Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife

Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife Josie Hughes, Sarah Endicott, David Lapins, Kyle Lochhead, Gregory Paradis This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6413286/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 03 Nov, 2025 Read the published version in Landscape Ecology → Version 1 posted 9 You are reading this latest preprint version Abstract Context Resource road networks have complex and varied impacts on wildlife and other forest values, yet spatial stochastic models forecasting impacts of forest disturbance rarely include automated road network projections. Hardy et al. ( 2023 ) partially addressed this need with a LANDIS-II extension, but there remains a need for tools that can be integrated into other modelling frameworks while identifying a pragmatic balance between achieving ecological relevancy and computational cost. Objectives Our goal is an open source resource road network projection tool that can be easily incorporated into modelling frameworks that assess the implications of forest change for wildlife. We compared the performance of several resource road network projection methods using ecologically relevant metrics. Methods We implemented simple iterative least cost path and minimum spanning tree methods with grade penalties in an open source R package. We assessed performance by comparing projections to observed resource road development since 1990 in a mountainous region of British Columbia. Results All resource road projection methods that we tested performed relatively well. Grade penalties improved performance, as did our minimum spanning tree method. However, the minimum spanning tree method required more computing time and memory, so users must weigh the benefits of improved performance against computational costs. Conclusions Our resource road network simulation methods can improve projections of anticipated resource development impacts on wildlife across large areas. Our open source implementation will be useful for improving projections of the cumulative effects of natural and anthropogenic disturbances on wildlife in an era of rapid change. forest roads forest dynamics spatially explicit modeling predictive ecology least cost paths Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The impacts of roads on wildlife are complex and varied. Mortality risk and stress on or near roads can be elevated due to vehicle collisions, hunting pressure, predation, noise and other stressors, but roads can also provide food, facilitate movement, and provide refuge (Trombulak and Frissell 2000 ; Fahrig and Rytwinski 2009 ; Hill et al. 2021 ). In Canadian forests, there is ample evidence of linear features attracting predators such as bears ( Ursus americanus , Ursus arctos ) and wolves ( Canis lupus ) seeking food or easy movement, while repelling prey such as woodland caribou ( Rangifer tarandus caribou ) and moose ( Alces americanus ) (Polfus et al. 2011 ; Laurian et al. 2012 ; Leblond et al. 2013 ; Beauchesne et al. 2013 ; Mumma et al. 2018 ; Dickie et al. 2020 ; St-Pierre et al. 2022 ); there is also evidence of predators avoiding features heavily used by humans (Muhly et al. 2011 ; Proctor et al. 2020 ; Whittington et al. 2022 ). Changes in mortality risks, habitat quality, and connectivity associated with anthropogenic disturbance can in turn alter use and movement patterns (Apps and McLellan 2006 ; Tucker et al. 2018 ; Proctor et al. 2020 ), demography and population viability (Fahrig and Rytwinski 2009 ; Pinard et al. 2012 ; Apps et al. 2013 ; DeCesare et al. 2014 ; Proctor et al. 2020 ; Johnson et al. 2020a ; Barrientos et al. 2021 ; Lochhead et al. 2022 ), and genetics (Habrich et al. 2021 ). Roads can also facilitate spread of invasive species (Riitters et al. 2017 ), and enable other development activities (Johnson et al. 2020b ). Impacts on wildlife persist after roads are no longer driveable as they continue to be used by wildlife and soil compaction impedes vegetation recovery (Lacerte et al. 2021 ; St-Pierre et al. 2021 ). Measures of road impacts are important for informing wildlife management decisions, and there is therefore a need to include roads in ecological forecasting models. Increasingly rapid environmental change, improvements in available cyberinfrastructure, and development of open source data, tools and methods have contributed to growing interest in ecological forecasting models (Dietze 2017 ; Bodner et al. 2021 ; McIntire et al. 2022 ). In particular, forest landscape models that can account for harvest, fire, and other disturbances are increasingly used to project the cumulative effects of climate change and forest management (Daniel et al. 2017 ; Snell et al. 2018 ; Tremblay et al. 2018 ; Albrich et al. 2020 ; Petter et al. 2020 ; Hof et al. 2021 ; Micheletti et al. 2021 ; Leblond et al. 2022 ; Barros et al. 2023 ; Labadie et al. 2023 ; Stewart et al. 2023 ; Bouderbala et al. 2023 ), building on a rich history of spatial-stochastic models of forest growth and succession following disturbance (Sturtevant and Fortin 2021 ). Despite the importance of roads for wildlife, very few of these projections have included road networks (Leston et al. 2020 ; Labadie et al. 2023 ). For example, in a recent collection of articles focused on “Using Landscape Simulation Models to Help Balance Conflicting Goals in Changing Forests” (Hof et al. 2021 ), only one includes a road projection (Leston et al. 2020 ). In the absence of road projection methods, researchers are left to conservatively infer the amount and randomly assign the location of roaded areas (e.g. Johnson et al. 2015 ). While road projections are not typically integrated with forest landscape models, methods have been developed to support harvest planning (Dean 1997 ; Clark et al. 2000 ; Anderson and Nelson 2004 ; Epstein et al. 2006 ; Stückelberger et al. 2007 ; Meignan et al. 2012 ), and some forest management planning software suites (e.g. Remsoft; Spatial Planning Systems) include road projection tools. These methods have occasionally been incorporated into forest landscape models (e.g. Leston et al. 2020 ) but they were not designed for that purpose. Existing road projection algorithms contained in proprietary closed-source software are difficult to fully integrate into stochastic models, which often require numerous replicates (De Pellegrin Llorente et al. 2017 ; Sturtevant and Fortin 2021 ). Further, costs of licensing are generally prohibitive, deploying large commercial software suites designed for desktops to cloud or high performance computing resources is difficult or impossible, and closed-source software is often not sufficiently transparent for use in research. At the operational scale, road planning methods and tools (Dean 1997 ; Clark et al. 2000 ; e.g. Anderson and Nelson 2004 ; Epstein et al. 2006 ; Stückelberger et al. 2007 ; Meignan et al. 2012 ; Remsoft; Spatial Planning Systems) may produce more accurate road projections but the costs of applying more complex methods across large spatial extents can be prohibitive in contexts where a coarser approximation of the road network may suffice. There is a need for open-source road network projection tools that can be integrated into the open modular platforms that are used for large-scale stochastic landscape simulations (Scheller et al. 2007 ; Daniel et al. 2016 ; Barros et al. 2023 ). Automated road projection methods often rely on more general purpose route finding algorithms that identify paths among points in a mathematical graph that minimize some measure of cost (Deo 2016 ). General-purpose open source tools for graph analysis (e.g. Csárdi and Nepusz 2006 ) include methods for building graphs from adjacency matrices, finding “least cost paths” between pairs of points, and finding a least cost network of routes (i.e. a minimum spanning tree) that connects a larger set of points (Deo 2016 ). More domain-specific tools (e.g. Etten 2017 ) can help make these general-purpose tools easier to use and apply in specific contexts. Promising simple graph-based options for integrating forest road network projections into large scale stochastic projections of forest change include an iterative least cost path method developed by Hardy et al (2019; 2023 ), and least cost path and minimum spanning tree algorithms used to reconstruct road development history in British Columbia (Lochhead and Muhly 2018 ; Lochhead et al. 2022 ). The former method is available as QGIS and LANDIS-II extensions, and the latter is embedded in a larger open source modelling project that is specific to British Columbia (FAIB 2025). The Hardy et al. ( 2023 ) tool is designed for integration with other components of the LANDIS-II modelling framework (Scheller et al. 2007 ; Labadie et al. 2023 ), but none of the existing implementations can be easily integrated into the full variety of modelling frameworks used for forecasting forest change in Canada and elsewhere (Daniel et al. 2016 ; e.g. Barros et al. 2023 ). We acknowledge that using graph analysis methods to simulate roads is not new, and that other variants of iterative least cost path (or source-to-closest-target) and minimum spanning tree algorithms have been proposed and explored (Clark et al. 2000 ; Anderson and Nelson 2004 ; Picard et al. 2006 ; Stückelberger et al. 2007 ). Anderson and Nelson ( 2004 ) considered horizontal and vertical alignment, turn radius and junction angles, and switchback spacing for a relatively small 7500 ha area. However, for wide-ranging wildlife such as boreal caribou projections at much larger spatial scales are needed, and constructing and analyzing graphs for large landscapes is memory intensive and computationally demanding. We therefore focus here on assessing the adequacy of methods that minimize memory and computational requirements by setting construction costs (represented as edge weights on a simple undirected graph) from values of adjacent locations on a single user-supplied input raster. This constraint permits penalization of steep grades as suggested by Anderson and Nelson ( 2004 ), but does not permit other aspects of their method. We compare a simple grade penalty approach to the method of calculating edge weights from a cost surface raster used by both Hardy et al. (2019; 2023 ) and Lochhead et al. ( 2022 ). We argue that if these simple algorithms can produce adequate outcomes for landscape-level stochastic ecological forecasting then the additional burden of more complex realistic algorithms is not warranted in that context. To help address the need for forest road network projection tools that can be easily integrated into large spatial scale stochastic models of forest change we implemented two alternative approaches in an open source R package: 1) iterative least cost paths and 2) a minimum spanning tree (Lochhead and Muhly 2018 ; Lochhead et al. 2022 ). We also consider two different methods for determining the cost of road building: 1) a simple elevation driven cost surface and 2) a simplified version of the grade penalty approach suggested by Anderson and Nelson ( 2004 ). Our intention was to produce outputs comparable to Hardy et al. (2019; 2023 ) with a tool that can be used in modelling contexts outside of QGIS/LANDIS, and can serve as a more easily modifiable starting point for further refinement and development. We compared the ability of these methods and Hardy’s method (2019; 2023) to reconstruct observed forest road network development since 1990 in a mountainous region of British Columbia. The adequacy of a road network projection method depends on the purpose of the projection, so it is important to select performance metrics that reflect the purpose. Given our interest in projecting impacts on wildlife we focused on several common predictors of the impacts of roads on wildlife, including distance to roads (Labadie et al. 2023 ), buffered disturbance footprint (Polfus et al. 2011 ; Ibisch et al. 2016 ; Johnson et al. 2020a ; Lochhead et al. 2022 ; Labadie et al. 2023 ; Stewart et al. 2023 ), and road density (Apps and McLellan 2006 ; Apps et al. 2013 ; Serrouya et al. 2017 ). We expect these simple graph-based road projection methods to represent a reasonable compromise between computational cost and realism for stochastic projections on large landscapes, and we expect our method for penalizing steep grades to yield more realistic road networks in steep terrain relative to our simple cost method. We do not have a similarly strong hypothesis about whether minimum spanning tree projections are likely to be more or less realistic than iterative least cost path projections. We expect minimum spanning tree projections to be more accurate when resource roads are designed to simultaneously access multiple sites at the lowest cost. However, real road networks are built over time, and it is not necessarily true that they are designed to minimize overall cost, so it is possible that larger more costly road networks built by iterative least-cost path algorithms will be more realistic. Methods Study Area and Data We chose the Revelstoke timber supply area (TSA 27) in British Columbia, Canada as an assessment area because it is mountainous (Fig. 1 ), containing approximately 38% non-forested land (British Columbia Ministry of Forests 2022 ) where road building is challenging. Thus, the area provides opportunities to observe the behaviour of road projection algorithms in steep complex terrain. Most of the road network was built to access forest harvest cutblocks, allowing projection of the road network from historic forest harvest data. The Revelstoke TSA includes portions of the Monashee and Selkirk mountain ranges, and is bisected by the Columbia River. At low to mid-elevations, forests of the interior cedar-hemlock biogeoclimatic zone (British Columbia Ministry of Forests 2022 ) include remnants of rare old growth inland temperate rainforest (Government of Canada Parks Canada Agency 2017 ). At higher elevations, forests are dominated by Engelmann spruce and subalpine fir (British Columbia Ministry of Forests 2022 ). The area is within the ancestral territories of the Ktunaxa Nation, Secwepemc Nation, Syilx and Sinixt (British Columbia Ministry of Forests 2022 ). Two of six southern mountain woodland caribou herds in the area have been extirpated, and the remaining herds are considered Endangered by the Committee on the Status of Endangered Wildlife in Canada and listed as threatened under the Canadian Species at Risk Act (Environment Canada 2014 ; Palm et al. 2020 ; British Columbia Ministry of Forests 2022 ). Mountain caribou avoid roads leading to a loss of available habitat (Apps and McLellan 2006 ; Polfus et al. 2011 ), suffer increased mortality near roads (Apps et al. 2013 ), and decline in abundance as the footprint of roads and cutblocks increases (Lochhead et al. 2022 ). Roads also alter grizzly bear habitat use, home range selection, movements, population fragmentation, survival, reproductive rates, population density, trends, and conservation status (Proctor et al. 2020 ). Road network data was obtained by combining the digital road atlas [Government of British Columbia ( 2021a ); a static dataset in our study area] and forest tenure road (Forests Lands Natural Resource Operations and Rural Development 2021) data sets for British Columbia. We treated the digital road atlas as a static dataset in our study area, while the forest tenure roads are a dynamic dataset, with new road segments added yearly as road permits are issued to licensees. We removed all roads from the digital road atlas that were within 50 m of forest tenure roads to avoid duplication. To obtain a starting point for projections we filtered out forest tenure roads with an award date greater than 1990, and assumed everything in the digital road atlas was built before 1990. Cutblock information was obtained from the Harvest Areas of BC (Consolidated Cutblocks) dataset (Government of British Columbia 2021b ), compiled from provincial Forest Cover, the RESULTS Reporting system, Forest Tenure applications, and satellite imagery using change detection processes. We used cutblocks harvested after 1990 as targets for the projections, and removed cutblocks that do not contain observed roads to reduce errors associated with open areas that have been misidentified as cutblocks. Following the Interior Appraisal Manual of BC, we assume the base cost of construction per kilometer of road is a linear function of slope (Government of British Columbia 2021c ), and that road construction across major water bodies is prohibitively expensive (Table 1 ). Existing roads are assigned a construction cost of 0. We compare two methods for integrating these costs into the cost of connections between adjacent raster cells. In the “simple cost” method, we first calculate a slope raster from a digital elevation model, then calculate an input cost raster that depends on slope (Table 1 , Online Resource 1–1). The cost of the connection between two adjacent cells is simply the mean of the cost raster (Fig. 1 a) value at each of those cells, adjusted for euclidean distance to penalize longer diagonal road segments. In the “grade penalty” method (Online Resource 1–1) we use the same construction costs (Table 1 ), but calculate the cost of a connection between neighbouring cells from the elevation difference between the cells, using an elevation raster as input (Fig. 1 b), and penalizing edge grades as suggested by Anderson and Nelson ( 2004 ). The most important difference between these two approaches is that grade is a property of the edge between two nodes in the grade penalty method, while slope is a property of each node in the simple cost method. When slope is a node property, the same slope penalty applies regardless of the alignment of the road segment. Calculating grade from the difference in elevation between pairs of nodes allows us to penalize roads that follow elevation contours differently from roads that go up or down. We examined sensitivity to pixel resolution by comparing results from 1 ha (1.7 million pixels) and 100 ha (17 thousand pixels) weight rasters. We did not consider finer resolutions because stochastic forest landscape models typically simulate dynamics at resolutions > = 1 ha. Table 1 Road construction costs. In the “grade penalty” method, we assign grades > 20% a more prohibitive cost of 65000, which is also assigned to major water bodies. These barriers are not set to NA because roads must cross barriers to reach a few isolated cutblocks in our example landscape. Cost Type Value ( $ per km) Source Slope (%) 16178 + 504*slope Interior appraisal manual Major lakes 65000 Major water bodies of BC Existing roads 0 Digital road atlas and forest tenure roads for BC All data sets were accessed using the bcdata R package and filtered to the extent of the Revelstoke Timber Supply Area (Teucher et al. 2021 ). These data sets are continuously updated so we have stored the versions used in our analysis in an Open Science Framework repository (Endicott and Hughes 2024a ). We were not granted permission to share the digital road atlas so it must be accessed from the B.C. Data Catalogue (Government of British Columbia 2021a ). Scripts used for data acquisition and analysis are available on GitHub (Endicott and Hughes 2024b ). Road Projection Methods Our road projection methods link multiple target locations (also known as landings within harvest cutblocks (i.e., log decks)) to an existing road network by finding minimum-cost locations for new roadways, described by Dean ( 1997 ) as the multiple target access problem (MTAP). Specifically, we compared iterative least cost path (ILCP) and minimum spanning tree approaches (MST) proposed by Lochhead et al (2018; 2022 ) for solving the MTAP to another iterative least cost path method developed by Hardy ( 2019 ; 2023) (Fig. 3 ). We used the QGIS version of Hardy’s method (2019), which was easier for us to apply and does not differ substantially from the LANDIS extension (Hardy, personal communication). The ILCP approaches