The default RPM uses distance to settlements and main roads as a proxy for accessibility, excluding smaller roads. Its weighting ignores large areas classified as poorly accessible and emphasises proximity to residential areas (Fig. 2A and C).

For the nationally adjusted approach, we used spatial data on hotels and free-time residences to create the accessibility layer. Overnight stays were distributed within buffer zones around accommodation buildings in two steps, reflecting different behavioural patterns in nature-based tourism. For example, recreation may be concentrated in the surroundings of a free-time residence or hotel, particularly if near a waterbody, whereas many nature-based tourism activities occur across a wider landscape, such as skiing routes or hiking trails within reasonable travel distance. To balance these patterns, we allocated half of the overnight stays to the surroundings of accommodations using a 200 m buffer and the other half to attractive areas within a 20 km buffer. The 20 km distance is consistent with the approach used in the nature attractiveness index developed by Parks and Wildlife Finland (2024). In the resulting accessibility map, only a few areas are marked as inaccessible, while hotspots emerge in regions with a high density of holiday accommodations, together with smaller-scale hotspots surrounding each individual accommodation location (Fig. 2B and D).

Trails and areas where facilities support outdoor activities are significant attractors for nature-based recreation. However, spatial data on recreation facilities are not available at the European level and is, thus, not included in the default input datasets. Some facility types are represented in the RPM attractiveness layer, such as national parks.

For the nationally adjusted approach, we integrated an extensive map of facilities (Fig. 3B and E). Relevant spatial features were extracted from the National Database of Sport Facilities (LIPAS; Lamberg and Laakso (2023)), which contains information on sport and recreation areas and facilities, such as national parks, nature trails and lookout towers. Raster map cells within recreational areas were assigned a weight of 10, implying ten times higher attractiveness than a random landscape. For trails and facility points, 100 m buffer zones were applied to include adjacent ecosystems. The cells within the buffers were assigned weights of 30 and 250, respectively, reflecting their relatively higher expected intensity of recreational use compared to broader recreational areas. The weighting scheme was designed to account for the strong differences in spatial extent between recreational areas, trails and facility points, so that all infrastructure types contribute meaningfully to the final aggregated result. When extent and weights are considered together, all three facility types have same magnitude of effect on the final aggregated ecosystem contribution.

The attractiveness layer of the default RPM (Fig. 3A and C) consists of various European datasets. Cells are assigned low, medium or high attractiveness values. For the nationally adjusted layer (Fig. 3B and E), we included similar attractive elements, such as waterbodies and shorelines, but used national data sources (Table 1) and assigned condition-based weights for all main waterbodies, not only swimming sites. We also adjusted the coastline buffers so that more emphasis was given to water ecosystems themselves and less for the terrestrial ecosystems of a coastline. This adjustment was made because recreational access and use of shorelines are already represented through the facilities layer, whereas the attractiveness layer aims to reflect the contribution of the waterbody itself as a landscape feature. Roadsides were more consistently designated as non-attractive for the nationally adjusted map.

As a novel element, we incorporated the effect of forest management, particularly clear-cuttings, which may locally have impacts on the recreation experience and potential in a country with freedom to roam in all forests. Forests of perceived natural appearance, without visible traces of forestry operations, are generally considered more pleasant and large-scale clear-cutting, in particular, is perceived as having a negative impact on the landscape and recreational experience (Tyrväinen et al. 2003, Gundersen and Frivold 2008, Silvennoinen et al. 2010, Tyrväinen et al. 2016, Silvennoinen 2017). Forest management influences ecosystem service potential, for example, in the surroundings of free-time residences. In our map, forests that lost tree-cover between 2003 and 2023 were set as non-attractive (see the fragmented black areas in Fig. 3) for nature-based recreation in 2023, using Global Forest Watch data on forest loss (Hansen et al. 2013). This dataset has previously been applied by PEOPLE-EA (2024) to demonstrate the use of earth observation data for improved RPMs.

To evaluate how well the default and nationally adjusted RPMs reflect the spatial potential for nature-based tourism, we compared them with the most spatially accurate reference data available: municipality-level estimates of nature-based overnight stays derived in Step 2. We assume that municipalities with higher actual recreation potential should attract larger volumes of nature-based tourism. Accordingly, better alignment between the RPM and observed overnight stays indicates stronger performance of the spatial proxy. For each municipality, we calculated the mean RPM score (separately for the default and national RPMs) and the intensity of nature-based overnight stays relative to municipality area. Spearman’s rank correlation was then used to assess the strength of the spatial relationship between the RPM values and nature-based tourism intensity.

The INCA tool automatically models spatial distribution of overnight stays, based on input data and overlays the output with an ET map to calculate contributions by ET. By default, INCA uses the pan-European CORINE Land Cover (CLC)-based accounting layer from the European Environment Agency (EEA) (2019).

For the nationally adjusted approach, we used the Finnish ET map (Vihervaara 2024), which has a higher 20-m resolution compared to the 100 m default map. It is based on the harmonised national High-Resolution CLC time series (HR-CLC; Syke (2023)), previously identified as the most suitable source for producing standardised and consistent ecosystem extent accounts in Finland (Costadone et al. 2024).

In the default accounting, artificial areas (1), croplands (2) and marine ecosystems (10, 12) are excluded with a weight of zero. In the nationally adjusted approach, cropland and marine ecosystems were considered to contribute to nature-based tourism and, therefore, only artificial areas were excluded.

