various statistical software along with GIS software.

Visual and overlay analysis
Visualization of spatial patterns is typically part of Explain phase, but as a contrast to Explore phase, visual analysis includes generation of multiple simultaneous views or overlay maps (e.g. Kyttä et al. 2013, Laatikainen et al. 2017, Brown et al. 2018b) ( Table 2). Overlay analysis, where multiple inputs are overlapped to generate new information, is common for viewing different mapped attributes or studying their relation to other geospatial data such as city plans, land use or ecologically valuable areas (e.g., Whitehead et al. 2014, 2018a, Rall et al. 2019.

Spatial pattern analysis
A broadly applied method in multiple studies to analyze spatial patterns of PPGIS data is to produce a spatially continuous intensity/density surface of mapped attributes through Kernel density estimation (Silverman 1986) (e.g., Alessa et al. 2008, Sherrouse et al. 2011, Pocewicz and Nielsen-Pincus 2013) ( Table 2). A simpler version of intensity surface can be calculated through point density analysis (Hausner et al. 2015, Kantola et al. 2018. Alessa et al. (2008) investigated spatial interpolation methods (e.g., kriging) for intensity surface mapping, but concluded that this method appears appropriate when the interpolated variable has a continuous spatial coverage across an area (e.g., air temperature). Clustering or dispersion of mapped attributes can be approached through methods developed for identifying statistically significant hot and cold spots, such as Getis-Ord Gi* statistics (Getis and Ord 1992). These methods have been applied extensively by Brown and Raymond (2014), Karimi et al. (2015), and Bagstad et al. (2017). Cluster identification through average nearest neighbor distance analysis has also been applied to create boundaries for clusters of mapped attributes (Raymond et al. 2016). Furthermore, Laatikainen et al. (2017) identified a corresponding distance band of the mean distance of the mapped points, based on which, points were aggregated to polygons to capture elongated clusters along the shoreline. Muñoz et al. (2019) implemented a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm (Ester et al. 1996) to find areas with highest density of mapped place values.

Proximity-related analysis
There are various types of analysis available for exploring the proximity among the attributes in mapped data or in relation to other GIS data ( Table 2). Distance of mapped attributes from domicile can be useful to explain variation in spatial patterns of mapped attributes (Brown et al. 2018b). Circular buffers at specified distances around home or mapped places allow for calculation of the amount of different land uses and other GIS variables describing urban structure (Kyttä et al. 2015, Laatikainen et al. 2017 or landscape characteristics (Ridding et al. 2018) within the neighborhood of a mapped place. Brown (2013) applied circular buffer analysis to multiple radii to analyze the cumulative proportion of mapped forest values located within the neighborhood of mapped forest use preferences. Similarly, Brown and Hausner (2017) examined the distribution of mapped ecosystem values in coastal and non-coastal zones using multiple distance bands from the coastline. Circular buffers are also applied as the first step to identify home range, a concept common in ecology (Burt 1943), to spatially identify individual place attachment or activity space in PPGIS data. In addition, minimum convex polygon around mapped attributes has been suggested as an operational model to define home ranges (Brown et al. 2015b, Hasanzadeh et al. 2017. Proximity analysis is also used in viewshed analysis to calculate the range of visible territory in all directions at specified distances from a mapped attribute based on topography, using elevation values from Digital Elevation Models. The viewsheds of specific landscape values may be compared to the average viewshed for that attribute in the data (Garcia-Martin et al. 2017) or may be used to calculate the proportion of landscape characteristics variables derived from various GIS data within the viewshed of mapped outdoor locations (Ridding et al. 2018).

Analysis across spatial scales
Interactions of mapped locations with the surrounding landscape at different spatial scales (Table 2) were examined by Pietilä and Fagerholm (2016), who analyzed mapped tourism impacts at destination, park zone, and site scale in the context of a national park and by Bijker and Sijtsma (2017), who analyzed important natural places on local, regional, national, and global scales. Moreover, Ridding et al. (2018) calculated landscape characteristic variables at different scales defined by 500 m and 5 km buffers around the mapped locations, whereas Ives et al. (2018) compared landscape values at suburb and municipality scales using metrics of value abundance and diversity.

