Canada’s human footprint reveals large intact areas juxtaposed against areas under immense anthropogenic pressure

To produce the Canadian human footprint, we adopted the methods originally developed by Sanderson et al. (2002) and later refined by Venter et al. (2016a, 2016b). The pressures we mapped for Canada were: the extent of built environments, crop land, pasture land, human population density, nighttime lights, railways, roads, navigable waterways, dams and associated reservoirs, mining activity, oil and gas, and forestry (Table 1). Each anthropogenic pressure was placed on a 0–10 scale to allow for comparison across pressures. Scoring methods were selected from pre-existing peer-reviewed articles following methods in Venter et al. (2016a, 2016b) and Woolmer et al. (2008) and the oil and gas pressure layer following methods used in Jarvis et al. (2010). After scoring, all noncompatible land uses were analysed and adjusted to avoid spatial overlap. Noncompatible land uses included built environments, crop land, mining, and pasture land. We eliminated any pixels from the given layers that overlapped with built environments, then we did the same operation for crop land and mining. To determine which layer would be kept if spatial overlap was present where it should not be, the layer with the highest score from the 0 to 10 scale was kept. To produce the final product of the terrestrial human footprint map of Canada, all the weighted layers were summed together. To compare different ecological areas in Canada, the COSEWIC national ecological areas were used (COSEWIC 2018). Individual pressures may overlap spatially and are therefore not mutually exclusive. Thus, each cell could range in value from 0 to 55 for any given grid cell, representing the observed maximum. The map was generated at a spatial resolution of 300 m, yielding over 99,000,000 pixels. We chose 300 m as a reasonable compromise between spatial precision and computational time, as well as the reference resolution of the land cover data we used. ArcGIS 10.5.1 and the Lambert Conformal Conic projection were used for all spatial analyses. Specific details on each of the pressure layers are provided in the following sections.

Built environments are lands that are constructed for human activity and include buildings, paved surfaces, and urban areas. Land transformation from built environments leads to habitat loss and fragmentation, changes in nutrient and hydrological flows, reduction of viable habitats for species, and decreased temperature regulation and carbon sequestration (Haase 2009; Tratalos et al. 2007).

We acquired data from the 2016 annual crop inventory (Government of Canada; Agriculture and Agri-Food Canada; Science and Technology Branch 2016), which provides a 30-m spatial resolution of land-use type and applied the subset of the “urban/developed” lands for the layer. The data do not include Yukon and Northwest and Nunavut territories; therefore, we captured the anthropogenic pressures for the northern territories through other layers such as: population density, nighttime lights, and roads. The data were a combination of satellite imagery: Landsat-8, Sentinel-2, and Gaofen-1 for optical imagery with RADARSAT-2 radar imagery, generating an accuracy of at least 85% (Government of Canada; Agriculture and Agri-Food Canada; Science and Technology Branch 2016). Built environments were assigned a score of 10 as built environments do not provide many ecosystem services or provide suitable habitat for many species of concern (Venter et al. 2016a, 2016b).

Human population density is linked to biodiversity loss (Cincotta and Engelman 2000). Presence of high human populations has led to over-hunting, deforestation, and introduced species (Prebble and Wilmshurst 2009). Though Canada generally has a low population density, (∼4 persons/km²) there have been significant increases in introduced species, over-exploitation, and pollution from 1999 to 2018 (Government of Canada; Statistics Canada 2017a; Woo-Durand et al. 2020).

Human population density was mapped using a subset of the 2016 Canadian Census Data (Government of Canada; Statistics Canada 2017a; Venter et al. 2016b). The vector layer used was the Census Dissemination Blocks, the smallest unit with an associated population, available through the Geo Suite 2016, a Statistics Canada tool we used for data retrieval (Government of Canada; Statistics Canada 2016). Following Venter et al. (2016a, 2016b), we calculated population density for each block; we assigned a pressure score of 10 for any block that had more than 1,000 people/km², as the assumption is that population density becomes saturated at that level (Venter et al. 2016a, 2016b). For more sparsely populated areas, we logarithmically scaled the pressure score as seen below to mimic how pressures on the system increase with population density:

Nighttime lights capture the sparser electric infrastructure found in rural, suburban, and working areas that have an associated pressure on natural environments (Venter et al. 2016a, 2016b). The Visible Infrared Imaging Radiometer Suite (VIIRS), mounted on the Sumo National Polar Partnership satellite, provides the means to collect and map low-light sources such as nighttime lights (Elvidge et al. 2013).

