Identifying differences in roadless areas in Canada based on global, national, and regional road datasets

Abstract

Roads are an overwhelming component of the global human footprint and their absence helps identify intact areas with high ecological value. Road‐free areas are decreasing globally, making accurate estimation of their location and size of great importance. Identification of such regions requires accurate data, but substantial variability exists in road network datasets created and maintained at different spatial scales. We compared estimates of road length, density, and roadless areas across Canada, which contains a high proportion of the world's remaining undisturbed and road‐free areas. Global‐ and national‐scale datasets included, on average, only 11%–14% of roads represented in regional‐scale data or volunteered geographic information (VGI), with the most pronounced differences in less‐developed areas. Regional‐scale datasets, with the lowest estimates of amount of roadless area and smallest mean roadless patch size, are likely the most complete road datasets but are not available for all jurisdictions, limiting their national‐scale utility. VGI provides a national‐scale alternative but still lacks many low‐use roads. Available global and national datasets have insufficient information for accurate assessments of roadless areas in Canada, which will require detailed, consistent subnational datasets assembled and maintained by each province and territory in a coordinated fashion to achieve national coverage.

INTRODUCTION

Colonization, mobility, and economic growth are universally accompanied by the development of road networks connecting geographic regions and providing access to resources, jobs, and markets (Meijer et al., 2018). Road building is driven by the desire to accelerate social integration and economic expansion, so it is no surprise that this often stimulates unsustainable development and environmental degradation (Laurance & Arrea, 2017; Strano et al., 2017). The creation of new road infrastructure has a growth-inducing effect on future development, providing access to untapped resources and driving land colonization, habitat disruption, and further road network expansion (Johnson et al., 2020; Laurance et al., 2014). This contagious development lies at the root of rising cumulative impacts over time (Ibisch et al., 2016; Johnson et al., 2020), resulting in significant environmental destruction in previously undisturbed areas. Much road building is poorly planned or conducted illegally; when environmental assessments do take place, they tend to focus only on the direct effects of the immediate road footprint and its use, while ignoring the likelihood of cumulative impacts and growth-inducing effects (Johnson et al., 2020; Laurance et al., 2014). Evidence demonstrates that projects can lead to societal hazards and negative effects on ecosystems and biodiversity, including habitat loss, pollution, wildlife mortality, barriers to species movement, and facilitation of biological invasions (Forman & Alexander, 1998; Ibisch et al., 2016; Laurance & Arrea, 2017; Trombulak & Frissell, 2000).

Careful management of road development, including keeping areas road-free, is recognized as a critical conservation strategy for avoiding the spread of road-induced environmental damage (Laurance et al., 2015; Selva et al., 2015). Roadless regions represent relatively undisturbed ecosystems that make substantial contributions to maintaining biodiversity (Crist et al., 2005; Selva et al., 2015). Roadless areas adjacent to protected areas have the potential to increase the connectivity and size of parks, designated wilderness, and wildlife refuges (Crist et al., 2005), and even smaller isolated road-free regions can provide habitat for rare species, increase habitat variation, and facilitate movement of organisms between larger patches (Selva et al., 2011; Strittholt & Dellasala, 2001). Areas without roads are also more resilient than fragmented regions to the effects of climate change (Selva et al., 2011). Analysis of existing datasets has revealed that over 80% of the Earth's surface consists of roadless areas (at least 1 km from the nearest road), but these are fragmented into almost 600,000 patches, more than half of which are <1 km² (Ibisch et al., 2016). The largest roadless areas of the world tend to correspond with regions such as intact ecosystems that have exceptional biodiversity, carbon, cultural, and health values, deserving special policy recognition (Watson, Evans, et al., 2018), which are becoming increasingly valuable as they dwindle (Watson et al., 2016). Given the significant global increase in the number and extent of roads anticipated within this century (Laurance & Arrea, 2017), identifying and maintaining current road-free regions is essential for preventing escalating environmental degradation stimulated by infrastructure expansion.

Available geospatial data on road networks include those that integrate various sources to achieve planet-wide coverage (CIESIN, 2013; Meijer et al., 2018); crowdsourced projects such as OpenStreetMap (OSM; www.openstreetmap.org) that maintain volunteered geographic information (VGI) on roads and trails in online databases (Barrington-Leigh & Millard-Ball, 2017); and national or subnational datasets created within countries or regions. While global-scale datasets provide broad coverage valuable for assessing road networks and roadless areas across large regions, these are often subject to geographical bias, with higher accuracy and completeness in more densely populated regions (Meijer et al., 2018). Global maps also tend to omit smaller roads in order to achieve consistency among input data sources. VGI efforts such as OpenStreetMap provide detailed data in areas where roads are accessible, but the completeness of OSM data varies, with smaller countries and more densely populated areas tending to be more complete (Barrington-Leigh & Millard-Ball, 2017). National-scale road network datasets are often more detailed but still not inclined to include lower-traffic roads such as resource extraction access routes, rural road networks without addresses, or roads built and maintained by private companies. Although these issues may be considered minor because any single route receives less vehicular use than a corresponding main road, they comprise the majority of the global road network (Coffin et al., 2021) and their collective contribution to environmental degradation and wildlife mortality is significant (Dabros et al., 2018; Shanley & Pyare, 2011; van Langevelde et al., 2009).