Calculation of indices across spatial units
Distribution of mapped attributes can also be analyzed across spatial units, such as landuse class, land management type, or grid cells using spatial indices (Table 2). Such indices, developed initially in landscape ecology (McGarigal and Marks 1995), have been modified for the purposes of PPGIS data and include, e.g. richness, diversity, abundance, dominance, rarity, and complementarity (Bryan et al. 2010, Brown andReed 2012a). Brown and Reed (2012a) presented a detailed elaboration of these 'social landscape metrics' and distinctions between boundary and inductive metrics. These spatial indices have been widely applied and further developed in PPGIS research in various application domains (Broberg et al. 2013, Hausner et al. 2015, Hasanzadeh 2019. One of the specific fields of interest to explain potential tensions between mapped place values and land-use preferences is the identification of conflict potential based on PPGIS data. Brown and Raymond (2014, Figure 1), proposed a conceptual model of landuse conflict potential as a function of the level of agreement on land-use preferences and place importance. This conceptualization yielded sampling grid-based methods for calculation of conflict indices using weighted and unweighted preferences, place values, and value compatibility scores. Application of these indices has been exemplified in practice, for example, by Brown and Raymond (2014) for residential and industrial development, Brown et al. (2017b) in natural resource management, and Plieninger et al. (2018) for land -and seascape values and development preferences. In addition, Karimi and Brown (2017) present an assessment of the different methods for conflict identification. Furthermore, Lechner et al. (2015) augmented the conflict identification approach to assess ecological connectivity.
Suitability analysis is another commonly used spatial analysis in land-use planning, wherein areas that are suitable for a specific land use are identified based on a set of decision criteria (see, e.g., Malczewski 2004). Suitability maps based on criteria (e.g., elevation, slope) are generated to provide individual data layers that are overlaid to identify areas of spatial intersection that can satisfy multiple criteria. Traditional suitability analysis has often relied on biophysical landscape features, but PPGIS data layers can be used to identify relevant social criteria, such as landscape values, to include in the analysis Brown 2003, Brown andReed 2012b). Consistency analysis is related to suitability analysis and seeks to identify the significant association of PPGIS mapped attribute's distribution (e.g., land-use preferences) with the current or proposed land uses (e.g., through zoning); it also assesses if the mapped attributes appear logically in accordance with the land use (Brown et al. 2018c). The consistency of PPGIS data with current or proposed land use can be interpreted with chi-square residual analysis, where spatial data are collected as frequencies.

Analysis of spatial associations
Spatial associations can be identified based on the relationships between mapped PPGIS attributes and physical or administrative land properties. These associations can be analyzed by tabulating the frequencies of mapped attributes within land units (cross tabulation) and calculating chi-square statistics and standardized residuals to examine the statistical association (Brown and Brabyn 2012a) (Table 2). In addition, calculation of Z scores reveals the statistically significant under -or over-representation of attributes in a given land unit as presented, for example, by Brown et al. (2015b), Brown and Hausner (2017) and Fagerholm et al. (2019a). Spearman's rank correlation analysis has been applied for the identification of spatial associations between pairs of mapped attributes such as ecosystem services or landscape values (e.g., Plieninger et al. 2013, Garcia-Martin et al. 2017. The spatial overlap between different-mapped attributes has been quantitatively measured using the phi correlation coefficient (Zhu et al. 2010, Rall et al. 2017, the Jaccard coefficient, and the Pearson's product moment correlations (Raymond and Brown 2011).

Cluster and multivariate association analysis
Creation of statistically significant clusters of mapped attributes, respondent groups, and/or other GIS data has been performed to identify bundles of perceived ecosystem services (i.e., sets of mapped ecosystem services that repeatedly appear together (Raudsepp-Hearne et al. 2010). Methods such as multiple correspondence analysis, hierarchical cluster analysis and principal component analysis have been used collectively to identify clusters/bundles of perceived ecosystem services in grid cells or land cover units (e.g., Plieninger et al. 2013, Rall et al. 2017. Several methods identifying statistical associations have been applied to the PPGIS data to find associations between mapped attributes, respondent groups and/or attributes from other GIS data. At a preliminary level, Spearman's correlation is useful to analyze relationships between spatial and non-spatial PPGIS data or other GIS data; for example, to explore the association between spatially explicit social values and ecological values (Bryan et al. 2011), the perceived impacts of tourism and visitor satisfaction (Pietilä and Fagerholm 2016), cognition of fearful places in the urban environment, and presence of day/night (Pánek et al. 2017), and to understand the general importance of mapped ecosystem services and their place-specific importance (Rall et al. 2017).
At a more advanced level, regression models are the prominent multivariate modeling methods to analyze associations. Logistic regression models have been used to assess, for example, the connections between urban structure, children's behavioral patterns, and environmental experiences, and health measures , the adjusted odds of walking a high share of estimated monthly trips and travel distance in an urban context (Kajosaari et al. 2019), landscape characteristics associated with outdoor places of personal importance for the delivery of cultural ecosystem services (Ridding et al. 2018), and whether communities favor, or oppose human activities in protected areas when controlling the landscape characteristics, accessibility, and demographics (Engen et al. 2018). Redundancy analysis, a multivariate analog of regression, has been applied to examine potential relations between mapped ecosystem services on different land covers, subjective well-being, and socio-demographic characteristics . Generalized linear models have been used to examine possible relationships between the mapped ecosystem services with frequency of green space use, affinity, and general importance of each service (Rall et al. 2017). Generalized linear-mixed models have been applied to quantify the relationship between biophysical landscape characteristics and mapped ES benefits across 13 study sites that showed grouped structure and spatial autocorrelation (Fagerholm et al. 2019a). Structural equation models have been used to assess contextual variation and mediation of different factors in linking urban structural characteristics with health and well-being outcomes (Kyttä et al. 2015, Laatikainen et al. 2019.

