Filter Publications
A Text-Messaging Chatbot to Support Outdoor Recreation Monitoring Through Community Science
Figure 1. Volunteer participation rates at sites grouped by parking lot size (small, medium, and large). Participation rates were statistically significantly higher at sites with small parking lots (a) compared to medium and large lots (b). There was no significant difference in participation rates between sites with large- and medium-sized parking lots (b). Public land managers depend on reliable and readily available data about outdoor recreation in parks and greenspaces. However, traditional recreation monitoring techniques including visitor surveying and counting cannot be implemented over large spatial and temporal scales, especially in remote and undeveloped settings where monitoring is costly. To fill these data gaps, and thereby inform decision-making, this study develops and tests the efficacy of a novel recreation monitoring technique that engages visitors in data collection using a chatbot and text-messages. Drawing on knowledge and methods from community science and crowdsourcing, we present a relatively low-cost and low-barrier approach to counting and characterizing recreational visits on public lands. In an 18-month pilot implementation on a national forest in Washington, USA, we found that crowdsourced data collected using the chatbot were consistent with results of controlled counts and in-person surveys. Furthermore, some sites received relatively high participation rates, up to 12% of recreating parties, regardless of cellular connectivity at the site. This study, which is the first to engage public land usersin community science using a text-messaging chatbot for the purposes of studying outdoor recreation, demonstrates the potential for technology to support new community science approaches that involve visitors in land stewardship and the development of recreation monitoring systems.
Read moreWhere Wilderness is Found: Evidence From 70,000 Trip Reports
Figure 1. The study area, showing the boundaries of the Mt. Baker-Snoqualmie National Forest, wilderness areas within it and the trailheads for the 470 hikes in the Washington Trails Association hiking guide that we included in the sample. Map created with the R programming language using the sf, ggspatial and cowplot packages (Dunnington, 2022; Pebesma & Bivand, 2023; R Core Team, 2022; Wilke, 2019). Data from USDA Forest Service, Washington Trails Association, Washington Department of Transportation, Environmental Systems Research Institute map service and Natural Earth, facilitated by the basemaps and rnaturalearth packages for R (ESRI, 2009; Massicotte & South, 2023; Schwalb-Willmann, 2022; USDA Forest Service, 2019; WSDOT, 2023; WTA, 2023c). 1. Outdoor recreation is an essential way many people engage with nature. The provision of public spaces for recreation intersects with conservation practices motivated by intertwined social and ecological values, such as strict practices associated with the concept of ‘wilderness’. Debates persist about how such concepts and management practices influence people’s recreation experiences. 2. Many US public land management agencies facilitate opportunities for outdoor recreation, relying on management frameworks and tools intended to foster specific experiential qualities. But these frameworks and tools assume simplistic relationships between settings and people’s experiences, and managers rarely assess these relationships. 3. This study uses a data set of nearly 70,000 crowdsourced trip reports from a hiking website to understand the qualities of visitors’ experiences on trails. We study the geographic distribution of experiential qualities commonly associated with US wilderness areas: aesthetics, awe, challenge, pristineness, quietness, solitude and timelessness. Using analytical methods that rely on machine learning and natural language processing, we identify these experiential qualities in trip reports from hundreds of routes, and use generalized linear models to analyse relationships between the frequency of each experiential quality and the route’s administrative, built, biophysical, geographic and social settings. 4. We find that four of the seven experiential qualities (aesthetics, awe, challenge and solitude) are commonly described in trip reports, each appearing in 15%–55% of manually coded reports. The extent to which setting characteristics explained variability in experiences differed, ranging from 34% of the variability in the proportion of trip reports describing aesthetics to 55% for awe. The setting characteristics associated with each experiential quality also differed, with characteristics such as trail mileage and summit destinations having stronger influences on experiential qualities than characteristics such as wilderness designation. 5. Synthesis and applications. Our findings suggest the need to consider more diverse variables in experience–setting relationships, develop more robust models to characterize those relationships and create new data sources to represent understudied variables. These advances would help empirically inform and improve frameworks and tools used for recreation and wilderness planning and monitoring, and potentially promote more responsive management to evolving social– ecological values.