As the second largest country in the world with low human population density that is largely concentrated near the border with the United States, Canada contains a significant amount of area free of industrialized human disturbance (Watson, Allan, et al., 2018). However, its natural resource extraction footprint extends much further than human settlement. Declines in intactness due to development are occurring in many ecosystems (Smith & Cheng, 2016a; Watson, Evans, et al., 2018), with aspirations for future development (Natural Resources Canada, 2021). Existing studies have highlighted Canada in global-scale analyses of roadless areas (Ibisch et al., 2016) and related indicators including intact forests (Potapov et al., 2017; Watson, Evans, et al., 2018), human footprint (Venter et al., 2016), and landscape integrity (Grantham et al., 2020), often emphasizing the large amount of undisturbed landscape remaining relative to other nations. However, these studies all used data derived at a global extent that do not fully represent the road network in Canada. Inaccuracies or missing data, therefore, have implications for conclusions on the location and size of intact areas or degree of human footprint from such research. Given that Canada holds a large proportion of the planet's remaining intact “wilderness areas” (Watson, Allan, et al., 2018), the nation has an outsized responsibility for stewardship of these areas as ecological intactness declines and wilderness area loss outpaces protection worldwide. There is therefore a need to understand how differences in road network data can affect the accuracy of roadless area identification in Canada and what consequences would arise from using data that do not accurately depict the Canadian road network when assessing ecological intactness.

The objectives of this paper are to identify whether landscape in Canada is as intact as global analyses portray and determine what data are required for accurate identification of roadless areas. We accomplish this by comparing estimates of the total length and density of roads and size and number of roadless area from available global-, national-, and provincial-scale road datasets across Canada in order to understand the magnitude of the differences between datasets and in which environments these differences are most pronounced. With roads being a valuable proxy for human impact and roadless areas for relative ecological integrity, our findings have important implications for conservation planning at various scales.

METHODS Road network datasets

We used free, publicly available datasets covering global, national, and when available, provincial extents in our comparison of road network data (Table 1; see Supporting Information for details on datasets and links to download). Five provinces – British Columbia (BC), Alberta (AB), Ontario (ON), Québec (QC), and Nova Scotia (NS) – have provincial road network datasets different from federal datasets. The remaining provinces and territories – Manitoba (MB), Saskatchewan (SK), Newfoundland and Labrador (NL), New Brunswick (NB), Prince Edward Island (PE), Yukon Territory (YT), the Northwest Territories (NTs), and Nunavut (NU) – did not have provincial/territorial datasets available.

TABLE 1 Scale, dataset name, acronym used in this paper, and sources of all road network datasets compared in analysis of road network attributes in Canada. A longer description and links to download each dataset are available in the Supporting Information. All datasets were downloaded on April 23, 2021

Scale	Dataset	Acronym	Source
Global	Global Roads Open Access Data Set	gROADS	Centre for International Earth Science Information Network
Global	Global Roads Inventory Project	GRIP	GLOBIO – Global biodiversity model for policy support
Global	OpenStreetMap	OSM	www.openstreetmap.org
National	Statistics Canada Road Network File	SC	Statistics Canada
National	Canada National Road Network	NRN	Statistics Canada
Provincial	British Columbia Digital Road Atlas	BCDRA	British Columbia Ministry of Forests, Lands, Natural Resource Operations and Rural Development
Provincial	Alberta Human Footprint	ABHFP	Alberta Biodiversity Monitoring Institute
Provincial	Ontario Road Network	ORN	Ontario Ministry of Natural Resources and Forestry
Provincial	Ontario Ministry of Natural Resources and Forestry Road Segments	MNRF	Ontario Ministry of Natural Resources and Forestry
Provincial	Addresses Québec AQréseau+	AQR	Géoboutique Québec
Provincial	Nova Scotia Topographic Database Roads, Rails, and Trails	NSTDB	Nova Scotia Geographic Data Directory

Road network comparison

Some road datasets included only roads that vehicles drive on, while others also contained pedestrian paths, hiking and biking trails, ferry routes, railways, small private driveways, aircraft runways, or proposed but not built roads. We focused only on vehicle access roads to ensure comparisons were limited to feature types common to all datasets. We selected and removed non-vehicular access features from any dataset that contained them based on road type or class values in the dataset feature attributes.

To compare the datasets, we first calculated total length of roads in kilometers and road density in kilometers of road per square kilometer of area across the entire extent of all provinces and territories for each global-, national-, and provincial-scale dataset (roads network comparison).

After creating a grid with 20 × 20 km cells that covered the extent of each province with provincial-scale road datasets, we calculated the area from each land cover type in the 2015 Land Cover of Canada dataset (Canada Centre for Remote Sensing, 2019) within each grid cell. We then calculated total road length and density from each dataset and averaged these across all cells that were composed of (1) at least 50% forest land cover; (2) at least 50% cropland land cover; (3) at least 20% urban land cover; (4) at least 30% grassland land cover; and (5) at least 30% shrubland land cover (land cover comparison). We used these five classes as proxies for areas with high (urban), medium (agricultural), and low (forests, grasslands, shrublands) levels of development; we were interested in comparing how roads were represented from each dataset in less-developed areas, which have lower human population numbers but are often targeted for natural resource extraction with more associated low-use roads versus areas that are more densely populated and developed for land uses other than natural resource extraction, with more high-use roads. Due to the small area and different land cover composition of Nova Scotia compared with the other provinces, there were no cells with at least 30% grassland or shrubland land cover, so these land cover types were not included for this province.

