Volume 4, Issue 5 p. 1216-1232
RESEARCH ARTICLE
Open Access

Monitoring public engagement with nature using Google Trends

Benjamin B. Phillips

Corresponding Author

Benjamin B. Phillips

Environment and Sustainability Institute, University of Exeter, Penryn, UK

Correspondence

Benjamin B. Phillips

Email: [email protected]

Search for more papers by this author
Katherine Burgess

Katherine Burgess

Natural England, Nobel House, London, UK

Search for more papers by this author
Cheryl Willis

Cheryl Willis

Natural England, Nobel House, London, UK

Search for more papers by this author
Kevin J. Gaston

Kevin J. Gaston

Environment and Sustainability Institute, University of Exeter, Penryn, UK

Search for more papers by this author
First published: 07 July 2022
Citations: 5
Handling Editor: Darragh Hare

Abstract

  1. How humans interact with nature affects the health of both people and ecosystems. Yet, long-term data on nature engagement are scarce because traditional survey methods are expensive, time consuming and require commitment over multiple years. Digital data sources (e.g. aggregated data from online searches) have major potential as a supplementary source of information and, in the absence of available data, as a proxy for more direct measurement of nature engagement.
  2. Using Google Trends, we created a list of refined and relevant search terms relating to diverse outdoor spaces and activities. We then compared trends in Google search volumes in England across both a 1-year and 10-year period to those from Google Community Mobility Reports, and from nationally representative survey data (Natural England's People and Nature Survey and the Monitor of Engagement with the Natural Environment).
  3. Search, survey and mobility data all support a general increase in public engagement with nature since 2009, and a more substantial increase during, or following, the initial national ‘lockdown’ period of the COVID-19 pandemic in England. Search volumes increased for many urban and rural outdoor green spaces (e.g. woodlands), blue spaces (e.g. reservoirs), exercise activities (e.g. walking, running and hiking) and explicitly nature-based activities (e.g. fishing, wild swimming and encouraging wildlife).
  4. Overall, volumes of Google searches were more closely related with longer-term (10-year) trends from survey data, than shorter-term changes during the COVID-19 pandemic. There were statistically significant relationships between search volumes, survey data (self-reported past behaviour) and mobility data (movement trends) for around half of comparisons. Of these, an average of 13–44% of variation in the data was explained.
  5. The findings show that Google Trends provides valuable information about public engagement with nature, which can help to supplement existing survey data by providing new insights about behavioural trends. The paper also provides a proof of concept for using Google Trends to understand changes in public engagement with nature, which could be applied to the many countries that lack long-term survey monitoring.

Read the free Plain Language Summary for this article on the Journal blog.

1 INTRODUCTION

How humans interact with the natural environment (henceforth ‘nature’) affects the health of both people and ecosystems. For people, interacting with nature can benefit mental and physical health and wellbeing (Bratman et al., 2019; Hunter et al., 2019; Twohig-Bennett & Jones, 2018). For nature, people's interactions can have both positive and negative influences on biodiversity. On the one hand, visitors (and their pets) to parks and nature reserves may disturb wildlife (Dertien et al., 2021; Larson et al., 2016; Weston et al., 2014), and alter vegetation communities, for example by leaving litter and dog excrement (De Frenne et al., 2022). On the other hand, exposure to nature is associated with pro-nature behaviours such as feeding birds, being a member of a conservation organisation, and environmental volunteering (Alcock et al., 2020; Martin et al., 2020; Richardson et al., 2020). For these reasons, it is important to monitor changes in how people are engaging with nature.

Traditionally, public engagement with nature has been monitored using methods such as questionnaire surveys. Despite online platforms substantially reducing the costs of surveys, long-term data on public engagement with nature remain scarce due to the scale and time commitment required to carry out a long-term survey. Surveys are also necessarily limited in how much information they can capture and are associated with various biases, including relying on self-reported, rather than actual, behaviour (Sedgwick, 2013). Using multiple methods can therefore help to triangulate true behavioural trends.

Digital data sources are increasingly being used in environmental research (Jarić et al., 2020; Ladle et al., 2016), including in studies of outdoor recreation. These have predominantly used data from Flickr (e.g. Graham & Eigenbrod, 2019; Lee et al., 2019; Mancini et al., 2019), as well as other platforms such as Reddit (Fox et al., 2021), Instagram (Grzyb et al., 2021) and Twitter (Tenkanen et al., 2017). Social media data has, for example, been used for studying drivers of outdoor recreation (Graham & Eigenbrod, 2019) and for understanding visitation patterns to national parks and protected areas (Wilkins et al., 2021).

Aggregated data from online searches offer another major, yet largely untapped, opportunity for understanding public engagement with nature. Specifically, a person's online searches relating to environmental spaces (e.g. woodlands) and associated activities (e.g. mountain biking) may indicate their interest, awareness and behavioural intentions to visit those spaces, and to engage in those activities. For example, a person carrying out an online search relating to camping (e.g. about potential camp sites, or about camping equipment) is likely to go camping soon, or to have been recently. Although this will not always be the case (and many people will engage in activities without carrying out an online search), the frequencies of such searches, across a population, may correlate with the frequencies of actual behaviours. If so, online search data could be used to supplement existing surveys. Online search data could also provide trends for public engagement with nature for the many countries that lack long-term monitoring, but which have prevalent internet usage. For example, there are more than 40 countries worldwide where at least 80% of the population use the internet (Roser et al., 2015). Yet, to our knowledge, no more than a handful of these countries are collecting long-term data on public engagement with nature.