solve the MTAP by decomposing the problem into several smaller single target access problems (STAPs). For each landing (target location), the least cost path to the existing road network was solved using a mathematical graph and Dijkstra’s shortest path algorithm (Dijkstra 1959 ). There are a variety of possible methods for building graphs, choosing branch locations, and ordering target locations (Anderson and Nelson 2004 ; Picard et al. 2006 ; Hardy 2019 ; Hardy et al. 2023 ). The ILCP algorithm we used (Lochhead and Muhly 2018 ; Lochhead et al. 2022 ) is parameterized with a weighted graph wherein each node corresponds to a raster pixel representing one of three components of the potential forest road network: a landing, a point on an existing road, or an intermediate point that is a plausible location for road construction. Each node is connected with up to eight neighbours, and the cost of construction along edges is either the average of the cost raster value at the source and destination nodes, or a more complex function of the difference in elevation between the source and destination nodes (Table 1 , Online Resource 1–1), which is then weighted by the distance to neighbouring nodes. Both the grade penalty and simple cost methods involve setting edge weights on a graph from adjacent locations on a single input grid. We used a greedy heuristic method that starts with the landing nearest to an existing road, builds a least cost path from the landing to the nearest (by Euclidean distance) point on the existing road, and updates the existing road network and weight raster with each new road segment before moving on to the next nearest landing. The method described in Hardy et al. (2019; 2023 ) is similar to our simple cost ILCP method in that it breaks the MTAP down into many STAPs and finds the least cost path for each with the ordering of road construction determined by the Euclidean distance from the existing road network. A key difference is that Hardy uses the Dijkstra algorithm to solve a “single-source many-targets shortest path problem”, which is a special case of STAPs that finds the path between the landing cell and one of multiple potential connection points on the road network. In contrast, our algorithm first finds the closest point on the road network to the selected landing point and then uses the Dijkstra algorithm to solve the least cost path between the two. However, since the cost of construction on existing roads is zero we expect that in practice these two approaches are similar and the algorithm will select the least cost location for the road to connect to the existing road network. Within cutblocks, the Hardy et al. method iterates over each cell in the harvest area and builds a road if it is not within skidding distance of an existing road, while we create landing points to serve as target locations for road construction. One of the limitations with ILCP methods is the sensitivity of the road projection to the ordering of the landings (Anderson and Nelson 2004 ). Minimum spanning trees (MST) offer an alternative approach to link landings and evaluate road branching opportunities (Clark et al. 2000 ; Picard et al. 2006 ). MST algorithms do not require users to select the order in which landings are processed, but it remains necessary to select a set of points on the existing road network to include in the tree. As with the ILCP approach, the first step is to construct a graph in which each node is connected with up to eight neighbours, and set the cost of construction along edges using either the grade penalty or simple cost methods. This graph is used to estimate the least cost paths that link each of the landings and the least cost paths that link each landing to the nearest point on the existing road network. These results are used to create a second weighted graph wherein nodes are either landings or nearest points on the existing road network, and edge weights correspond to the total cost of the least cost path between pairs of nodes. The minimum spanning tree for this graph is solved using Kruskal’s algorithm (Kruskal 1956 ), which links all nodes with a tree that minimizes the sum of edge weights. This method differs from Clark et al. ( 2000 ) who used Euclidean distances for edge weights, and thus did not incorporate spatial variation in road construction costs. Note that the evolution of a road network over time can be modeled by iteratively applying any of the methods, but in this study we focus on comparing a single application of each of the methods to a network of forest roads developed since 1990. To help make these methods accessible, we implemented the ILCP and MST methods in the open source R package roads (Endicott et al. 2023 ) which depends on several other R packages: igraph for calculating least cost paths and minimum spanning trees (Csárdi and Nepusz 2006 ; Csárdi et al. 2024 ); terra for manipulating spatial raster data (Hijmans et al. 2023 ); and sf for manipulating spatial vector data (Pebesma 2018 ; Pebesma et al. 2022 ). We experimented with using graph construction methods in the package gdistance (Etten 2017 ), but opted to use Lochhead et al’s (2018; 2022 ) more computationally efficient methods (Online Resource 1–7) by default. The package is available on CRAN (The Comprehensive R Archive Network) and the source code, tutorials and documentation are available on GitHub (Endicott et al. 2023 ). Comparison of Method Variants The ILCP and MST methods connect landings within each cutblock to the existing road network, so an important first step is to select one or more landing locations in each block. We investigated the sensitivity of results to the pattern and density of landing locations by comparing results derived from the centroid of each cutblock to regular or randomly selected landing densities of 1 (low) and 10 (high) points per km 2 (Fig. 3 ). We used default settings in the Hardy QGIS plugin except for skidding distance which was set to 200 m which gives ~ 9 pts/km 2 . We compared two resolutions of the weight raster (1 ha and 100 ha). In total, we compared 12 road projection method variants to a cutblocks only scenario with no projected roads (Table 2 ). We recorded compute time and peak RAM usage for each run, and ran some projections at 25 ha to show how speed and memory requirements vary with number of landings and number of nodes in the graph (Online Resource 1–5), but did not assess other aspects of performance at that resolution. Table 2 Road projection method variants. We examined sensitivity of minimum spanning tree (MST) and iterative least cost path (ILCP) algorithms to the method of setting edge weights (grade penalty or simple cost), the placement of landings within cutblocks (regular or random), the density of landings within cutblocks, and map resolution. Projection method Sample type Sample density (pts/km 2 ) Resolutions MST grade penalty & simple cost Regular 1 1 & 100 ha MST grade penalty & simple cost Regular 10 1 & 100 ha MST grade penalty & simple cost Random 1 1 & 100 ha MST grade penalty & simple cost Random 10 1 & 100 ha MST grade penalty & simple cost Centroid N/A 1 & 100 ha ILCP grade penalty & simple cost Regular 10 1 ha Hardy QGIS N/A ~ 9 1 ha Cutblocks only N/A N/A 1 ha Performance Metrics To assess performance we calculated four characteristics of the projected and observed road networks that are relevant for wildlife: road density, measured as density of roads in meters per hectare (m/ha); distance to nearest road, measured as distance from each raster cell to the nearest road; road disturbance footprint, measured with a 1 km buffer around roads (Ibisch et al. 2016 ); and forestry disturbance footprint, measured with a 500 m buffer around roads and cutblocks. The forestry disturbance footprint is a variant of the buffered anthropogenic disturbance metric that is used as an indicator of boreal caribou critical habitat (Johnson et al. 2020a ) and to quantify impacts of disturbance on Southern Mountain caribou (Polfus et al. 2011 ; Lochhead et al. 2022 ). The other metrics are also often used in studies of anthropogenic impacts on wildlife (Apps and McLellan 2006 ; Apps et al. 2013 ; Serrouya et al. 2017 ; e.g. Labadie et al. 2023 ; Stewart et al. 2023 ). We also calculated road presence (i.e. whether or not roads are present in each pixel), but note that accurate projection of the exact locations of roads is not generally a requirement in stochastic ecological forecasts wherein the exact locations of future disturbances are often not known. Forestry disturbance footprint and road density were calculated using the disturbanceMetrics and rasterizeLineDensity functions in the caribouMetrics R package (Hughes et al. 2022 ). Aspatial performance of the projections was assessed by comparing projected and observed mean values of each road network characteristic across the timber supply area and within cutblocks. As an example, distance to roads within cutblocks is the average distance from non-road pixels within cutblocks to the nearest roads. Spatial performance was assessed in two different ways. First, difference maps were created by subtracting the projected distance to the nearest road from the observed in order to highlight discrepancies between the road networks. The observed and projected road networks were then overlaid on the difference maps and regions of major discrepancies were investigated visually. Second, performance according to the three binary metrics (road presence, road disturbance footprint, and forestry disturbance footprint) was assessed assigning pixels to 5 categories: pre-existing footprint; agree roadless, meaning that both projected and observed show no footprint; agree roaded, meaning that both projected and observed show new footprint; false positive, meaning the projection shows a footprint but no footprint was observed; and false negative, meaning the projection shows no footprint but a footprint was observed. We used these categories to calculate sensitivity, precision and F-measure of each metric (Lee et al. 2021 ). Sensitivity, also known as recall, gives the proportion of the true positives that were projected, while precision gives the proportion of positives projected that are true. Sensitivity will be lower when observed roads are missed and precision will be lower when roads are projected that were not observed. F-measure attempts to balance sensitivity and precision to give an overall score. Results According to all aspatial performance measures, the grade penalty method for setting edge weights yields better results than our simple cost methods (compare Fig. 2 and Fig. S1 ). We therefore recommend using the grade penalty method, and focus on further examining variation in performance among grade penalty method variants (Fig. 2 ). The MST regular high density method yields the best projections of road density overall (projected density: 2.79 m roads/ha, observed density: 2.9 m roads/ha), while the MST random high density method had the best projections within cutblocks (projected density: 38.41 m roads/ha, observed density: 37.69 m roads/ha), and all MST variants yield comparably good projections outside cutblocks (projected density: 1.84–1.85 m roads/ha, observed density: 2 m roads/ha). All MST variants yield better projections of the forestry disturbance footprint (projected: 0.3, observed: 0.3) than the Hardy or ILCP methods (0.29), and the Hardy and MST regular high density methods yield comparably accurate projections of road presence and distance to road. Maps (Figs. 3 and 4 ) help clarify exactly how the projection methods differed from one another, and from observed roads. While the Hardy et al. ( 2023 ) method and the grade penalty MST and ILCP methods projected similar mean road densities within cutblocks, they created distinctly different road patterns within cutblocks. In particular, the grade penalty method incentivized roads to follow slope contours, while the Hardy method created more road curvatures (Fig. 3 ). Although the grade penalty method improved realism by following elevation contours, all these methods created dead ends, intersections, and abrupt turns that would be avoided in the engineering of operational roads in steep terrain (Figs. 3 and 4 c). All of the projected road methods also crossed valley bottoms far more often than observed roads which often run parallel to one another on each side of rivers (Fig. 4 a), but this discrepancy could likely be remedied by improved costs of crossing streams, rivers and other wet areas in valley bottoms. Finally, it is important to note that some discrepancies between observed and projected road networks were caused by errors in the observed data, such as cutblocks that were incorrectly classified (e.g. Figure 4 b). We omitted cutblocks with no associated access roads from our analysis, and note that failure to do this would distort results (Figs. S3 and S4). Projections of the forestry disturbance footprint, the road disturbance footprint, and, to a lesser extent, distance to road were more accurate than projections of road density and road presence (Fig. 2 ). However, all of the road projection methods underestimate road presence and forestry disturbance footprint (Fig. 2 ), projecting fewer roads than we observe. One reason for underestimation of road density was that all these methods generate straight road segments within a pixel. As raster resolution increases, the discrepancy within pixels between real roads and straight projected segments increased, making coarse resolution projections less accurate (Fig. S6). Although these methods underestimate road presence, they represented a substantial improvement over the “cutblocks only” scenario, which assumed no new roads were built (Fig. 2 ). The methods were not designed to project the exact spatial location of new roads (within 100 m), so it was not surprising that they did not yield accurate spatial projections of road presence (F-Measure: 0.007–0.257). It was somewhat more surprising that none of the methods provided better information about the exact location of the forestry disturbance footprint than the locations of cutblocks only (F-Measure: 0.927–0.956). This was likely due to many of the projected roads being within 500 m of either the existing road network or a cutblock so the false negatives created by failing to project roads are balanced by the lack of false positives. All methods other than “cutblocks only” provided comparably good projections of the road disturbance footprint (F-Measure: 0.859–0.883), though higher sampling densities yielded fewer false negatives and more false positives (Fig. 5 ). Although the MST method performed better than the alternatives we tested according to several important wildlife habitat metrics (Fig. 2 ), it was also more costly to compute (Fig. S7). All of these methods required construction of a mathematical graph with a node for each pixel in the weight raster where road construction was possible (i.e. not NA), and memory requirements increase with the number of nodes in the graph (Fig. S7). Compute time also increased with the number of landing targets (Fig. S7). Lochhead et al’s (2018; 2022 ) method of graph construction was more computationally efficient than the adjacency matrix approach used in the R package gdistance (Etten 2017 ) (Fig. S8), in part because it avoided the additional cost of constructing a transition matrix, and in part because it ignored variation in the size of raster cells that is accounted for in the gdistance geoCorrection function. For our example landscape the Hardy QGIS plugin was 65 times faster than the MST and ILCP methods implemented in R (Fig. S7). Discussion Despite a rapidly growing body of evidence of complex and varied impacts of road networks on wildlife and other values, and increasing use of spatial stochastic models for forecasting impacts of forest disturbance across regions, very few spatial stochastic projections of forest disturbance dynamics have included road network projections, in part because tools suited for this purpose have not been available. Hardy et al. (2019; 2023 ) partially addressed this need with an extension that can be integrated with other components of the LANDIS-II modelling framework (e.g. Labadie et al. 2023 ). Our implementations of iterative least cost path and minimum spanning tree methods (Lochhead and Muhly 2018 ; Lochhead et al. 2022 ) in the open source R package roads (Endicott et al. 2023 ) provide an alternative to Hardy’s method that improved predicted road patterns in our example landscape, and can be easily integrated into R workflows and modelling frameworks used for forecasting forest change (Daniel et al. 2016 ; e.g. Barros et al. 2023 ). Although projecting roads within a stochastic ecological forecasting model was beyond the scope of this study, our analysis of performance using metrics relevant to wildlife can help guide method selection in that context. We evaluated performance of forest road network projection algorithms in a topographically complex region of British Columbia where we anticipated projection would be challenging, and implemented a simplified version of Anderson and Nelson’s ( 2004 ) grade penalty method for setting graph edge weights that improves performance in this mountainous region. The simplified grade penalty method does not substantially increase memory or computational requirements, and we recommend its use in regions where slope is an important determinant of road construction cost. However, we also acknowledge that we have used a very simple cost model, and that the important determinants of road construction costs may vary among landscapes. We have therefore also allowed the possibility for users of the roads R package to specify their own method for determining edge weights from adjacent node values in a weight function. A more complex cost model could include variation among substrate types (e.g. peat, clay, gravel) and more nuanced consideration of the costs of various types of water crossings (e.g. culverts and bridges). Road construction costs and wildlife impacts also vary substantially among types of roads (e.g. long-term vs temporary haul roads). If distinguishing among types of roads is important, an effective approach might be to first project the paths of lengthy long-term haul roads using a cost model and a set of targets that are appropriate for this type of road, and then project the paths of shorter temporary haul roads or spur roads from this long-term network using a different cost model and set of landing locations. Among the method variants we considered, the minimum spanning tree (MST) regular high landing density method yielded the best projections of average road density and forestry disturbance footprint (Fig. 2 ). Aspatial summaries of road presence, distance to road, and road disturbance footprint were comparable to Hardy’s method (2023) (Fig. 2 ). However, the MST algorithm was also more computationally demanding than the other methods we tested (Fig. S7), so users will need to consider whether the benefits of improved performance outweigh the increased computational costs in their particular circumstance. Aspatial metrics of overall projection accuracy can be difficult to interpret. To more deeply understand variation in the behaviour of road network projection algorithms we found it helpful to distinguish between performance within cutblocks and outside of them, and to examine spatial patterns. This examination highlighted some errors that could be fixed by better accounting for the cost of water crossings in our weight raster, and other more fundamental limitations and differences among methods. Even in cases where the Hardy and grade penalty methods project similar mean road densities, the projected pattern of roads within cutblocks is quite different, in that the grade penalty roads tend to follow elevation contours, while Hardy et al’s method produces curves (Fig. 3 ). We suspect the latter behaviour is an artifact of searching for nearest destinations > 200 m away on a 100 m grid. Although the grade penalty