Lastly, we compared roadless areas using provincial datasets for five provinces against a global map of roadless areas that used OSM data as input (Ibisch et al., 2016) (roadless area comparison). We applied a 1-km buffer to both sides of all roads in each dataset in order to evaluate how the number and size of roadless area patches varied among datasets. This matched the buffer size used by Ibisch et al. (2016) to represent an estimate of the area near roads that is significantly impacted by negative ecological road effects. The road-effect zone, the spatial region over which significant ecological effects extend outward from the physical edges of a road, is typically much wider than the road surface plus roadsides (Forman & Alexander, 1998). The width of this zone can vary depending on the impact in question, in addition to characteristics of the road and its traffic volume, features of the terrain and surrounding landscape, and the taxa being considered (Ibisch et al., 2016). Actual road-effect zones can also be asymmetrical on either side of the road and vary in area of ecological influence over space and time, sometimes extending five or more kilometers from road edges (Ibisch et al., 2016). However, based on their review of publications assessing the spatial influence of roads on ecological attributes, Ibisch et al. (2016) chose a distance of 1 km from road edges to define roadless areas that are less influenced by negative road effects. In order to make our conclusions comparable with a global-scale analysis, we used the same buffer distance around roads to delineate roadless areas, and recreated the fields in Table S4 from Ibisch et al. (2016) to summarize roadless patch metrics in the five provinces with provincial datasets.

We made our calculations of length and area in the geographic coordinate system North America Data 1983 CSRS and projected coordinate system Lambert Conformal Conic. We conducted all analyses in ArcMap 10.7.

RESULTS Road network comparison

In all jurisdictions in Canada, GRIP and gROADS had less total road length and lower road density compared with all available national or regional datasets (Table 2). On average, GRIP represented 14% and gROADS 11% of the length of roads from the datasets with the most road length in each jurisdiction, which was always the provincial dataset when available and OSM otherwise. Figure S1 shows roads from each dataset in each province at a national scale, while Figure S2 uses Nova Scotia as an example to demonstrate detailed differences in provincial road coverage from each dataset.

TABLE 2 Estimates of total road length, road density, and percentage of length of roads in the dataset with the highest total road length within that region (comprehensive dataset) represented by each dataset in each province and territory in Canada. Dataset in each region with the longest total road length represented is highlighted in bold

Province or territory	Area (sq. km)	Scale	Dataset	Total road length (km)	Total road density (km/km2)	% of roads relative to comp. dataset
BC	917,732	Global	GRIP	14,225	0.02	2%
		Global	gROADS	15,535	0.02	2%
		Global	OSM	197,963	0.22	25%
		National	SC	124,845	0.14	16%
		National	NRN	91,219	0.1	12%
		Provincial	BCDRA	788,946	0.86	100%
AB	639,936	Global	GRIP	37,260	0.06	10%
		Global	gROADS	30,118	0.05	8%
		Global	OSM	267,261	0.42	68%
		National	SC	214,843	0.34	55%
		National	NRN	234,610	0.37	60%
		Provincial	ABHFP	391,825	0.61	100%
SK	632,214	Global	GRIP	33,293	0.05	13%
		Global	gROADS	17,638	0.03	7%
		Global	OSM	258,391	0.41	100%
		National	SC	237,396	0.38	92%
		National	NRN	249,477	0.39	97%
MB	627,595	Global	GRIP	17,944	0.02	16%
		Global	gROADS	17,393	0.02	16%
		Global	OSM	110,289	0.12	100%
		National	SC	96,445	0.11	87%
		National	NRN	88,417	0.1	80%
ON	980,243	Global	GRIP	70,682	0.07	13%
		Global	gROADS	45,675	0.05	8%
		Global	OSM	293,634	0.3	53%
		National	SC	229,215	0.23	41%
		National	NRN	261,714	0.27	47%
		Provincial	ORN + MNRF	552,538	0.56	100%
QC	1,476,348	Global	GRIP	38,799	0.03	6%
		Global	gROADS	24,036	0.02	4%
		Global	OSM	222,905	0.15	35%
		National	SC	147,586	0.1	23%
		National	NRN	160,225	0.11	25%
		Provincial	AQR	644,738	0.44	100%
NB	74,525	Global	GRIP	8324	0.11	15%
		Global	gROADS	7163	0.1	13%
		Global	OSM	55,662	0.75	100%
		National	SC	32,216	0.43	58%
		National	NRN	35,694	0.48	64%
NS	57,534	Global	GRIP	6911	0.12	6%
		Global	gROADS	6146	0.11	6%
		Global	OSM	55,417	0.96	51%
		National	SC	38,489	0.67	36%
		National	NRN	50,915	0.88	47%
		Provincial	NSTDB	107,867	1.87	100%
PE	6023	Global	GRIP	3343	0.56	38%
		Global	gROADS	1573	0.26	18%
		Global	OSM	8796	1.46	100%
		National	SC	6915	1.15	79%
		National	NRN	6980	1.16	79%
NL	397,598	Global	GRIP	7042	0.02	18%
		Global	gROADS	5012	0.01	13%
		Global	OSM	39,006	0.1	100%
		National	SC	24,744	0.06	63%
		National	NRN	20,102	0.05	52%
NU	2,010,567	Global	GRIP	0	0	0%
		Global	gROADS	213	0	9%
		Global	OSM	2268	0	100%
		National	SC	862	0	38%
		National	NRN	912	0	40%
YK	455,867	Global	GRIP	3455	0.01	26%
		Global	gROADS	2707	0.01	20%
		Global	OSM	13,489	0.03	100%
		National	SC	8847	0.02	66%
		National	NRN	6406	0.01	47%
NT	1,277,219	Global	GRIP	2008	0	27%
		Global	gROADS	2239	0	30%
		Global	OSM	6261	0	85%
		National	SC	6243	0	84%
		National	NRN	7401	0.01	100%

Abbreviations: ABHFP, Alberta human footprint road network; AQR, addresses Quebec road network; BCDRA, British Columbia digital road atlas road network; GRIP, global roads inventory project; gROADS, global roads open access dataset; MNRF, Ontario Ministry of Natural Resources road network; NRN, National Road Network; NSTDB, Nova Scotia topographic database road network; ORN, Ontario road network; OSM, OpenStreetMap; SC, statistics Canada road network.

