Introduction

Developing forest roads is a complex and multifaceted process that plays an important role in the socioeconomic development of mountainous and forest regions (Bournaris et al. 2014; Kantartzis et al. 2021; Yildirim and Kadi 2022). Factors like location, structure, and functionality are crucial in determining the success of these roads (Abdi et al. 2009; Epstein et al. 2001; Makhdoum 2008; Xie et al. 2010). To ensure effectiveness, forest road networks must consider technical, economic, environmental, and social factors (Epstein et al. 2007; Gumus 2009; Parsakhoo et al. 2017; Tampekis et al. 2018). Connectivity with city road networks is also important for transporting forest products and enabling forest operations (Kweon 2019). Additionally, well-developed forest road networks can promote tourism by improving accessibility to tourist resources, while, on the other hand, poor maintenance or the downgrading of these roads can lead to increased costs, environmental and social problems, and the isolation of mountainous regions (Dragan and Cocean 2015; Girardin et al. 2022).

To ensure the sustainability of forests, it is crucial to develop infrastructure in natural and forest environments in a reasonable and planned manner, as forest road networks should serve multiple purposes and manage conflicting land uses, making reducing their construction costs a priority (Enache et al. 2013; Caliskan 2013). In addition, it is important to consider factors such as climate change, location, and other geographical specifications in forest road network planning (Dodson 2021). The region’s topography should be taken into account when designing the network to provide the necessary information for sustainable land-use management while minimizing costs (de Faria et al. 2022). The forest road network should also be designed to support fire prevention and accessibility in the event of destruction (Caliskan and Sevim 2022), the transportation of forest products, recreational activities, and sustainable forest management (Laschi et al. 2016; Picchio et al. 2018). Further, to ensure the sustainable development of the population of a mountainous region and of the wider region, integrated forest road network design should consider various factors such as silviculture, agriculture, livestock farming, and mountain tourism requirements. However, the construction and optimum forest opening of a forest road network can be expensive, and related investment and budget allocation policies should be integrated to manage costs (Tampekis et al. 2018; Kantartzis et al. 2021, 2023). Forest road network planning should also take into account critical factors influencing forest road construction and protect a region’s timber, soil, and landscape. A well-functioning forest road network should be designed in a thematic context that incorporates all aspects of the environment (Meng 2017; Gatti et al. 2021; Lee et al. 2021), including the atmosphere, lithosphere, hydrosphere, biosphere, and anthroposphere.

When it comes to planning forest road networks, various methods are commonly used, such as spatial analysis, geographical information system (GIS)-based techniques, hierarchical evaluation methods, and expert knowledge and experience (Gumus 2009; Abdi et al. 2009; Caliskan 2013; Laschi et al. 2016; Mo et al. 2017; Moulogianni 2022). For example, Gumus (2009) used a GIS-based approach to design a wood harvesting network in Turkey, while Abdi et al. (2009) used a GIS and multi-criteria evaluation (MCE) method to design a forest road network in Iran with the lowest construction cost. Hayati et al. (2013) proposed a three-stage methodology that combined the Delphi method and a spatial multi-criteria fuzzy map evaluation method to identify the lowest-impact road network alternative in forest road network planning in Iran. Other researchers, such as Caliskan (2013) and Mo et al. (2017), also utilized GIS technology and spatial analysis methods to support forest road network planning in Turkey and China, respectively. These approaches provide insights into different predictors of forest change categories and aim to minimize negative environmental impacts while meeting technical requirements and cost minimization, while the work of Hu et al. (2016) applied a multi-model inference method to test different sampling strategies for socioeconomic and biophysical variables that illuminated potentially different predictors of forest change categories.

Despite the challenges posed by factors such as geography and climate, forest road network planning and construction have seen significant progress thanks to new technologies, big data management, and optimization techniques (Gulci et al. 2017; Heinimann and Breschan 2012; Stuckelberger et al. 2006, 2007; Moulogianni 2022). However, current approaches do not yet utilize the network science paradigm (Barabasi 2013; Brandes et al. 2013; Tsiotas 20192020), which studies complex communication systems through graphs and statistical mechanics and has contributed to spatial analysis by describing the structure of geographical networks beyond their geometry (Tsiotas and Polyzos 2018; Ducruet et al. 2011; Tsiotas and Ducruet 2021; Tsiotas 2021). This approach has incorporated topological variables into the modeling of transportation and other infrastructure networks (Tsiotas and Polyzos 2018; Tsiotas 2021), but integrating topological analysis into current spatial analysis protocols for forest road networks requires still more research and development.