method improves performance by aligning roads with elevation contours, all these methods create dead ends, intersections, and abrupt turns that are avoided in real road networks. We chose not to implement Anderson and Nelson’s methods (2004) for penalizing sharp curves, large junction angles and switchback spacing because these complications would increase algorithmic complexity and computational costs. Simple algorithms can recreate some aspects of observed road networks but not others, and more complex realistic algorithms are generally more costly to develop and use. Before investing in additional complexity, it is important to recognize that adequacy depends on the goal. For some purposes, projections from relatively crude cost models and simple algorithms may be adequate. For example, models of caribou movement often include distance to roads as a predictor (e.g. Labadie et al. 2023 ), and a combined metric of buffered anthropogenic disturbance best explains variation in boreal caribou demographic rates among ranges (Johnson et al. 2020a ). In our example landscape, projections of the disturbance footprints (500 m and 1km buffer) and distance to road are reasonably accurate, and for wildlife response models informed by these measures our somewhat crude road projections are likely good enough. On the other hand, these methods do not predict the exact spatial locations of roads, so outcomes and impacts should not be reported or interpreted at the level of individual pixels; for example, these projections should not guide selection of sample locations for studies of the local (pixel level) impacts of anticipated roads. Examination of spatial mismatch between observed and projected roads also highlighted errors in our input data that are not unique to our study area; there is considerable variation in the consistency and accuracy of available resource road information across Canada and elsewhere (Poley et al. 2022 ; Forests Lands Natural Resource Operations and Rural Development) that users should keep in mind when interpreting discrepancies between observed and projected road networks. These road projection methods require construction of large graphs, with a node for each pixel in which road construction is possible. In our example landscape we were able to project at 100 m resolution in a reasonable amount of time (< 2.4 hours) using modest amounts of memory (< 3 GB RAM). However, our example landscape is fairly small, and memory requirements increase rapidly with landscape size (Online Resource 1–5). Although it is possible to use machines with more RAM, the computational cost of adding road projections to spatial stochastic simulations of forest change relevant for wide-ranging wildlife is not trivial. For example, in a study of the effects of forest management and climate change scenarios on wildlife in a single boreal caribou range (100 stochastic realizations of 6 scenarios), we projected roads at 50 m resolution using graph with ~ 21 million nodes. Each ILCP road projection required 2.7 hours and 11 GB of RAM, for a total of 1620 compute hours. Options for reducing computational requirements include restricting the area where road construction is possible (by setting some portions of the weight raster to NA), reducing landing density within cutblocks (by using the centroid method), increasing raster resolution, or using Hardy’s more efficient implementations. The first three options are available to users of the roads R package. However, those contemplating reducing computational cost by decreasing raster resolution should be cautious of the results given that all of these methods become less accurate as raster cell size increases (Figs. S5 and S6). The distance to road metric is less sensitive to changes in cell resolution, and distance metrics may also be more ecologically meaningful (Rowland et al. 2000 ). Those considering reducing computational cost by reducing landing density within cutblocks should be clear that their projected road network will be incomplete and metrics such as total road length and road density will be underestimated. More involved possibilities for reducing computational costs include adopting faster routing algorithms (Larmet 2022 ), implementing slow components of the R package (Fig. S8) in a faster compiled language, parallel processing, and implementing a multi-scale approach in which coarse resolution results are used to limit the area considered in finer resolution projections. Several of these options have been implemented in the cppRouting R package (Larmet 2022 ), and future versions of the roads R package could borrow and build on cppRouting functionality. There are many metrics of connectivity and fragmentation (Francis et al. 2021 ) we did not consider in this paper. In general, performance will vary among metrics, and it is difficult to determine what is good enough without reference to particular circumstances (i.e. objectives, species, landscapes). For example, if the outcome of interest is space use or connectivity projected from more (e.g. Whittington et al. 2022 ; Labadie et al. 2023 ) or less (e.g. Diniz et al. 2020 ; Hughes et al. 2023 ) species-specific movement models it may be helpful to test sensitivity to the difference between observed and retrospectively projected roads to confirm that projections are adequate for the intended purpose. When constructing new wildlife response models intended for projection, it may also be helpful to select metrics such as distance to road or buffered disturbance footprint that are more easily projectable (Bodner et al. 2021 ). Rather than testing additional performance metrics, we have focused our efforts on the development of an open source R package (Endicott et al. 2023 ) and workflows (Endicott and Hughes 2024b ) that enable users to devise performance tests appropriate to their particular circumstances and goals. We recognize that acquiring data to retrospectively project roads can be challenging, so we also provide output maps (Endicott and Hughes 2024a ) and an example script (analysis/scripts/4_calc_any_metric.R) to enable calculation of alternative performance metrics of interest for our Revelstoke example, along with a data preparation script (analysis/scripts/1_prepareData.R) that could be modified to extract required input data for other areas in British Columbia (Endicott and Hughes 2024b ). In this study our focus was developing tools capable of integrating forest road development into projections of the cumulative effects of disturbance on wildlife. However, we note there are other possible uses for these road network projection methods. Cost minimizing methods are not suitable for projecting development of seismic lines (e.g. Fig. S9), but for other circumstances such as mine access where the objective is to minimize the cost of reaching a target location these methods may be applicable. Lochhead et al. ( 2022 ) used the least cost path algorithm to reconstruct the likely history of road network development in a case where harvest dates were known but road construction dates were not. It would also be interesting to account for persistent impacts of resource roads on vegetation, productivity and carbon sequestration (Braham et al. 2023 ). Conclusion Our open source automated forest road projection tools can be integrated into spatial stochastic models of the implications of forest change for wide-ranging wildlife such as woodland caribou. In a mountainous test landscape, a high density of landings, a method for penalizing road construction on steep grades, and a minimum spanning tree algorithm all improved projection accuracy, and we recommend these options. We hope that these tools will reduce barriers to integrating resource roads into spatial stochastic models of forest dynamics, and thereby help improve projections of anticipated resource development impacts on wildlife across large areas. Declarations Acknowledgements This work was motivated by discussions with Colin Daniel (Apex Resource Management Solutions) and needs of the Western Boreal Initiative, an effort to project cumulative effects of disturbances on forests and wildlife led by Eliot McIntire (Canadian Forest Service, Natural Resources Canada) and Samuel Haché (Canadian Wildlife Service, Environment and Climate Change Canada). Questions from Steve Cumming (Université Laval) and Lisa Venier (Canadian Forest Service, Natural Resources Canada) helped refine our focus and approach. We thank Clément Hardy for answering our questions about his method. Funding Funding for this work was provided by Environment and Climate Change Canada. Competing interests The authors declare there are no competing interests. Author Contributions Conceptualization: JH, SE, DL, GP, KL Data curation: SE, DL Formal analysis: JH, SE, DL Funding acquisition: JH Investigation: JH, SE Methodology: JH, SE, DL, KL Project administration: JH Resources: JH Software: JH, SE, KL Supervision: JH Visualization: SE Writing – original draft: JH, SE, DL Writing – review & editing: JH, SE, DL, GP, KL Data availability Data generated or analyzed during this study are available in the Open Science Framework repository, https://osf.io/8vmqh/?view_only=d168967961914f22af89e44c19aec019. The code to reproduce the analysis and figures in this paper is available on GitHub at https://github.com/LandSciTech/RoadPaper. The roads package can be installed from CRAN and the code and a user guide are available on GitHub at https://github.com/LandSciTech/roads. References Albrich K, Rammer W, Turner MG, et al (2020) Simulating forest resilience: A review. Global Ecol Biogeogr 29:2082–2096. https://doi.org/10.1111/geb.13197 Anderson AE, Nelson J (2004) Projecting vector-based road networks with a shortest path algorithm. Can J For Res 34:1444–1457. https://doi.org/10.1139/x04-030 Apps CD, McLellan BN (2006) Factors influencing the dispersion and fragmentation of endangered mountain caribou populations. Biol Conserv 130:84–97. https://doi.org/10.1016/j.biocon.2005.12.004 Apps CD, Mclellan BN, Kinley TA, et al (2013) Spatial factors related to mortality and population decline of endangered mountain caribou. J Wildl Manage 77:1409–1419. https://doi.org/10.1002/jwmg.601 Barrientos R, Ascensão F, D’Amico M, et al (2021) The lost road: Do transportation networks imperil wildlife population persistence? Perspect Ecol Conserv 19:411–416. https://doi.org/10.1016/j.pecon.2021.07.004 Barros C, Luo Y, Chubaty AM, et al (2023) Empowering ecological modellers with a PERFICT workflow: Seamlessly linking data, parameterisation, prediction, validation and visualisation. Methods Ecol Evol 14:173–188. https://doi.org/10.1111/2041-210X.14034 Beauchesne D, Jaeger JA, St-Laurent M-H (2013) Disentangling woodland caribou movements in response to clearcuts and roads across temporal scales. PLoS One 8:e77514. https://doi.org/10.1371/journal.pone.0077514 Bodner K, Rauen Firkowski C, Bennett JR, et al (2021) Bridging the divide between ecological forecasts and environmental decision making. Ecosphere 12:e03869. https://doi.org/10.1002/ecs2.3869 Bouderbala I, Labadie G, Béland J-M, et al (2023) Long-term effect of forest harvesting on boreal species assemblages under climate change. PLOS Clim 2:e0000179. https://doi.org/10.1371/journal.pclm.0000179 Braham N, Valeria O, Imbeau L (2023) Characterization of vegetation dynamics on linear features using airborne laser scanning and ensemble learning. Forests 14:511. https://doi.org/10.3390/f14030511 British Columbia Ministry of Forests (2022) Revelstoke timber supply area timber supply review data package Clark MM, Meller RD, McDonald TP (2000) A three-stage heuristic for harvest scheduling with access road network development. For Sci 46:204–218. https://doi.org/10.1093/forestscience/46.2.204 Csárdi G, Nepusz T (2006) The igraph software package for complex network research. InterJournal Complex Systems:1695 Csárdi G, Nepusz T, Traag V, et al (2024) Igraph: Network analysis and visualization in R Daniel CJ, Frid L, Sleeter BM, Fortin M-J (2016) State-and-transition simulation models: A framework for forecasting landscape change. Methods Ecol Evol 7:1413–1423. https://doi.org/10.1111/2041-210X.12597 Daniel CJ, Ter-Mikaelian MT, Wotton BM, et al (2017) Incorporating uncertainty into forest management planning: Timber harvest, wildfire and climate change in the boreal forest. For Ecol Manage 400:542–554. https://doi.org/10.1016/j.foreco.2017.06.039 De Pellegrin Llorente I, Hoganson HM, Carson MT, Windmuller-Campione M (2017) Recognizing spatial considerations in forest management planning. Curr For Rep 3:308–316. https://doi.org/10.1007/s40725-017-0068-x Dean DJ (1997) Finding optimal routes for networks of harvest site access roads using GIS-based techniques. Can J For Res 27:11–22. https://doi.org/10.1139/x96-144 DeCesare NJ, Hebblewhite M, Bradley M, et al (2014) Linking habitat selection and predation risk to spatial variation in survival. J Anim Ecol 83:343–352. https://doi.org/10.1111/1365-2656.12144 Deo N (2016) Graph theory with applications to engineering and computer science. Dover Publications Dickie M, McNay SR, Sutherland GD, et al (2020) Corridors or risk? Movement along, and use of, linear features varies predictably among large mammal predator and prey species. J Anim Ecol 89:623–634. https://doi.org/10.1111/1365-2656.13130 Dietze M (2017) Ecological Forecasting. Princeton University Press, New Jersey Dijkstra EW (1959) A note on two problems in connexion with graphs. Numer Math (Heidelb) 1:269–271. https://doi.org/10.1007/BF01386390 Diniz MF, Cushman SA, Machado RB, De Marco Júnior P (2020) Landscape connectivity modeling from the perspective of animal dispersal. Landsc Ecol 35:41–58. https://doi.org/10.1007/s10980-019-00935-3 Endicott S, Hughes J (2024b) RoadPaper. https://github.com/LandSciTech/RoadPaper Endicott S, Hughes J (2024a) Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife. https://osf.io/8vmqh/?view_only=d168967961914f22af89e44c19aec019 Endicott S, Lochhead KD, Hughes J, Kirby P (2023) Roads: Road network projection Environment Canada (2014) Recovery strategy for the woodland caribou, southern mountain population (Rangifer tarandus caribou) in Canada. Environment Canada, Ottawa, Ontario Epstein R, Weintraub A, Sapunar P, et al (2006) A combinatorial heuristic approach for solving real-size machinery location and road design problems in forestry planning. Oper Res 54:1017–1027. https://doi.org/10.1287/opre.1060.0331 Etten J van (2017) R package gdistance: Distances and routes on geographical grids. J Stat Softw 76:1–21. https://doi.org/10.18637/jss.v076.i13 Fahrig L, Rytwinski T (2009) Effects of roads on animal abundance: An empirical review and synthesis. Ecol Soc 14: https://doi.org/10.5751/ES-02815-140121 FAIB (Forest Analysis and Inventory Branch) (2025) Castor - forest and land use simulator. https://github.com/bcgov/castor. Accessed 4 Apr 2024 Forests Lands Natural Resource Operations and Rural Development BCM of (2021) Forest tenure road lines - Data catalogue. https://catalogue.data.gov.bc.ca/dataset/243c94a1-f275-41dc-bc37-91d8a2b26e10. Accessed 13 Oct 2021 Forests Lands Natural Resource Operations and Rural Development BCM of Resource roads spatial data update. https://www2.gov.bc.ca/gov/content/data/geographic-data-services/topographic-data/roads/resource-roads. Accessed 20 Apr 2023 Francis RA, Millington JDA, Perry GLW, Minor ES (eds) (2021) The Routledge handbook of landscape ecology. Routledge, London Government of British Columbia (2021c) Interior appraisal manual - Province of British Columbia. https://www2.gov.bc.ca/gov/content/industry/forestry/competitive-forest-industry/timber-pricing/interior-timber-pricing/interior-appraisal-manual. Accessed 23 Nov 2021 Government of British Columbia (2021a) Digital Road Atlas (DRA) - Master Partially-Attributed Roads - Data Catalogue. https://catalogue.data.gov.bc.ca/dataset/bb060417-b6e6-4548-b837-f9060d94743e. Accessed 13 Oct 2021 Government of British Columbia (2021b) Harvested areas of BC (consolidated cutblocks) - Data catalogue. https://catalogue.data.gov.bc.ca/dataset/harvested-areas-of-bc-consolidated-cutblocks-. Accessed 15 Jul 2021 Government of Canada Parks Canada Agency (2017) Old Growth Inland Rainforest. https://parks.canada.ca/pn-np/bc/glacier/nature/naturelle-natural/pluviale-rainforest. Accessed 19 Apr 2023 Habrich AK, Lawrence ER, Fraser DJ (2021) Varying genetic imprints of road networks and human density in North American mammal populations. Evol Appl 14:1659–1672. https://doi.org/10.1111/eva.13232 Hardy C (2019) Forest roads network: A QGIS plugin. https://github.com/Klemet/ForestRoadNetworkPluginForQGIS. Accessed 10 Jan 2023 Hardy C, Messier C, Valeria O, Filotas E (2023) A LANDIS-II extension for simulating forest road networks. Can J For Res. https://doi.org/10.1139/cjfr-2022-0306 Hijmans RJ, Bivand R, Pebesma E, Sumner MD (2023) Terra: Spatial data analysis Hill JE, DeVault TL, Belant JL (2021) A review of ecological factors promoting road use by mammals. Mamm Rev 51:214–227. https://doi.org/10.1111/mam.12222 Hof AR, Montoro Girona M, Fortin M-J, Tremblay JA (2021) Editorial: Using landscape simulation models to help balance conflicting goals in changing forests. Front Ecol Evol 9:818. https://doi.org/10.3389/fevo.2021.795736 Hughes J, Endicott S, Simpkins C, et al (2022) caribouMetrics: Models and metrics of boreal caribou responses to forest landscapes Hughes J, Lucet V, Barrett G, et al (2023) Comparison and parallel implementation of alternative moving-window metrics of the connectivity of protected areas across large landscapes. Landsc Ecol. https://doi.org/10.1007/s10980-023-01619-9 Ibisch P, Hoffmann M, Kreft S, et al (2016) A global map of roadless areas and their conservation status. Science (1979) 354:1423–1427. https://doi.org/10.1126/science.aaf7166 Johnson CA, Sutherland GD, Neave E, et al (2020a) Science to inform policy: Linking population dynamics to habitat for a threatened species in Canada. J Appl Ecol 57:1314–1327. https://doi.org/10.1111/1365-2664.13637 Johnson CJ, Ehlers LPW, Seip DR (2015) Witnessing extinction – Cumulative impacts across landscapes and the future loss of an evolutionarily significant unit of woodland caribou in Canada. Biol Conserv 186:176–186. https://doi.org/10.1016/j.biocon.2015.03.012 Johnson CJ, Venter O, Ray JC, Watson JEM (2020b) Growth-inducing infrastructure represents transformative yet ignored keystone environmental decisions. Conserv Lett 13:e12696. https://doi.org/10.1111/conl.12696 Kruskal JB (1956) On the shortest spanning subtree of a graph and the traveling salesman problem. Proc Am Math Soc 7:48–50. https://doi.org/10.2307/2033241 Labadie G, Hardy C, Boulanger Y, et al (2023) Global change risks a threatened species due to alteration of predator–prey dynamics. Ecosphere 14:e4485. https://doi.org/10.1002/ecs2.4485 Lacerte R, Leblond M, St-Laurent M-H (2021) Determinants of vegetation regeneration on forest roads following restoration treatments: Implications for boreal caribou conservation. Restor Ecol 29:e13414. https://doi.org/10.1111/rec.13414 Larmet V (2022) cppRouting: Algorithms for routing and solving the traffic assignment problem Laurian C, Dussault C, Ouellet J-P, et al (2012) Interactions between a large herbivore and a road network. Écoscience 19:69–79. https://doi.org/10.2980/19-1-3461 Leblond M, Boulanger Y, Pascual Puigdevall J, St-Laurent M-H (2022) There is still time to reconcile forest management with climate-driven declines in habitat suitability for boreal caribou. Glob Ecol Conserv 39:e02294. https://doi.org/10.1016/j.gecco.2022.e02294 Leblond M, Dussault C, Ouellet J-P (2013) Avoidance of roads by large herbivores and its relation to disturbance intensity. J Zool 289:32–40 Lee S, Kim J, Lee G, et al (2021) Prediction of aquatic ecosystem health indices through machine learning models using the wgan-based data augmentation method. Sustainability 13:10435. https://doi.org/10.3390/su131810435 Leston L, Bayne E, Dzus E, et al (2020) Quantifying long-term bird population responses to simulated harvest plans and cumulative effects