National-scale datasets NRN and SC had greater road length and higher density than global-scale datasets, but never as many roads as OSM data or provincial-scale datasets, with the exception of the NTs where NRN had the most road length of any dataset. Percentage of total kilometers of road in the NRN dataset compared with the most comprehensive dataset in each jurisdiction (other than the NTs) ranged from 12% (British Columbia) to 97% (Saskatchewan). Percentage of total kilometers of roads in the SC dataset compared with the most comprehensive dataset ranged from 16% (British Columbia) to 92% (Saskatchewan). NRN road length on average represented 56% of the roads in the OSM or provincial datasets, while SC road length represented 55% on average.

In the five provinces with provincial datasets (BC, AB, ON, QC, and NS), provincial data always had the highest total length of roads and OSM always had the second-highest total length of roads. OSM data represented 25% (British Columbia) to 68% (Alberta) of road lengths in the provincial datasets with an average of 55%. In the other jurisdictions where regional datasets were not available, OSM had the most road length of any dataset, with the exception of the NTs, as noted above.

Land cover comparison

The length and density of roads from each dataset varied in environments with different levels of development (Table 3). Provincial datasets had higher road length and road density in less-developed (forest, grassland, and shrubland) land cover types than national or global datasets. On average, in 20 × 20-km regions with at least 50% forested landcover, OSM had 48% of the mean road length from the provincial datasets, GRIP and gROADS had 6%, NRN had 38%, and SC had 24%. On average, in at least 30% grassland regions, OSM had 27% of the mean road length from the provincial datasets, GRIP and gROADS had 4% and 3%, respectively, NRN had 29%, and SC had 17%. In areas with at least 30% shrubland land cover, OSM mean road length estimates were on average 48% of road length from provincial datasets, GRIP and gROADS were each 7%, NRN was 39%, and SC was 23%.

TABLE 3 Mean total road length (kms of road with each grid cell averaged across all grid cells), road density (km of road per km² of area averaged across all grid cells), and percentage of the mean length of roads represented by the dataset with the highest road length, calculated from each global-, national-, and provincial-scale dataset within a square grid with 20 × 20 km cells covering each of five provinces. Bolded values show the dataset with the highest estimates in each province for each landcover type

Province	Dataset	Forest Landcover > 50%			Cropland Landcover > 50%			Urban Landcover > 20%			Grassland Landcover > 30%			Shrub Landcover > 30%
		Road length	Road density	% top	Road length	Road density	% top	Road length	Road density	% top	Road length	Road density	% top	Road length	Road density	% top
BC	gROADS	7.3	0.0	2%	23.9	0.1	3%	171.4	0.9	10%	10.7	0.0	3%	6.5	0.0	4%
	GRIP	8.1	0.0	2%	16.0	0.0	2%	145.0	0.8	8%	14.2	0.0	3%	6.1	0.0	3%
	OSM	108.1	0.3	31%	281.0	0.9	37%	1753.7	9.1	100%	108.6	0.3	26%	29.9	0.1	16%
	NRN	44.2	0.1	12%	272.9	0.9	36%	1298.4	7.0	74%	69.2	0.2	17%	24.3	0.1	13%
	SC	48.9	0.1	14%	194.3	0.7	26%	932.4	5.1	53%	88.0	0.2	21%	20.8	0.1	11%
	BCDRA	354.1	1.0	100%	750.7	2.3	100%	1281.1	7.0	73%	413.6	1.1	100%	184.8	0.6	100%
AB	gROADS	9.0	0.0	8%	72.6	0.2	14%	244.5	0.6	7%	28.5	0.1	9%	5.4	0.0	10%
	GRIP	9.0	0.0	8%	60.0	0.2	12%	1115.6	2.8	33%	26.1	0.1	8%	4.8	0.0	9%
	OSM	100.8	0.3	87%	477.5	1.2	95%	3362.7	8.4	100%	190.1	0.5	59%	41.6	0.1	76%
	NRN	74.2	0.2	64%	452.3	1.2	90%	2255.8	5.6	67%	181.9	0.5	56%	32.8	0.1	60%
	SC	39.4	0.1	34%	338.0	0.9	67%	1691.0	4.2	50%	132.5	0.4	41%	16.3	0.0	30%
	ABHFP	115.9	0.3	100%	502.1	1.3	100%	1789.6	4.5	53%	323.0	0.9	100%	54.6	0.1	100%
ON	gROADS	10.8	0.0	4%	159.8	0.5	26%	201.6	1.1	12%	0.0	0.0	0%	5.6	0.0	7%
	GRIP	10.6	0.0	4%	128.7	0.4	21%	824.0	3.5	47%	0.0	0.0	0%	5.2	0.0	7%
	OSM	86.2	0.3	30%	607.3	1.9	100%	1740.1	9.0	100%	0.0	0.0	0%	49.1	0.2	62%
	NRN	83.8	0.2	29%	539.4	1.7	89%	1242.6	6.9	71%	0.4	0.0	32%	44.6	0.2	56%
	SC	50.1	0.1	18%	458.8	1.4	76%	1020.4	5.7	59%	0.0	0.0	0%	28.4	0.1	36%
	ORN + MNRF	285.4	0.8	100%	540.6	1.7	89%	1245.6	6.9	72%	1.3	0.0	100%	78.9	0.3	100%
QC	gROADS	8.4	0.0	4%	124.5	0.4	21%	220.9	0.7	8%	0.0	0.0	0%	0.6	0.0	6%
	GRIP	8.1	0.0	4%	85.8	0.4	20%	1424.3	4.3	48%	0.2	0.0	4%	1.0	0.0	10%
	OSM	81.4	0.2	35%	553.7	1.8	100%	2681.2	9.0	100%	1.3	0.0	24%	3.6	0.0	36%
	NRN	55.4	0.2	24%	474.2	1.6	87%	2068.9	7.0	78%	0.7	0.0	12%	2.6	0.0	26%
	SC	39.6	0.1	18%	392.7	1.3	71%	1710.2	5.7	64%	0.4	0.0	7%	1.4	0.0	15%
	AQR	249.4	0.7	100%	404.3	1.3	73%	1713.8	5.8	64%	5.6	0.0	100%	9.9	0.0	100%
NS	gROADS	29.8	0.3	10%	109.0	0.3	9%	251.6	0.6	10%	-	-	-	-	-	-
	GRIP	28.6	0.2	10%	127.0	0.3	10%	293.7	0.7	12%	-	-	-	-	-	-
	OSM	170.4	2.1	59%	841.8	2.1	68%	2431.0	6.1	100%	-	-	-	-	-	-
	NRN	165.4	2.2	58%	743.0	1.9	60%	1699.0	4.2	70%	-	-	-	-	-	-
	SC	103.3	1.3	36%	648.9	1.6	52%	1636.0	4.1	67%	-	-	-	-	-	-
	NSTDB	287.4	4.0	100%	1244.4	3.1	100%	1966.0	4.9	81%	-	-	-	-	-	-