Google Search is by far the most popular search engine globally, holding around 90% of the market share over the past decade (Statcounter, 2021). The Google Trends website is a platform that provides data on the relative volume of Google searches over time for a given search term or string. Data are available from 2004 to present, providing a valuable long-term dataset that extends beyond the duration of any direct surveys of nature engagement that we are aware of. Furthermore, Google Trends provides relative search volumes for different topics and related queries. This makes it possible to understand the context in which a search term is being used, and to create a refined search string that excludes irrelevant searches. Google Trends has previously been used for diverse research purposes (Jun et al., 2018), including in studies of epidemiology (Mavragani et al., 2018), politics (Mellon, 2014) and tourism (Li et al., 2018), as well as for assessing interest in, and awareness of, nature, protected areas and environmental issues (Cooper et al., 2019; Correia et al., 2018; Fukano et al., 2020; Le Nghiem et al., 2016; Mccallum & Bury, 2013; Proulx et al., 2014; Rousseau & Deschacht, 2020; Soriano-Redondo et al., 2017; Souza et al., 2021). However, no study has yet tested the use of online search data for exploring public engagement with diverse spaces and activities relating to nature.

Several other sources of data provide information about public engagement with nature with which to compare data from Google Trends (Figure 1). First, Google Community Mobility Reports provide movement trends for different regions, and for different types of spaces (including ‘parks’), since the onset of the COVID-19 pandemic (Google, 2021a). These are produced using aggregated, anonymized sets of data from smart phone users who have turned on the Location History setting. Second, Natural England's People and Nature Survey (PANS)—a nationally representative online questionnaire survey—provides information about attitudes and behaviours relating to the natural environment in England, since April 2020 (Natural England, 2021a). The preceding Monitor of Engagement with the Natural Environment (MENE) provides similar information from in-person surveys for 2009 to 2019, with nearly half a million responses across that period (Natural England, 2019). These data sources measure subtly different things and have different methods and sample biases, but are likely to be correlated. Whilst Google Trends data about environmental spaces and activities can be assumed broadly to reflect behavioural intentions, Google Mobility data reflect actual behaviour, and the PANS and MENE data reflect self-reported past behaviour. Furthermore, each of these data sources has its own associated biases and limitations. Google Trends data reflect those using Google Search (and therefore the internet). They therefore underrepresent people aged 55+, especially those aged 65+, who use the internet less than younger age groups (ONS, 2020). Google Mobility data are limited to people owning a smart phone with location data turned on. This may have similar socio-demographic biases. PANS and MENE are apparently the most representative data sources, but likely retain more subtle biases associated with response rates. Triangulating these different data sources can therefore provide greater confidence that biases associated with each method are not obscuring true behavioural trends.

Details are in the caption following the image
A summary of the paper, outlining the approach, the data sources that were used, and the comparisons that were made between these.
This study examines the potential of Google Trends for monitoring public engagement with nature (Figure 1). We address the following research questions:
  • Q1. How have volumes of Google searches relating to outdoor spaces and activities changed between 2009 and 2021?
  • Q2. How do trends on public engagement with nature from search data compare to those from survey and mobility data?
We use England as a study area because this is, to our knowledge, the only country for which long-term survey data are available on public engagement with nature. However, the implications of the findings are particularly important for countries and regions where no long-term survey data exist.

2 MATERIALS AND METHODS

We used the following process to examine the potential of Google Trends for monitoring public engagement with nature (Figure 1). 1. We created a broad initial list of possible search terms relating to different outdoor spaces and activities. 2. We used Google Trends to refine search terms to ensure that they identified relevant Google searches. 3. We extracted data from Google Trends for the final set of search terms across both a 1-year and 10-year period. 4. We assessed correlations between trends for the different search terms to identify shared patterns in their usage. 5. We compared trends to those from Google Community Mobility Reports, the People and Nature Survey (PANS), and the Monitor of Engagement with the Natural Environment (MENE).

2.1 Q1. How have volumes of Google searches relating to outdoor spaces and activities changed between 2009 and 2021?

We created an initial list of search terms covering a broad range of different outdoor spaces (henceforth ‘spaces’), including urban green spaces (e.g. parks, gardens, allotments), rural green spaces (e.g. woodlands, nature reserves, national parks), blue spaces (e.g. rivers, lakes, beaches), and associated practices (henceforth ‘activities’) based around those mentioned in PANS and MENE (Appendix S1). These were structured around the cultural ecosystem service framework by Fish et al. (2016), whereby activities are divided into ‘playing and exercising’, ‘creating and expressing’, ‘producing and caring’ and ‘gathering and consuming’. We subdivided environmental spaces into ‘urban’, ‘rural’ and ‘blue’ spaces. Where appropriate, search terms included common synonyms, alternative spellings and plural forms. As many activities can be carried out either indoors or outdoors, and in relation to nature or not (e.g. different types of photography), we added additional terms to specify relevant context for ambiguous search terms (e.g. ‘photography’ became ‘landscape photography + nature photography + wildlife photography’). Search terms therefore almost always consisted of a search string but, for simplicity, we refer to these here as search terms.

