Humans are modifying natural systems on an unprecedented scale (Corlett 2015). As a result, more than one million species are now threatened with extinction (Tollefson 2019). Protected areas (PAs) represent a key strategy for halting biodiversity loss. The number of PAs has increased exponentially over the last few decades (Guan et al. 2021), now covering approximately 17% of the planet's terrestrial area (UNEP-WCMC & IUCN 2021). In December 2022, the Parties to the Convention on Biological Diversity are expected to agree to a further increase in PAs, covering 30% of the planet by 2030. Yet, for biodiversity to be conserved, PAs must also be effective (Geldmann et al. 2019). Recent studies have shown that many of the world's PAs continue to be impacted by human activities (Jones et al. 2018; Anderson & Mammides 2020), putting the biodiversity in those areas at risk. To further improve effectiveness, we must develop the right tools to assess human pressure within PAs in its entirety (Schulze et al. 2018). This remains a challenging task considering the vast number of PAs worldwide (> 250,000; UNEP-WCMC and IUCN, 2021) and the fact that we often lack basic information regarding many of the human activities occurring in them.
To overcome this challenge, researchers are increasingly relying on remote sensing information to quantify human threats within PAs. Recent advancements in computing power and remote sensing technology have made such data exceedingly useful for assessing human pressure on biodiversity (Petrou et al. 2015; Luque et al. 2018). Remote sensing datasets are now often available on a global scale and are being updated regularly to the extent that, in some cases, we are provided with near real-time information (Radočaj et al. 2020; Ye et al. 2021). Importantly, remote sensing data are standardized and comparable across countries and regions and can capture a range of human activities impacting biodiversity (G. Pricope et al. 2019; Radočaj et al. 2020). Such activities include, for example, agriculture and urbanization. Combined with other pressures, they can be used to estimate the cumulative human pressure on natural systems (Kennedy et al. 2019). Consequently, it is not surprising that remote sensing information now plays a central role in our efforts to assess human threats within PAs. Results from remote assessments have proven exceedingly informative, revealing regions in which PAs have failed to mitigate human impact (Jones et al. 2018; Geldmann et al. 2019; Mammides 2020) and drivers of human pressure within PAs (Elleason et al. 2021; Guan et al. 2021).
Two global datasets that have been used extensively to assess human pressure on natural systems—and which are largely based on remote sensing techniques (Appendix A)—are the Human Footprint (HFP) and the Human Modification (HM) indices (Venter et al. 2016; Kennedy et al. 2019). Both these indices map human pressure on a global level using eight and thirteen threats, respectively; these include built-up areas, road networks, agriculture, human population densities, and others (Appendix A). The majority of the threats are based on data derived from remotely sensed imagery, although a few are derived from ground-based inventories, e.g., those based on data retrieved from the OpenStreetMap platform (Venter et al. 2016; Kennedy et al. 2019). The threats are quantified and combined to provide a cumulative human pressure score ranging from 0–50 in the case of HFP and 0–1 in the case of HM. The higher the score, the greater the potential impact of humans on biodiversity. Studies have already linked HFP to biodiversity impacts (Watson & Venter 2019), such as increased extinction risk (Di Marco et al. 2018) and reduced animal movements (Tucker et al. 2018).
In addition to the above two indices, researchers have been also evaluating the effectiveness of PAs using other broad-scale datasets that are largely built using remote sensing techniques, such as human settlements (Guan et al. 2021) and human population densities (Gillespie et al. 2019) as measured by LandScan Global (Rose et al. 2021). LandScan Global, for example, disaggregates subnational population censuses into 1km2 grids using a multi-variable dasymetric modeling approach and high-resolution satellite imagery (Rose et al. 2021).
Although such data are undeniably crucial for evaluating human pressure on biodiversity, it should be recognized that they are, for the most part, limited to human threats and activities that can be observed remotely. Consequently, they often exclude other key threats, common in PAs and known to affect biodiversity, which can only be recorded through in-situ field surveys. Examples of such threats include hunting (Schulze et al. 2018; Benítez-López et al. 2019), which can result in species extirpations (Sreekar et al. 2015), and invasive species (Hulme 2018).
An underlying assumption is that higher human pressure, as documented and quantified remotely, represents a good proxy for these other threats that are best captured through field surveys. This is a reasonable assumption, especially considering that studies have already documented such relationships. For example, researchers have long shown that road expansion facilitates the spread of invasive species (Mortensen et al. 2009) and that higher human population densities within PAs correlate with higher species richness of invasive species (Spear et al. 2013). However, the extent to which this assumption is entirely valid has not been investigated in detail. Consequently, this remains a key knowledge gap, potentially biasing our policy decisions and conservation efforts.
In this study, we address this gap using a unique European Union (EU) dataset, which documents human threats in more than 27,000 PAs (known as Natura 2000 sites) through field surveys. The threats are reported by the EU Member States responsible for surveying their Natura 2000 sites using a standardized typology (Appendix B) similar to that used in other international initiatives (Salafsky et al. 2008; Stolton et al. 2019). Threats are classified into multiple broad categories (Appendix B), including categories that can also be captured remotely, such as agriculture and urbanization, and categories that are better recorded using field surveys, such as the use of biological resources and invasive species.