Patterns were more variable in agricultural regions (at least 50% cropland land cover). OSM road length estimates ranged from 37% to 100% of the dataset of the highest road length estimates with a mean of 80% and were the highest in Quebec and Ontario. Provincial dataset road length estimates ranged from 73% to 100% of the dataset with the highest road length estimates with a mean of 92% and were the highest in BC, Alberta, and Nova Scotia. GRIP and gROADS estimates of mean total road length compared with datasets with the highest mean road length were on average 13% and 15%, respectively. NRN mean road length estimates had an average of 72% of road length from the datasets with the highest road lengths, while SC had an average of 58%.

In the most densely populated and developed regions (at least 20% urban land cover), VGI-based dataset OSM had the highest mean total road length estimates in all provinces. Provincial dataset road length estimates were on average 69% of OSM estimates in urban land cover. GRIP and gROADS mean total road length estimates were 30% and 9% respectively of OSM road length. NRN estimates were on average 68% and SC road length estimates were on average 59% of OSM road length.

Roadless area comparison

Analysis of roadless areas (patches at least 1 km from any road) revealed differences in amounts and sizes among provinces and datasets (Table 4). For all five provinces with provincial datasets, the provincial dataset had the smallest mean roadless area patch size, least amount of total roadless area, and the lowest percentage of provincial area that was roadless, while global-scale datasets always had the largest mean roadless patch sizes and most roadless area. Figure 1 compares roadless areas from a global-scale dataset (gROADS) with the provincial datasets for these provinces (roadless areas from the other datasets are depicted in Figure S3). Particularly large differences in roadless area estimates between regional and global datasets were in Nova Scotia, where the provincial dataset found only 9% of the province's area was roadless compared with 83% and 79% from the global datasets. In all provinces, OSM had the second-least amount of roadless area and percentage of total provincial area that was roadless, followed by the national datasets.

TABLE 4 Metrics related to the number and size of roadless area patches created by buffering road features in each dataset 1 km on both sides in five provinces of Canada

Prov.	Total area (km2)	Dataset	Total roadless area (km2)	% roadless cover	Mean roadless area patch size (km2)	Maximum roadless patch size (km2)	Median roadless patch size (km2)	Total # roadless patches	# roadless patches > 1 km2	# roadless patches > 5 km2	# roadless patches > 10 km2	# roadless patches > 50 km2	# roadless patches > 100 km2	# roadless patches > 1000 km2
BC	917,732	gROADS	890,779	97%	322.4	219,326	0.09	2763	475	281	226	138	106	44
		GRIP	892,033	97%	313.3	219,255	0.09	2847	496	284	227	132	104	45
		OSM	743,268	81%	129.2	174,663	0.23	5752	1972	1182	914	456	310	54
		NRN	823,822	90%	225.4	195,231	0.14	3655	1018	652	543	313	240	66
		SC	785,742	86%	152.0	205,505	0.25	5169	1862	1144	891	387	259	54
		BCDRA	480,517	52%	40.7	206,313	0.20	11,800	3432	1550	1008	363	224	34
AB	639,936	gROADS	584,791	91%	950.9	173,897	176.61	615	580	548	532	471	413	65
		GRIP	581,808	91%	826.4	173,829	162.15	704	654	616	595	519	451	55
		OSM	63,598	57%	44.3	88,028	1.26	8217	4430	1692	1096	352	199	34
		NRN	384,246	60%	52.3	116,076	1.19	7352	3893	1435	950	305	167	29
		SC	401,053	63%	59.7	64,502	1.47	6723	3887	1409	933	322	181	29
		ABHFP	99,382	47%	32.1	91,101	0.44	9332	3383	1296	805	265	151	32
ON	980,243	gROADS	906,488	92%	175.0	608,520	0.22	5179	1919	1380	1109	372	169	29
		GRIP	02,546	92%	164.3	606,289	0.22	5495	2046	1449	1137	363	163	29
		OSM	743,338	76%	78.2	530,104	0.09	9509	2156	1071	772	325	213	44
		NRN	744,409	76%	74.4	482,854	0.09	10,004	2271	1129	775	325	218	48
		SC	776,054	79%	84.2	507,977	0.08	9213	2083	1025	712	287	192	41
		ORN + MNRF	618,531	63%	45.7	445,507	0.12	13,538	3438	1552	990	318	185	27
QC	1,476,348	gROADS	1,435,130	97%	485.5	922,796	0.15	2956	740	494	414	264	188	29
		GRIP	1,433,219	97%	447.7	924,202	0.15	3201	835	532	450	271	183	26
		OSM	1,273,366	86%	158.2	818,325	0.40	8048	2987	1403	933	409	274	55
		NRN	1,318,150	89%	192.2	811,660	0.42	6858	2581	1192	806	397	282	76
		SC	1,344,479	91%	198.9	878,511	0.38	6761	2442	998	612	223	137	30
		AQR	1,038,052	70%	77.8	748,536	0.27	13,342	4028	1464	873	236	119	18
NS	57,534	gROADS	47,765	83%	44.4	8880	0.16	1076	276	170	139	65	50	10
		GRIP	45,350	79%	34.2	5386	0.16	1326	351	223	183	95	70	11
		OSM	11,545	20%	4.7	1266	0.24	2459	711	273	163	33	20	1
		NRN	12,475	22%	5.2	1925	0.25	2418	704	294	183	37	15	2
		SC	21,441	37%	10.3	3394	0.34	2088	725	360	241	58	30	3
		NSTDB	5297	9%	4.4	928	0.14	1196	281	100	68	16	10	0