To help serve this demand, this paper introduces a new multilayer network model to analyze forest roads in the Kilkis Prefecture in Greece. Forest road planning and development involve natural resources, the environment, and economic factors (Hu et al. 2016; Laschi et al. 2016; Mo et al. 2017). The complexity of this development includes climate, land quality, accessibility, land use, and economic variables (Enache et al. 2013; Laschi et al. 2016; Dodson 2021; Caliskan and Sevim 2022; de Faria et al. 2022). While network science has been effective in transportation networks (Barthelemy 2011; Ducruet et al. 2011; Tsiotas and Ducruet 2021; Tsiotas 2021), it has not yet been applied to forest transportation. Motivated by this demand, this paper proposes integrating the total environment conceptualization with the network paradigm to develop a total environment multilayer network model for forest road transportation. The proposed multilayer model goes beyond current modeling approaches that are mainly defined in GIS-based spatial analysis or a hierarchical evaluation context, as it incorporates hierarchical modeling, accessibility, and environmental considerations (Gumus 2009; Abdi et al. 2009; Caliskan 2013; Laschi et al. 2016; Mo et al. 2017) into a multilayer network model conceived in the context of the total environment (Meng 2017; Gatti et al. 2021; Lee et al. 2021). The proposed multilayer network model is composed of one layer per total environment sphere and suggests an integrated methodological framework for dealing with the total environment’s complexity, where community detection analysis is applied to define zones of land use and environmental similarity in the context of our case study. The overall approach is expected to be particularly insightful for Greece, where forests are mainly located in inaccessible mountainous areas with intense geomorphologies, unfavorable climate conditions, and uneven vegetation, which pose challenges in the planning, mapping, and construction of forest road networks (Ministry of Agriculture 1992).

Methodology and data

The methodological framework of this study uses a multilayer network approach (Boccaletti et al. 2014; Kivelä et al. 2014), along with total environment conceptual decomposition (Meng 2017; Gatti et al. 2021; Lee et al. 2021) into the spheres of atmosphere, lithosphere, hydrosphere, biosphere, and anthroposphere, to construct a multilayer graph model integrating land and environmental uses for sustainable forest transportation development. The overall approach aims to provide a framework integrating forest road network planning and development based on a case study of a local forest road network in Greece. The methodological framework consists of eight steps divided into the four stages of modeling, analysis, policy implementation, and evaluation-conclusion making, as illustrated in Fig. 1. To construct the total environment multilayer network model, the first step (Step#1) involves collecting as many variables as possible that express land uses, environmental conditions, relevant functionality, and socioeconomic attributes of the forest road network. The available variables are then converted into edge variables by assigning their information to edge weights, which are proportional to the edge (road segment) length and the land uses of the area in the case of divisible attributes (e.g., timber volume, etc.). The available edge variables are next classified based on the total environment spheres, and a graph model per variable is constructed (where each variable’s information is available in edge weight format). In the second step (Step#2), the numerical scale is removed from the edge weights (per graph model) by normalizing them to the interval [0, 1] to facilitate the merging performed next. In the third step (Step#3), the available graphs within each sphere are merged into a single-layer graph model by averaging the group’s edge weights. This approach results in one graph model per sphere (atmosphere, lithosphere, hydrosphere, biosphere, and anthroposphere) composing the total environment multilayer network. In the fourth step (Step#4), the weights across the five sphere layers (each including normalized average weights) are summed to construct an aggregate (single-layer) total environment graph model to which a community detection analysis is applied (Step#5), dividing the total environment network into consistent communities (that are dense within and sparse between) in terms of their land use and environmental functionality. In the next two steps, we formulate land use and environmental profiles (Step#6) of the communities to conceive a mix of best practices and proper planning and development policies (Step#7) that are specialized to each community. In the final step (Step#8), we discuss the overall approach and formulate conclusions.

Fig. 1
figure 1

Source: own elaboration

The total environment methodological framework developed for the analysis of the forest road network of the Koupa area (Kilkis Prefecture, Region of Central Macedonia, Greece).

The forest road network of Koupa, Kilkis, Greece

The Koupa Forest is located in northern Greece; specifically, in the Kilkis district of the Region of Central Macedonia (Fig. 2). It lies southwest of the Municipal Department of Skra and is 6 km away by road. The forest contains the settlement of Koupa. The certified logging tables indicate that the Koupa Forest was privately owned and managed between 1913 and 1915. It was proposed as a classified forest in 1922 and, in 1962, land titles were applied for and a forest map was drawn up. In 1989, the forest area was determined and found to exceed 28,000 acres (2800 ha). There are boundary disputes with the adjacent public forests of Grippa-Fanou and Sechovo. The forest occupies the northeastern slopes of Mount Paiko and has varying directions, including north (N), east (E), and south (S), and different micro-locations with directions ranging from N, northeast (NE), northwest (NW), E, southeast (SE), and S. The forest’s elevation range is between 1540 and 380 m, and its average slope is 19.4%. The forest area has numerous geological folds, with deep trenches in all directions, mainly from west to east, and these alternate on both sides of the cliffs, which have gentle to strong gradients. Many of these gullies form the boundaries of the forest fragments and stands or of the forest itself with the rest of the public forest. The area enclosed within the boundaries has been measured using GIS Map Info, and totals 27,521.9 acres, with 82.9% being forested, 2.5% partially forested, 13.2% fields, and 1.5% bare land.