2.1.1 Refining search terms

From the initial list of search terms, we used the following process to create a refined shortlist. We ensured that search terms were those that were likely to be used, in the context of Google Search, to refer to engaging with nature (face validity) by rejecting those that we considered to be vague and often used in irrelevant contexts (e.g. ‘farmland’, ‘greenspace’ and ‘ocean’, which we assumed that people are unlikely to search for in relation to visiting those spaces).

We ensured that search terms predominantly represented relevant searches (content validity) using Google Trends to check the relevance of the most popular related queries (following Correia, 2019; Mellon, 2014). Google Trends' ‘interest over time’ is a measure of relative search volume. This is provided on a scale from 0 to 100, where 100 is the maximum search volume for the search term, as a proportion of all searches, within the specified criteria. Other values are scaled linearly as a proportion of this (Google, 2021b). We checked the most popular related queries for the past 1 year, 5 years and since 2004. We used an acceptance criterion that required the most popular related queries to be relevant to the intended subject. This included any of the ten most popular related queries that had ≥20 relative searches. For example, irrelevant related searches for ‘river’ included ‘river island’ (the high street shop), ‘river levels’ (relating to flooding) and ‘the river’ (the album by Bruce Springsteen). If the criterion was not met, we refined the search term to exclude irrelevant related searches (e.g. changing ‘river’ to ‘river—island—levels—the’). If this was still insufficient to meet the criterion, we made the search term more specific (e.g. changing ‘river’ to ‘river uk’, to exclude searches about rivers in other countries), then similarly refined this search term by excluding irrelevant related searches. For example, the final search term for ‘river’ was ‘river uk—island—levels—the—weather—google—longest—cottage—news—is—map’. In several cases, this process still resulted in too many irrelevant related queries, so we added the word ‘walks’ (e.g. changing ‘woodland’ to ‘woodland walks’) as this is the most frequent method by which people visit and experience most of these outdoor spaces (Natural England, 2019, 2021a). If these refinements did not result in the criterion being met, we did not use the search term. The complexity of each search term was limited by Google Trends 100-character limit. This meant that final search terms sometimes could not include all synonyms and plural forms, whilst sufficiently excluding irrelevant results. We acknowledge that there are multiple different search strings that could be used for each search term to meet our acceptance criterion, and that these do not necessarily provide similar trends.

The refinement process resulted in a total of 46 search terms: 16 relating to outdoor spaces and 30 relating to outdoor activities. The initial set of search terms, the refinement process and the final set of search terms are provided in Appendix S1.

2.1.2 Data extraction

For each search term, we extracted data from Google Trends using the ‘gtrendsR’ package (Massicotte & Eddelbuettel, 2020) in R 3.6.2 (R Core Team, 2021). Data were extracted for searches within England across two time periods: short-term (1 year: 1st April 2020 to 31st March 2021) and long-term (10 years: 1 March 2009 to 28 February 2019), to match the availability of survey and mobility data with which to compare trends (Figure 1). Whilst Google Trends provides data for each day for searches within the past 90 days, searches across a 1-year or 10-year period are provided for each week or month, respectively. As described above, we used Google Trends' ‘interest over time’ as a measure of relative search volume. As a control, we also extracted trends for 20 randomly-generated words that are unrelated to outdoor spaces and activities, which we used as search terms (following Ficetola, 2013). Finally, we assessed linear correlations between trends for each Google Trends search term to identify groups of search terms with similar usage. We provide a sample code for extracting, refining and plotting data from Google Trends in R in Appendix S4.

2.2 Q2. How do trends on public engagement with nature from search data compare to those from survey and mobility data?

We made the following comparisons between datasets (Figure 1): Comparison 1. Google Trends vs Google Community Mobility Reports data; Comparison 2. Google Trends vs the People and Nature Survey (PANS); Comparison 3. Google Community Mobility Reports with PANS; and Comparison 4. Google Trends vs the Monitor of Engagement with the Natural Environment (MENE).

2.2.1 Mobility data

Google Community Mobility Reports are aggregated, anonymized sets of data that provide movement trends for different regions and types of spaces (Google, 2021a). The data come from smart phone users who have turned on the Location History setting, which is turned off by default. Data are available for each day since 15th February 2020 and for each country, as well as for component cities and districts. The data provide the percent change relative to the baseline, which is ‘the median value, for the corresponding day of the week, during the 5-week period Jan 3–Feb 6’ (Google, 2021a). Google Mobility data are provided for the United Kingdom as a whole, rather than England (as per the other data sources) – though England makes up 84% of the population (ONS, 2021). Mobility data are split by six different categories that group together places with similar characteristics: ‘grocery & pharmacy’, ‘parks’, ‘transit stations’, ‘retail & recreation’, ‘residential’ and ‘workplaces’. We used data for the United Kingdom within the ‘parks’ category, which provides mobility trends for ‘places like local parks, national parks, public beaches, marinas, dog parks, plazas, and public gardens’ (Google, 2021a). Given that Google Trends data are provided for each week, we calculated the mean for each week of the year for the Google Mobility data.