Predict/Model
The final analysis phase, Predict/Model, aims to generalize and predict mapped attributes to other places and contexts (prediction) or produce a representation of a system to make inferences (model). Analysis methods in this phase require multiple data sources in addition to PPGIS data and involve multivariate modeling. Performing analysis in Predict/Model phase requires in-depth expertise in applying GIS and statistical software. The phase may also demand skills in computer coding.

Predictive analysis
In the absence of empirical PPGIS data, quantitative relationships with physical landscape variables can be used to extrapolate, i.e.value transfer, mapped PPGIS attributes spatially for wider regions or even at national scale (Table 3). Brown and Brabyn (2012b) extrapolated regional landscape values to a national scale using empirical relationships between physical landscape character and mapped PPGIS attributes. Brown et al. (2015c) and Brown et al. (2016) used the percent of mapped ecosystem values, spatially associated with land cover classes, as value transfer coefficients to assess the similarity between actual-mapped ecosystem values and value transfer spatial distributions.
For assessment of land use consistency, the quantitative relationship between existing land classifications and perceived landscape values has also been applied to build predictive discriminant functions to classify prospective lands for conservation purposes (Raymond and Brown 2006). Generated land classes can be mapped and overlaid with expert-derived classifications to estimate agreements in land use.
To identify and predict how conservation priorities change with the inclusion of PPGIS data, Whitehead et al. (2014) used the open-source spatial conservation prioritization software, Zonation (Moilanen 2007), to identify areas where there were synergies and/or conflicts between species distributions and social values derived from mapped data.

Modeling
Regression models have been applied to estimate probabilities of mapped positive and negative experiences in places (Table 3). Snizek et al. (2013) applied a logistic multinomial regression model on urban cyclists to estimate the probability of a positive experience versus no experience, and the probability of a negative experience versus no experience, depending on variables such as road environment, cycling facilities, and environmental factors. Using PPGIS survey data on positive and negative experiences in the city of Stockholm, Samuelsson et al. (2018) predicted probabilities that if an experience was to occur at the location, it is positive rather than negative, as modeled through spatial logistic regression on environment attributes such as residential or workplace density and closeness to water or major roads.
An open-source statistical modeling application for social value prediction, SolVES (Social Value of Ecosystem Services, http://solves.cr.usgs.gov/), developed by U.S. Geological Survey, quantifies the relationship between density of perceived social values Table 3. PPGIS analysis methods, purposes, example tools and application domains in Predict/Model phase. Goals in Predict/Model phase include generalizing and predicting mapped attributes to other places and contexts (prediction) or producing a representation of a system to make inferences (model). mapped through PPGIS and explanatory environmental variables (e.g. elevation, slope, distance to roads or water) using multiple regression modeling. SolVES was developed in the context of national forest planning (Sherrouse et al. 2011). Later, SolVES was integrated with the Maxent maximum entropy modeling software (Elith et al. 2010) to generate comprehensive social value maps and to produce robust models (Sherrouse et al. 2014, van Riper et al. 2017. PPGIS data have also been used to model people-based environmental exposure in urban context. Hasanzadeh et al. (2018) developed an individualized residential exposure model (IREM) to estimate local activity space of an individual, recognizing that the place exposure not only varies from one person to another in its geographical extents, but also from place to place in its magnitude. Mathematical models have also been applied to relate stated residential housing preferences with revealed preferences for the same individuals using empirical data describing the urban structure .
Sensitivity analysis refers to a set of methods that can be applied to either the Explain or Predict/Model phases, to identify the uncertainty of a model or system, typically by varying the inputs and then examining the effects on the outcomes. Sensitivity analysis has not yet achieved widespread use in PPGIS applications but would be useful given the ofteninherent limitations in the quantity and/or quality of PPGIS data used as input. For example, varying the quantity of PPGIS data entered in the spatial analysis through simulation or subsampling can demonstrate how the quantity of PPGIS data influences the spatial outcomes Pullar 2012, Brown et al. 2014b) or how the choice of parameters can affect the measurements derived from PPGIS data modeling (Hasanzadeh 2019).