Fig. 2
figure 2

Source: Own elaboration from data extracted from the Forest Directorate of Central Macedonia Region—FODICER (2008)

a, b Location maps and c and forestry map of the Koupa Forest Complex

The Koupa Forest is primarily composed of beech and oak trees, which yield technical coarse timber as well as fine timber, such as round wood, fine roundwood, fine-dimensioned chips, and firewood from all forest species. Management studies indicate that in the 20 years between 1963 and 1980, 159.42 m3 (17%) of beech wood, 39.66 m3 (4%) of oak wood, 175.37 m3 (19%) of beech chips, 161.50 m3 (17%) of beech firewood, and 402.05 m3 (43%) of oak firewood were produced annually. The main consumer centers for usable timber are Thessaloniki and other cities in Macedonia, followed by other urban centers in the country. The primary consumer center for firewood is the province of Paionia, with some demand from Thessaloniki. The forest farm of Koupa has a varied geomorphological relief, with altitudes ranging from 380 to 1540 m. Many streams and ridges are present within this wide range of altitudes, forming two large slopes facing each other. One slope faces north, while the other faces south. On each slope, there are basins with ridges and streams of a lesser extent. The forest contains many springs, which are favored by the nature of the rocks (aesstolite—slate). The dense forest cover regulates the quantity and speed of run-off water and contributes to the creation of many contact springs. Although there are no meteorological stations within the Koupa Forest, the weather station at the city of Giannitsa provides information on the climatic conditions of the area.

Graph modeling and data

This section describes the modeling of the Koupa Forest Road Network (KFRN) in Kilkis, Greece, using a total environment multilayer graph approach. The approach consists of five layers (atmosphere, lithosphere, hydrosphere, biosphere, and anthroposphere) without interlayer connections. Each layer is constructed using land use and geographical variables. In particular, the KFRN is modeled as a total environment multilayer graph \(\mathcal{M}(\mathcal{G},\mathcal{C}\, = \,\emptyset )\) (Boccaletti et al. 2014; Kivela et al. 2014) composed of five layers \(\mathcal{G}\, = \,\left\{ {G_{p} } \right\}\, = \,\left\{ {V_{p} ,E_{p} } \right\}\) , where V is the node set, E is the edge (links) set, and p = A is the atmosphere, p = L is the lithosphere, p = H is the hydrosphere, p = B is the biosphere, and p = AN is the anthroposphere layer. Between layers, no interlayer connections \(\mathcal{C}\, = \,\{ {\rm E}_{ij} \subseteq V_{i} \times V_{j}\) | with p = A, L, H, B, AN, and with i,j as node indicators} = Ø exist. The multilayer \(\mathcal{M}\) is a multiplex network; namely, all its layers have the same node (Vp = V) and edge (Ep = E) sets Gp = (V, E), composed of n = 245 vertices (nodes) and m = 260 links (edges), respectively. Nodes i \(\in\) V express points of specific interest (either intersections or turns) in the road network’s path, and edges ij \(\in\) E express the road segments defined by two successive nodes. The resulting five total environment layers {atmosphere: GA, lithosphere: GL, hydrosphere: GH, biosphere: GB, and athroposphere: GAN} were constructed based on the set of available land-use and geographical variables, which are briefly described as follows:

The atmosphere layer (GA) includes three variables related to weather conditions and meteorological data. The first variable measures the aeolic or wind potential (A1), which is the average annual wind speed per road segment retrieved from the closest available meteorological station and is measured in meters per second (m/s). Data for this variable were retrieved from the open geodatabase of the Greek Government (GEODATA 2022). The second variable measures the average annual rain precipitation (A2) per road segment retrieved from the closest meteorological station in millimeters of mercury (mmHg). Finally, the third variable measures the average annual temperature (A3) per road segment retrieved from the closest meteorological station in degrees Celsius (°C). Data for variables A2 and A3 were extracted from the Hellenic National Meteorology Service (HNMS 2022). The atmosphere layer variables provide information about the local weather conditions in the KFRN area and can help to analyze the impact of the weather on the road network’s performance and accessibility.