2.2.2 Survey data

Natural England's People and Nature Survey (PANS) is a nationally representative online questionnaire survey of people in England which, since April 2020, provides information on attitudes and behaviours relating to the natural environment (Natural England, 2021a). The preceding Monitor of Engagement with the Natural Environment (MENE)—an in-person survey—provides similar information for 2009 to 2019, with nearly half a million responses across that period (Natural England, 2019). While the questions contained in these two surveys mostly differ, they have some overlap. We identified key survey questions that provide relevant comparisons with the Google Trends and Google Mobility data (Appendices S1 and S2). We recoded all multiple-choice questions as binary (Appendix S2). For each question, we calculated either the weekly mean (for PANS) or the monthly mean (for MENE) to match the interval of Google Trends data over the associated time period across which the data were being compared (1 year or 10 years, respectively) (Appendix S2).

2.2.3 Statistical analyses

Statistical analyses were carried out in R 3.6.2 (R Core Team, 2021). We tested the relationships between search volumes for Google Trends search terms and the other data sources using generalised additive mixed models (GAMM; ‘mgcv’ package; Wood, 2011). We modelled each variable separately because: (i) there was a high level of correlation between variables, (ii) we were interested in the predictive power of search volumes from individual Google Trends search terms, and (iii) the number of data points was relatively low (a maximum of 52 data points for the 1-year period, representing each week of the year). In all cases, we used an AR(1) autoregressive model to account for spatial autocorrelation, a random factor of month to account for seasonal trends, and the default thin plate regression spline as the smooth class, with a maximum of 3 degrees of freedom. Models were checked visually to meet assumptions. A summary of the main statistical comparisons is provided below, whilst a complete list of comparisons, models and results is provided in Appendix S3.

We compared short-term trends from Google Trends data, Google Mobility data and PANS data across a 1-year period (1 April 2020 to 31 March 2021), which was during the COVID-19 pandemic. This period was used because it covers the dates for which both Google Mobility and PANS data are available. For the first set of models (Comparison 1), we used Google Mobility data as the response variable (a single trend) and Google Trends search terms as explanatory variables. For the second set of models (Comparison 2), we used PANS data as response variables and Google Trends search terms as explanatory variables. Both PANS and Google Trends data consisted of many possible trends that could be compared. We therefore paired PANS questions with related Google Trends search terms, e.g. relating to the same outdoor space or activity (Appendix S1). Not all questions in PANS were asked on a weekly basis, so sample sizes varied between comparisons (Appendix S3). We also tested the relationship between each of these Google Trends search terms and two general PANS questions, relating to the frequency of time spent outside in green and natural spaces in the past 12 months, and the number of visits in the past 14 days (Appendix S3). For the third set of models (Comparison 3), we used Google Mobility data as the response variable (a single trend) and PANS data as explanatory variables. We compared the trend of every PANS question from Appendix S2 that related to outdoor spaces and activities that are typically associated with being away from home.

To provide a longer-term assessment, we compared trends from Google Trends to those from MENE across a 10-year period (1 March 2009 to 28 February 2019). For this fourth set of models (Comparison 4), we used MENE data as response variables and Google Trends search terms as explanatory variables. Again, both MENE and Google Trends data consisted of many possible trends that could be compared, so we paired MENE questions with related Google Trends search terms, e.g. relating to the same outdoor space or activity (Appendix S1). As for Comparison 2, we also compared all Google Trends search terms to two general MENE questions about the frequency of time spent outdoors in the past 12 months, and the number of visits in the past 7 days (Appendix S3). MENE questions q4 and q5 were changed in April 2016 from participants being asked about up to 10 recent visits, to being asked about a single, randomly selected visit (Appendix S2). This resulted in an artificial disruption to several trends (Figure S5.9; Natural England, 2019; see p. 22, point 3.17–3.19). We accounted for this in statistical models by including an additional random effect—a binary variable representing whether the time period was before or after April 2016, which was when the changes to the survey questions were made.

3 RESULTS

3.1 Q1. How have volumes of Google searches relating to outdoor spaces and activities changed between 2009 and 2021?

For most outdoor spaces and activities, relative search volumes from Google Trends were fairly stable between years, from 2009 to 2020, but showed strong seasonal trends (Figure 2; Figure S5.1–5.6). In most cases, trends for the 20 randomly-generated search terms, which were unrelated to outdoor spaces and activities, did not show seasonal trends (Figure S5.7). Seasonal changes in search volumes are therefore likely to reflect real changes in interest relating to the search terms, rather than seasonal changes in the use of Google Search. Search volumes for some spaces and activities have gradually increased over time, for example searches relating to gardens, walking, and drawing/painting nature/wildlife/landscapes. However, search volumes for around half of the randomly-generated search terms also showed a gradual increase (Figure S5.7). These trends could therefore partly be explained by changing patterns in the use of Google Search. Nonetheless, search volumes for some outdoor spaces and activities have seen much greater increases, for example searches relating to dog walking, hiking, wild swimming, encouraging wildlife, foraging and litter picking. Others have decreased, for example searches relating to zoos, mountain biking and photography of nature/wildlife/landscapes.

Details are in the caption following the image
Google Trends data for some example search terms relating to different outdoor spaces and activities, from 2009 to 2021. Lines were fitted using loess smoothing curves (span = 0.1) ± 95% confidence intervals. The grey area highlights the COVID-19 restrictions in England, with the darker grey blocks representing national lockdowns. The relative search volume is not comparable between different search terms. In most cases, search terms are paraphrased versions of a more extensive search string that was developed to exclude popular irrelevant searches (Appendix S1). Trends for the full list of search terms are provided in Appendix S5.