Abbreviations: ABHFP, Alberta human footprint road network; AQR, addresses Quebec road network; BCDRA, British Columbia digital road atlas road network; GRIP, global roads inventory project; gROADS, global roads open access dataset; MNRF, Ontario Ministry of Natural Resources road network; NRN, National Road Network; NSTDB, Nova Scotia topographic database road network; ORN, Ontario road network; OSM, OpenStreetMap; SC, statistics Canada road network.

FIGURE 1. Roadless areas (green) that are at least 1 km from the nearest road feature in each of five provinces in Canada (British Columbia, Alberta, Ontario, Quebec, and Nova Scotia) resulting from creating a 1-km buffer around all road features in (a) the global roads open access inventory project (gROADS) global-scale road dataset compared with (b) provincially developed road datasets. Data deficient provinces not considered for roadless area analysis are shown with gray hatching

Mean roadless area patch size from all provincial datasets was <80 km²; in BC, Alberta, and Ontario, mean roadless patch size was <50 km² and in Nova Scotia, <5 km². Analysis of provincial datasets resulted in the lowest number of roadless patches >1000 km² and the highest number of roadless area patches, except in Nova Scotia, where there are no roadless patches larger than 1000 km² and almost 77% of roadless patches are <1 km². In Alberta and British Columbia, around half of the provincial area is within 1 km of a road and 71% of roadless patches in British Columbia and 64% in Alberta are <1 km². The larger provinces of Ontario and Quebec have slightly more roadless area at 63% and 72%, but 75% of roadless patches in Ontario and 70% in Quebec are <1 km². In all five provinces, less than half a percent of roadless patches were larger than 1000 km², indicating few large roadless areas remain in any of these provinces.

DISCUSSION

There are at least 10 datasets with various spatial extents available that depict roads in part or all of Canada, but there is little consistency among these in the number and types of roads included, the metadata included to describe road attributes, the spatial location of the road features themselves, and the completeness and coverage of roads in different regions and geographies across the country. This significant variation in roads represented in global, national, and provincial datasets results in widely different estimates of the size and number of roadless patches in Canadian provinces. Use of global, and to a lesser extent national, datasets in Canadian provinces resulted in both greater roadless area and larger mean roadless patches than the more detailed provincial or OSM datasets, with estimates of the percentage of provincial area that is roadless ranging from 27% to 74% less in provincial compared with global-scale road datasets, and from 13% to 33% less in provincial compared with national-scale datasets. It is clear that using datasets derived at broad spatial scales to map roadless cover in Canada results in an overestimation of areas without roads, implying a higher degree of ecological intactness in some parts of the country.

Inaccuracies, challenges, and opportunities in road network mapping across Canada

Where available, provincial-scale road datasets had the highest estimates of road length and density and the lowest estimates of roadless area mean size out of all available datasets, with the largest differences in less-developed areas compared with more developed agricultural and urban regions. These differences are likely driven by the inclusion of more low-use roads, such as unpaved resource extraction or recreation routes, in provincial datasets compared with national or non-VGI global ones. These lower-use roads, while receiving less traffic than highways, main arteries, or residential routes, nonetheless make up a large portion of the global road network and can have significant negative environmental effects (Coffin et al., 2021; Laurance et al., 2009; van Langevelde et al., 2009). Minor- and low-use roads can also provide access to previously inaccessible regions, inducing the growth of further infrastructure and natural resource exploitation (Johnson et al., 2020; Laurance et al., 2014). The lack of inclusion of resource extraction routes and other low-use roads in less-developed regions of Canada thus represents a major drawback to the use of global- or national-scale datasets for accurately representing the road network and road-free regions across the country (see Hirsh-Pearson et al. [in press], for example).

It should be noted while drawing these comparisons that a number of factors make it impossible to determine the relative accuracy of each road dataset with respect to the true current footprint of roads on the landscape. First, most do not have any timestamps on dates of road creation or decommissioning, which compromises the accuracy of road density estimates or understanding of the pace and scale of road-building processes. Second, there is often little, infrequent, or patchy maintenance of the datasets, especially those at broader spatial scales, and lack of consistency of what qualifies as decommissioning and regular ground-truthing of road datasets compared with the true footprint of passable roads on the landscape. Hence, these layers may represent a cumulative road footprint over time rather than the current drivable road network, especially in remote regions.

