From national recreation statistics and mobile data to local estimates of recreational activity in Finland

The focus in examining the impacts of multi-local living on recreational activity is on second-home mobility because it comprises the largest multi-local population group in Finland (Rannanpää et al., 2022). The potential local recreational use of the area is calculated based on data from the National Inventory of Recreational Uses of Nature (LVVI, 2020), which received survey responses from nearly 8,700 Finns. The national inventory is utilised in research in the field, and it promotes the construction and management of outdoor recreation services. The participants in the survey were randomly extracted from Statistics Finland’s (Tilastokeskus, 2023) population database concerning the population of Finland. These data were used to calculate the potential monthly close-to-residence recreation use in grid i (RActivity_ji) based on the following calculation: (1), RActivityij=Recreationij*Populationij {RActivity_{ij}} = {Recreation_{ij}}*{Population_{ij}} where the variable Recreation denotes the annual number of participations in outdoor recreational activity of the region k based on the definition in the LVVI survey. The average value for outdoor recreational activity (Recreation) was 182 times in 2020, of which 77% occurred within walking distance from home used in this study to account for activities (LVVI, 2020). Since no information is available about the monthly recreational activity, the annual recreational activity is used to estimate monthly activity. This estimation is based on recreational activity statistics from another monthly recreational activity study (Strain et al., 2021) and on the statistics of monthly use of second homes in Finland (Voutilainen et al., 2021), which were used to calculate monthly activity and further divide the volume of annual recreation activities into monthly ones. The calculation of the monthly population statistics in grid i (Population_ij) is described in sections 3.2 and 3.3. The original spatial units in the mobile phone data are described in Figure 1 along with urban-rural categories and division into regions. To facilitate a comparison between the different spatial units, the results of the grid data are summarized based on the grid used in the mobile phone data.

Due to the strong association between multi-local and urban-rural living styles, the results of the local variations in recreational activity are described using urban-rural typology (see Figure 2), which divides areas into seven categories: Three urban and four rural categories (Helminen et al., 2020). In the typology, urban categories contain inner urban areas, outer urban areas, and peri-urban areas. Local centres in rural areas are population centres which are located outside of urban areas. Rural areas close to urban areas are areas that have a rural character but are functionally connected and are close to urban areas. Core rural areas have intensive land use, with a relatively dense population and a diverse economic structure at the local level. Sparsely populated rural areas consist of small, dispersed settlements that are located at a distance from each other, and most of the land area is forested. Because the typology is implemented using a nationwide 250m × 250m cell grid, the urban-rural category of the mobile phone grids was determined based on the largest population of the typology on a grid.

Figure 2:

Urban-rural categories, regions, and mobile phone data grids in Finland. In the Figure, urban classes have been combined into a single class to make urban areas stand out more clearly.

We estimate local recreation activity with a population data that are compiled from the YKR database (Monitoring System of Spatial Structure and Urban Form), which is created and maintained by the Finnish Environment Institute, and it comprises data on population, housing, workplaces, and travel to work from the year 2019 for each of 319,841 square-shaped populated grid cells (250m × 250m) into which the territory of Finland is divided (Figure 2). The YKR database was used to estimate the volumes of seasonal populations by population grids (see equation 1, Population_ij).

The average monthly population (Population_ij) describes the population after considering the monthly use patterns of primary homes and summer cottages (see Adamiak et al., 2017). For 2019 it was calculated as: (2), Populationij=Popi+sh2019i*users*usej−RP2019i*access*usej \matrix{{{Population_{ij}}} \hfill & { = {Pop_i} + \left( {sh{{2019}_i}*users*{use_j}} \right)} \hfill \cr {} \hfill & { - \left( {RP{{2019}_i}*access*{use_j}} \right)} \hfill \cr } where the variable Pop_i denotes the registered population of a given grid i in the year 2019. The variable sh2019_i stands for the number of summer cottages in grid i in 2020 and the constant users denotes the average number of people using one summer cottage, which is 4.7, according to the Finnish Free Time Residence Barometer (Voutilainen et al. 2021). The variable access describes the shares of grid populations that had access to a summer cottage in 2020, and based on the LVVI (2020) study it was set to 0.43. The variable use describes the monthly use of summer cottages in 2021 and it is based on the Finnish Free-Time Residence Barometer (Voutilainen et al., 2021). The parameters used in the calculation are derived from different years because comprehensive surveys on recreational use are rarely carried out in Finland. The Finnish Free Time Residence Barometer was carried out in March and April 2021, with a random sample of 5000 second-home owners.