During the COVID-19 pandemic, long-term trends in search volumes were disrupted for many outdoor spaces and activities (Figure 2; Figure S5.1–5.6). During the initial national lockdown (March–May 2020), search volumes: (i) increased (relative to previous years) for many urban and rural outdoor green and blue spaces, especially gardens, allotments, canals, reservoirs, lakes, the countryside, woodlands and nature reserves, (ii) increased for many exercise activities such as walking, running and cycling, (iii) increased for some simple nature-based activities such as fishing, foraging and wild swimming, (iv) increased for activities relating to caring for nature, namely encouraging wildlife, feeding birds, gardening and growing vegetables and (v) increased for activities relating to eating and socialising outdoors, namely barbecuing and picnicking.

Most of the increases in search volumes during the initial national COVID-19 lockdown remained at unseasonably high levels throughout much of 2020. Unsurprisingly, search volumes were lower than usual for outdoor spaces that were closed during the initial national lockdown (e.g. zoos and National Trust sites), and for activities that typically require travel (e.g. diving, hiking and climbing). Whilst search volumes for most of these spaces and activities subsequently returned to pre-COVID levels, diving and climbing remained at unseasonably low levels (possibly due to travel restrictions), though hiking increased dramatically to unseasonably high levels during the summer, despite search volumes for national parks remaining unusually low across the entire year. Search volumes relating to drawing/painting nature/wildlife/landscapes increased temporarily during the initial lockdown period, whilst search volumes relating to nature/wildlife/landscape photography, and reading about nature/wildlife, showed relatively little change.

For the 20 randomly-generated search terms, there was no consistent change in trends during the COVID-19 pandemic (Figure S5.7). However, long-term trends were clearly disrupted (in most cases temporarily increasing during the initial lockdown period) for around one third of search terms (e.g. for the search terms: enter, combine and dilute). Search volume trends for the full list of outdoor spaces and activities, and for the 20 randomly-generated search terms, are provided in Figures S5.1–S5.7.

3.1.1 Correlations between the search volumes for different outdoor spaces and activities

There were consistent strong, positive correlations between relative search volumes from Google Trends for most outdoor spaces, especially over the 10-year period (Figure 3). However, there were some exceptions. Search volumes for woodlands and the countryside both had relatively weak positive correlations with search volumes for other outdoor spaces in both the short-term trends (1 year: 2020–2021) and long-term trends (10 years: 2009–2019). For the short-term trends, which cover the COVID-19 pandemic, search volumes relating to gardens and allotments only weakly correlated with search volumes for most other outdoor spaces. There were much weaker correlations, overall, between search volumes for different outdoor activities, than for different outdoor spaces (Figure 3). However, there were also some exceptions to this. Search volumes for many activities relating to playing and exercising (e.g. walking, golf and fishing) were strongly correlated with one another, and with search volumes for many outdoor spaces. In particular, search volumes for many activities associated with blue spaces (e.g. sailing, kayaking and surfing) were strongly correlated with one another and, unsurprisingly, with search volumes for some blue spaces (particularly for beaches).

Details are in the caption following the image
Correlations between the relative search volumes from Google Trends of different search terms over both the 1-year (top right) and 10-year (bottom left) period. These show consistent, strong positive correlations between many outdoor spaces and activities, especially over the 10-year period, and for those that are related (e.g. water-based activities). The colour and size of each square indicates the correlation coefficient (red = negative, blue = positive, larger and darker = stronger). Boxes are blank where p > 0.05.

There were relatively weak correlations between search volumes for activities relating to producing and caring (e.g. gardening, encouraging wildlife and litter picking), or relating to creating and expressing (e.g. drawing/painting and photography), with search volumes for other outdoor spaces and activities. However, search volumes for activities relating to gathering and consuming (e.g. barbecuing and picnicking) were relatively strongly correlated with search volumes for many outdoor spaces and activities. For the short-term trends (2020–2021), search volumes for activities that typically relate to being at or near home (e.g. gardening, feeding birds and running) were much less strongly positively correlated (and were sometimes negatively correlated) with search volumes for other outdoor spaces and activities that typically relate to being away from home (e.g. hiking, horse riding and sailing).

3.2 Q2. How do trends on public engagement with nature from search data compare to those from survey and mobility data?

3.2.1 Short-term trends (1 year: Comparisons 1–3)

Search, survey and mobility data all support a substantial increase in public engagement with nature during, or following, the initial lockdown period (Figure 2; Figure S5.8). For example, PANS found that 40% of adults surveyed in April–June 2020 reported that they had spent more time outside since coronavirus restrictions began, and 31% reported that they were exercising more in outdoor spaces (Natural England, 2021b).

Search volumes for many outdoor spaces and activities significantly correlated with Google Mobility data for parks (Comparison 1; p < 0.05 for 15/46 comparisons; mean adjusted R2 for statistically significant models = 0.44) (Figure 4; Appendix S3). The Google Trends search terms that most strongly predicted the Google Mobility data were for zoos, beaches, forests, national parks, kayaking, sailing, surfing, camping, climbing, fishing and horse riding. In each case, the adjusted R2 was ≥0.35. Unsurprisingly, for most outdoor spaces and activities that are typically associated with being at or near home (e.g. gardening, dog walking, feeding birds, reading), Google Trends search volumes were very poor predictors of Google Mobility data for parks.

