Conservation spillover effect of UNESCO World Heritage Sites into surrounding landscapes

Study scale

We evaluated UNESCO World Heritage Sites (WHS) and their buffer zones of different scales at a global extent (Fig. 1). We used multiple spatially continuous global datasets (land use and landcover (LULC), human footprint (HF), and forest landscape integrity index (FLII)) to determine the spatial variability of the effects of the WHS on the surrounding communities at three scales (1, 11, and 100 km). We chose these measurements because they represent different types of indicators as described by Groves & Game (2016); specifically natural, proxy and constructed indicators. Natural indicators reflect a direct measure of the system. A proxy indicator can be used to capture values without a natural scale. Lastly, a constructed indicator will combine multiple measures to reflect a broader concept or idea. In this project, global LULC data is a natural indicator that describes the trends of the landscape (ESA GlobCover, 2009). FLII acts as a proxy indicator for the landscape as it measures forest integrity, a proxy of conservation success (Grantham et al., 2020). HF is a constructed indicator because it encapsulates different measurements in one scale (Wildlife Conservation Society—WCS, and Center for International Earth Science Information Network—CIESIN—Columbia University, 2005). Further details and decision making on the indicators are discussed below (Table S1).

Data collection

We obtained primary WHS site location and PA data from the UNESCO World Heritage Site and the World Database on Protected Areas (WDPA) data downloads respectively (UNESCO, 2022; UNEP-WCMC IUCN, 2022). The original WHS data contained 1,154 WHS points and the original WDPA data consisted of 254,526 polygons. We subsetted the sites by selecting polygons where the WHS points and the WDPA polygons intersected using the select by location tool in ArcGIS Pro. This resulted in 390 WHS points and 740 WDPA polygons. We then used the dissolved boundary tool in ArcGIS to dissolve the boundaries of the selected WDPA polygons to have an accurate representation of the protected area to the fullest extent as there were many overlapping polygons from the WDPA dataset. This process resulted in 7,244 individual polygons. To add the WHS attributes back to the WDPA polygons, we spatially joined the dissolved polygons to the 740 original WDPA polygons that were subsetted by location with WHS points. This resulted in 382 WDPA polygons. Since there were 390 WHS points that were within WDPA polygons, we used the spatial join tool in ArcGIS to join the WDPA polygons to many WHS points, as there are some polygons that contain more than one WHS point. This resulted in 390 polygons that included attributes from both WHS and WDPA. This final set of 390 polygons includes WHS from all three WHS categories. Although cultural WHS mainly protect cultural assets, the ones included in this study are associated with a PA and contribute to protecting biodiversity. For instance, the English Lake District in the UK is classified as a cultural WHS, but it contains a significant portion of land and lakes managed for conservation (UNESCO, 2017).

Indicators used in the general literature to assess the impact of PAs include human appropriated net primary production (HANPP, Vačkář et al., 2016) and forest loss (Allan et al., 2017; Leberger et al., 2020; Barber et al., 2014; Ford et al., 2020). To differentiate from these studies, we used global land use and land cover (LULC, ESA GlobCover, 2009), human footprint (HF, Wildlife Conservation Society—WCS, and Center for International Earth Science Information Network—CIESIN—Columbia University, 2005) and forest landscape integrity index (FLII, Grantham et al., 2020) as indicators of human and environmental change associated with the classification of a WHS. HF is measured in global hectares and FLII is measured by index values ranging from 0–10,000. We then downloaded the available indicator datasets from their already established online databases in shapefile or raster format (Table S1). We used ArcGIS Pro 2.8.0 and R v1.4.1717 for further analysis.

Data processing

We imported the WHS shapefile and indicator rasters into R. We used the st_buffer tool in the sf package (Pebesma, 2018) to create buffers of 1, 10, and 100 km around each individual WHS polygon. We selected these three distances for logarithmic scale and to encapsulate different human and natural drivers that might impact change including policy, ecosystems gradients, or land use history (Ordway et al., 2021). We used the exact_extract function in the exactextractr package (Baston, 2022) to calculate the mean and variance of HF and FLII and mode of the global land use and land cover raster in the different scales and core WHS polygon.