We used an annual composite from 2016 generated by the National Oceanic and Atmospheric Administration (NOAA) to assess nighttime lights. The spatial resolution of the data is 589 m (15 arc-second geographic grids). For areas above 67°N, that were not included in the annual composite, we randomly selected a single date of imagery to fill the northern section and visually compared it with other dates to make sure it was not an outlier (NOAA 2019). We then rescaled the data on a 0–10 scale using an equal quintile approach (Venter et al. 2016a, 2016b).

Agriculture is recognised as one of the most important pressures to biodiversity globally (Ricketts and Imhoff 2003). For the Canadian human footprint, we used the 2016 annual crop inventory that includes pasture, agricultural land, cereals, pulses, oil seeds, vegetables, fruits, and other crops (Government of Canada; Agriculture and Agri-Food Canada; Science and Technology Branch 2016). Satellite imagery from optical (Landsat-8, Sentinel-2 and Geifen-1) and radar (RADARSAT-2) was used to obtain a spatial resolution of 30 m. The provincial accuracy for crop class had a minimum of 86.27% and a maximum accuracy of 94.51% (Government of Canada; Agriculture and Agri-Food Canada; Science and Technology Branch 2016). We assigned crops a pressure score of 7, as some species can still use crop lands unlike most built environments (Venter et al. 2016a, 2016b).

Pasture lands are areas that are grazed by domesticated livestock. Pastures are often associated with fences, soil compaction, intensive browsing, invasive species, and altered fire regimes (Kauffman and Krueger 1984). Using the annual crop inventory (30-m spatial resolution) (Government of Canada; Agriculture and Agri-Food Canada; Science and Technology Branch 2016), we assigned pastures a pressure score of 4, as it is less intensive than crop lands; however, pasture lands are associated with fencing that would no longer allow for the land to be intact (Venter et al. 2016a, 2016b).

Roads are linear features that directly convert and fragment habitats. Roads can alter the immediate physical and chemical environments, provide access for human recreation into intact areas, allow for the spread of invasive species, and be a sink for populations through vehicle collisions and mortality from construction (Trombulak and Frissell 2000).

We used the publicly available 2016 National Road Network vector layer produced by Statistics Canada (Government of Canada; Statistics Canada 2017b). The data are divided into different categories of use: Trans-Canada highway, national highway system, major highway, secondary highways, major streets, and all other streets. Following Woolmer et al. (2008) we assessed roads from the lens of access points into intact areas and gave different weights to linear road features and associated buffers based on road type (Table 2).

Railways provide a direct pressure to the ecosystems that host them; however, in terms of access they differ from roads. For roads and railways, direct pressures exist as a result of the actual footprint such as physical removal of viable habitat or reduction in the quality of it; indirect pressures may present themselves in the form of altering ecological functions, edge effect, reducing connectivity, or other human pressures made possible by the direct pressure (Burton et al. 2014). However, discontinued rail lines provide an indirect pressure as they can be used as a means of dispersal of humans and their activities into landscapes. Conversely, operational rails only allow for human access from individual rail stations. We used the publicly available National Railway Network vector layer (Government of Canada; Natural Resources Canada 2016) and adapted the methods from Woolmer et al. (2008) using operational and discontinued rails, going out a maximum distance of 900 m using the same intervals as roads (Table 3).

Navigable waterways like roads and rails act as means of access to wilderness areas. Canada’s waterways have a long history of human use as they have enabled travel from sea to sea (Brine 1995). Once the people’s “highway”, settlements were formed along the waterways to allow movement and access. Used by First Nations in precolonial times, the knowledge was shared when the first European explorers arrived. These waterways were later instrumental in the fur trade (Brine 1995; O’Donnell 1989).

We used the data set generated for navigable coasts for 2009 from the global human footprint with a 1 km² spatial resolution (Venter et al. 2016b). The layer included the Great Lakes, as they can act like inland seas, and was generated using distance to settlements, stream depth, and hydrological data (Venter et al. 2016b). We found the centreline of the waterway then weighted them to follow the other access-based layers (Table 4).

Dams directly change hydrology of the areas and they modify the environment, often producing human-made flooded reservoirs (Woolmer et al. 2008). The vector data set was obtained from “Large Dams and Reservoirs of Canada” (Global Forest Watch Canada 2010). We mapped the physical dam as we would a built environment, scoring it as 10 (Venter et al. 2016a, 2016b; Woolmer et al. 2008). We scored the associated reservoirs in the same manner as navigable waterways given that they can provide additional access to areas by watercraft (Table 4).