Read moreAn Open‐Source Image Classifier for Characterizing Recreational Activities Across Landscapes
Environmental management increasingly relies on information about ecosystem services for decision-making. Compared with regulating and provisioning services, cultural ecosystem services (CES) are particularly challenging to characterize and measure at management-relevant spatial scales, which has hindered their consideration in practice. Social media are one source of spatially explicit data on where environments support various types of CES, including physical activity. As tools for automating social media content analysis with artificial intelligence (AI) become more commonplace, studies are promoting the potential for AI and social media to provide new insights into CES. Few studies, however, have evaluated what biases are inherent to this approach and whether it is truly reproducible. This study introduces and applies a novel and open-source convolutional neural network model that uses computer vision to recognize recreational activities in the content of photographs shared as social media. We train a model to recognize 12 common recreational activities to map one aspect of recreation in a national forest in Washington, USA, based on images uploaded to Flickr. The image classifier performs well, overall, but varies by activity type. The model, which is trained with data from one region, performs nearly as well in a novel region of the same national forest, suggesting that it is broadly applicable across similar public lands. By comparing the results from our CNN model with an on-site survey, we find that there are apparent biases in which activities visitors choose to photograph and post to social media. After considering potential issues with underlying data and models, we map activity diversity and find that natural features (such as rivers, lakes and higher elevations) and some built infrastructure (campgrounds, trails, roads) support a greater diversity of activities in this region. We make our model and training weights available in open-source software, to facilitate reproducibility and further model development by researchers who seek to understand recreational values at management-relevant scales—and more broadly provide an example of how to build, test and apply AI to understand recreation and other types of CESs.
Read moreNo Walk in the Park: the Viability and Fairness of Social Media Analysis for Parks and Recreational Policy Making
Figure 1. Social Vulnerability Index classes by census tract in Seattle (WA). The yellow regions depict the selected parks. Recent years have seen an increase in the use of social media for various decision-making purposes in the context of urban computing and smart cities, including management of public parks. However, as such decision-making tasks are becoming more autonomous, a critical concern that arises is the extent to which such analysis are fair and inclusive. In this article, we examine the biases that exist in social media analysis pipelines that focus on researching recreational visits to urban parks. More precisely, we demonstrate the potential biases that exist in different data sources for estimating the number and demographics of visitors through a comparison of image content shared on Instagram and Flickr from 10 urban parks in Seattle, Washington. We draw a comparison against a traditional intercept survey of park visitors and a multi-modal city-wide survey of residents. We evaluate the viability of using more complex AI facial recognition algorithms and its capabilities for removing some of the presented biases. We evaluate the AI algorithm through the lens of algorithmic fairness and its impact on sensitive demographic groups. We show that despite the promising results, there are new sets of concerns regarding equity that arise when we use AI algorithms.
Read moreNext-Generation Visitation Models using Social Media to Estimate Recreation on Public Lands
Figure 1. Locations of geotagged social media posts made by visitors to public lands in WWA and NNM. Points represent the latitude and longitude where a Flickr photograph (purple) or tweet (green) was created. For Instagram, points represent places to which images were assigned by users (blue). Larger points represent a greater number of Instagram images from the location. Study sites are contained within the management units (shaded grey). Figure created using R48 version 3.5.3. Outdoor and nature-based recreation provides countless social benefits, yet public land managers often lack information on the spatial and temporal extent of recreation activities. Social media is a promising source of data to fill information gaps because the amount of recreational use is positively correlated with social media activity. However, despite the implication that these correlations could be employed to accurately estimate visitation, there are no known transferable models parameterized for use with multiple social media data sources. This study tackles these issues by examining the relative value of multiple sources of social media in models that estimate visitation at unmonitored sites and times across multiple destinations. Using a novel dataset of over 30,000 social media posts and 286,000 observed visits from two regions in the United States, we compare multiple competing statistical models for estimating visitation. We find social media data substantially improve visitor estimates at unmonitored sites, even when a model is parameterized with data from another region. Visitation estimates are further improved when models are parameterized with on-site counts. These findings indicate that while social media do not fully substitute for on-site data, they are a powerful component of recreation research and visitor management.