Data analysis

We observed the frequency distribution of land use and land cover to descriptively compare between scales, regions, and categories. For FLII and HF, we used linear models and AICc model selection (Burnham & Anderson, 2002) to determine what combinations of buffer distance, WHS category, and IUCN category had the best support to explain the variation in each indicator. We considered individual models for each variable, combinations of variables, and interactions between variables for FLII and HF. We used the top model (ΔAIC <2) and 95% confidence intervals to understand the direction, magnitude, and confidence in model estimates.

Descriptive patterns of land use and land cover

The most common LULC in the WHS varied as a function of distance (Fig. 1B), region, WHS category, and IUCN category (Fig. 2). For example, as distance increased, cropland frequencies were greater, and forest vegetation lower. These differences become varied further when considering the effect of the other variables, region, WHS category, and IUCN category with distance on landcover.

Patterns of LULC types across the three scales were different in the five regions. For example, of WHS in the Arab States, inside the border and beyond share bare areas and water bodies LULC types; however, the other types change with buffer distance. Within the PA boundary, mosaic vegetation was the only modal landcover type besides bare areas and water bodies (Fig. 2A). However, this changes to mosaic croplands within 1 km of the WHS border and to only bare areas within 100 km. Moreover, for the WHS in Africa, there was no modal artificial surface and associated areas within or beyond the WHS, whereas in Europe and North America, artificial surfaces were present inside and within 10 km of WHS (Fig. 2A).

The IUCN categories lend themselves to different frequencies of LULC types across spatial scales. For instance, around Ib, one of the stricter conservation IUCN categories, the most frequent LULC type was open needle leaved deciduous or evergreen forest. Interestingly, in this IUCN category, there was higher heterogeneity of LULC within the WHS than in the surrounding landscape (Fig. 2B). In Ia, the most frequent landcover beyond the WHS boundary was artificial surfaces and associated areas 1 and 10 km from the WHS boundary but was not present 100 km from the border (Fig. 2B). These two categories, in addition to category III, have the lowest diversity of LULC types. Category II, the most commonly reported IUCN category, had much more variation. In this category, forest vegetation decreases just beyond the WHS border, but remains relatively consistent across buffer distances. In addition, it is only in the 10 km buffer and beyond that there is a cropland presence. Interestingly, category VI, which allows for sustainable resource use, demonstrated an increasing frequency of forested habitats as buffer distance increased (Fig. 2B). IUCN category Not Applicable was the most common category in this study and has the highest variation of landcover across the buffer distances. It also includes the highest frequency of bare areas and permanent snow and ice.

Similar to the other variables, the three WHS categories illustrated varying patterns in LULC frequencies across the buffer distances (Fig. 2C). One noticeable result is that natural WHS have a consistent frequency of permanent snow and ice across the buffer distances (Fig. 2C). Beside natural WHS, only inside the mixed WHS was there a presence of permanent snow and ice. The four types of croplands also varied over buffer distance in natural WHS (Fig. 2C). Mixed WHS have the highest frequencies of agricultural landcover, especially within 10 and 100 km of the WHS border (Fig. 2C). Another notable pattern is a slight increase in bare areas as buffer distance increases in cultural WHS (Fig. 2C). This may highlight the remote nature of some cultural WHS. Artificial surfaces and associated areas remain at a similar frequency between inside the cultural WHS and 1 km beyond its boundary but is greater 10 km from the boundary. Despite more bare areas and artificial surfaces in cultural WHS, the frequency of forested areas is greater. This pattern is also present in mixed WHS; as buffer distance increases, forested area is greater (Fig. 2C).

Statistical analysis of human footprint and forest landscape integrity index

For both HF and FLII mean and variance, the model that was supported best by the data was buffer distance along with the interaction between IUCN and WHS categories (Table 1). As buffer distance increased, mean HF was higher, and the variance was lower (Table 2, Fig. 3A). The 95% confidence intervals for both the mean and variance estimates were bound by zero, giving strength to the observed pattern. FLII was highest inside the WHS and lower with greater distance of the buffer; variance was lowest inside WHS (Table 2, Fig. 3C). Estimates of the mean were less certain, with only FLII inside the WHS not overlapping zero. Estimates of variance were more certain.