Details are in the caption following the image
Modelled relationships (GAMMs) between Google Mobility data (% change in mobility for the ‘parks’ category) and Google Trends data (relative search volume [RSV] for different outdoor spaces and activities). Search volumes for many outdoor spaces and activities were significantly positively correlated with Google Mobility data for parks (plots are presented for the [15/46] trends where p < 0.05). The negative trend for ‘allotment’ may be because it is a much more locally-based outdoor space, so was likely searched about more when people were less ‘mobile’ overall (according to Google Mobility), for example during the COVID-19 lockdown periods. Points are model residuals for each week from April 2020 to April 2021. Lines are model estimates ± standard error. R2 is adjusted R-squared. The full set of plots is provided in Figure S5.10.

When comparing to PANS data (Comparison 2), search volumes rarely correlated with Google Trends data for the same outdoor spaces and activities (p < 0.05 for 5/36 comparisons; mean adjusted R2 for statistically significant models = 0.13) (Appendix S3; Figure S5.11). A similar amount of variation was explained by Google Trends search terms for PANS data relating to the frequency of time spent outside in green and natural spaces in the past 12 months (p < 0.05 for 13/36 comparisons; mean adjusted R2 for statistically significant models = 0.14) (Appendix S3; Figure S5.12). However, search volumes for outdoor spaces and activities were much stronger predictors of the mean weekly number of visits to green and natural spaces (p < 0.05 for 15/36 comparisons; mean adjusted R2 for statistically significant models = 0.28) (Figure 5; Appendix S3). The mean weekly number of visits to green and natural spaces was particularly well explained by search volumes for National Trust sites, forests, reservoirs, beaches, nature reserves, national parks, picnicking, fishing, sailing, and kayaking. In each case, the adjusted R2 was ≥0.25.

Details are in the caption following the image
Modelled relationships (GAMMs) between People and Nature Survey (PANS) data (mean number of visits to green and natural spaces in the past 14 days) and Google Trends data (relative search volume [RSV] for different outdoor spaces and activities). Search volumes for outdoor spaces and activities were often positively correlated with the number of visits to green and natural spaces (plots are presented for the 15/36 trends where p < 0.05). Points are model residuals for each week from April 2020 to April 2021. Lines are model estimates ± standard error. R2 is adjusted R-squared. The full set of plots is provided in Figure S5.13.

Data from PANS and Google Mobility for parks (Comparison 3) were rarely significantly correlated (p < 0.05 for 3/20 comparisons; mean adjusted R2 for statistically significant models = 0.31) (Appendix S3; Figure S5.14).

3.2.2 Long-term trends (10 years: Comparison 4)

Survey data provide evidence for a similar general increase in public engagement with nature since 2009 to that which is suggested by search data (Figure S5.9). For example, the MENE survey suggests that the number of visits to the natural environment increased by 38% between 2009 and 2019 (primarily due to increasing visits to urban parks) (Natural England, 2019). However, there are also some discrepancies between the two data sources that are difficult to explain. For example, there were substantial increases in search volumes relating to encouraging wildlife yet, according to survey data, the proportion of adults doing so in their gardens did not increase (Natural England, 2019).

Search volumes for outdoor spaces and activities frequently correlated with MENE data for the same spaces and activities (p < 0.05 for 22/37 comparisons; mean adjusted R2 for statistically significant models = 0.27) (Figure 6; Appendix S3). The Google Trends search terms that most strongly predicted the data for the associated MENE survey question (see Appendix S1) were for beaches, fishing, cycling, running, swimming, wild swimming, dog walking, kayaking, sailing, picnicking and hiking. In each case, the adjusted R2 was ≥0.30. There were a similar number of statistically significant relationships between Google Trends search terms and MENE data for the mean weekly number of visits to green and natural spaces, but the amount of variation explained was lower (p < 0.05 for 22/37 comparisons; mean adjusted R2 for statistically significant models = 0.14) (Appendix S3; Figure S5.16). This was similarly the case for the MENE data on the frequency of time spent outdoors, away from home, in the last 12 months (p < 0.05 for 20/37 comparisons; mean adjusted R2 for statistically significant models = 0.17) (Appendix S3; Figure S5.17). The Google Trends search terms that were the strongest predictors of these MENE survey questions also differed slightly. These were for woodland, dog walking, climbing and hiking, and all had an adjusted R2 of ≥0.20. There were also strong negative associations between some search terms and the number and frequency of outdoor visits, namely for sailing, surfing and mountain biking (Figures S5.16 and S5.17). This reflects the fact that search volumes for some activities have decreased closely aligned with the general increase in the number and frequency of visits.