We also estimate local recreational activity using Telia’s Crowd Insights commercial mobile phone activity data, which reports the number of people dwelling in an area over a night, making it possible to distinguish multi-local living from day-to-day mobility. The dataset covers all of Finland and estimates the average number of people per month living in the grids (abbreviation Population_ij in equation 1), as visualised in Figure 1. Mobile phone data have been successfully utilised earlier in studies mapping the extent of multi-local living (Rannanpää et al., 2022) and related mobility in Finland during the Covid-19 pandemic (Willberg et al., 2021). The utilised mobile phone data are based on the period from March 2020 to February 2021, which allows it to be compared with data from the Finnish Free-Time Residence Barometer (Voutilainen et al., 2021). Mobile phone data contain information on the connection of mobile phones to base stations and this approach makes the material more comprehensive compared to, for example, localized call data, because the material also includes subscriptions that were not used (see Burkhard et al., 2017).

Despite the dynamic aspects of mobile phone data, its use entails several challenges. In Finland, Telia enjoys roughly one-third of the market share (Data.Traficom, 2023) and therefore, in mobile phone data, the number of people has been weighted to represent the total population based on the operator’s market share in individuals’ place of residence. Second, it should be noted that estimating multi-local living with the monthly average population is an approximate variable, as it includes all overnight mobility that crosses municipal boundaries. According to estimates, a significant part of mobility is related to second homes, about 90,000,000 annual overnight stays and leisure tourism, about 45,000,000 annual overnight stays which together cover more than 60% of all mobility in Finland (Rannanpää et al. 2022).

Our second objective was to develop a model that could predict local recreational activity using mobile phone data and other explanatory variables that could be easily compiled across many places. These covariates include urban-rural category (size of the grid), location (distance to urban centres of various sizes, services, and national parks), land use statistics (including water access, forest cover, and biodiversity of forest areas), basic infrastructure (broadband availability), and other statistics (population density, teleworking potential) (Table 1). We estimated a varied set of candidate regression models, including several functional forms where we defined linear relationship between the mobile phone data and the population statistics and other regressors in R. The changes in the estimated local recreational activity in mobile phone data were analysed using a regression model that explains the monthly difference DPopulation_ij from average monthly recreational activity of the grid based on the following model: (3), DPopulationij=α+βCategory+βLocation+βLand+ βInfrastructure+βPopulation+ε \matrix{{{DPopulation_{ij}}} \hfill & { = \alpha + \beta Category + \beta Location + \beta Land} \hfill \cr {} \hfill & { + \;\beta Infrastructure + \beta Population + \varepsilon } \hfill \cr } where:

Category refers to urban-rural category of the grid;

Regression models were estimated using spatial regressions (SAR) because the monthly differences in the recreational activity were spatially autocorrelated indicating that the values of variables at nearby grids are not independent from each other (Tobler, 1970). Main source of spatial autocorrelation in the recreational activity relates to human activity, because the data are affected by processes that connect different places, including spatial interaction and spatial diffusion processes; or by phenomena that extend over space to occupy regions rather than point locations (Odland, 1988). Based on the spatial dependence diagnostics functions from R’s spdep package (Bivand, 2022), the SAR model is estimated using a spatial error model, where the error terms across different spatial units are correlated. The spatial error model is defined as follows: (4). y=Xβ+ε, where ε=λWε+ξ y = X\beta + \varepsilon ,\;{\rm{where}}\;\varepsilon = \lambda W\varepsilon + \xi

In the regression model, λ is the spatial autoregressive coefficient for the error lag (to distinguish the notation from the spatial autoregressive coefficient ρ in a spatial lag model), and ξ is an uncorrelated and homoscedastic error term. Spatial error dependence may be interpreted as a nuisance (and the parameter λ as a nuisance parameter) in the sense that it reflects spatial autocorrelation in measurement errors or in variables that are otherwise not crucial to the model. The spatial regression is used because traditional nonspatial models might give biased or inefficient estimators of regression coefficients if spatial autocorrelation is omitted because it violates the assumption of independently and identically distributed (i.i.d.) errors of most standard statistical procedures (Anselin & Bera, 1998).

The change in the average population and local recreation activity during the year occurs in large areas (Figure 3). The grid and mobile phone data have remarkable differences because the latter detects monthly variations in average population much better than static gird data which ignores mobility. This different ability of the databases to estimate the local recreation use is most evident in January and March. During these months the grid data cannot detect temporal variation in the average population around the tourism centres in Lapland (Figure 3). In July, when people spend more time in their second homes than in January, the differences between databases decrease. The geographical changes in the recreation activity are remarkable, especially in July (Figure 3) when local recreational activity increases mostly on the Baltic coast, Lakeland Finland, and the Eastern and Northern parts of Finland.

Figure 3:

Differences in the local recreational activities by grid data and mobile phone data approaches.

The total local recreational activity in Finland is 654–683 million visits, depending on the type of database. In the urban-rural classes, the local recreation activity statistics vary considerably between calculations. The widest differences are found between mobile phone data and permanent population statistics (Table 2). With mobile phone data, the total annual recreation activity of the sparsely populated rural areas is 50.5% higher than the corresponding figure for the permanent population statistics. The change in recreational activity demonstrates the high impact of multi-local living on temporal recreational use of areas. Moreover, in the local centre of rural areas and in the core rural areas, the permanent population statistics underestimate the recreational activity, as the difference between permanent population statistic and mobile phone data estimations are 13.6% and 7.1% (Table 2). In contrast, the permanent population statistics of the other urban-rural categories tend to overestimate recreational activities (Table 2).