of disturbance. Front Ecol Evol 8:252. https://doi.org/10.3389/fevo.2020.00252 Lochhead KD, Kleynhans EJ, Muhly TB (2022) Linking woodland caribou abundance to forestry disturbance in southern British Columbia, Canada. J Wildl Manage 86:e22149. https://doi.org/10.1002/jwmg.22149 Lochhead KD, Muhly TB (2018) A demo of road simulation in CLUS. https://github.com/bcgov/clus/blob/master/reports/roads/draft-CLUS-roads.md. Accessed 2 Dec 2021 McIntire EJB, Chubaty AM, Cumming SG, et al (2022) PERFICT: A re-imagined foundation for predictive ecology. Ecol Lett 25:1345–1351. https://doi.org/10.1111/ele.13994 Meignan D, Frayret J-M, Pesant G, Blouin M (2012) A heuristic approach to automated forest road location. Can J For Res 42:2130–2141. https://doi.org/10.1139/x2012-140 Micheletti T, Stewart FEC, Cumming SG, et al (2021) Assessing pathways of climate change effects in SpaDES: An application to boreal landbirds of Northwest Territories Canada. Front Ecol Evol 9:654. https://doi.org/10.3389/fevo.2021.679673 Muhly TB, Semeniuk C, Massolo A, et al (2011) Human activity helps prey win the predator-prey space race. PLoS One 6:e17050. https://doi.org/10.1371/journal.pone.0017050 Mumma MA, Gillingham MP, Parker KL, et al (2018) Predation risk for boreal woodland caribou in human-modified landscapes: Evidence of wolf spatial responses independent of apparent competition. Biol Conserv 228:215–223. https://doi.org/10.1016/j.biocon.2018.09.015 Palm EC, Fluker S, Nesbitt HK, et al (2020) The long road to protecting critical habitat for species at risk: The case of southern mountain woodland caribou. Conserv Sci Pract 2:e219. https://doi.org/10.1111/csp2.219 Pebesma E (2018) Simple features for R: Standardized support for spatial vector data. R J 10:439–446. https://doi.org/10.32614/RJ-2018-009 Pebesma E, Bivand R, Racine E, et al (2022) Sf: Simple features for R Petter G, Mairota P, Albrich K, et al (2020) How robust are future projections of forest landscape dynamics? Insights from a systematic comparison of four forest landscape models. Environ Model Softw 134:104844. https://doi.org/10.1016/j.envsoft.2020.104844 Picard N, Gazull L, Freycon V (2006) Finding optimal routes for harvesting tree access. Int J For Eng 17:35–50. https://doi.org/10.1080/14942119.2006.10702534 Pinard V, Dussault C, Ouellet J-P, et al (2012) Calving rate, calf survival rate, and habitat selection of forest-dwelling caribou in a highly managed landscape. J Wildl Manage 76:189–199. https://doi.org/10.1002/jwmg.217 Poley LG, Schuster R, Smith W, Ray JC (2022) Identifying differences in roadless areas in Canada based on global, national, and regional road datasets. Conserv Sci Pract n/a:e12656. https://doi.org/10.1111/csp2.12656 Polfus JL, Hebblewhite M, Heinemeyer K (2011) Identifying indirect habitat loss and avoidance of human infrastructure by northern mountain woodland caribou. Biol Conserv 144:2637–2646. https://doi.org/10.1016/j.biocon.2011.07.023 Proctor MF, McLellan BN, Stenhouse GB, et al (2020) Effects of roads and motorized human access on grizzly bear populations in British Columbia and Alberta, Canada. Ursus 2019:16–39. https://doi.org/10.2192/URSUS-D-18-00016.2 Remsoft Remsoft road optimizer. https://remsoft.com/. Accessed 12 Apr 2023 Riitters K, Potter K, Iannone BV, et al (2017) Landscape correlates of forest plant invasions: A high-resolution analysis across the eastern United States. Divers Distrib 24:274–284. https://doi.org/10.1111/ddi.12680 Rowland MM, Wisdom MJ, Johnson BK, Kie JG (2000) Elk distribution and modeling in relation to roads. J Wildl Manage 64:672–684. https://doi.org/10.2307/3802737 Scheller RM, Domingo JB, Sturtevant BR, et al (2007) Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecol Modell 201:409–419. https://doi.org/10.1016/j.ecolmodel.2006.10.009 Serrouya R, Kellner A, Pavan G, et al (2017) Time vs. Distance: Alternate metrics of animal resource selection provide opposing inference. Ecosphere 8:e01730. https://doi.org/10.1002/ecs2.1730 Snell RS, Elkin C, Kotlarski S, Bugmann H (2018) Importance of climate uncertainty for projections of forest ecosystem services. Reg Environ Change Spatial Planning Systems Patchworks. https://spatial.ca/. Accessed 10 Jan 2023 Stewart FEC, Micheletti T, Cumming SG, et al (2023) Climate-informed forecasts reveal dramatic local habitat shifts and population uncertainty for northern boreal caribou. Ecol Appl 33:e2816. https://doi.org/10.1002/eap.2816 St-Pierre F, Drapeau P, St-Laurent M-H (2021) Drivers of vegetation regrowth on logging roads in the boreal forest: Implications for restoration of woodland caribou habitat. For Ecol Manage 482:118846. https://doi.org/10.1016/j.foreco.2020.118846 St-Pierre F, Drapeau P, St-Laurent M-H (2022) Stairway to heaven or highway to hell? How characteristics of forest roads shape their use by large mammals in the boreal forest. For Ecol Manage 510:120108. https://doi.org/10.1016/j.foreco.2022.120108 Stückelberger J, Heinimann HR, Chung W (2007) Improved road network design models with the consideration of various link patterns and road design elements. Can J For Res 37:2281–2298. https://doi.org/10.1139/X07-036 Sturtevant BR, Fortin M-J (2021) Understanding and modeling forest disturbance interactions at the landscape level. Front Ecol Evol 9:636. https://doi.org/10.3389/fevo.2021.653647 Teucher AC, Albers SJ, Hazlitt SL (2021) Bcdata: An R package for searching and retrieving data from the B.C. Data Catalogue. J Open Source Softw 6:2927. https://doi.org/10.21105/joss.02927 Tremblay JA, Boulanger Y, Cyr D, et al (2018) Harvesting interacts with climate change to affect future habitat quality of a focal species in eastern Canada’s boreal forest. PLoS One 13:e0191645. https://doi.org/10.1371/journal.pone.0191645 Trombulak SC, Frissell CA (2000) Review of ecological effects of roads on terrestrial and aquatic communities. Conserv Biol 14:18–30. https://doi.org/10.1046/j.1523-1739.2000.99084.x Tucker MA, Böhning-Gaese K, Fagan WF, et al (2018) Moving in the Anthropocene: Global reductions in terrestrial mammalian movements. Science (1979). https://doi.org/10.1126/science.aam9712 Whittington J, Hebblewhite M, Baron RW, et al (2022) Towns and trails drive carnivore movement behaviour, resource selection, and connectivity. Mov Ecol 10:17. https://doi.org/10.1186/s40462-022-00318-5 Additional Declarations No competing interests reported. Supplementary Files supplement.pdf Online Resource 1. Detailed methods and additional results and figures, including: detailed methods for the grade penalty function; results using simple cost, all cutblocks, and a coarse resolution raster; projection method benchmarking and profiling; and a map of roads in Fort Nelson timber supply area. Cite Share Download PDF Status: Published Journal Publication published 03 Nov, 2025 Read the published version in Landscape Ecology → Version 1 posted Editorial decision: Revision requested 15 Aug, 2025 Reviews received at journal 12 Aug, 2025 Reviews received at journal 03 Aug, 2025 Reviewers agreed at journal 23 Jul, 2025 Reviewers agreed at journal 21 Jul, 2025 Reviewers invited by journal 22 Apr, 2025 Editor assigned by journal 09 Apr, 2025 Submission checks completed at journal 09 Apr, 2025 First submitted to journal 09 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6413286","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":440867263,"identity":"73eae6c4-981b-42e0-8a61-3d5f308da872","order_by":0,"name":"Josie Hughes","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYBACxgYGhgNgJAHkfSBZC+MMEiyDaGHmIUYtc3v7wwMfGO4kzo9uPvzZpsKOQd79AOOHH/gc1nPG4OAMhmeJG+8cS5POOZPMYHgmgVmyB5+WGTkMh3kYDidunJFjxpzbdoDBcAYDGwM+FzLOSH9w+A9YS/7nz5b/IFoY/+DVkmBwmAGoZb5EDoM0Y8MBBnkJBja84QD2S4/BYeMNEmlmkj3HknkMeBKbpWXwaDFsb3/84UfFYdn5M5KBjBo7Ofn2wwc/vsGnpQFEGgDRAYgAj8EBxgY8GhgY5OGMBnTGKBgFo2AUjAIoAAAzClKoYOFp4AAAAABJRU5ErkJggg==","orcid":"","institution":"National Wildlife Research Center, Environment and Climate Change Canada","correspondingAuthor":true,"prefix":"","firstName":"Josie","middleName":"","lastName":"Hughes","suffix":""},{"id":440867264,"identity":"80dbe43a-86df-47f4-94ad-8d26a9832c82","order_by":1,"name":"Sarah Endicott","email":"","orcid":"","institution":"National Wildlife Research Center, Environment and Climate Change Canada","correspondingAuthor":false,"prefix":"","firstName":"Sarah","middleName":"","lastName":"Endicott","suffix":""},{"id":440867265,"identity":"1cf4acae-00a7-46a4-83ae-d9c000d82cdd","order_by":2,"name":"David Lapins","email":"","orcid":"","institution":"National Wildlife Research Center, Environment and Climate Change Canada","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"","lastName":"Lapins","suffix":""},{"id":440867266,"identity":"c596ab38-aeda-499e-9fdd-629d8a2918b0","order_by":3,"name":"Kyle Lochhead","email":"","orcid":"","institution":"Ministry of Forests, British Columbia","correspondingAuthor":false,"prefix":"","firstName":"Kyle","middleName":"","lastName":"Lochhead","suffix":""},{"id":440867267,"identity":"95d03820-04ae-4681-9b10-285511e53041","order_by":4,"name":"Gregory Paradis","email":"","orcid":"","institution":"The University of British Columbia","correspondingAuthor":false,"prefix":"","firstName":"Gregory","middleName":"","lastName":"Paradis","suffix":""}],"badges":[],"createdAt":"2025-04-09 15:23:22","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6413286/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6413286/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10980-025-02232-8","type":"published","date":"2025-11-03T15:57:37+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80367928,"identity":"2e6e700e-3f57-49c2-a834-558412f452c0","added_by":"auto","created_at":"2025-04-11 06:02:54","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":535155,"visible":true,"origin":"","legend":"\u003cp\u003eRoad construction cost surface and an elevation surface used to calculate grade penalties. Costs are calculated using a simple model informed by slope and major barriers (Table 1, Online Resource 1-1) in the Revelstoke timber supply area in British Colombia, Canada. In both layers, pixels containing existing roads are assigned a value of 0\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/06d3c977254ef2938c0565b1.png"},{"id":80366801,"identity":"125caa6e-aa6d-406a-a690-5e88e13c9adf","added_by":"auto","created_at":"2025-04-11 05:38:54","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":899979,"visible":true,"origin":"","legend":"\u003cp\u003eProportional difference of projected mean from observed mean road metrics using different projection method variants within cutblocks, outside cutblocks, and overall (1 ha weight raster). Values greater than zero indicate overprojection and negative values indicate underprojection. Cutblocks not accessible via the observed road network were excluded from analysis. Performance varies among metrics (columns), projection method variants (colours), and within and outside cutblocks. Projection method variants are defined in Table 2. Note that extreme values for the cutblocks only method within cutblocks are cut off to ensure patterns in the other methods to remain visible. Bars that are cutoff are labeled with the total value\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/450486695679335826f086b9.png"},{"id":80366803,"identity":"7ca5145d-85ae-4203-9681-7e864c4cc637","added_by":"auto","created_at":"2025-04-11 05:38:54","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":983895,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of road projection method variants (Table 2). The large four sided cutblock in the top right was added to more clearly show algorithm behaviour within cutblocks.The grade penalty method for setting edge weights (top row) encourages roads to follow slope contours, and yields fewer obviously unrealistic 90 degree turns in steep terrain than our simple cost method (middle row). Selecting a single landing at the centroid of each cutblock (bottom left) creates too few roads within cutblocks. In this 1 ha resolution example the Hardy QGIS plugin (bottom middle) creates more curved roads than our simple cost method (middle row), and does not follow slope contours like the grade penalty method (top row). Landing density is high for all except the centroid and Hardy variants\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/b930bee046ef3ff67a72de80.png"},{"id":80366804,"identity":"d5f5e764-607d-4711-937d-fe8d8ea08ce8","added_by":"auto","created_at":"2025-04-11 05:38:54","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2238481,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of the minimum spanning tree projection with regular high density sampling to the observed road network. The main map shows observed distance to the nearest road minus the projected distance. Insets highlight areas of interest with satellite imagery (Google 2022) for context: (a) Projected and observed road networks follow valleys but projected roads cross valley bottoms more frequently; (b) some open areas are misidentified as cutblocks; (c) projected roads follow contours but do not avoid turns and dead ends like observed switchbacks\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/49b65b0218f264a7cef6f86b.png"},{"id":80367930,"identity":"999fc81f-c964-44af-a4bd-58f7ff592450","added_by":"auto","created_at":"2025-04-11 06:02:54","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":927738,"visible":true,"origin":"","legend":"\u003cp\u003eVariation in spatial performance among projection method variants (1 ha weight raster). F-measure is the harmonic mean of precision and sensitivity, precision is the proportion of predicted pixels that were observed and sensitivity is the proportion of observed pixels that were correctly predicted. Values closer to one indicate better performance. Cutblocks not accessible via the observed road network were excluded from analysis. Projection method variants are defined in Table 2\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/55d28f4dd128bd5dfcad4d72.png"},{"id":95564135,"identity":"d5b6130f-9bd4-4e2c-9582-e5c4dc9dd96e","added_by":"auto","created_at":"2025-11-10 16:08:13","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6311593,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/516896a6-b7b6-4adc-afc7-61edd66d36d4.pdf"},{"id":80367937,"identity":"16023b49-32da-40a3-8024-25653f2a9003","added_by":"auto","created_at":"2025-04-11 06:02:56","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":8107580,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eOnline Resource 1.\u003c/em\u003e Detailed methods and additional results and figures, including: detailed methods for the grade penalty function; results using simple cost, all cutblocks, and a coarse resolution raster; projection method benchmarking and profiling; and a map of roads in Fort Nelson timber supply area.\u003c/p\u003e","description":"","filename":"supplement.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6413286/v1/d95ceb0375ce3b8ebbedbcd4.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe impacts of roads on wildlife are complex and varied. Mortality risk and stress on or near roads can be elevated due to vehicle collisions, hunting pressure, predation, noise and other stressors, but roads can also provide food, facilitate movement, and provide refuge (Trombulak and Frissell \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Fahrig and Rytwinski \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Hill et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In Canadian forests, there is ample evidence of linear features attracting predators such as bears (\u003cem\u003eUrsus americanus\u003c/em\u003e, \u003cem\u003eUrsus arctos\u003c/em\u003e) and wolves (\u003cem\u003eCanis lupus\u003c/em\u003e) seeking food or easy movement, while repelling prey such as woodland caribou (\u003cem\u003eRangifer tarandus caribou\u003c/em\u003e) and moose (\u003cem\u003eAlces americanus\u003c/em\u003e) (Polfus et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Laurian et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Leblond et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Beauchesne et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Mumma et al. \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Dickie et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; St-Pierre et al. \u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e2022\u003c/span\u003e); there is also evidence of predators avoiding features heavily used by humans (Muhly et al. \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Proctor et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Whittington et al. \u003cspan citationid=\"CR93\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Changes in mortality risks, habitat quality, and connectivity associated with anthropogenic disturbance can in turn alter use and movement patterns (Apps and McLellan \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Tucker et al. \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Proctor et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), demography and population viability (Fahrig and Rytwinski \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Pinard et al. \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Apps et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; DeCesare et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Proctor et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Johnson et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e; Barrientos et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and genetics (Habrich et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Roads can also facilitate spread of invasive species (Riitters et al. \u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), and enable other development activities (Johnson et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e). Impacts on wildlife persist after roads are no longer driveable as they continue to be used by wildlife and soil compaction impedes vegetation recovery (Lacerte et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; St-Pierre et al. \u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Measures of road impacts are important for informing wildlife management decisions, and there is therefore a need to include roads in ecological forecasting models.