Oil and gas production have a number of associated pressures to nature such as wildlife mortality, habitat fragmentation and loss, noise and light pollution, introduction of invasive species, and sedimentation of waterways (Brittingham et al. 2014; Jones et al. 2015). The mines and minerals data set, updated in 2015, was used as it lists active oil and gas fields. The data were discrete points in vector format (Government of Canada; Natural Resources Canada 2017). The direct pressures from oil and gas have been found to be highly localized; therefore, we adapted the scoring method using a 10 to 0 scale to score the linear circular decay out to 5 km away from the site centre (Jarvis et al. 2010).

Following the methods used by Venter et al. (2016a, 2016b), a single independent person used high-resolution satellite imagery to visually identify human pressures within 5,000 randomly located, 1-km² sample plots, after training and obtaining “almost perfect agreement” for 400 random plots between trainer and independent validator. Using World Imagery, available through ArcGIS, the 5,000 plots had a median resolution of 0.5 m and a median acquisition year of 2014 (ArcGIS. n.d.).

We used methods from Venter et al. (2016a, 2016b) to develop a standardized key to visually interpret the pressures. For the eight pressures that both our Canadian human footprint and the global human footprint had in common we mimicked their scoring, but for the new pressures included in our study we simply followed their standards for linear or polygons features (Supplementary Material, S1). Interpretations were marked if they were “certain” or “uncertain”; in our case 254 plots were uncertain and therefore discarded, leaving 4,746 validation plots. Generally, plots were classified as uncertain for two main reasons: inadequate resolution of the imagery (15 m) so it was not clear if there were any pressures present on the land or cloud cover obscuring some of or all the image. The plots that were retained for the visual scoring were all certain and we therefore consider the in-situ pressures for the plot as true. The mean human footprint score for the 1-km² plots were determined in ArcGIS, then both the visual and human footprint scores were normalized on a 0–1 scale.

The root mean squared error (Chai and Draxler 2014) and the Cohen kappa statistic of agreement (Viera and Garrett 2005) were used to quantify the level of agreement between the Canadian footprint map and the validation data set. The root mean squared error measures the differences between the values calculated in the human footprint and the visual scores from the validation. As the error is squared, outliers are emphasized with this statistical calculation. The Cohen kappa statistic of agreement expresses the agreement between the human footprint scores and the visual interpretation scores considering the potential that agreement or disagreement may occur by chance. Following previous analyses (Venter et al. 2016a, 2016b), visual plots that were within 20% of the human footprint plots scores were considered a match for the Cohen kappa statistic.

Canada has an area-weighted average human footprint score of 1.48, and the maximum observed score for the country is 55 out of a theoretical 66. The pressures across Canada display strong spatial patterns, showing higher values in Southern Canada where most of the country’s population lives (Fig. 1). With the 12 pressures included, we found that 82% of Canada’s land areas had a human footprint score of less than 1 and therefore were considered intact (Allan et al. 2017). In this context, intact is defined as landscapes that are mostly free of the 12 human pressures we mapped. To conceptualize this definition of intact, cells that had a population density of one or more people per square kilometre obtained a pressure score of 1 or above and were therefore not considered intact. However, pressures such as seismic lines, pollution, or invasive species were not mapped and may be present in areas that we identified as intact. The low human footprint state was defined as areas where the human footprint score was between 1 and 4. The upper limit was determined based on the assignment of a score of 4 for pasture land, which would often have fences fragmenting the connectedness (Venter et al. 2016a). Approximately 5% of Canada was classified in the low human footprint state. The moderate human footprint areas had scores between 4 and 10 and covered 7% of the country. The areas of high human footprint, with a value of 10 or higher, covered 6% of Canada and highlighted areas with multiple overlapping pressures to biodiversity (Fig. 1).

We used national ecological areas defined by COSEWIC as a means of comparing the different ecological regions of Canada (COSEWIC 2018). The human footprint differs markedly across those areas, with 84% of the Boreal ecological area, which covers the largest extent of Canada, still being intact. The Great Lakes Plains, the smallest ecological area, has 76% in the high human footprint category, being the largest percentage in the high category compared with all other ecological areas. The Prairies follow the Great Lakes Plains as the second largest values in the high human footprint category with 57%. The Great Lakes Plains has the smallest percentage in the intact category with a value of 0.6% followed by the Prairies with 8%. Conversely, the Arctic, which is the second largest ecological area, is over 99% intact.

The pressure layer that contributes the most towards the mean human footprint of Canada is roads with a mean human footprint score of 0.72 (Fig. 2) and covering over 1,000,000 km². Crop land is the second most prevalent pressure with a mean human footprint score of 0.27. The only other pressure above 0.10 was population density with a value of 0.20. In terms of extent, population density covers just under one-third of Canada, an area of 3,200,000 km². While nighttime lights cover over 200,000 km², they have a relatively small mean human footprint of 0.01.

