Mapping roadless areas in regions with contrasting human footprint

In an increasingly human- and road-dominated world, the preservation of functional ecosystems has become highly relevant. While the negative ecological impacts of roads on ecosystems are numerous and well documented, roadless areas have been proposed as proxy for functional ecosystems. However, their potential remains underexplored, partly due to the incomplete mapping of roads. We assessed the accuracy of roadless areas identification using freely available road-data in two regions with contrasting levels of anthropogenic influence: boreal Canada and temperate Central Europe (Poland, Slovakia, Czechia, and Hungary). Within randomly selected circular plots (per region and country), we visually examined the completeness of road mapping using OpenStreetMap 2020 and assessed whether human influences affect mapping quality using four variables. In boreal Canada, roads were completely mapped in 3% of the plots, compared to 40% in Central Europe. Lower Human Footprint Index and road density values were related to greater incompleteness in road mapping. Roadless areas, defined as areas at least 1 km away from any road, covered 85% of the surface in boreal Canada (mean size ± s.d. = 272 ± 12,197 km2), compared to only 0.4% in temperate Central Europe (mean size ± s.d. = 0.6 ± 3.1 km2). By visually interpreting and manually adding unmapped roads in 30 randomly selected roadless areas from each study country, we observed a similar reduction in roadless surface in both Canada and Central Europe (27% vs 28%) when all roads were included. This study highlights the urgent need for improved road mapping techniques to support research on roadless areas as conservation targets and surrogates of functional ecosystems.


Roadless areas definition and data processing
This study is based on the OpenStreetMap (OSM) road dataset, a dynamic and openly accessible database that is updated daily and allows copying, reproduction, redistribution, and modification with proper attribution to OSM and its contributors [1].Roads were obtained from OSM for 2020 and included 36 different road categories [2,3].All categories have been incorporated into our analysis, given their characteristic of facilitating human and (motorized) access, a factor recognized as a potential threat and risk to various species and ecosystems, as revised by Ibisch et al. [4].Each road was buffered with a geodesic buffer of 1 km on each side of every road (Figure S5).After creating the buffer around each road, the area of boreal Canada and temperate Central Europe were extracted to obtain a layer containing only roadless areas located 1 km away from the nearest road.'Natural Earth data' provided the layer for lakes, which were extracted from the country layer to exclude large water bodies from this analysis [5].Wetlands, smaller lakes, and streams were included in the roadless area assessment.

Anthropogenic influences and road mapping within the circular plots randomly selected
We hypothesized that regions with higher anthropogenic influences would exhibit better mapping compared to regions with lower anthropogenic impact.To investigate this, we analyzed various datasets related to road mapping completeness, including road density, Travel time to major cities, Human Footprint Index, and Human Modification Index [2,6,7,8].The distribution of all four explanatory variables and the completeness of roads was examined and a correlation matrix was created (Fig. S1, Table S3).We then examined only circular plots containing roads (mapped and unmapped) for statistical testing (Tables S4,   S5).

Forest cover and roadless areas
The Canadian boreal region comprises approximately 2.2 million km² of forest, with 1.4 million km² designated as forested roadless areas [9].Conversely, Central Europe's forest cover spans 187,447 km², as per Copernicus data, with a mere 821 km² representing forested roadless areas [10].In both regions the forest cover within the 30 selected roadless Table S2.Summary of the visual interpretation of the randomly selected circular plots (n = 1000 per country, 3.14 km² each) for each of the Central European countries.It shows the number of circular plots within the following categories: plots with all roads completely mapped, plots with roads partially mapped, plots with all roads unmapped, and plots without roads.

Figure S1 :
Figure S1:Histograms showing the frequency distribution of road completeness, road density, travel time to major cities, Human Footprint Index and Human Modification Index for boreal Canada and temperate Central Europe in the randomly selected 1000 circular plots of 3.14 km 2 each (as indicated in Table1, road completeness refers to plots with all roads completely mapped = 1, plots with roads partially mapped = 2, plots with all roads

Figure S2 .
Figure S2.Comparison of (standardized) model coefficients of Generalized Least Squares models without (black) and with (grey) accounting for spatial autocorrelation (exponential correlation structure) in the response variable (road completeness).Road completeness encompassed the following three categories, fully mapped = 3, partially mapped = 2, not mapped = 1.The explanatory variables were road density, Human Footprint Index, travel time to major cities and Human Modification Index.The reference country category was boreal Canada.

Figure S3 .
Figure S3.Comparison of 30 randomly selected roadless areas in boreal Canada and the four temperate Central European countries before and after visual interpretation and road mapping.The size order was based on the largest roadless area obtained post-mapping.Open circles denote roadless areas before manual mapping, while stars represent the generated largest roadless patches after mapping.Note the different scales.

Figure S4 .
Figure S4.The spatial distribution of the 30 randomly selected roadless areas in the boreal region of Canada with their corresponding ID number.See TableS7for details on each roadless area.This figure was created using ArcGIS Pro 3.2 (https://www.esri.com/enus/arcgis/products/arcgis-pro/overview).

Figure S5 .
Figure S5.The spatial distribution of the 30 randomly selected roadless areas in each of the four selected countries of temperate Central Europe represented by Poland, Slovakia, Czechia, and Hungary are indicated with ID numbers.See Tables S8-S11 for details on each roadless area.This figure was created using ArcGIS Pro 3.2 (https://www.esri.com/enus/arcgis/products/arcgis-pro/overview).

Table S3 .
Correlation matrix between variables of human influences in the 1000 randomly selected plots in Central Europe and boreal Canada.Correlation coefficients (Spearman) range from -1 to 1.

Table S4 .
Model selection table and coefficient values of the Generalized Least Squares models.Models ranked by AICc.The response variable is road completeness (fully mapped =

Table S5 .
Model selection table of the Ordinal Regression models accounting for country differences in Europe in relation to road completeness.Models ranked by AICc.

Table S6 .
Extent and amount of roadless areas (calculated by creating a 1km geodesic buffer around each road and extracting the remaining area) across boreal Canada and each of the selected Central European countries.The table provides information on the number of roadless areas in different size classes, along with the total roadless surface.

Table S7 .
Outcome of the visual interpretation of 30 randomly selected roadless areas in boreal Canada, showing the assigned ID, the surface of the largest roadless area after road mapping, the sum of all roadless areas created after road mapping, the reduction in size of roadless areas after road mapping, the median size of the newly identified areas after road mapping and the main landcover type.

Table S8 .
Outcome of the visual interpretation of 30 randomly selected roadless areas in Poland, showing the assigned ID, the surface of the largest roadless area after road mapping, the sum of all roadless areas created after road mapping, the reduction in size of roadless areas after road mapping, and the main landcover type.

Table S9 .
Outcome of the visual interpretation of 30 randomly selected roadless areas in Slovakia, showing the assigned ID, the surface of the largest roadless area after road mapping, the sum of all roadless areas created after road mapping, the reduction in size of roadless areas after road mapping, and the main landcover type.

Table S10 .
Outcome of the visual interpretation of 30 randomly selected roadless areas in Czechia, showing the assigned ID, the sum of all roadless areas created after road mapping, the reduction in size of roadless areas after road mapping, and the main landcover type.

Table S11 .
Outcome of the visual interpretation of 30 randomly selected roadless areas in Hungary, showing the assigned ID, the surface of the largest roadless area after road mapping, the sum of all roadless areas created after road mapping, the reduction in size of roadless areas after road mapping, and the main landcover type.