The discrepancy between estimates of the national total supply of the ecosystem service is mainly explained by the differing coverage of the tourism statistics and the prominence of non-commercial accommodation in Finnish travel culture. Finnish statistics on domestic travel are based on population-level sample surveys, which allow all types of travel and accommodation to be captured, including non-commercial accommodation such as stays at free-time residences (46% of domestic holiday travel in 2023). In contrast, Eurostat statistics rely on commercial accommodation data, which leads to a significant under-representation of domestic tourism in Finland. Finnish travel statistics also show that the proportion of nature-motivated trips is particularly high for stays at own free-time residences. The results highlight the importance of including non-commercial accommodation types in nature-based tourism accounting in countries where use of free-time residences is common, such as Norway and Sweden (Aubrunner et al. 2025, Svae et al. 2025).

Although the estimated national-level nature-attribution percentages are similar in both approaches (around 50%), they show considerable disagreement at the regional level, as noted in the Methods section (Fig. 1). The Recreation Potential Map (RPM) used in the default approach appears to underestimate contribution of remote regions to nature-based tourism and over-emphasise more densely populated areas, likely due to strong reliance on accessibility. For example, large areas of estimated low accessibility in Lapland substantially reduce its RPM-based nature-attribution percentage, despite the presence of several major nature-based tourism hotspots. In the Finnish case, the use of national survey data instead of the default RPM had a significant impact on the results. Similar findings have been reported in Norway, where dedicated surveys were used to compile nature-based tourism accounts, leading to improved accuracy (Svae et al. 2025). This is consistent with earlier recommendations that nature-based tourism accounting should be supported by detailed national datasets and tourism surveys to improve the representation of local conditions and visitor behaviour (Zulian and La Notte 2022).

As expected, the ecosystem service map resulting from the nationally adjusted approach (Fig. 4) appears more plausible, based on expert judgement, reflecting a more detailed understanding of the characteristics of nature-based tourism and its institutional context (Barton 2022). In the default output map, accessibility is strongly emphasised, with hotspots clustered around towns and cities. Large zero-contribution areas appear where accessibility is assumed to be low, although the national data show that many of these areas include dense clusters of free-time residences that are accessible via smaller roads.

In contrast, the nationally adjusted results show tourism more evenly distributed across the landscape, with concentrations of overnight stays in areas of good environmental condition, recreational facility areas and local hotspots around free-time residences and hotels. Unlike in the default approach, accessibility is not limited to proximity to main roads. Key nature tourism regions – particularly free-time residence areas – as well as known attraction hotspots are well represented in the adjusted map, whereas the default map is not able to capture most of these features. National data inputs in this case seem to provide crucial additional information.

Additional confirmation of default vs. adjusted RPM reliability was gained by using Finnish Travel Statistics as a reference. Municipalities with high numbers of nature-based overnight stays in the statistics are likely to have high nature-based recreation potential in reality. A comparison between the reference statistics and mean RPM values calculated as municipality-level averages (Fig. 6) shows a stronger Spearman correlation for the nationally adjusted RPM (0.54) than for the default RPM (0.27), indicating that the adjusted RPM might better capture the spatial variation in nature-based tourism potential.

The results and their evaluation support the idea that nationally adjusted data inputs can bring additional value to modelling of nature-based recreation in national contexts. Recreation facilities information seems to be an integral national input for modelling and also using contextually relevant data for the accessibility component appears essential for nationally adjusted accounting. While the RPM may be suitable for estimating accessibility for daily recreation (occurring in proximity of residential areas), accessibility for nature-based tourism is better represented using spatial information on recreation accommodation locations. In addition, municipality-level statistics, with higher spatial accuracy than NUTS-3 level data, better identify the municipalities with genuinely high volumes of domestic tourism, whereas the default approach relies on RPM-modelling for within-region distributions. Our results indicate that replacing and complementing generic default maps with nationally relevant data – covering key aspects such as accessibility from accommodation, recreation facilities and nationally relevant attractiveness proxies – produces spatial distribution estimates that better reflect existing knowledge, statistics and national travel characteristics. These findings are consistent with Zulian et al. (2024), who conclude that European-scale biophysical maps cannot fully capture national-scale characteristics and that national implementations should incorporate locally relevant data and knowledge. They further identify traveller behaviour and the representation of ecosystem types, such as lakes, wetlands, coastal areas and marine ecosystems, as important areas for future development.

Accurate spatial estimation is critical for ecosystem accounting, as INCA attributes overnight stays to ecosystem types (ETs), based on the modelled spatial distribution. Given that the spatial estimates appeared more accurate under the nationally adjusted approach, we assume that the resulting ET distribution (Fig. 5) also better reflects reality.

The choice of ecosystem extent data (national HR-CLC, 20 m, vs. pan-European CLC, 100 m) had only a minor effect on relative ET contributions at the national scale (differences of 1–2 percentage points), which is expected given that the CLC is an aggregated version of the HR-CLC. At finer spatial scales, however, the higher resolution and smaller minimum mapping unit of the HR-CLC allow a more realistic and detailed representation of ET distribution, particularly in heterogeneous landscapes, where spatial resolution can strongly influence ecosystem mapping and accounting outputs (Kokkoris et al. 2024).

The default accounting approach excludes marine ecosystems from the modelling process entirely. Although ecosystem service reporting is not mandatory for marine ecosystems, we argue that they should, nevertheless, be included in the modelling phase and excluded later if required, consistent with recent recommendations to improve the representation of marine and coastal ecosystems in nature-based tourism accounting (Zulian et al. 2024). Otherwise, municipalities with substantial marine areas risk severely skewed representations of nature-based tourism and ET contributions.