\u003c/p\u003e \u003cp\u003eIncreasingly rapid environmental change, improvements in available cyberinfrastructure, and development of open source data, tools and methods have contributed to growing interest in ecological forecasting models (Dietze \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Bodner et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; McIntire et al. \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In particular, forest landscape models that can account for harvest, fire, and other disturbances are increasingly used to project the cumulative effects of climate change and forest management (Daniel et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Snell et al. \u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Tremblay et al. \u003cspan citationid=\"CR90\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Albrich et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Petter et al. \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Hof et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Micheletti et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Leblond et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Barros et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Stewart et al. \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Bouderbala et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), building on a rich history of spatial-stochastic models of forest growth and succession following disturbance (Sturtevant and Fortin \u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Despite the importance of roads for wildlife, very few of these projections have included road networks (Leston et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For example, in a recent collection of articles focused on \u0026ldquo;Using Landscape Simulation Models to Help Balance Conflicting Goals in Changing Forests\u0026rdquo; (Hof et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), only one includes a road projection (Leston et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In the absence of road projection methods, researchers are left to conservatively infer the amount and randomly assign the location of roaded areas (e.g. Johnson et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWhile road projections are not typically integrated with forest landscape models, methods have been developed to support harvest planning (Dean \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Clark et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Anderson and Nelson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Epstein et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; St\u0026uuml;ckelberger et al. \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Meignan et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), and some forest management planning software suites (e.g. Remsoft; Spatial Planning Systems) include road projection tools. These methods have occasionally been incorporated into forest landscape models (e.g. Leston et al. \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) but they were not designed for that purpose. Existing road projection algorithms contained in proprietary closed-source software are difficult to fully integrate into stochastic models, which often require numerous replicates (De Pellegrin Llorente et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Sturtevant and Fortin \u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Further, costs of licensing are generally prohibitive, deploying large commercial software suites designed for desktops to cloud or high performance computing resources is difficult or impossible, and closed-source software is often not sufficiently transparent for use in research. At the operational scale, road planning methods and tools (Dean \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Clark et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; e.g. Anderson and Nelson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Epstein et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; St\u0026uuml;ckelberger et al. \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Meignan et al. \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Remsoft; Spatial Planning Systems) may produce more accurate road projections but the costs of applying more complex methods across large spatial extents can be prohibitive in contexts where a coarser approximation of the road network may suffice. There is a need for open-source road network projection tools that can be integrated into the open modular platforms that are used for large-scale stochastic landscape simulations (Scheller et al. \u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Daniel et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Barros et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAutomated road projection methods often rely on more general purpose route finding algorithms that identify paths among points in a mathematical graph that minimize some measure of cost (Deo \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). General-purpose open source tools for graph analysis (e.g. Cs\u0026aacute;rdi and Nepusz \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e) include methods for building graphs from adjacency matrices, finding \u0026ldquo;least cost paths\u0026rdquo; between pairs of points, and finding a least cost network of routes (i.e. a minimum spanning tree) that connects a larger set of points (Deo \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). More domain-specific tools (e.g. Etten \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) can help make these general-purpose tools easier to use and apply in specific contexts. Promising simple graph-based options for integrating forest road network projections into large scale stochastic projections of forest change include an iterative least cost path method developed by Hardy et al (2019; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and least cost path and minimum spanning tree algorithms used to reconstruct road development history in British Columbia (Lochhead and Muhly \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The former method is available as QGIS and LANDIS-II extensions, and the latter is embedded in a larger open source modelling project that is specific to British Columbia (FAIB 2025). The Hardy et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) tool is designed for integration with other components of the LANDIS-II modelling framework (Scheller et al. \u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), but none of the existing implementations can be easily integrated into the full variety of modelling frameworks used for forecasting forest change in Canada and elsewhere (Daniel et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; e.g. Barros et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe acknowledge that using graph analysis methods to simulate roads is not new, and that other variants of iterative least cost path (or source-to-closest-target) and minimum spanning tree algorithms have been proposed and explored (Clark et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Anderson and Nelson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Picard et al. \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; St\u0026uuml;ckelberger et al. \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Anderson and Nelson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e) considered horizontal and vertical alignment, turn radius and junction angles, and switchback spacing for a relatively small 7500 ha area. However, for wide-ranging wildlife such as boreal caribou projections at much larger spatial scales are needed, and constructing and analyzing graphs for large landscapes is memory intensive and computationally demanding. We therefore focus here on assessing the adequacy of methods that minimize memory and computational requirements by setting construction costs (represented as edge weights on a simple undirected graph) from values of adjacent locations on a single user-supplied input raster. This constraint permits penalization of steep grades as suggested by Anderson and Nelson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e), but does not permit other aspects of their method. We compare a simple grade penalty approach to the method of calculating edge weights from a cost surface raster used by both Hardy et al. (2019; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and Lochhead et al. (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). We argue that if these simple algorithms can produce adequate outcomes for landscape-level stochastic ecological forecasting then the additional burden of more complex realistic algorithms is not warranted in that context.\u003c/p\u003e \u003cp\u003eTo help address the need for forest road network projection tools that can be easily integrated into large spatial scale stochastic models of forest change we implemented two alternative approaches in an open source R package: 1) iterative least cost paths and 2) a minimum spanning tree (Lochhead and Muhly \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). We also consider two different methods for determining the cost of road building: 1) a simple elevation driven cost surface and 2) a simplified version of the grade penalty approach suggested by Anderson and Nelson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). Our intention was to produce outputs comparable to Hardy et al. (2019; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) with a tool that can be used in modelling contexts outside of QGIS/LANDIS, and can serve as a more easily modifiable starting point for further refinement and development. We compared the ability of these methods and Hardy\u0026rsquo;s method (2019; 2023) to reconstruct observed forest road network development since 1990 in a mountainous region of British Columbia. The adequacy of a road network projection method depends on the purpose of the projection, so it is important to select performance metrics that reflect the purpose. Given our interest in projecting impacts on wildlife we focused on several common predictors of the impacts of roads on wildlife, including distance to roads (Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), buffered disturbance footprint (Polfus et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Ibisch et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Johnson et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Stewart et al. \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and road density (Apps and McLellan \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Apps et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Serrouya et al. \u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe expect these simple graph-based road projection methods to represent a reasonable compromise between computational cost and realism for stochastic projections on large landscapes, and we expect our method for penalizing steep grades to yield more realistic road networks in steep terrain relative to our simple cost method. We do not have a similarly strong hypothesis about whether minimum spanning tree projections are likely to be more or less realistic than iterative least cost path projections. We expect minimum spanning tree projections to be more accurate when resource roads are designed to simultaneously access multiple sites at the lowest cost. However, real road networks are built over time, and it is not necessarily true that they are designed to minimize overall cost, so it is possible that larger more costly road networks built by iterative least-cost path algorithms will be more realistic.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Area and Data\u003c/h2\u003e \u003cp\u003eWe chose the Revelstoke timber supply area (TSA 27) in British Columbia, Canada as an assessment area because it is mountainous (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), containing approximately 38% non-forested land (British Columbia Ministry of Forests \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) where road building is challenging. Thus, the area provides opportunities to observe the behaviour of road projection algorithms in steep complex terrain. Most of the road network was built to access forest harvest cutblocks, allowing projection of the road network from historic forest harvest data. The Revelstoke TSA includes portions of the Monashee and Selkirk mountain ranges, and is bisected by the Columbia River. At low to mid-elevations, forests of the interior cedar-hemlock biogeoclimatic zone (British Columbia Ministry of Forests \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) include remnants of rare old growth inland temperate rainforest (Government of Canada Parks Canada Agency \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). At higher elevations, forests are dominated by Engelmann spruce and subalpine fir (British Columbia Ministry of Forests \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The area is within the ancestral territories of the Ktunaxa Nation, Secwepemc Nation, Syilx and Sinixt (British Columbia Ministry of Forests \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Two of six southern mountain woodland caribou herds in the area have been extirpated, and the remaining herds are considered Endangered by the Committee on the Status of Endangered Wildlife in Canada and listed as threatened under the Canadian Species at Risk Act (Environment Canada \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Palm et al. \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; British Columbia Ministry of Forests \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Mountain caribou avoid roads leading to a loss of available habitat (Apps and McLellan \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Polfus et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), suffer increased mortality near roads (Apps et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), and decline in abundance as the footprint of roads and cutblocks increases (Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Roads also alter grizzly bear habitat use, home range selection, movements, population fragmentation, survival, reproductive rates, population density, trends, and conservation status (Proctor et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eRoad network data was obtained by combining the digital road atlas [Government of British Columbia (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2021a\u003c/span\u003e); a static dataset in our study area] and forest tenure road (Forests Lands Natural Resource Operations and Rural Development 2021) data sets for British Columbia. We treated the digital road atlas as a static dataset in our study area, while the forest tenure roads are a dynamic dataset, with new road segments added yearly as road permits are issued to licensees. We removed all roads from the digital road atlas that were within 50 m of forest tenure roads to avoid duplication. To obtain a starting point for projections we filtered out forest tenure roads with an award date greater than 1990, and assumed everything in the digital road atlas was built before 1990. Cutblock information was obtained from the Harvest Areas of BC (Consolidated Cutblocks) dataset (Government of British Columbia \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2021b\u003c/span\u003e), compiled from provincial Forest Cover, the RESULTS Reporting system, Forest Tenure applications, and satellite imagery using change detection processes. We used cutblocks harvested after 1990 as targets for the projections, and removed cutblocks that do not contain observed roads to reduce errors associated with open areas that have been misidentified as cutblocks.\u003c/p\u003e \u003cp\u003eFollowing the Interior Appraisal Manual of BC, we assume the base cost of construction per kilometer of road is a linear function of slope (Government of British Columbia \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2021c\u003c/span\u003e), and that road construction across major water bodies is prohibitively expensive (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Existing roads are assigned a construction cost of 0. We compare two methods for integrating these costs into the cost of connections between adjacent raster cells. In the \u0026ldquo;simple cost\u0026rdquo; method, we first calculate a slope raster from a digital elevation model, then calculate an input cost raster that depends on slope (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Online Resource 1\u0026ndash;1). The cost of the connection between two adjacent cells is simply the mean of the cost raster (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea) value at each of those cells, adjusted for euclidean distance to penalize longer diagonal road segments. In the \u0026ldquo;grade penalty\u0026rdquo; method (Online Resource 1\u0026ndash;1) we use the same construction costs (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), but calculate the cost of a connection between neighbouring cells from the elevation difference between the cells, using an elevation raster as input (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb), and penalizing edge grades as suggested by Anderson and Nelson (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). The most important difference between these two approaches is that grade is a property of the edge between two nodes in the grade penalty method, while slope is a property of each node in the simple cost method. When slope is a node property, the same slope penalty applies regardless of the alignment of the road segment. Calculating grade from the difference in elevation between pairs of nodes allows us to penalize roads that follow elevation contours differently from roads that go up or down. We examined sensitivity to pixel resolution by comparing results from 1 ha (1.7\u0026nbsp;million pixels) and 100 ha (17 thousand pixels) weight rasters. We did not consider finer resolutions because stochastic forest landscape models typically simulate dynamics at resolutions\u0026thinsp;\u0026gt;\u0026thinsp;=\u0026thinsp;1 ha.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRoad construction costs. In the \u0026ldquo;grade penalty\u0026rdquo; method, we assign grades\u0026thinsp;\u0026gt;\u0026thinsp;20% a more prohibitive cost of 65000, which is also assigned to major water bodies. These barriers are not set to NA because roads must cross barriers to reach a few isolated cutblocks in our example landscape.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCost Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eValue (\u003cspan\u003e$\u003c/span\u003e per km)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSource\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSlope (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e16178\u0026thinsp;+\u0026thinsp;504*slope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInterior appraisal manual\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMajor lakes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e65000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMajor water bodies of BC\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExisting roads\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDigital road atlas and forest tenure roads for BC\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAll data sets were accessed using the bcdata R package and filtered to the extent of the Revelstoke Timber Supply Area (Teucher et al. \u003cspan citationid=\"CR89\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). These data sets are continuously updated so we have stored the versions used in our analysis in an Open Science Framework repository (Endicott and Hughes \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024a\u003c/span\u003e). We were not granted permission to share the digital road atlas so it must be accessed from the B.C. Data Catalogue (Government of British Columbia \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2021a\u003c/span\u003e). Scripts used for data acquisition and analysis are available on GitHub (Endicott and Hughes \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024b\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eRoad Projection Methods\u003c/h3\u003e\n\u003cp\u003eOur road projection methods link multiple target locations (also known as landings within harvest cutblocks (i.e., log decks)) to an existing road network by finding minimum-cost locations for new roadways, described by Dean (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1997\u003c/span\u003e) as the multiple target access problem (MTAP). Specifically, we compared iterative least cost path (ILCP) and minimum spanning tree approaches (MST) proposed by Lochhead et al (2018; \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) for solving the MTAP to another iterative least cost path method developed by Hardy (\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; 2023) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We used the QGIS version of Hardy\u0026rsquo;s method (2019), which was easier for us to apply and does not differ substantially from the LANDIS extension (Hardy, personal communication).