Details are in the caption following the image
Modelled relationships (GAMMs) between the Monitor of Engagement with the Natural Environment survey (MENE) data and Google Trends data (relative search volume [RSV] for related outdoor spaces and activities). Search volumes for outdoor spaces and activities frequently correlated with MENE data for the same spaces and activities (plots are presented for the 22/37 trends where p < 0.05). The causes of negative correlations in three cases are unclear. The most prominent (for RSV: hiking) is likely because the ‘walking (inc. hiking)’ MENE category is dominated by people ‘walking’ in an outdoor space (e.g. in a local park), rather than ‘hiking’. The proportion of outdoor visits that focus on walking is likely lower at times of the year when hiking is more popular (e.g. in summer), because people are more engaged in other activities (e.g. water sports, camping, fishing). This reflects a limitation of the available data sources with which to compare to search volume trends. Points are model residuals for each month from March 2009 to March 2019. Lines are model estimates ± standard error. R2 is adjusted R-squared. The full set of plots is provided in Figure S5.15.

4 DISCUSSION

Our study has shown that Google Trends provides valuable, long-term information about public engagement with nature. Its particular strengths include being relatively simple to use, and its ability to provide information on a diverse range of outdoor spaces and activities. Below, we discuss the insights provided by Google Trends and consider the reasons for similarities and differences with survey and mobility data. We then provide some general guidance and next steps for using Google Trends to monitor public engagement with nature.

4.1 What does Google Trends tell us about public engagement with nature?

Google Trends provides further evidence for a general increase in public engagement with nature since 2009 (Natural England, 2019), and a more substantial increase during the COVID-19 pandemic (Natural England, 2021b). Other recent studies using Google Trends have similarly reported that search volumes for nature-related topics (forest, birds, nature, biodiversity, gardening, and vegetable plot) increased in many European countries during the COVID-19 pandemic (Rousseau & Deschacht, 2020), but that search volumes for national parks declined globally due to travel restrictions (Souza et al., 2021).

Our study goes beyond the findings provided by survey data, mobility data and previous studies, by comparing across a diverse range of outdoor spaces and activities. For example, we have shown that search volumes have increased for many outdoor activities, such as for dog walking, hiking and litter picking. This also includes activities that have recently become popular, but which are not covered within PANS, such as foraging and wild swimming. Furthermore, we have been able to distinguish between many similar activities that are coarsely grouped together within PANS, such as ‘boating, water sports or swimming outdoors’ (Natural England, 2021a). This is a particular strength of Google Trends, and there is potential to go even further. For example, another study found that search volumes for boredom, loneliness, worry and sadness increased in Europe and the USA during the start of the COVID-19 pandemic, suggesting that people's mental health may have been affected (Brodeur et al., 2021). The outdoors was, in many cases, likely being used to alleviate negative mental health and wellbeing impacts of the pandemic, given that PANS respondents most commonly reported visiting natural spaces for fresh air, physical and mental health, and to connect with wildlife/nature (Natural England, 2021b). Future research could try to link search volumes for nature engagement with search volumes relating to mental and physical wellbeing.

What might search volumes for outdoor spaces and activities tell us about the impacts of people on nature? Evidence from Google Trends for a general increase in outdoor recreation may suggest negative impacts because recreational activities (e.g. hiking/walking/dog walking/running, and wildlife watching) affect the behaviour and physiology of wild animals (Larson et al., 2016). Available studies indicate that these typically begin to exhibit behavioural impacts (mostly on flight initiation, or abundance) at distances of less than 100 m for wading and passerine birds, >400 m for hawks and eagles, 50 m for small rodents, and 1 km for large ungulates (Dertien et al., 2021). Such negative impacts are likely to be most pronounced in nature reserves and national parks. Indeed, disturbance from recreational activities is considered to be one of the main threats to biodiversity in terrestrial protected areas (Schulze et al., 2018). On the other hand, search volumes for some pro-nature behaviours have increased, such as those relating to encouraging wildlife, which consisted of searches about nest boxes, bug hotels, hedgehog houses and wildflower meadows. These could be having positive (but also potentially negative; Shutt & Lees, 2021) impacts on nature in people's gardens and local areas.

Volumes of online searches likely reflect not only changes in the intentions of people to engage with nature, but also their recent experiences. For example, the frequencies with which people encounter wildlife, which is affected by both a person's behaviour and the abundance and behaviours of wildlife, may affect subsequent Google searches about those species by people seeking to identify them, or to learn more about them. Indeed, a recent study found that changes in the relative volume of Google searches relating to bird species in the USA were well aligned with reported bird observations from citizen science data (Schuetz & Johnston, 2021).

4.2 How do trends from search data compare to those from mobility and survey data?

Overall, volumes of Google searches were more closely associated with longer-term trends from survey data, than with shorter-term changes during the COVID-19 pandemic. There were statistically significant relationships with behavioural trends from survey and mobility data for around half of comparisons and, of these, an average of 13%–44% of variation in the data was explained by each variable.