Discussion
In this article, we propose a methodological framework to describe and guide method application for data gathered through PPGIS approaches (including PGIS and collecting VGI) to guide both research and practice. The reviewed analysis methods can be grouped in three phases, beginning from Explore and advancing to Explain and Predict/Model, highlighting the depth and breadth of tools and methods applied in spatial PPGIS data analysis. The past decade has witnessed transitions in this field from practical and demonstrative participatory mapping in different planning contexts to a focus on analytical possibilities and challenges associated with PPGIS data aggregation.
Although the cursory analysis, restricted to the Explore phase, is often sufficient to support practical management and planning needs, more complex methods have been developed within academia to drive scientific advancements. We encourage a shift towards more evidence-based, or knowledge-informed planning (Rydin 2007, Davoudi 2012 by integrating more complex methods of spatial analysis into the planning process, including elements of exploration, explanation, and prediction. This would entail greater attention to: different sampling strategies for eliciting spatial attributes; different approaches to aggregating spatial attributes (including overlap and conflict analyses); possibilities for integrating different forms of spatial attributes; addressing issues of commensurability and compatibility; and the development of automated analyses tools targeted for practitioners (building on Raymond et al. 2014, Brown 2017, Pietilä and Fagerholm 2019, Kenter et al. 2019). In addition, an essential future direction relates to determining which methods are most suitable in the context of planning; adapting analyses methods to different phases of planning and decision-making processes (Kahila-Tani et al. 2019), each with their different purposes and intended outcomes.
Researchers have an important role in ensuring that the PPGIS data and outputs can be readily applied in planning decisions by advancing methods that account for uncertainty. Identifying data thresholds and confidence intervals is basic to most scientific data, but the current analyses methods for PPGIS data do not estimate the validity of results, thus impeding the greater influence and impact of data and outputs on planning (see, e.g., Kyttä 2014, Brown andFagerholm 2015). New analysis methods indicating the level of certainty associated with the spatial PPGIS data and derived results are now required, especially for planning or management decision support (building on the spatial uncertainty classes described by Lechner et al. 2014).
Furthermore, as online PPGIS approaches are basically a questionnaires, questions like sampling strategy, sample size, and response rate are critical for interpretation of analytical outputs and also the possible analysis methods. A few important trends need to be highlighted here. First, most countries over the past decades have generally experienced reduced response rates leading to small and possibly biased sample sizes, even in PPGIS surveys (Brown et al. 2014a, Brown 2017. Second, partly due to the decreasing response rates, online panels are being increasingly used in PPGIS surveys (e.g. Bijker and Sijtsma 2017). Finally, PPGIS approaches are increasingly being used in citizen science projects to generate big data (Kelling et al. 2015, See et al. 2016. These trends raise questions about possible analyses on both exceedingly small or very large datasets, and the possible application of data weighting in all phases of PPGIS data analyses. Along with these, attention needs to be paid to 'what constitutes genuine collaboration in PPGIS studies?', highlighted by Kahila-Tani et al. (2019) in their review of over 200 urban and regional planning cases. In the field of public health, this has been addressed successfully in a few 'strongly participatory science' processes. The public not only participated in survey development and data collection, but also in the subsequent data analysis in a form of knowledge justice (Allen 2018). Similarly, Gray et al. (2018) highlighted the importance of inclusion of stakeholders and standardized communication about participatory socio-environmental modeling for potential innovation and new insights to collectively reason the environmental problems. In support of the datainformation-knowledge-wisdom hierarchy (Rowley 2007), we encourage collected PPGIS data to be made publicly available (following data protection regulations) to make it accessible for analysis and review by the wider public.
Our review indicates that PPGIS data analysis methods are heavily focused on mobilizing knowledge but limited in terms of methods for synthesizing and translating insights across knowledge systems into actionable insights. However, it is a fallacy to assume that more emphasis on analytical methods and tools alone will improve the communicability and usability of PPGIS data. We assert that the coupling of advanced analytical methods with sophisticated knowledge co-creation and deliberative valuation processes, otherwise referred to as pragmatic paradigm including negotiation , can facilitate communication and uptake of results (Ramirez-Gomez et al. 2017, Fagerholm et al. 2019b. Analytical methods and co-creation processes need to be developed in tandem to understand how and to what extent individual values articulated in PPGIS surveys become shared and social in collective environmental decision-making processes and how PPGIS approaches can build coalitions for social change (e.g., Kenter et al. 2019). Combining analytical tools and processes of knowledge co-creation in this way necessitates a detailed consideration of how issues of conflict, power, and equity are articulated and elicited in PPGIS studies.
PPGIS data analysis methods mostly describe the current state and often overlook the temporal dimension. In particular, there is a lack of methods that would embrace the interrelationships between changes in socio-ecological or urban planning regime/intervention, or changes caused by sudden shocks in systems such as by storms (i.e., place change), and changes in people's place-related values, perceptions, behavior, and preferences (Kendal and Raymond 2018). Longitudinal studies by Brown and Weber (2012) and Brown and Donovan (2014) measured PPGIS values in two surveys with 6 and 14 years in between, respectively, but to cater for the dynamism in global challenges, we need new PPGIS data analysis methods to understanding how people's place-based values form and change at various scales to comprehensively incorporate such dynamics.
The new technological developments offer wide possibilities to extend PPGIS data analysis, for example, to 3D and virtual environments, to elicit indoor values and preferences, or to engage with new forms of geo-navigation. From a visualization perspective, in the field of PPGIS, it is still rare to apply web-based and dynamic tools for visualization (such as open access tools GeoServer (http://geoserver.org/) and OpenLayers (https:// openlayers.org/)), which could assist in the creation of potentially effective PPGIS data visualizations. Moreover, the possibilities to harness artificial intelligence, automation, Internet of Things, big data collected through social media, and machine learning to PPGIS data analysis remains unexplored and presents opportunities for new enquiry.
The analysis methods presented here focus on spatial PPGIS data analysis possibilities, but we acknowledge that web-based mapping surveys also yield a rich source of qualitative analysis in terms of the non-spatial PPGIS data. In fact, mixed method approaches are prominently featured in the literature highlighting the advantage of linking participatory mapping, for example, with narrative analysis techniques to elicit landscape values and development preferences (Plieninger et al. 2018), social media to share memories of a place (Nummi 2018), or route tracking to monitor mountain bikers (Wolf et al. 2018).