\u003c/p\u003e \u003cp\u003eThe ILCP approaches solve the MTAP by decomposing the problem into several smaller single target access problems (STAPs). For each landing (target location), the least cost path to the existing road network was solved using a mathematical graph and Dijkstra\u0026rsquo;s shortest path algorithm (Dijkstra \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e1959\u003c/span\u003e). There are a variety of possible methods for building graphs, choosing branch locations, and ordering target locations (Anderson and Nelson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Picard et al. \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Hardy \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Hardy et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The ILCP algorithm we used (Lochhead and Muhly \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) is parameterized with a weighted graph wherein each node corresponds to a raster pixel representing one of three components of the potential forest road network: a landing, a point on an existing road, or an intermediate point that is a plausible location for road construction. Each node is connected with up to eight neighbours, and the cost of construction along edges is either the average of the cost raster value at the source and destination nodes, or a more complex function of the difference in elevation between the source and destination nodes (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Online Resource 1\u0026ndash;1), which is then weighted by the distance to neighbouring nodes. Both the grade penalty and simple cost methods involve setting edge weights on a graph from adjacent locations on a single input grid. We used a greedy heuristic method that starts with the landing nearest to an existing road, builds a least cost path from the landing to the nearest (by Euclidean distance) point on the existing road, and updates the existing road network and weight raster with each new road segment before moving on to the next nearest landing.\u003c/p\u003e \u003cp\u003eThe method described in Hardy et al. (2019; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) is similar to our simple cost ILCP method in that it breaks the MTAP down into many STAPs and finds the least cost path for each with the ordering of road construction determined by the Euclidean distance from the existing road network. A key difference is that Hardy uses the Dijkstra algorithm to solve a \u0026ldquo;single-source many-targets shortest path problem\u0026rdquo;, which is a special case of STAPs that finds the path between the landing cell and one of multiple potential connection points on the road network. In contrast, our algorithm first finds the closest point on the road network to the selected landing point and then uses the Dijkstra algorithm to solve the least cost path between the two. However, since the cost of construction on existing roads is zero we expect that in practice these two approaches are similar and the algorithm will select the least cost location for the road to connect to the existing road network. Within cutblocks, the Hardy et al. method iterates over each cell in the harvest area and builds a road if it is not within skidding distance of an existing road, while we create landing points to serve as target locations for road construction.\u003c/p\u003e \u003cp\u003eOne of the limitations with ILCP methods is the sensitivity of the road projection to the ordering of the landings (Anderson and Nelson \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). Minimum spanning trees (MST) offer an alternative approach to link landings and evaluate road branching opportunities (Clark et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Picard et al. \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). MST algorithms do not require users to select the order in which landings are processed, but it remains necessary to select a set of points on the existing road network to include in the tree. As with the ILCP approach, the first step is to construct a graph in which each node is connected with up to eight neighbours, and set the cost of construction along edges using either the grade penalty or simple cost methods. This graph is used to estimate the least cost paths that link each of the landings and the least cost paths that link each landing to the nearest point on the existing road network. These results are used to create a second weighted graph wherein nodes are either landings or nearest points on the existing road network, and edge weights correspond to the total cost of the least cost path between pairs of nodes. The minimum spanning tree for this graph is solved using Kruskal\u0026rsquo;s algorithm (Kruskal \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e1956\u003c/span\u003e), which links all nodes with a tree that minimizes the sum of edge weights. This method differs from Clark et al. (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2000\u003c/span\u003e) who used Euclidean distances for edge weights, and thus did not incorporate spatial variation in road construction costs.\u003c/p\u003e \u003cp\u003eNote that the evolution of a road network over time can be modeled by iteratively applying any of the methods, but in this study we focus on comparing a single application of each of the methods to a network of forest roads developed since 1990.\u003c/p\u003e \u003cp\u003eTo help make these methods accessible, we implemented the ILCP and MST methods in the open source R package roads (Endicott et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) which depends on several other R packages: igraph for calculating least cost paths and minimum spanning trees (Cs\u0026aacute;rdi and Nepusz \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Cs\u0026aacute;rdi et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e); terra for manipulating spatial raster data (Hijmans et al. \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2023\u003c/span\u003e); and sf for manipulating spatial vector data (Pebesma \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Pebesma et al. \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). We experimented with using graph construction methods in the package gdistance (Etten \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), but opted to use Lochhead et al\u0026rsquo;s (2018; \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) more computationally efficient methods (Online Resource 1\u0026ndash;7) by default. The package is available on CRAN (The Comprehensive R Archive Network) and the source code, tutorials and documentation are available on GitHub (Endicott et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eComparison of Method Variants\u003c/h3\u003e\n\u003cp\u003eThe ILCP and MST methods connect landings within each cutblock to the existing road network, so an important first step is to select one or more landing locations in each block. We investigated the sensitivity of results to the pattern and density of landing locations by comparing results derived from the centroid of each cutblock to regular or randomly selected landing densities of 1 (low) and 10 (high) points per km\u003csup\u003e2\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We used default settings in the Hardy QGIS plugin except for skidding distance which was set to 200 m which gives\u0026thinsp;~\u0026thinsp;9 pts/km\u003csup\u003e2\u003c/sup\u003e. We compared two resolutions of the weight raster (1 ha and 100 ha). In total, we compared 12 road projection method variants to a cutblocks only scenario with no projected roads (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). We recorded compute time and peak RAM usage for each run, and ran some projections at 25 ha to show how speed and memory requirements vary with number of landings and number of nodes in the graph (Online Resource 1\u0026ndash;5), but did not assess other aspects of performance at that resolution.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRoad projection method variants. We examined sensitivity of minimum spanning tree (MST) and iterative least cost path (ILCP) algorithms to the method of setting edge weights (grade penalty or simple cost), the placement of landings within cutblocks (regular or random), the density of landings within cutblocks, and map resolution.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProjection method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSample type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSample density (pts/km\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eResolutions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMST grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRegular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 \u0026amp; 100 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMST grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRegular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 \u0026amp; 100 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMST grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRandom\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 \u0026amp; 100 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMST grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRandom\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 \u0026amp; 100 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMST grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCentroid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 \u0026amp; 100 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eILCP grade penalty \u0026amp; simple cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRegular\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHardy QGIS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCutblocks only\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 ha\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e\n\u003ch3\u003ePerformance Metrics\u003c/h3\u003e\n\u003cp\u003eTo assess performance we calculated four characteristics of the projected and observed road networks that are relevant for wildlife: road density, measured as density of roads in meters per hectare (m/ha); distance to nearest road, measured as distance from each raster cell to the nearest road; road disturbance footprint, measured with a 1 km buffer around roads (Ibisch et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2016\u003c/span\u003e); and forestry disturbance footprint, measured with a 500 m buffer around roads and cutblocks. The forestry disturbance footprint is a variant of the buffered anthropogenic disturbance metric that is used as an indicator of boreal caribou critical habitat (Johnson et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e) and to quantify impacts of disturbance on Southern Mountain caribou (Polfus et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The other metrics are also often used in studies of anthropogenic impacts on wildlife (Apps and McLellan \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Apps et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Serrouya et al. \u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; e.g. Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Stewart et al. \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). We also calculated road presence (i.e. whether or not roads are present in each pixel), but note that accurate projection of the exact locations of roads is not generally a requirement in stochastic ecological forecasts wherein the exact locations of future disturbances are often not known. Forestry disturbance footprint and road density were calculated using the \u003cem\u003edisturbanceMetrics\u003c/em\u003e and \u003cem\u003erasterizeLineDensity\u003c/em\u003e functions in the caribouMetrics R package (Hughes et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAspatial performance of the projections was assessed by comparing projected and observed mean values of each road network characteristic across the timber supply area and within cutblocks. As an example, distance to roads within cutblocks is the average distance from non-road pixels within cutblocks to the nearest roads. Spatial performance was assessed in two different ways. First, difference maps were created by subtracting the projected distance to the nearest road from the observed in order to highlight discrepancies between the road networks. The observed and projected road networks were then overlaid on the difference maps and regions of major discrepancies were investigated visually. Second, performance according to the three binary metrics (road presence, road disturbance footprint, and forestry disturbance footprint) was assessed assigning pixels to 5 categories: pre-existing footprint; agree roadless, meaning that both projected and observed show no footprint; agree roaded, meaning that both projected and observed show new footprint; false positive, meaning the projection shows a footprint but no footprint was observed; and false negative, meaning the projection shows no footprint but a footprint was observed. We used these categories to calculate sensitivity, precision and F-measure of each metric (Lee et al. \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Sensitivity, also known as recall, gives the proportion of the true positives that were projected, while precision gives the proportion of positives projected that are true. Sensitivity will be lower when observed roads are missed and precision will be lower when roads are projected that were not observed. F-measure attempts to balance sensitivity and precision to give an overall score.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eAccording to all aspatial performance measures, the grade penalty method for setting edge weights yields better results than our simple cost methods (compare Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). We therefore recommend using the grade penalty method, and focus on further examining variation in performance among grade penalty method variants (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The MST regular high density method yields the best projections of road density overall (projected density: 2.79 m roads/ha, observed density: 2.9 m roads/ha), while the MST random high density method had the best projections within cutblocks (projected density: 38.41 m roads/ha, observed density: 37.69 m roads/ha), and all MST variants yield comparably good projections outside cutblocks (projected density: 1.84\u0026ndash;1.85 m roads/ha, observed density: 2 m roads/ha). All MST variants yield better projections of the forestry disturbance footprint (projected: 0.3, observed: 0.3) than the Hardy or ILCP methods (0.29), and the Hardy and MST regular high density methods yield comparably accurate projections of road presence and distance to road.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eMaps (Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) help clarify exactly how the projection methods differed from one another, and from observed roads. While the Hardy et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) method and the grade penalty MST and ILCP methods projected similar mean road densities within cutblocks, they created distinctly different road patterns within cutblocks. In particular, the grade penalty method incentivized roads to follow slope contours, while the Hardy method created more road curvatures (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Although the grade penalty method improved realism by following elevation contours, all these methods created dead ends, intersections, and abrupt turns that would be avoided in the engineering of operational roads in steep terrain (Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec). All of the projected road methods also crossed valley bottoms far more often than observed roads which often run parallel to one another on each side of rivers (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea), but this discrepancy could likely be remedied by improved costs of crossing streams, rivers and other wet areas in valley bottoms. Finally, it is important to note that some discrepancies between observed and projected road networks were caused by errors in the observed data, such as cutblocks that were incorrectly classified (e.g. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). We omitted cutblocks with no associated access roads from our analysis, and note that failure to do this would distort results (Figs. S3 and S4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eProjections of the forestry disturbance footprint, the road disturbance footprint, and, to a lesser extent, distance to road were more accurate than projections of road density and road presence (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). However, all of the road projection methods underestimate road presence and forestry disturbance footprint (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), projecting fewer roads than we observe. One reason for underestimation of road density was that all these methods generate straight road segments within a pixel. As raster resolution increases, the discrepancy within pixels between real roads and straight projected segments increased, making coarse resolution projections less accurate (Fig. S6).\u003c/p\u003e \u003cp\u003eAlthough these methods underestimate road presence, they represented a substantial improvement over the \u0026ldquo;cutblocks only\u0026rdquo; scenario, which assumed no new roads were built (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The methods were not designed to project the exact spatial location of new roads (within 100 m), so it was not surprising that they did not yield accurate spatial projections of road presence (F-Measure: 0.007\u0026ndash;0.257). It was somewhat more surprising that none of the methods provided better information about the exact location of the forestry disturbance footprint than the locations of cutblocks only (F-Measure: 0.927\u0026ndash;0.956). This was likely due to many of the projected roads being within 500 m of either the existing road network or a cutblock so the false negatives created by failing to project roads are balanced by the lack of false positives. All methods other than \u0026ldquo;cutblocks only\u0026rdquo; provided comparably good projections of the road disturbance footprint (F-Measure: 0.859\u0026ndash;0.883), though higher sampling densities yielded fewer false negatives and more false positives (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAlthough the MST method performed better than the alternatives we tested according to several important wildlife habitat metrics (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), it was also more costly to compute (Fig. S7). All of these methods required construction of a mathematical graph with a node for each pixel in the weight raster where road construction was possible (i.e. not NA), and memory requirements increase with the number of nodes in the graph (Fig. S7). Compute time also increased with the number of landing targets (Fig. S7). Lochhead et al\u0026rsquo;s (2018; \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) method of graph construction was more computationally efficient than the adjacency matrix approach used in the R package gdistance (Etten \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) (Fig. S8), in part because it avoided the additional cost of constructing a transition matrix, and in part because it ignored variation in the size of raster cells that is accounted for in the gdistance \u003cem\u003egeoCorrection\u003c/em\u003e function. For our example landscape the Hardy QGIS plugin was 65 times faster than the MST and ILCP methods implemented in R (Fig. S7).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eDespite a rapidly growing body of evidence of complex and varied impacts of road networks on wildlife and other values, and increasing use of spatial stochastic models for forecasting impacts of forest disturbance across regions, very few spatial stochastic projections of forest disturbance dynamics have included road network projections, in part because tools suited for this purpose have not been available. Hardy et al. (2019; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) partially addressed this need with an extension that can be integrated with other components of the LANDIS-II modelling framework (e.g. Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Our implementations of iterative least cost path and minimum spanning tree methods (Lochhead and Muhly \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Lochhead et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) in the open source R package roads (Endicott et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) provide an alternative to Hardy\u0026rsquo;s method that improved predicted road patterns in our example landscape, and can be easily integrated into R workflows and modelling frameworks used for forecasting forest change (Daniel et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; e.g. Barros et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Although projecting roads within a stochastic ecological forecasting model was beyond the scope of this study, our analysis of performance using metrics relevant to wildlife can help guide method selection in that context.\u003c/p\u003e \u003cp\u003eWe evaluated performance of forest road network projection algorithms in a topographically complex region of British Columbia where we anticipated projection would be challenging, and implemented a simplified version of Anderson and Nelson\u0026rsquo;s (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2004\u003c/span\u003e) grade penalty method for setting graph edge weights that improves performance in this mountainous region. The simplified grade penalty method does not substantially increase memory or computational requirements, and we recommend its use in regions where slope is an important determinant of road construction cost. However, we also acknowledge that we have used a very simple cost model, and that the important determinants of road construction costs may vary among landscapes. We have therefore also allowed the possibility for users of the roads R package to specify their own method for determining edge weights from adjacent node values in a weight function. A more complex cost model could include variation among substrate types (e.g. peat, clay, gravel) and more nuanced consideration of the costs of various types of water crossings (e.g. culverts and bridges). Road construction costs and wildlife impacts also vary substantially among types of roads (e.g. long-term vs temporary haul roads). If distinguishing among types of roads is important, an effective approach might be to first project the paths of lengthy long-term haul roads using a cost model and a set of targets that are appropriate for this type of road, and then project the paths of shorter temporary haul roads or spur roads from this long-term network using a different cost model and set of landing locations.\u003c/p\u003e \u003cp\u003eAmong the method variants we considered, the minimum spanning tree (MST) regular high landing density method yielded the best projections of average road density and forestry disturbance footprint (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Aspatial summaries of road presence, distance to road, and road disturbance footprint were comparable to Hardy\u0026rsquo;s method (2023) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). However, the MST algorithm was also more computationally demanding than the other methods we tested (Fig. S7), so users will need to consider whether the benefits of improved performance outweigh the increased computational costs in their particular circumstance.\u003c/p\u003e \u003cp\u003eAspatial metrics of overall projection accuracy can be difficult to interpret. To more deeply understand variation in the behaviour of road network projection algorithms we found it helpful to distinguish between performance within cutblocks and outside of them, and to examine spatial patterns. This examination highlighted some errors that could be fixed by better accounting for the cost of water crossings in our weight raster, and other more fundamental limitations and differences among methods. Even in cases where the Hardy and grade penalty methods project similar mean road densities, the projected pattern of roads within cutblocks is quite different, in that the grade penalty roads tend to follow elevation contours, while Hardy et al\u0026rsquo;s method produces curves (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We suspect the latter behaviour is an artifact of searching for nearest destinations\u0026thinsp;\u0026gt;\u0026thinsp;200 m away on a 100 m grid.\u003c/p\u003e \u003cp\u003eAlthough the grade penalty method improves performance by aligning roads with elevation contours, all these methods create dead ends, intersections, and abrupt turns that are avoided in real road networks. We chose not to implement Anderson and Nelson\u0026rsquo;s methods (2004) for penalizing sharp curves, large junction angles and switchback spacing because these complications would increase algorithmic complexity and computational costs. Simple algorithms can recreate some aspects of observed road networks but not others, and more complex realistic algorithms are generally more costly to develop and use. Before investing in additional complexity, it is important to recognize that adequacy depends on the goal. For some purposes, projections from relatively crude cost models and simple algorithms may be adequate. For example, models of caribou movement often include distance to roads as a predictor (e.g. Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and a combined metric of buffered anthropogenic disturbance best explains variation in boreal caribou demographic rates among ranges (Johnson et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e). In our example landscape, projections of the disturbance footprints (500 m and 1km buffer) and distance to road are reasonably accurate, and for wildlife response models informed by these measures our somewhat crude road projections are likely good enough. On the other hand, these methods do not predict the exact spatial locations of roads, so outcomes and impacts should not be reported or interpreted at the level of individual pixels; for example, these projections should not guide selection of sample locations for studies of the local (pixel level) impacts of anticipated roads.\u003c/p\u003e \u003cp\u003eExamination of spatial mismatch between observed and projected roads also highlighted errors in our input data that are not unique to our study area; there is considerable variation in the consistency and accuracy of available resource road information across Canada and elsewhere (Poley et al. \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Forests Lands Natural Resource Operations and Rural Development) that users should keep in mind when interpreting discrepancies between observed and projected road networks.\u003c/p\u003e \u003cp\u003eThese road projection methods require construction of large graphs, with a node for each pixel in which road construction is possible. In our example landscape we were able to project at 100 m resolution in a reasonable amount of time (\u0026lt;\u0026thinsp;2.4 hours) using modest amounts of memory (\u0026lt;\u0026thinsp;3 GB RAM). However, our example landscape is fairly small, and memory requirements increase rapidly with landscape size (Online Resource 1\u0026ndash;5). Although it is possible to use machines with more RAM, the computational cost of adding road projections to spatial stochastic simulations of forest change relevant for wide-ranging wildlife is not trivial. For example, in a study of the effects of forest management and climate change scenarios on wildlife in a single boreal caribou range (100 stochastic realizations of 6 scenarios), we projected roads at 50 m resolution using graph with ~\u0026thinsp;21\u0026nbsp;million nodes. Each ILCP road projection required 2.7 hours and 11 GB of RAM, for a total of 1620 compute hours.\u003c/p\u003e \u003cp\u003eOptions for reducing computational requirements include restricting the area where road construction is possible (by setting some portions of the weight raster to NA), reducing landing density within cutblocks (by using the centroid method), increasing raster resolution, or using Hardy\u0026rsquo;s more efficient implementations. The first three options are available to users of the roads R package. However, those contemplating reducing computational cost by decreasing raster resolution should be cautious of the results given that all of these methods become less accurate as raster cell size increases (Figs. S5 and S6). The distance to road metric is less sensitive to changes in cell resolution, and distance metrics may also be more ecologically meaningful (Rowland et al. \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). Those considering reducing computational cost by reducing landing density within cutblocks should be clear that their projected road network will be incomplete and metrics such as total road length and road density will be underestimated. More involved possibilities for reducing computational costs include adopting faster routing algorithms (Larmet \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), implementing slow components of the R package (Fig. S8) in a faster compiled language, parallel processing, and implementing a multi-scale approach in which coarse resolution results are used to limit the area considered in finer resolution projections. Several of these options have been implemented in the cppRouting R package (Larmet \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and future versions of the roads R package could borrow and build on cppRouting functionality.\u003c/p\u003e \u003cp\u003eThere are many metrics of connectivity and fragmentation (Francis et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) we did not consider in this paper. In general, performance will vary among metrics, and it is difficult to determine what is good enough without reference to particular circumstances (i.e. objectives, species, landscapes). For example, if the outcome of interest is space use or connectivity projected from more (e.g. Whittington et al. \u003cspan citationid=\"CR93\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Labadie et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) or less (e.g. Diniz et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Hughes et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) species-specific movement models it may be helpful to test sensitivity to the difference between observed and retrospectively projected roads to confirm that projections are adequate for the intended purpose. When constructing new wildlife response models intended for projection, it may also be helpful to select metrics such as distance to road or buffered disturbance footprint that are more easily projectable (Bodner et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Rather than testing additional performance metrics, we have focused our efforts on the development of an open source R package (Endicott et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and workflows (Endicott and Hughes \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024b\u003c/span\u003e) that enable users to devise performance tests appropriate to their particular circumstances and goals. We recognize that acquiring data to retrospectively project roads can be challenging, so we also provide output maps (Endicott and Hughes \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024a\u003c/span\u003e) and an example script (analysis/scripts/4_calc_any_metric.R) to enable calculation of alternative performance metrics of interest for our Revelstoke example, along with a data preparation script (analysis/scripts/1_prepareData.R) that could be modified to extract required input data for other areas in British Columbia (Endicott and Hughes \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024b\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn this study our focus was developing tools capable of integrating forest road development into projections of the cumulative effects of disturbance on wildlife. However, we note there are other possible uses for these road network projection methods. Cost minimizing methods are not suitable for projecting development of seismic lines (e.g. Fig. S9), but for other circumstances such as mine access where the objective is to minimize the cost of reaching a target location these methods may be applicable. Lochhead et al. (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) used the least cost path algorithm to reconstruct the likely history of road network development in a case where harvest dates were known but road construction dates were not. It would also be interesting to account for persistent impacts of resource roads on vegetation, productivity and carbon sequestration (Braham et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur open source automated forest road projection tools can be integrated into spatial stochastic models of the implications of forest change for wide-ranging wildlife such as woodland caribou. In a mountainous test landscape, a high density of landings, a method for penalizing road construction on steep grades, and a minimum spanning tree algorithm all improved projection accuracy, and we recommend these options. We hope that these tools will reduce barriers to integrating resource roads into spatial stochastic models of forest dynamics, and thereby help improve projections of anticipated resource development impacts on wildlife across large areas.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eThis work was motivated by discussions with Colin Daniel (Apex Resource Management Solutions) and needs of the Western Boreal Initiative, an effort to project cumulative effects of disturbances on forests and wildlife led by Eliot McIntire (Canadian Forest Service, Natural Resources Canada) and Samuel Hach\u0026eacute; (Canadian Wildlife Service, Environment and Climate Change Canada). Questions from Steve Cumming (Universit\u0026eacute; Laval) and Lisa Venier (Canadian Forest Service, Natural Resources Canada) helped refine our focus and approach. We thank Cl\u0026eacute;ment Hardy for answering our questions about his method.\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eFunding for this work was provided by Environment and Climate Change Canada.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare there are no competing interests.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eConceptualization: JH, SE, DL, GP, KL Data curation: SE, DL Formal analysis: JH, SE, DL Funding acquisition: JH Investigation: JH, SE Methodology: JH, SE, DL, KL Project administration: JH Resources: JH Software: JH, SE, KL Supervision: JH Visualization: SE Writing \u0026ndash; original draft: JH, SE, DL Writing \u0026ndash; review \u0026amp; editing: JH, SE, DL, GP, KL\u003c/p\u003e\n\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eData generated or analyzed during this study are available in the Open Science Framework repository, https://osf.io/8vmqh/?view_only=d168967961914f22af89e44c19aec019. The code to reproduce the analysis and figures in this paper is available on GitHub at https://github.com/LandSciTech/RoadPaper. The roads package can be installed from CRAN and the code and a user guide are available on GitHub at https://github.com/LandSciTech/roads.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAlbrich K, Rammer W, Turner MG, et al (2020) Simulating forest resilience: A review. Global Ecol Biogeogr 29:2082\u0026ndash;2096. https://doi.org/10.1111/geb.13197\u003c/li\u003e\n\u003cli\u003eAnderson AE, Nelson J (2004) Projecting vector-based road networks with a shortest path algorithm. Can J For Res 34:1444\u0026ndash;1457. https://doi.org/10.1139/x04-030\u003c/li\u003e\n\u003cli\u003eApps CD, McLellan BN (2006) Factors influencing the dispersion and fragmentation of endangered mountain caribou populations. Biol Conserv 130:84\u0026ndash;97. https://doi.org/10.1016/j.biocon.2005.12.004\u003c/li\u003e\n\u003cli\u003eApps CD, Mclellan BN, Kinley TA, et al (2013) Spatial factors related to mortality and population decline of endangered mountain caribou. J Wildl Manage 77:1409\u0026ndash;1419. https://doi.org/10.1002/jwmg.601\u003c/li\u003e\n\u003cli\u003eBarrientos R, Ascens\u0026atilde;o F, D\u0026rsquo;Amico M, et al (2021) The lost road: Do transportation networks imperil wildlife population persistence? Perspect Ecol Conserv 19:411\u0026ndash;416. https://doi.org/10.1016/j.pecon.2021.07.004\u003c/li\u003e\n\u003cli\u003eBarros C, Luo Y, Chubaty AM, et al (2023) Empowering ecological modellers with a PERFICT workflow: Seamlessly linking data, parameterisation, prediction, validation and visualisation. Methods Ecol Evol 14:173\u0026ndash;188. https://doi.org/10.1111/2041-210X.14034\u003c/li\u003e\n\u003cli\u003eBeauchesne D, Jaeger JA, St-Laurent M-H (2013) Disentangling woodland caribou movements in response to clearcuts and roads across temporal scales. PLoS One 8:e77514. https://doi.org/10.1371/journal.pone.0077514\u003c/li\u003e\n\u003cli\u003eBodner K, Rauen Firkowski C, Bennett JR, et al (2021) Bridging the divide between ecological forecasts and environmental decision making. Ecosphere 12:e03869. https://doi.org/10.1002/ecs2.3869\u003c/li\u003e\n\u003cli\u003eBouderbala I, Labadie G, B\u0026eacute;land J-M, et al (2023) Long-term effect of forest harvesting on boreal species assemblages under climate change. PLOS Clim 2:e0000179. https://doi.org/10.1371/journal.pclm.0000179\u003c/li\u003e\n\u003cli\u003eBraham N, Valeria O, Imbeau L (2023) Characterization of vegetation dynamics on linear features using airborne laser scanning and ensemble learning. Forests 14:511. https://doi.org/10.3390/f14030511\u003c/li\u003e\n\u003cli\u003eBritish Columbia Ministry of Forests (2022) Revelstoke timber supply area timber supply review data package\u003c/li\u003e\n\u003cli\u003eClark MM, Meller RD, McDonald TP (2000) A three-stage heuristic for harvest scheduling with access road network development. For Sci 46:204\u0026ndash;218. https://doi.org/10.1093/forestscience/46.2.204\u003c/li\u003e\n\u003cli\u003eCs\u0026aacute;rdi G, Nepusz T (2006) The igraph software package for complex network research. InterJournal Complex Systems:1695\u003c/li\u003e\n\u003cli\u003eCs\u0026aacute;rdi G, Nepusz T, Traag V, et al (2024) Igraph: Network analysis and visualization in R\u003c/li\u003e\n\u003cli\u003eDaniel CJ, Frid L, Sleeter BM, Fortin M-J (2016) State-and-transition simulation models: A framework for forecasting landscape change. Methods Ecol Evol 7:1413\u0026ndash;1423. https://doi.org/10.1111/2041-210X.12597\u003c/li\u003e\n\u003cli\u003eDaniel CJ, Ter-Mikaelian MT, Wotton BM, et al (2017) Incorporating uncertainty into forest management planning: Timber harvest, wildfire and climate change in the boreal forest. For Ecol Manage 400:542\u0026ndash;554. https://doi.org/10.1016/j.foreco.2017.06.039\u003c/li\u003e\n\u003cli\u003eDe Pellegrin Llorente I, Hoganson HM, Carson MT, Windmuller-Campione M (2017) Recognizing spatial considerations in forest management planning. Curr For Rep 3:308\u0026ndash;316. https://doi.org/10.1007/s40725-017-0068-x\u003c/li\u003e\n\u003cli\u003eDean DJ (1997) Finding optimal routes for networks of harvest site access roads using GIS-based techniques. Can J For Res 27:11\u0026ndash;22. https://doi.org/10.1139/x96-144\u003c/li\u003e\n\u003cli\u003eDeCesare NJ, Hebblewhite M, Bradley M, et al (2014) Linking habitat selection and predation risk to spatial variation in survival. J Anim Ecol 83:343\u0026ndash;352. https://doi.org/10.1111/1365-2656.12144\u003c/li\u003e\n\u003cli\u003eDeo N (2016) Graph theory with applications to engineering and computer science. Dover Publications\u003c/li\u003e\n\u003cli\u003eDickie M, McNay SR, Sutherland GD, et al (2020) Corridors or risk? Movement along, and use of, linear features varies predictably among large mammal predator and prey species. J Anim Ecol 89:623\u0026ndash;634. https://doi.org/10.1111/1365-2656.13130\u003c/li\u003e\n\u003cli\u003eDietze M (2017) Ecological Forecasting. Princeton University Press, New Jersey\u003c/li\u003e\n\u003cli\u003eDijkstra