Although the comparisons between the search, mobility and survey data were non-significant in around half of cases, there are several factors that can help to account for this. In general, Google Trends was a better predictor of Google Mobility data, than of PANS survey data. This may be because Google Trends, which is biased towards frequent internet users, and Google Mobility data, which are limited to smartphone users with location history turned on, reflect a more similar socio-demographic. Alternatively, it may be because much of the survey data provides an imperfect comparison. For example, the survey questions about outdoor activities reflect the proportion of people having engaged in each activity during a specific recent visit (see Appendix S2). This is not the same as the proportion of people having recently engaged in each activity, which would provide a better comparison with Google Trends data. For other survey questions, there is a mismatch in timeframes, with some referring to behaviour across the past year as a whole. There is some evidence that this may have had an impact. Namely, Google Trends was a better predictor of the mean number of recent visits to outdoor spaces, than the frequency of visits in the past year. The previously mentioned groupings of activities in the survey data (e.g. ‘boating, water sports or swimming outdoors’) also mean that categories likely reflect the most popular component activity (e.g. for this category, we found a much stronger relationship with search volumes for sailing than for other water-based activities). Also, although Google Mobility data reflect actual behaviour rather than self-reported past behaviour, the available information about where activity is measured is rather opaque and somewhat contradictory. For example, the ‘parks’ category is described as including ‘places like local parks, national parks, public beaches, marinas, dog parks, plazas and public gardens’ (Google, 2021a), whereas elsewhere it states that ‘note, Parks typically means official national parks and not the general outdoors found in rural areas’ (Google, 2021c).

Given these mismatches and imperfect comparisons between data sources, it is encouraging that there was nonetheless good explanatory power overall. Whilst it is tempting to assume that the survey data are the most reliable source of information due to their nationally-representative socio-demographic sample, there are inherent biases associated with survey data, for example with the accuracy of self-reported behaviour, and with selection bias—particularly for online surveys (Schonlau et al., 2009). As described previously, some of the questions in the MENE data also subtly changed in 2016. This was considered to be an improvement to the survey (Natural England, 2019). However, it affected the way that people responded, and impacted the long-term trends. This highlights another challenge of long-term surveys, namely balancing opposing demands to improve question quality versus maintaining consistency over time. The findings therefore highlight the power of data triangulation for providing greater confidence in behavioural trends.

4.3 Key considerations when using Google Trends to monitor public engagement with nature

There are some reasons for uncertainty when trying to decipher exactly what information Google Trends can and cannot tell us. In the first instance, we can be confident that Google Trends data were based predominantly on relevant searches because (i) we used a multi-step validation process that tested different types of validity (following Mellon, 2014); (ii) we checked trends for some randomly-generated search terms to account for general changes in patterns of use in Google Search over time and (iii) we correlated search volumes for different outdoor spaces and activities and found logically consistent relationships, for example strong positive correlations between search volumes for blue spaces and water-based activities, and between gardens, allotments and garden-based activities (Figure 3). In light of criticisms of some previous research using Google Trends (e.g. see Correia, 2019; Ficetola, 2013), we encourage using a rigorous validation process as standard practice to ensure a greater level of confidence in resulting insights.

Beyond the reasons already outlined as to why Google Trends data are not likely to correlate perfectly with mobility or survey data, it is also important to be aware that in many or most cases, people will not search for outdoor spaces and activities beforehand if they are already familiar with them (such as for local or frequently visited spaces). Neither will Google searches for these spaces and activities be a representative sample of actual behaviour. In particular, Google searches likely reflect ‘new’ engagement in activities and spaces, and ‘destination’ recreation rather than ‘day-to-day’ recreation such as walking the dog. This was recently found to be the case in a similar study using Flickr (Graham & Eigenbrod, 2019). As our findings show, there has been very clear growth in search volumes for activities that have become increasingly popular in the UK in recent years, such as wild swimming and foraging. Subsequent declines in search volumes for such activities, however, do not necessarily reflect decreasing engagement in them. It could simply be that people have become sufficiently familiar with these activities that they no longer search for them online. Of course, it does depend on the activity. For some activities, even frequent participants will use Google Search to find tips, and to visit related websites and forums.

5 CONCLUSION AND NEXT STEPS

Our study shows that Google Trends provides valuable information about public engagement with nature, can help to supplement existing survey data by providing new insights, and can be combined to triangulate true behavioural trends. However, Google Trends is just one additional tool, with its own set of biases and limitations, which users should be aware of. We encourage applying a rigorous validation process when creating search terms to ensure a greater level of confidence in resulting insights. Nonetheless, this paper provides a proof of concept for using Google Trends to understand changes in public engagement with nature. This could be applied to the many other countries that lack long-term monitoring—particularly for the more than 40 countries worldwide where at least 80% of the population use the internet (Roser et al., 2015). Doing so should involve translating search terms, identifying different local forms of nature engagement and accounting for changing patterns of usage of Google Search, alongside a rigorous validation process such as that which we have described.

AUTHORS' CONTRIBUTIONS

B.B.P. conceived the ideas, collected and analysed the data, and led the writing of the manuscript; K.B., C.W. and K.J.G. contributed to the ideas and interpretation of the results. All authors contributed to manuscript drafts and gave final approval for publication.

ACKNOWLEDGEMENTS

The research was initiated whilst B.B.P. was carrying out an internship with Natural England as part of a NERC GW4+ Doctoral Training Partnership studentship from the Natural Environment Research Council [NE/L002434/1], with additional funding from the Cornwall Area of Outstanding Natural Beauty unit. The manuscript was improved by feedback from Kai Chan, Alison Johnston, Nathan Fox and the Associate Editor.

    CONFLICT OF INTEREST

    K.J.G. is an Editor for People and Nature, but was not involved in the peer review and decision making process.

    DATA AVAILABILITY STATEMENT

    The extracted data and associated code are available from the University of Exeter's institutional repository https://doi.org/10.24378/exe.4003 (Phillips et al., 2022). The raw data are freely available from the specified sources.