The differences in monthly recreational activity in urban and rural areas confirm the changing geographical pattern of recreational activity (Table 2). In the mobile phone data, the largest changes in recreational activity occur in sparsely populated rural areas, where the activity in July increased by 97.7% compared to the activity in January (Table 2). Furthermore, in the core rural and local centre of the rural area, the activity increases by almost 65.8% and in core rural areas by 57.8% (Table 2). The corresponding changes in urban areas and rural areas close to urban areas are smaller. Although, even in these categories the amount of recreational activity increases in July because mobility activity is higher then. The smallest increase occurs in the inner urban areas, where the growth in the activity is about 34% (Table 2). However, this is less than the increase in mobility activity between January and July (42%), so overall multi-local living reduces recreational activity in urban areas and in rural areas close to urban areas.

The volumes of the recreational activity in regions are further investigated with mobile phone data by using the regional recreational activity profiles. These profiles demonstrate that multi-local living changes the volumes of recreational activities between the regions and between urban–rural areas within the regions (Figure 4). In particular, the mobility from urban to rural areas during the summer can be detected from the recreation activity profiles. For instance, the volume of recreational activities in the Etelä-Savo region increased by 88.7% from January to July, while in the Uusimaa region the volume increased by 35.4% during the same time period. These activity changes relate to the change in the average population of the regions. The biggest change in the average population occurs in Uusimaa region in July, when the average population drops by 64,616 inhabitants compared to January. In contrast, the average population in Etelä-Savo region increases by 52,842 inhabitants in July compared to January. However, there are also temporal variations in the recreational activities because, for example, one of the highest peaks in the recreational activities in Lappi region is in spring (Figure 4). The regional profiles of recreational activities also highlight the mobility of people between cities and rural areas. For instance, in the Etelä-Savo region the change in recreational activity concerns main sparsely populated rural areas as the increase corresponds to 58.3% and in all rural areas it is 79% of the total increase of recreational activities.

Figure 4:

Monthly variation of recreational activity by region. Calculation is based on mobile phone data.

The local recreational monthly activity changes in average populations are analysed using regression analysis to detect the kind of areas where the recreational activity varies during the year (Table 3). The ability of the models to explain variation in recreational activity varies between months. The best goodness of fit measured with r-squared value is obtained in July when the models explain 48% of the total variation in recreational activity (Table 3). The r-squared value is about 10% in March and 6% in January, which indicates that the changes in recreation activity are more random during the winter and spring than in the summer. The following interpretation of the regression coefficients focuses on the SAR model as based on AIC values and Lagrange multiplier diagnostics (LMerr 3431,7, p-value <0,001) this model seems to explain the recreation activity and allocation of the average population better than the OLS model does (Table 3).

The regression models explain the changes in the local recreational activity of the grids during the year. The comparisons of regression coefficients between models reveal the differences in population changes in different months. For an urban variable, the smaller negative coefficient in January model than in July model (Table 3), shows that there will be more recreational use in urban areas in January than in July (see Table 3). Similarly, a location close to services, such as schools, or the high availability of broadband increases the growth of the recreation activity in January and March compared with July (Table 3). However, the increasing distance to the city, which has over 200,000 inhabitants, increases the recreation activity in January and July.

Respectively, recreation activity increases in July in sparsely populated rural areas, while it decreases in these areas in January. Moreover, long distances to services and poor availability of broadband are associated with increasing recreational activity in July (Table 3). These variables imply that the increase in local recreation activity is somehow linked to the residential environment, as people move from denser urban areas to rural areas to compensate for the lack of the nature in urban environments (Strandell & Hall, 2015). This interpretation is strengthened by other variables because the recreational activity increases in July also in areas where forests and water bodies account for a large proportion of the surface area, whereas these variables decrease the recreational activity in January (Table 3). In March, only water coverage decreases the recreational activity. The activity growth that takes place in July also occurs in areas where the number of second homes is high based on the proportion of building stock (Table 3).

Compared with July, in March, the increase of the recreational activity is explained by the high proportion of second homes in the land use, which indicates that activity growth takes place especially in tourism centres in the Eastern and Northern parts of Finland (Table 3). The interpretation of the growing recreational activity around the tourism centres is partly supported by the regression coefficients of the proportion of nature reserve from the total grid area as the increasing proportion of nature reserves decrease the recreational activity in July (Table 3). The largest nature reserves in Finland are located in the Eastern and Northern parts of the country. The interpretation of the activity increase in March is also supported by the coefficient of the zonation index, that is, biodiversity value of the forests, which indicates that activity growth in March occurs in areas with low zonation values (Table 3).