Conclusions
For data gathered through PPGIS approaches, development of methods has reached a high level of maturity. In this article, we have summarized the depth and breadth of these methods. We provide a framework for scholars interested in PPGIS approaches to guide their thinking, observations, and interpretations. Our framework is based on the categorization of existing methods into three phases, Explore, Explain, and Predict/Model, aiming at different depths of understanding and knowledge discovery. We believe the framework is particularly useful for researchers new to the field, to rapidly appraise the different analytical methods available and to guide planning practitioners in using the appropriate techniques to address specific questions or problems.
We urge a renaissance in the field involving: 1) the development of methods considering knowledge-informed planning; 2) the development of easy to understand decision heuristics; 3) PPGIS data analysis in genuine collaboration with the public; 4) coupling analytical methods with deliberative valuation and knowledge co-creation processes, enabling the synthesis and translation of PPGIS insights across knowledge systems into actionable insight; 5) addressing temporal dimensions and dynamics in analysis; and 6) embracing recent technological developments. Considering PPGIS approaches in fields where it has not been applied yet, a more interdisciplinary PPGIS approach would support the emergence of further novel analysis methods. PPGIS approaches provide an operational bridge between social and natural/technical/engineering sciences, thereby offering considerable opportunity to address societal challenges and thus, provide integrated solutions to sustainability problems, as increasingly called for, for example, in terms of biodiversity, nature-based solutions, and climate resilience agendas (Kabisch et al. 2017, Diaz et al. 2019. This paper provides a solid platform for both understanding existing methods and the development of new methods for addressing such integrated sustainability challenges. Tiina Rinne (nee Laatikainen) is a Post-doctoral Researcher at the Department of Built Environment, Aalto University. Her research focuses on health promoting aspects of the built environment, the human-environment interactions primarily from the active living perspective and particularly to the endless possibilities of participatory mapping methods.

Kamyar Hasanzadeh is a Post-doctoral researcher at the Department of Built Environment, Aalto
University. His research is mainly focused on geospatial tool development and modeling for environmental health promotion studies.
Anna Broberg is COO and co-founder at Maptionnaire, where she commercializes research results on digital community engagement within the information technology and service industry. Her academic background lies in urban planning, transportation studies, and GIS. Marketta Kyttä works as a Professor in Land use planning at the Department of Built Environment, Aalto University, Finland. Her work concentrates on research with PPGIS methodology studying themes like environmental health promotion, social sustainability, age-and child-friendly environments and participatory planning.