Visually comparing the global human footprint (Venter et al. 2016a) with the national version at a broad scale shows similarities in the spatial patterns of anthropogenic pressures (Supplementary Material, S2). Closer examination shows a number of variations in the details. In agricultural areas, such as the prairies ecological region, the Canadian human footprint shows a higher concentration of pressures than the global one (Figs. 3a–3c). For urban areas, the Canadian human footprint captures the distinction between areas such as parks, urban areas, and industrial areas showing a lower human footprint score than the global one (Figs. 3d–3f). In natural resource intensive areas, higher scores for the Canadian human footprint are present compared with the global product that missed these features across Canada. For example, in the boreal ecological area, forestry harvest and infrastructure from oil and gas could be included with the Canadian human footprint (Figs. 3g–3i).

When mapping nationally explicit data, the greatest improvements to the global data sets were found with the National Roads Network and the Annual Crop Inventory. The global human footprint scores roads within Canada as 50% less of a pressure than the Canadian human footprint. The Annual Crop Inventory that was used for mapping crop land for Canada captured over 285,000 km² more than the global product (Fig. 2).

Our validation shows a strong agreement between the Canadian human footprint measure of pressures and the pressures scored using visual interpretation of high-resolution images. The root mean squared error for 4,746 validation 1 km² plots was 0.08 on a normalized 0–1 scale (Chai and Draxler 2014). The Cohen Kappa statistic was 0.91, signifying “almost perfect” agreement between the human footprint and the validation data set (Landis and Koch 1977; Viera and Garrett 2005). We scored 40 of the validation plots as having a pressure score 20% higher than the initial visual interpretation (false positive) and 113 were 20% lower (false negative). The remaining 4,593 plots (96.8%) were within 20% agreement. While the results from the validation represent almost perfect agreement, it appears from the higher false-negative rate that the human footprint map may be underrepresenting the pressure scores across some proportion of the country. The maps should therefore be considered as conservative estimates of anthropogenic pressures on the environment (Fig. 4). When applying a more rigorous threshold for agreement, within 15% of one another, we found that the Cohen Kappa statistic was of substantial strength with a score of 0.77. When applying a less rigorous threshold of 25%, the Cohen Kappa statistic increased to 0.95 (almost perfect strength) (Supplementary Material, S3).

We compared the validation results for the Canadian human footprint with those of the global human footprint clipped to Canada. The global human footprint obtained a root mean squared error of 0.10 on a normalized 0–1 scale for the same validation plots (Chai and Draxler 2014). For the Cohen Kappa statistic, the value was 0.76 using 20% agreement, which is considered substantial agreement between the human footprint and the validation data set, demonstrating lower agreement than the Canadian product (Landis and Koch 1977; Viera and Garrett 2005).

Intact areas worldwide are crucial for conserving threatened biodiversity (Di Marco et al. 2019), yet they experience increasing pressures from human land use. There is therefore a need to protect intact areas to help conserve biodiversity and ecosystem services. Furthermore, the importance of large-scale and intact ecosystems is increasing as these areas become rarer (Watson et al. 2016). When applying the 12 anthropogenic pressures to the eight national ecological areas in Canada, we find that five of those areas have a terrestrial land mass that is over 50% in an intact state. In particular, the Arctic has over 99%, the Northern Mountains has over 95%, and the Boreal follows with over 83%. This demonstrates promising opportunities for the three largest ecological areas in Canada. However, the Boreal is experiencing significant forest loss and degradation from natural resource exploration, industrial forestry, rapid climate change, and anthropogenic fires (Watson et al. 2016). Although faced with criticism, the Canadian Boreal Forest Conservation Framework provides an outline on how to protect at least 50% of the forest through a network of connected protected areas, to prevent excessive degradation, which is crucial to protect wilderness areas of Canada (Boreal Leadership Council 2003; Nishnawbe Aski Nation, n.d.).

Canada has few intact areas left in three of the eight ecological areas (Great Lakes Plains, Prairies, and Atlantic). Our human footprint shows where species are experiencing the most anthropogenic pressures and would likely have the least intact natural ecosystem for disturbance sensitive species. However, it is known that certain species can thrive in large cities and built environments (Sanderson et al. 2002). As mentioned previously, Canada’s Target 1 of 17% conservation of terrestrial lands requires those areas to be representative of the county’s ecosystems (Convention on Biological Diversity 2020). With the Atlantic, Prairies, and Great Lakes Plains areas containing less than 24%, 8%, and 1% of intact lands, respectively, it is unclear how Canada will develop protected areas that represent the ecosystems in those regions, without restoration.