EW (1959) A note on two problems in connexion with graphs. Numer Math (Heidelb) 1:269\u0026ndash;271. https://doi.org/10.1007/BF01386390\u003c/li\u003e\n\u003cli\u003eDiniz MF, Cushman SA, Machado RB, De Marco J\u0026uacute;nior P (2020) Landscape connectivity modeling from the perspective of animal dispersal. Landsc Ecol 35:41\u0026ndash;58. https://doi.org/10.1007/s10980-019-00935-3\u003c/li\u003e\n\u003cli\u003eEndicott S, Hughes J (2024b) RoadPaper. https://github.com/LandSciTech/RoadPaper\u003c/li\u003e\n\u003cli\u003eEndicott S, Hughes J (2024a) Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife. https://osf.io/8vmqh/?view_only=d168967961914f22af89e44c19aec019\u003c/li\u003e\n\u003cli\u003eEndicott S, Lochhead KD, Hughes J, Kirby P (2023) Roads: Road network projection\u003c/li\u003e\n\u003cli\u003eEnvironment Canada (2014) Recovery strategy for the woodland caribou, southern mountain population (Rangifer tarandus caribou) in Canada. Environment Canada, Ottawa, Ontario\u003c/li\u003e\n\u003cli\u003eEpstein R, Weintraub A, Sapunar P, et al (2006) A combinatorial heuristic approach for solving real-size machinery location and road design problems in forestry planning. Oper Res 54:1017\u0026ndash;1027. https://doi.org/10.1287/opre.1060.0331\u003c/li\u003e\n\u003cli\u003eEtten J van (2017) R package gdistance: Distances and routes on geographical grids. J Stat Softw 76:1\u0026ndash;21. https://doi.org/10.18637/jss.v076.i13\u003c/li\u003e\n\u003cli\u003eFahrig L, Rytwinski T (2009) Effects of roads on animal abundance: An empirical review and synthesis. Ecol Soc 14: https://doi.org/10.5751/ES-02815-140121\u003c/li\u003e\n\u003cli\u003eFAIB (Forest Analysis and Inventory Branch) (2025) Castor - forest and land use simulator. https://github.com/bcgov/castor. Accessed 4 Apr 2024\u003c/li\u003e\n\u003cli\u003eForests Lands Natural Resource Operations and Rural Development BCM of (2021) Forest tenure road lines - Data catalogue. https://catalogue.data.gov.bc.ca/dataset/243c94a1-f275-41dc-bc37-91d8a2b26e10. Accessed 13 Oct 2021\u003c/li\u003e\n\u003cli\u003eForests Lands Natural Resource Operations and Rural Development BCM of Resource roads spatial data update. https://www2.gov.bc.ca/gov/content/data/geographic-data-services/topographic-data/roads/resource-roads. Accessed 20 Apr 2023\u003c/li\u003e\n\u003cli\u003eFrancis RA, Millington JDA, Perry GLW, Minor ES (eds) (2021) The Routledge handbook of landscape ecology. Routledge, London\u003c/li\u003e\n\u003cli\u003eGovernment of British Columbia (2021c) Interior appraisal manual - Province of British Columbia. https://www2.gov.bc.ca/gov/content/industry/forestry/competitive-forest-industry/timber-pricing/interior-timber-pricing/interior-appraisal-manual. Accessed 23 Nov 2021\u003c/li\u003e\n\u003cli\u003eGovernment of British Columbia (2021a) Digital Road Atlas (DRA) - Master Partially-Attributed Roads - Data Catalogue. https://catalogue.data.gov.bc.ca/dataset/bb060417-b6e6-4548-b837-f9060d94743e. Accessed 13 Oct 2021\u003c/li\u003e\n\u003cli\u003eGovernment of British Columbia (2021b) Harvested areas of BC (consolidated cutblocks) - Data catalogue. https://catalogue.data.gov.bc.ca/dataset/harvested-areas-of-bc-consolidated-cutblocks-. Accessed 15 Jul 2021\u003c/li\u003e\n\u003cli\u003eGovernment of Canada Parks Canada Agency (2017) Old Growth Inland Rainforest. https://parks.canada.ca/pn-np/bc/glacier/nature/naturelle-natural/pluviale-rainforest. Accessed 19 Apr 2023\u003c/li\u003e\n\u003cli\u003eHabrich AK, Lawrence ER, Fraser DJ (2021) Varying genetic imprints of road networks and human density in North American mammal populations. Evol Appl 14:1659\u0026ndash;1672. https://doi.org/10.1111/eva.13232\u003c/li\u003e\n\u003cli\u003eHardy C (2019) Forest roads network: A QGIS plugin. https://github.com/Klemet/ForestRoadNetworkPluginForQGIS. Accessed 10 Jan 2023\u003c/li\u003e\n\u003cli\u003eHardy C, Messier C, Valeria O, Filotas E (2023) A LANDIS-II extension for simulating forest road networks. Can J For Res. https://doi.org/10.1139/cjfr-2022-0306\u003c/li\u003e\n\u003cli\u003eHijmans RJ, Bivand R, Pebesma E, Sumner MD (2023) Terra: Spatial data analysis\u003c/li\u003e\n\u003cli\u003eHill JE, DeVault TL, Belant JL (2021) A review of ecological factors promoting road use by mammals. Mamm Rev 51:214\u0026ndash;227. https://doi.org/10.1111/mam.12222\u003c/li\u003e\n\u003cli\u003eHof AR, Montoro Girona M, Fortin M-J, Tremblay JA (2021) Editorial: Using landscape simulation models to help balance conflicting goals in changing forests. Front Ecol Evol 9:818. https://doi.org/10.3389/fevo.2021.795736\u003c/li\u003e\n\u003cli\u003eHughes J, Endicott S, Simpkins C, et al (2022) caribouMetrics: Models and metrics of boreal caribou responses to forest landscapes\u003c/li\u003e\n\u003cli\u003eHughes J, Lucet V, Barrett G, et al (2023) Comparison and parallel implementation of alternative moving-window metrics of the connectivity of protected areas across large landscapes. Landsc Ecol. https://doi.org/10.1007/s10980-023-01619-9\u003c/li\u003e\n\u003cli\u003eIbisch P, Hoffmann M, Kreft S, et al (2016) A global map of roadless areas and their conservation status. Science (1979) 354:1423\u0026ndash;1427. https://doi.org/10.1126/science.aaf7166\u003c/li\u003e\n\u003cli\u003eJohnson CA, Sutherland GD, Neave E, et al (2020a) Science to inform policy: Linking population dynamics to habitat for a threatened species in Canada. J Appl Ecol 57:1314\u0026ndash;1327. https://doi.org/10.1111/1365-2664.13637\u003c/li\u003e\n\u003cli\u003eJohnson CJ, Ehlers LPW, Seip DR (2015) Witnessing extinction \u0026ndash; Cumulative impacts across landscapes and the future loss of an evolutionarily significant unit of woodland caribou in Canada. Biol Conserv 186:176\u0026ndash;186. https://doi.org/10.1016/j.biocon.2015.03.012\u003c/li\u003e\n\u003cli\u003eJohnson CJ, Venter O, Ray JC, Watson JEM (2020b) Growth-inducing infrastructure represents transformative yet ignored keystone environmental decisions. Conserv Lett 13:e12696. https://doi.org/10.1111/conl.12696\u003c/li\u003e\n\u003cli\u003eKruskal JB (1956) On the shortest spanning subtree of a graph and the traveling salesman problem. Proc Am Math Soc 7:48\u0026ndash;50. https://doi.org/10.2307/2033241\u003c/li\u003e\n\u003cli\u003eLabadie G, Hardy C, Boulanger Y, et al (2023) Global change risks a threatened species due to alteration of predator\u0026ndash;prey dynamics. Ecosphere 14:e4485. https://doi.org/10.1002/ecs2.4485\u003c/li\u003e\n\u003cli\u003eLacerte R, Leblond M, St-Laurent M-H (2021) Determinants of vegetation regeneration on forest roads following restoration treatments: Implications for boreal caribou conservation. Restor Ecol 29:e13414. https://doi.org/10.1111/rec.13414\u003c/li\u003e\n\u003cli\u003eLarmet V (2022) cppRouting: Algorithms for routing and solving the traffic assignment problem\u003c/li\u003e\n\u003cli\u003eLaurian C, Dussault C, Ouellet J-P, et al (2012) Interactions between a large herbivore and a road network. \u0026Eacute;coscience 19:69\u0026ndash;79. https://doi.org/10.2980/19-1-3461\u003c/li\u003e\n\u003cli\u003eLeblond M, Boulanger Y, Pascual Puigdevall J, St-Laurent M-H (2022) There is still time to reconcile forest management with climate-driven declines in habitat suitability for boreal caribou. Glob Ecol Conserv 39:e02294. https://doi.org/10.1016/j.gecco.2022.e02294\u003c/li\u003e\n\u003cli\u003eLeblond M, Dussault C, Ouellet J-P (2013) Avoidance of roads by large herbivores and its relation to disturbance intensity. J Zool 289:32\u0026ndash;40\u003c/li\u003e\n\u003cli\u003eLee S, Kim J, Lee G, et al (2021) Prediction of aquatic ecosystem health indices through machine learning models using the wgan-based data augmentation method. Sustainability 13:10435. https://doi.org/10.3390/su131810435\u003c/li\u003e\n\u003cli\u003eLeston L, Bayne E, Dzus E, et al (2020) Quantifying long-term bird population responses to simulated harvest plans and cumulative effects of disturbance. Front Ecol Evol 8:252. https://doi.org/10.3389/fevo.2020.00252\u003c/li\u003e\n\u003cli\u003eLochhead KD, Kleynhans EJ, Muhly TB (2022) Linking woodland caribou abundance to forestry disturbance in southern British Columbia, Canada. J Wildl Manage 86:e22149. https://doi.org/10.1002/jwmg.22149\u003c/li\u003e\n\u003cli\u003eLochhead KD, Muhly TB (2018) A demo of road simulation in CLUS. https://github.com/bcgov/clus/blob/master/reports/roads/draft-CLUS-roads.md. Accessed 2 Dec 2021\u003c/li\u003e\n\u003cli\u003eMcIntire EJB, Chubaty AM, Cumming SG, et al (2022) PERFICT: A re-imagined foundation for predictive ecology. Ecol Lett 25:1345\u0026ndash;1351. https://doi.org/10.1111/ele.13994\u003c/li\u003e\n\u003cli\u003eMeignan D, Frayret J-M, Pesant G, Blouin M (2012) A heuristic approach to automated forest road location. Can J For Res 42:2130\u0026ndash;2141. https://doi.org/10.1139/x2012-140\u003c/li\u003e\n\u003cli\u003eMicheletti T, Stewart FEC, Cumming SG, et al (2021) Assessing pathways of climate change effects in SpaDES: An application to boreal landbirds of Northwest Territories Canada. Front Ecol Evol 9:654. https://doi.org/10.3389/fevo.2021.679673\u003c/li\u003e\n\u003cli\u003eMuhly TB, Semeniuk C, Massolo A, et al (2011) Human activity helps prey win the predator-prey space race. PLoS One 6:e17050. https://doi.org/10.1371/journal.pone.0017050\u003c/li\u003e\n\u003cli\u003eMumma MA, Gillingham MP, Parker KL, et al (2018) Predation risk for boreal woodland caribou in human-modified landscapes: Evidence of wolf spatial responses independent of apparent competition. Biol Conserv 228:215\u0026ndash;223. https://doi.org/10.1016/j.biocon.2018.09.015\u003c/li\u003e\n\u003cli\u003ePalm EC, Fluker S, Nesbitt HK, et al (2020) The long road to protecting critical habitat for species at risk: The case of southern mountain woodland caribou. Conserv Sci Pract 2:e219. https://doi.org/10.1111/csp2.219\u003c/li\u003e\n\u003cli\u003ePebesma E (2018) Simple features for R: Standardized support for spatial vector data. R J 10:439\u0026ndash;446. https://doi.org/10.32614/RJ-2018-009\u003c/li\u003e\n\u003cli\u003ePebesma E, Bivand R, Racine E, et al (2022) Sf: Simple features for R\u003c/li\u003e\n\u003cli\u003ePetter G, Mairota P, Albrich K, et al (2020) How robust are future projections of forest landscape dynamics? Insights from a systematic comparison of four forest landscape models. Environ Model Softw 134:104844. https://doi.org/10.1016/j.envsoft.2020.104844\u003c/li\u003e\n\u003cli\u003ePicard N, Gazull L, Freycon V (2006) Finding optimal routes for harvesting tree access. Int J For Eng 17:35\u0026ndash;50. https://doi.org/10.1080/14942119.2006.10702534\u003c/li\u003e\n\u003cli\u003ePinard V, Dussault C, Ouellet J-P, et al (2012) Calving rate, calf survival rate, and habitat selection of forest-dwelling caribou in a highly managed landscape. J Wildl Manage 76:189\u0026ndash;199. https://doi.org/10.1002/jwmg.217\u003c/li\u003e\n\u003cli\u003ePoley LG, Schuster R, Smith W, Ray JC (2022) Identifying differences in roadless areas in Canada based on global, national, and regional road datasets. Conserv Sci Pract n/a:e12656. https://doi.org/10.1111/csp2.12656\u003c/li\u003e\n\u003cli\u003ePolfus JL, Hebblewhite M, Heinemeyer K (2011) Identifying indirect habitat loss and avoidance of human infrastructure by northern mountain woodland caribou. Biol Conserv 144:2637\u0026ndash;2646. https://doi.org/10.1016/j.biocon.2011.07.023\u003c/li\u003e\n\u003cli\u003eProctor MF, McLellan BN, Stenhouse GB, et al (2020) Effects of roads and motorized human access on grizzly bear populations in British Columbia and Alberta, Canada. Ursus 2019:16\u0026ndash;39. https://doi.org/10.2192/URSUS-D-18-00016.2\u003c/li\u003e\n\u003cli\u003eRemsoft Remsoft road optimizer. https://remsoft.com/. Accessed 12 Apr 2023\u003c/li\u003e\n\u003cli\u003eRiitters K, Potter K, Iannone BV, et al (2017) Landscape correlates of forest plant invasions: A high-resolution analysis across the eastern United States. Divers Distrib 24:274\u0026ndash;284. https://doi.org/10.1111/ddi.12680\u003c/li\u003e\n\u003cli\u003eRowland MM, Wisdom MJ, Johnson BK, Kie JG (2000) Elk distribution and modeling in relation to roads. J Wildl Manage 64:672\u0026ndash;684. https://doi.org/10.2307/3802737\u003c/li\u003e\n\u003cli\u003eScheller RM, Domingo JB, Sturtevant BR, et al (2007) Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecol Modell 201:409\u0026ndash;419. https://doi.org/10.1016/j.ecolmodel.2006.10.009\u003c/li\u003e\n\u003cli\u003eSerrouya R, Kellner A, Pavan G, et al (2017) Time vs. Distance: Alternate metrics of animal resource selection provide opposing inference. Ecosphere 8:e01730. https://doi.org/10.1002/ecs2.1730\u003c/li\u003e\n\u003cli\u003eSnell RS, Elkin C, Kotlarski S, Bugmann H (2018) Importance of climate uncertainty for projections of forest ecosystem services. Reg Environ Change\u003c/li\u003e\n\u003cli\u003eSpatial Planning Systems Patchworks. https://spatial.ca/. Accessed 10 Jan 2023\u003c/li\u003e\n\u003cli\u003eStewart FEC, Micheletti T, Cumming SG, et al (2023) Climate-informed forecasts reveal dramatic local habitat shifts and population uncertainty for northern boreal caribou. Ecol Appl 33:e2816. https://doi.org/10.1002/eap.2816\u003c/li\u003e\n\u003cli\u003eSt-Pierre F, Drapeau P, St-Laurent M-H (2021) Drivers of vegetation regrowth on logging roads in the boreal forest: Implications for restoration of woodland caribou habitat. For Ecol Manage 482:118846. https://doi.org/10.1016/j.foreco.2020.118846\u003c/li\u003e\n\u003cli\u003eSt-Pierre F, Drapeau P, St-Laurent M-H (2022) Stairway to heaven or highway to hell? How characteristics of forest roads shape their use by large mammals in the boreal forest. For Ecol Manage 510:120108. https://doi.org/10.1016/j.foreco.2022.120108\u003c/li\u003e\n\u003cli\u003eSt\u0026uuml;ckelberger J, Heinimann HR, Chung W (2007) Improved road network design models with the consideration of various link patterns and road design elements. Can J For Res 37:2281\u0026ndash;2298. https://doi.org/10.1139/X07-036\u003c/li\u003e\n\u003cli\u003eSturtevant BR, Fortin M-J (2021) Understanding and modeling forest disturbance interactions at the landscape level. Front Ecol Evol 9:636. https://doi.org/10.3389/fevo.2021.653647\u003c/li\u003e\n\u003cli\u003eTeucher AC, Albers SJ, Hazlitt SL (2021) Bcdata: An R package for searching and retrieving data from the B.C. Data Catalogue. J Open Source Softw 6:2927. https://doi.org/10.21105/joss.02927\u003c/li\u003e\n\u003cli\u003eTremblay JA, Boulanger Y, Cyr D, et al (2018) Harvesting interacts with climate change to affect future habitat quality of a focal species in eastern Canada\u0026rsquo;s boreal forest. PLoS One 13:e0191645. https://doi.org/10.1371/journal.pone.0191645\u003c/li\u003e\n\u003cli\u003eTrombulak SC, Frissell CA (2000) Review of ecological effects of roads on terrestrial and aquatic communities. Conserv Biol 14:18\u0026ndash;30. https://doi.org/10.1046/j.1523-1739.2000.99084.x\u003c/li\u003e\n\u003cli\u003eTucker MA, B\u0026ouml;hning-Gaese K, Fagan WF, et al (2018) Moving in the Anthropocene: Global reductions in terrestrial mammalian movements. Science (1979). https://doi.org/10.1126/science.aam9712\u003c/li\u003e\n\u003cli\u003eWhittington J, Hebblewhite M, Baron RW, et al (2022) Towns and trails drive carnivore movement behaviour, resource selection, and connectivity. Mov Ecol 10:17. https://doi.org/10.1186/s40462-022-00318-5\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"landscape-ecology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"land","sideBox":"Learn more about [Landscape Ecology](https://www.springer.com/journal/10980)","snPcode":"10980","submissionUrl":"https://submission.nature.com/new-submission/10980/3","title":"Landscape Ecology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"forest roads, forest dynamics, spatially explicit modeling, predictive ecology, least cost paths","lastPublishedDoi":"10.21203/rs.3.rs-6413286/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6413286/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eContext\u003c/h2\u003e \u003cp\u003eResource road networks have complex and varied impacts on wildlife and other forest values, yet spatial stochastic models forecasting impacts of forest disturbance rarely include automated road network projections. Hardy et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) partially addressed this need with a LANDIS-II extension, but there remains a need for tools that can be integrated into other modelling frameworks while identifying a pragmatic balance between achieving ecological relevancy and computational cost.\u003c/p\u003e\u003ch2\u003eObjectives\u003c/h2\u003e \u003cp\u003eOur goal is an open source resource road network projection tool that can be easily incorporated into modelling frameworks that assess the implications of forest change for wildlife. We compared the performance of several resource road network projection methods using ecologically relevant metrics.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe implemented simple iterative least cost path and minimum spanning tree methods with grade penalties in an open source R package. We assessed performance by comparing projections to observed resource road development since 1990 in a mountainous region of British Columbia.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAll resource road projection methods that we tested performed relatively well. Grade penalties improved performance, as did our minimum spanning tree method. However, the minimum spanning tree method required more computing time and memory, so users must weigh the benefits of improved performance against computational costs.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eOur resource road network simulation methods can improve projections of anticipated resource development impacts on wildlife across large areas. Our open source implementation will be useful for improving projections of the cumulative effects of natural and anthropogenic disturbances on wildlife in an era of rapid change.\u003c/p\u003e","manuscriptTitle":"Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-11 05:38:49","doi":"10.21203/rs.3.rs-6413286/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-15T10:05:47+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-12T13:03:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-03T13:17:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"147859736696024893927251249514630669127","date":"2025-07-24T00:32:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"237931693668076051112607751284281708757","date":"2025-07-21T07:28:14+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-22T13:16:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-10T01:24:07+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-10T01:22:27+00:00","index":"","fulltext":""},{"type":"submitted","content":"Landscape Ecology","date":"2025-04-09T15:20:19+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"landscape-ecology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"land","sideBox":"Learn more about [Landscape Ecology](https://www.springer.com/journal/10980)","snPcode":"10980","submissionUrl":"https://submission.nature.com/new-submission/10980/3","title":"Landscape Ecology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"390b5136-3bae-41c4-bfec-0675d0d2e46b","owner":[],"postedDate":"April 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-10T16:04:40+00:00","versionOfRecord":{"articleIdentity":"rs-6413286","link":"https://doi.org/10.1007/s10980-025-02232-8","journal":{"identity":"landscape-ecology","isVorOnly":false,"title":"Landscape Ecology"},"publishedOn":"2025-11-03 15:57:37","publishedOnDateReadable":"November 3rd, 2025"},"versionCreatedAt":"2025-04-11 05:38:49","video":"","vorDoi":"10.1007/s10980-025-02232-8","vorDoiUrl":"https://doi.org/10.1007/s10980-025-02232-8","workflowStages":[]},"version":"v1","identity":"rs-6413286","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6413286","identity":"rs-6413286","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}