Indices Of Human Pressure Developed Largely Based On Remote Sensing Techniques
We focused our analysis on three indices used often to assess human pressure on natural systems: (a) HFP (Williams et al. 2020), (b) HM (Kennedy et al. 2019), and (c) human population densities, as measured by LandScan Global (Rose et al., 2021). We first quantified the three indices within each Natura 2000 site and then compared them to the threats in each site reported in the EU's database. Since the spatial resolution of the three indices is 1km2, we only included in our analysis Natura 2000 sites that are ≥ 5km2 to avoid spatial discrepancies. We also excluded sites for which information has not been updated since 2012 to reduce temporal mismatches (Appendix A). In total, we analyzed 8,210 PAs situated in 25 EU Member States.
First, we assessed the relationships between the various threat categories using the Phi coefficient—a statistical test suitable for measuring pairwise relationships between dichotomous variables. Phi values can range from − 1 to 1, indicating a strong negative and positive relationship; values close to zero suggest no relationship. Second, we evaluated whether a larger number of threats, as reported in the EU's database, is associated with higher HFP and HM values and higher human population densities. Third, we evaluated whether the presence of each threat category is reflected by higher values of the indices examined.
Relationships Between The Threat Categories And The Three Indices
We found that agriculture-related threats were the most common, occurring in 65% of the Natura 2000 sites (n = 5,299; Appendix C). Conversely, mining-related threats were the least common, occurring in 21% of the sites (n = 1,723). Phi coefficients ranged from 0.07 to 0.31, suggesting that most of the threats did not often co-occur with each other (Fig. 1). Some threats, such as those associated with urbanization and transportation (recorded in 34% and 45% of sites, respectively), were expectedly more related than others. Still, even in those cases, relationships were, at best moderate (Fig. 1). Other threats, such as those related to agriculture and forestry, were only weakly associated (0.14) despite being observed in most sites (65% and 51%, respectively).
The total number of threat categories recorded in each site was only weakly related to the three indices examined (Fig. 2). Although Natura 2000 sites with no threats tended to have lower HFP and HM values and human densities, the rest of the sites had, on average, similar values regardless of the total number of threats recorded in the field (Fig. 2). When each category was analyzed separately, our results indicated that the indices reflected mostly the threats associated with urbanization, agriculture, and pollution but not others (Fig. 3). Natura 2000 sites in which these three threat categories were recorded tended to have, on average, higher HFP and HM values and human densities compared to sites in which those threats were absent (Fig. 3). In addition, sites impacted by "human intrusions and disturbances" (e.g., those resulting from recreational activities; Appendix B) tended to have higher human densities but not higher HFP or HM values (Fig. 3). In the case of the other threat categories, such as those related to forestry and silvicultural activities, Natura 2000 sites tended to have lower and not higher HFP and HM values and human densities (Fig. 3). This is not surprising since silvicultural activities are likely taking place in remote sites with larger forest expanses and lower human presence. Collectively, these findings suggest that remote assessments of human pressure on biodiversity may be missing important human activities and threats.
Advantages And Disadvantages Of Field Surveys Vs. Remote Assessments
One could counterargue that remote assessments can be complemented with further information, potentially capturing additional threats, such as deforestation (Hansen et al. 2013). However, even in those cases, studies have shown that remotely-sensed forest loss may miss smaller-scale logging, which could still significantly impact local biodiversity (Peres et al. 2006; Mammides 2018). In general, it is not uncommon for various types of human threats to go undetected when using remote-sensing methods (Peres et al. 2006; Schulze et al. 2018). This could explain, for example, why numerous Natura 2000 sites in our dataset had relatively low HFP, HM values, and low human population densities but were still threatened by multiple types of human activities according to the field surveys (Figs. 2 and 3).
Another important distinction between the two approaches is that field data tend to reflect actual biodiversity impacts, as confirmed by the experts during the surveys. For example, when it comes to agriculture-related threats, field experts can distinguish between agricultural activities that are harmful to local biodiversity versus those that are not. On the contrary, analyses based on remotely sensed information must assume that the observed agricultural category (e.g., irrigated croplands) is equally impactful across all sites. Moreover, experts in the field can record more subtle impacts that cannot be captured remotely, such as those related to the use of pesticides (Appendix B). However, a major drawback of field surveys is that they are incredibly expensive and time-consuming (Rhodes et al. 2015). Moreover, they often involve multiple experts with varying expertise and experience, leading to inter-observer biases and errors (Brown & Williams 2016). Perhaps more importantly, field data are extremely difficult to update frequently. For example, in our analysis, we had to exclude hundreds of Natura 2000 sites for which information was outdated (> 10 years old). Such constraints seriously limit the utility of field data for conservation purposes.
Consequently, data collected remotely will rightly remain crucial for evaluating human pressure on biodiversity. However, researchers and practitioners must be aware of their limitations, particularly concerning the fact that (a) they do not necessarily represent an actual impact on local biodiversity (since this needs to be assessed in conjunction with biodiversity observations on the ground) and (b) they often exclude key threats that cannot be observed remotely but can be equally detrimental to biodiversity (Schulze et al. 2018).