However, there is no universal agreement on when a road is no longer considered a road and road decommissioning is often unsuccessful (Ray, 2014), especially without deactivation to prevent use (Hunt & Hupf, 2014) or active restoration following deactivation (Hesselink, 2019). Even when backcountry roads are no longer used for resource extraction, recreational users are often interested in keeping them open (Hunt et al., 2009). Because the provincial datasets are produced at least in part by labor-intensive interpretation of high-resolution imagery and/or ground-based mapping by dedicated teams and are updated more frequently, they are likely the most accurate available depiction of roads in Canada, even when considering the possibility of depicting some roads as present that are no longer functional.

A challenge to the labor-intensive methods used to create provincial road datasets, despite their increased accuracy, is that they are difficult to scale up to national extents. The significant effort required to map roads at this scale may also be the reason regional datasets do not exist in all jurisdictions, resulting in patchy coverage of the country, which limits the utility of regional maps for national-scale analyses. Previous mapping projects, including the Human Access of Canada's Landscapes (Global Forest Watch Canada, 2014) and the Boreal Ecosystem Anthropogenic Disturbances (BEAD; Environment and Climate Change Canada, 2012) datasets, produced broad-scale maps of roads and other disturbances using interpretation of imagery, but manual maps of roads are challenging to create and consistently maintain at national scales. There have been some efforts to conduct automated extraction and mapping of roads (Clode et al., 2007; Doucette et al., 2004; Jin & Davis, 2005) and other linear features (Queiroz et al., 2020), but not at spatial extents such as the entirety of Canada or even a single province. Semi-automated mapping techniques using artificial intelligence (AI) approaches, such as machine learning algorithms applied to high-resolution satellite imagery, show promise for supplementing the fully manual approaches used in VGI datasets like OSM to fill gaps in global road network maps (Basu et al., 2019). However, even AI approaches require some level of validation by expert review, and are likely still to be time-consuming and challenging to train and validate at the national scale in Canada.

Even using a labor-intensive methodology, none of the provincial datasets represents every road on the landscape in Canada, as can be seen when comparing spatial road locations to roads visible in recent satellite imagery (Figure S3). This is likely due to new roads being built and old roads becoming overgrown and unused between dataset updates in addition to the margin of error introduced by the complexities of manual road identification and mapping. One analysis by Global Forest Watch Canada in the Castle region of Alberta, for example, identified approximately 172 km roads that were either not mapped or categorized as roads in the provincial ABMI dataset (Smith & Cheng, 2016b). Improvements are needed to accurately represent the entire Canadian road network in spatial datasets, especially for roads such as resource access routes, which are often built quickly to provide short-term access for natural resource extraction but can lead to the further development of ecologically intact landscapes (Johnson et al., 2020).

Global Forest Watch Canada's mapping of logging roads and other linear features in the Broadback watershed of Quebec illustrates this trend well; they identified an expansion of roads from 339 km in 1980 to almost 3954 km by 2015 (Smith & Cheng, 2016c). In Canada, where short-term economic priorities drive a resource-intensive economy (Ray et al., 2021), publicly available data on roads servicing resource extraction industries are vital for identifying remaining ecologically intact regions and protecting them from further degradation. Until logistically feasible automated or semi-automated road mapping is possible at broad scales, provincial and territorial road datasets using manual imagery interpretation and/or data concatenation from regional sources provide the best option for intact area identification. Up-to-date, detailed datasets, with adequate metadata, are required for understanding the actual extent of road networks, for conducting analyses, and for land management purposes.

The lack of comprehensive, open datasets for linear infrastructure in Canada is not limited to roads however; Global Forest Watch Canada's 2017 analysis of pipeline infrastructure illustrated a very incomplete and variable quality of open datasets on pipeline infrastructure in Canada, at both the national and provincial levels (Smith, 2017). Given constitutional responsibilities, provinces and territories have the responsibility to ensure such good quality data on roads and other linear infrastructure exist for their jurisdiction, ideally with coordination from the federal government to ensure consistency in coverage, quality, and data types included across the country.

Comparison of the utility of road network datasets in Canada

The VGI-based dataset OSM applications such as humanitarian mapping following natural disasters, hydrological modeling, and research on subjects ranging from urban morphology to drivers of deforestation (Barrington-Leigh & Millard-Ball, 2017) represent on average 78% of the road length depicted by provincial datasets: an underestimation similarly noted by Grantham et al. (2020) in their global analysis of forest landscape integrity and by Ibisch et al. (2016) in other countries (e.g., Malaysia). However, in urban and sometimes agricultural environments in our study, OSM consistently had the highest estimates of road length and density of any road dataset, suggesting that in more densely populated regions with many potential mappers and reliable internet access, VGI is the most complete source of road information: a result that has been observed globally (Barrington-Leigh & Millard-Ball, 2017). OSM also more closely matched the amount of roadless area from provincial datasets than global- or national-scale data, although this was still underestimated in less-developed regions with more forested, grassland, and shrubland land cover, where low-use roads dominate. Increasing the completeness of VGI in less-mapped areas of Canada and improving the coverage of this dataset in areas with an extensive rural road network relies on active engagement by road users, such as backcountry recreators, or road builders such as resource extraction companies in more remote regions.