Read moreResearch Note: Residential Distance and Recreational Visits to Coastal and Inland Blue Spaces in Eighteen Countries
Figure 1. Given residential locations (correct to three decimal degrees) of the 15,216 respondents included in analysis. The map of Spain includes respondents resident in the autonomous city of Melilla. Respondents resident in the Canary Islands, Azores, and Madeira are not displayed. Varied categorisations of residential distance to bluespace in population health studies make comparisons difficult. Using survey data from eighteen countries, we modelled relationships between residential distance to blue spaces (coasts, lakes, and rivers), and self-reported recreational visits to these environments at least weekly, with penalised regression splines. We observed exponential declines in visit probability with increasing distance to all three environments and demonstrated the utility of derived categorisations. These categories may be broadly applicable in future research where the assumed underlying mechanism between residential distance to a blue space and a health outcome is direct recreational contact.
Read moreRecreational Use in Dispersed Public Lands Measured Using Social Media Data and On-Site Counts
Figure 1. MBSNF (boundary in gray) in western WA. Left: The 15 trail areas observed in this study are dispersed across 4 ranger districts (MTB: Mt. Baker, DAR: Darrington, SKY: Skykomish, SNO: Snoqualmie). Right: Geotagged Flickr photos (purple points) taken in and around the MBSNF. Sources: USFS and Flickr. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Outdoor recreation is one of many important benefits provided by public lands. Data on recreational use are critical for informing management of recreation resources, however, managers often lack actionable information on visitor use for large protected areas that lack controlled access points. The purpose of this study is to explore the potential for social media data (e.g., geotagged images shared on Flickr and trip reports shared on a hiking forum) to provide land managers with useful measures of recreational use to dispersed areas, and to provide lessons learned from comparing several more traditional counting methods. First, we measure daily and monthly visitation rates to individual trails within the Mount Baker-Snoqualmie National Forest (MBSNF) in western Washington. At 15 trailheads, we compare counts of hikers from infrared sensors, timelapse cameras, and manual on-site counts, to counts based on the number of shared geotagged images and trip reports from those locations. Second, we measure visitation rates to each National Forest System (NFS) unit across the US and compare annual measurements derived from the number of geotagged images to estimates from the US Forest Service National Visitor Use Monitoring Program. At both the NFS unit and the individual-trail scales, we found strong correlations between traditional measures of recreational use and measures based on user-generated content shared on the internet. For national forests in every region of the country, correlations between official Forest Service statistics and geotagged images ranged between 55% and 95%. For individual trails within the MBSNF, monthly visitor counts from on-site measurements were strongly correlated with counts from geotagged images (79%) and trip reports (91%). The convenient, cost-efficient and timely nature of collecting and analyzing user-generated data could allow land managers to monitor use over different seasons of the year and at sites and scales never previously monitored, contributing to a more comprehensive understanding of recreational use patterns and values. Related News Univ. of Washington program Nature and Health studies link between environment and well-being (December 30, 2021)
Read morePhotos, Tweets, and Trails: Are Social Media Proxies for Urban Trail Use?