One of the most prevalent pressures to intactness and biodiversity are roads. The existence and expansion of roads to connect communities and resource areas are direct and indirect pressures to ecosystems, such as fragmenting habitats and by providing a means of access into intact areas (Lee and Cheng 2014; Sanderson et al. 2002). For conservation efforts, roads are one of the important pressures to address (van der Marel et al. 2020), especially in Canada where we find roads are the most prevalent pressure (Fig. 2).

Conservation planning recognises the need to understand the patterns of pressures and how they interact (Margules and Pressey 2000; Tulloch et al. 2015). When assessing national at-risk species, the Canadian Living Planet Index found a 59% decline in these at-risk populations between 1970 and 2016 (WWF Canada 2020). A global analysis of over 8,000 threatened or near-threatened species found that overexploitation was the most important pressure, followed by agricultural activities and then urban development (Maxwell et al. 2016). For our analysis, those pressures were represented by the footprint of crop land, forestry, built environments, and pasture land, all of which fall within the top six mean human footprint scores for Canada. With Canada’s at-risk species facing more than one pressure (Woo-Durand et al. 2020), the utility of the Canadian human footprint, which includes the pressures that are most affecting biodiversity (Maxwell et al. 2016), is an important tool for conservation planning and the mitigation of such pressures.

While the overall intensity, or the mean human footprint score, remained low for Canada, we found several differences when comparing global and national human footprints. The most significant difference in the mean human footprint score was found in nighttime lights. Nighttime lights for Canada had a mean human footprint score 18 times less than in the global human footprint. The reduction in score from the global to the national product is the result of using more recent and higher-resolution imagery that addresses saturation and spillage observed with the global product (Elvidge et al. 2013).

Producing the human footprint for Canada allows us to include data sets that are nationally relevant and offer more information and detail than many of the global footprint maps. The largest increase in mean human footprint score comes from crop land that has a mean human footprint score over six times more extensive than that in the global product. The improved accuracy for mapping crop land could be part of the reason we see higher footprint values in the Prairies when compared with the global product, as the Prairies are a large agricultural centre for the country. Furthermore, the Canadian data set for roads allowed for the inclusion of minor roads that the global data set could not include (Venter et al. 2016b). The Canadian data also led to a near doubling of the mean human footprint score for roads when compared with Canada’s score with the global data. However, there is still room for improvement in mapping linear infrastructure in Canada. When we compare the national roads with some provincial road data, we find that the national data do not capture all the resource roads and some of the smaller roads that are mapped at a provincial or territorial scale.

The global human footprint and the Canadian human footprint show the same overall patterns of pressures. However, we find more disagreements in areas where there are more cumulative pressures. By developing a finer resolution national product with Canadian data, we can measure the improvements from global human footprints and confirm the soundness of our human footprint with the almost perfect validation score. This demonstrates the importance of national studies for conservation of biodiversity and ecosystem services (Woolmer et al. 2008).

This is the first national product for the Canadian human footprint, with room for future refinements. Firstly, linear features besides roads, such as seismic lines or outdoor recreation such as trails, should be included if possible, in future revisions. These features appeared in approximately 1% of the validation plots but were not mapped as there were no national data sets for oil and gas exploration and recreation. Also, recreation more broadly can have significant impacts (Mullins and Wright 2016) and should be included as data become available. These data should be a priority for future improvements to our work. Other pressures such as extreme weather and introduced species are important, but inherently difficult to map (Venter et al. 2006; Woo-Durand et al. 2020).

Secondly, the built environments data set did not cover the full extent of the country (only to 59°N); therefore, the theoretical max north of 59° latitude is 10 lower for the human footprint, leading to a potential underestimation of anthropogenic pressure. This is unlikely to be a major omission, as these pressures are sparse or absent above this latitude. When the Cohen Kappa statistic was calculated for the 2003 validation points above 59° latitude, we still have an “almost perfect agreement” with 0.91. Despite lower population density in the north, natural resource exploration has increased, bringing with it more temporary workers and work camps (Ensign et al. 2014). Thirdly, the data sets we used to map mining and oil and gas only provided point features. Further efforts are needed to develop complete polygon boundary and associated linear features that more accurately represent the geographic extent of the pressure. Lastly, our product is not immune to the limitations of spatial analyses such as mixed pixel problems that arise when resampling to the resolution of the project and the assumption of linear and consistent responses of ecosystems to pressures (Halpern and Fujita 2013). While there is certainly scope for further refinement, we do note that the validation of the Canadian human footprint revealed a much closer agreement between our data set and actual observable pressures than previous efforts in other jurisdictions (Venter et al. 2016a; Kennedy et al. 2019; Williams et al. 2020).