HF for both cultural and natural WHS was highest in IUCN category Ia (Fig. 3B, Table 2) followed by Ib in cultural and V in natural. The former result aligns with the difference seen in the interaction between IUCN category and landcover where Ia had the high artificial and associated land frequencies (Fig. 2B). Ia also has the lowest FLII for cultural WHS but was similar to other IUCN categories in natural sites (Fig. 3D). Ia was not present for mixed WHS for either HF or FLII. The lowest HF in natural sites fell under IUCN category Ib, which also had the highest FLII (Figs. 3B and 3C), though there was high variance in Ib for cultural WHS. FLII was high in IUCN category VI in all three WHS categories and had high variance in each Not Applicable category. The former result also aligns well with the IUCN and landcover results as VI had the highest frequency of mosaic forest-shrubland across the three buffer distances (Fig. 2B). Additionally, IUCN type II had high FLII values across WHS categories. Though there were fewer mixed WHS, across the IUCN categories, HF is low and FLII is high apart from a small number of outliers in each.

Future research

The observed patterns in LULC associated with the WHS suggest testable relationships for future research. For example, a temporal analysis could test if the pattern of increasing cropland landcover beyond natural WHS is a consequence of lack of development within many natural WHS or if agriculture in the region has been misplaced due to WHS/PA establishment. This dichotomy is important to explore given the evidence that WHS/PA can serve as sources for food for local people (Nakamura & Hanazaki, 2017; Oldekop et al., 2016). A lack of consideration of these interactions can result in increasing tensions between the PA and local communities and translate into poor conservation success.

In our subset of WHS there are more natural WHS than there are cultural WHS, the inverse of the full WHS network. The different needs for cultural WHS than for natural suggest interesting questions. Comer (2012) states that preservation should be emphasized for cultural/archeological sites rather than conservation in natural sites because cultural sites are not able to bounce back like natural sites. Therefore, the impacts of tourism from the WHS label may be more drastic on cultural sites than natural. Indeed, in our data, though uncertain, HF was higher in the natural than cultural (and lowest in the mixed) and FLII was highest in mixed.

While there is clear value to a global analysis, it is still important to also consider the individual relationships a WHS has with the boarding landscape as these results may not be true for each WHS. In a case study analysis, the positive conservation spillover effect was not consistent at each site (Hyland, 2021). For instance, FLII decreased steadily as distance from WHS boundary increased in the Río Plátano Biosphere Reserve, whereas, in the Great Smoky Mountains, FLII dropped drastically 1 km from boundary, but then increased again in the 10 km buffer zone (Hyland, 2021). The decrease of FLII in the 100 km buffer zone and increase in HF in the 100 km buffer zone may point to a larger problem at hand (Hyland, 2021; Radeloff et al., 2010). The isolation of PAs globally. This isolation makes it difficult for species to travel resulting in less genetic diversity, more fatalities, and overall habitat loss (Fahrig, 2003). This will be especially felt with climate change as species ranges shift in response to changing temperatures and precipitation while the PAs will remain constant (Rannow et al., 2014; Elsen et al., 2020).

Study limitations

Because of the global nature of study, there were limitations in available datasets for indicators. High quality, fine grain, and easily accessible global data is lacking for environmental indicators. For instance, we had completed the analysis for human appropriated net primary production (HANPP), but then later decided to remove it from the study because the grain size was much larger than the other indicators relative to the size of our buffers. Even though it was removed from the study, HANPP showed similar patterns to FLII and HF with increasing averages as distance increased from the WHS boundary.

In addition, better reporting on IUCN categories should be a priority for the UNESCO WHS convention for current and future WHS as the most common type in this study is Not Applicable. These categories are crucial to highlighting trends in the relationship between protection and environmental integrity of the PA itself and its neighboring landscape. These patterns may be the key to highlighting and implementing effective conservation methods and efforts of these culturally and naturally significant sites and their surrounding landscapes.