There is a slightly larger mismatch between datasets with a national scope relative to provincial datasets compared with OSM. National-scale datasets NRN and SC represent on average 58% and 57% respectively of roads in Canada and overestimate the amount and average size of roadless areas compared with provincial datasets. While national-scale datasets are a significant improvement on global-scale road network data, they still do not fully represent roads in Canada and appear especially lacking in their coverage of smaller low-volume roads compared with VGI or provincial data. These datasets more closely matched provincial and VGI estimates in urban and agricultural regions compared with less-developed forest, grassland, or shrubland regions where there are fewer main roads and more low-volume rural roads. Natural resource extraction, which often occurs in more remote regions, is the constitutional responsibility of individual provinces/territories in Canada (Ray et al., 2021), so roads servicing these industries are not necessarily contained in federal datasets. These smaller roads, despite receiving less use than the main road network, constitute the majority of roads both in Canada and globally and represent the frontier of road–ecology interactions (Coffin et al., 2021), with a high potential for growth-inducing effects as they allow access further into intact areas (Johnson et al., 2020). Using global- or national-scale road data to identify roadless regions will result in an underestimation of the true extent of the road network and overestimation of the amount of roadless area remaining, especially when making decisions about land-use management in areas with a high amount of low-volume rural roads.

Conversely, global-scale non-VGI datasets gROADS and GRIP represent on average only 12% and 15%, respectively, of roads across Canada, with underestimation in all land cover types compared with provincial or VGI datasets. These datasets also overestimated the amount of roadless area, portraying far more roadless area compared with other datasets in the five provinces examined. Consequently, using global-scale road datasets makes Canada appear to have more conservation opportunity on the global stage. This is relevant for both target setting and performance management related to ecological integrity, for example, in the developing post-2020 Global Biodiversity Framework under the Convention on Biological Diversity (https://www.cbd.int/conferences/post2020). gROADS and GRIP can provide relative information on global road distribution, but their use for detailed identification of ecosystem intactness or roadless areas in Canada is limited.

Other important road attributes

Other aspects of roads not investigated in this analysis are also important in assessments of roadless areas. For example, decommissioning unused or unpaved roads can eventually lead to habitat recovery in surrounding areas (D'Amico et al., 2016), causing a previously disturbed region to become effectively roadless once more, so identifying and removing decommissioned roads from datasets would be valuable. However, complete information on date of road creation/closing was not available for all road locations in any dataset we examined, making an evaluation of whether roads are no longer passable very challenging, although not impossible in small areas (Smith & Cheng, 2016c). Roads also have varying impacts on the surrounding landscape depending on their type and amount of use (Ibisch et al., 2016; Jaeger et al., 2005), but information on traffic volume or even road type is lacking from most datasets, further challenging evaluation of their impacts. Road mapping efforts require a standardized approach to both spatial and metadata creation to be the most useful for mapping areas that are presently free of roads and to assess changes in roadless area number and extent over time as roads are created and decommissioned.

CONCLUSIONS

Roads are a significant and pervasive global disturbance that facilitate further development and have an ecological impact that extends far beyond the roads themselves. Roadless areas are important for biodiversity preservation and maintenance of natural ecosystem functions, making them valuable conservation targets; keeping areas road-free is the best strategy for preserving ecological integrity and avoiding environmental degradation (Laurance et al., 2014). As a nation holding a significant portion of the world's remaining ecologically intact areas (Watson, Allan, et al., 2018), Canada has a global responsibility for effective stewardship and maintenance of road-free areas within the context of its biodiversity protection and climate change commitments.

Our analysis has demonstrated that estimates of roadless areas are widely variable and almost universally incomplete. Road network datasets created at different spatial scales produce widely different estimates of road length and density, and thus roadless areas, as illustrated in the analyses of five Canadian provinces. Regional (provincial) datasets appear to provide the most accurate depiction of the length and density of the road network and resulting roadless patches in Canada, but when these datasets are not available, VGI from crowdsourced road mapping projects provides the best alternative. Global- and even national-scale road network maps underestimate the true footprint of roads in Canada and overestimate roadless area amount and size, making them of little use for accurately identifying roadless regions, especially in less-developed environments. However, even the provincial datasets did not contain all roads visible in high-resolution satellite imagery, with gaps in the coverage of low-traffic volume rural roads and resource extraction routes, so even the best estimates of roadless areas in Canada are likely to overestimate the extent of intact regions at least slightly, especially where low-use roads are dense. Additionally, all datasets were lacking in metadata that would be useful in ecological analyses, such as information on dates of road construction or decommissioning.

To effectively identify roadless areas – increasingly limited in the world – and monitor their changes over time, a standardized approach to road network mapping that includes spatial and metadata on individual roads of all types and that can be updated frequently is necessary. Future Canadian road mapping efforts should be led by provinces and territories with the goal of developing consistent, accurate, timely, open datasets using logistically feasible approaches to road mapping that combined covers the country. There is also a key opportunity for the Canadian federal government to take a leading role in ensuring consistency of data quality and coverage and to work toward amalgamating smaller jurisdictional datasets into a single, open dataset for the entire country. Improving data sharing processes among provincial, territorial, and federal jurisdictions will increase national road network completeness, providing information needed to map and monitor roadless areas as ecologically valuable intact landscapes that continue to decline globally and nationally.

ACKNOWLEDGMENTS

Funding to support this work was provided by a contribution agreement with Environment and Climate Change Canada, project number GCEX21S036.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS

Richard Schuster, Justina C. Ray, and Lucy G. Poley conceptualized the research; Lucy G. Poley conducted the analysis, created figures, and led the writing of the initial manuscript draft; all authors contributed content to and edited subsequent manuscript drafts and the final version.

DATA AVAILABILITY STATEMENT

Data used in this paper are available here: https://osf.io/28ua4/. Road datasets can also be downloaded directly from the sources given in the Supporting Information.

ETHICS STATEMENT

All ethical guidelines were followed in the conduct of this research.