Figure 1: Maps of annual average daily trail traffic (AADTT), annual average photo-user days (AAPUD), and annual averageTwitter-user days (AATUD) Decision makers need information on the use of, and demand for, public recreation and transportation facilities. Innovations in monitoring technologies and diffusion of social media enable new approaches to estimation of demand. We assess the feasibility of using geo-tagged photographs uploaded to the image-sharing website Flickr and tweets from Twitter as proxy measures for urban trail use. We summarize geo-tagged Flickr uploads and tweets along 80 one-mile segments of the multiuse trail network in Minneapolis, Minnesota, and correlate results with previously published estimates of annual average daily trail traffic derived from infrared trail monitors. Although heat maps of Flickr images and tweets show some similarities with maps of variation in trail traffic, the correlation between photographs and trail traffic is moderately weak (0.43), and there is no meaningful statistical correlation between tweets and trail traffic. Use of a simple log-log bivariate regression to estimate trail traffic from photographs results in relatively high error. The predictor variables included in published demand models for the same trails explain roughly the same amount of variation in photo-derived use, but some of the neighborhood socio-demographic and built-environment independent variables have different effects. Taken together, these findings show that both Flickr images and tweets have limitations as proxies for demand for urban trails, and that neither can be used to develop valid, reliable estimates of trail use. These results differ from previously published results that indicate social media may be useful in assessing relative demand for recreational destinations. This difference may be because urban trails are used for multiple purposes, including routine commuting and shopping, and that trail users are less inclined to use social media on trips for these purposes.
Read moreMeasuring Recreational Visitation at US National Parks with Crowd-Sourced Photographs
Figure 1. Average monthly visitation in each park, from 2007 to 2012, expressed as the percent of total visits measured by NPS and Flickr PUD. Land managers rely on visitation data to inform policy and management decisions. However, visitation data is often costly and burdensome to obtain, and provides a limited depth of information. In this paper, we assess the validity of using crowd-sourced, online photographs to infer information about the habits and preferences of recreational visitors by comparing empirical data from the National Park Service to photograph data from the online platform Flickr for 38 National Parks in the western United States. Using multiple regression analysis, we find that the number of photos posted monthly in a park can reliably indicate the number of visitors to a park in a given month. Through additional statistical testing we also find that the home locations of photo-takers, provided voluntarily on an online profile, accurately show the home origins of park visitors. Together, these findings validate a new method for measuring recreational visitation, opening an opportunity for land managers worldwide to track and understand visitation by augmenting current data collection methods with crowd-sourced, online data that is easy and inexpensive to obtain. In addition, it enables future research on how visitation rates change with changes in access, management or infrastructure, weather events, or ecosystem health, and facilitates valuation research, such as travel cost studies.
Read moreSpatial and Temporal Dynamics and Value of Nature-Based Recreation, Estimated via Social Media
Figure 1. Conserved lands in Vermont. Conserved lands provide multiple ecosystem services, including opportunities for nature-based recreation. Managing this service requires understanding the landscape attributes underpinning its provision, and how changes in land management affect its contribution to human wellbeing over time. However, evidence from both spatially explicit and temporally dynamic analyses is scarce, often due to data limitations. In this study, we investigated nature-based recreation within conserved lands in Vermont, USA. We used geotagged photographs uploaded to the photo-sharing website Flickr to quantify visits by in-state and out-of-state visitors, and we multiplied visits by mean trip expenditures to show that conserved lands contributed US $1.8 billion (US $0.18–20.2 at 95% confidence) to Vermont’s tourism industry between 2007 and 2014. We found eight landscape attributes explained the pattern of visits to conserved lands; visits were higher in larger conserved lands, with less forest cover, greater trail density and more opportunities for snow sports. Some of these attributes differed from those found in other locations, but all aligned with our understanding of recreation in Vermont. We also found that using temporally static models to inform conservation decisions may have perverse outcomes for nature-based recreation. For example, static models suggest conserved land with less forest cover receive more visits, but temporally dynamic models suggest clearing forests decreases, rather than increases, visits to these sites. Our results illustrate the importance of understanding both the spatial and temporal dynamics of ecosystem services for conservation decision-making.
Read more