The lithosphere layer (GL) includes variables providing information on the characteristics of the road network and the physical terrain surrounding it. In particular, the first variable, road length (L1) refers to the actual curved length of the road segments in kilometers. The second variable, road slope (L2), is defined as the height difference between the starting and end points of a road segment in degrees. The third variable, site quality (L3), is a categorical variable that describes the potential for wood production in an area and for certain silvicultural types, and is quantized based on the frequency of the categories. Next, the fourth variable, geological structure (L4), is also a categorical variable. It includes information on the geological structure (schist, limestone, and their mixes) and is also quantized based on the frequency of the categories. Finally, the fifth variable, road straight length (L5), is defined by the straight (Euclidean distance) length between the starting and end points of a road segment in kilometers. Data for all variables included in this sphere were retrieved from the Forest Directorate of Central Macedonia Region (FODICER 2008) in Greece.

The hydrosphere layer (GH) includes two variables. The first one describes the distance from nearest spring (H1), is conceived as a water table proxy for the road segments, and measures the distance (km) of a road segment from its nearest spring. The second variable is spring dynamic (H2), which measures the height difference (km) between a road segment’s centroid and the level of its nearest spring. This variable is also conceived as a water table proxy for the road segments. Both variables were retrieved from the FODICER (2008) in Greece.

The biosphere layer (GH) includes four variables related to the forest ecosystem adjacent to the road segments. The first variable, called logging volume (B1), represents the overall logging volume of the forest area per unit area. The second and the third variables are, respectively, the chestnut (B2) and cherry tree (B3) logging volumes of the adjacent forest area per unit area (measured in m3/m2). The fourth variable, called vegetation map (B4), is a categorical variable that represents the volume of different silvicultural types (forest species, rangelands, agricultural cultivations) adjacent to a road segment forest area per unit area. The data source for the variables B1, B2, and B3 is the FODICER (2008) in Greece, while the data source for B4 is the Greek Payment Authority of Common Agricultural Policy (OPEKEPE 2022).

Finally, the anthroposphere layer (GAN) includes six variables related to transportation and economic activities and their impact on the road network. The first variable is the road type (AN1), which is an ordinal variable expressing the quality of the road surface. The second variable is the road width (AN2), which is conceived as a measure of resistance to traffic and expresses the width of the available road segments. The third variable, called distance from nearest settlement (AN3), is conceived as a cost to the place of demand and expresses the distance (km) of the road segment’s centroid from the nearest settlement. The fourth variable, named distance from nearest production unit (AN4), is conceived as a spatial aspect of production cost and expresses the distances of the road segments from their nearest production unit. The fifth variable is the coefficient of variation (CV) of azimuths (AN5), which is conceived as an aspect of a road segment’s spatial impedance and expresses the azimuths' homogeneity of the linear compartments composing a road segment. The sixth variable is the direction change rate (AN6), which is the average of the azimuth change rates computed for the linear compartments composing a road segment, and captures an aspect of a road segment’s spatial impedance. Data for all variables included in this sphere were retrieved from the FODICER (2008) in Greece.

Edge weights in each collective sphere's layer (p = A, L, H, B, and AN) are computed by averaging the edge weights of the variables included in a sphere's family. For variables with a negative analogy (where lower scores are preferable to higher ones), we inverse edge weights to positively contribute to averages. Due to the normalization and averaging of the weights, the asymmetry in the number of variables composing each sphere does not cause any computational bias and thus does not suggest any concern, and provides the model with flexibility in relation to data availability. Within this context, the edge weights wij of the aggregate (overlaid layer) total environment graph model \(G_{{{\text{TEN}}}} = \cup_{p = A,L,H,B,AN} G_{p}\) are computed by the summation of the sphere layer weights.

Community detection based on modularity optimization

After the network analysis, we apply community detection analysis to the KFRN to detect concrete network parts. This part of the analysis uses the modularity optimization algorithm proposed by Blondel et al. (2008). In conceptual terms, modularity is an objective function defined as the sum of the actual minus the expected number of edges falling within a network community. It assigns zeros to concrete networks (e.g., a star graph has zero modularity) and values approximating 1 (→ 1) to highly divisible networks (Fortunato 2010), whereas an entirely disconnected network has infinite modularity. The modularity optimization algorithm (Blondel et al. 2008) is a greedy algorithm that requires the maximization of the intra-community connections (those within communities) and the consequent minimization of the inter-community connections (those between communities). The algorithm is applied in two stages (Blondel et al. 2008; Fortunato 2010). In the first stage, each node is assigned to a separate community and nodes are swept and placed step by step in collective communities if such a merge increases the gain in the weighted modularity function (Qw) of the initial graph. In the second stage, the resulting communities are considered as nodes, and the procedure is repeated until convergence of the modularity function is reached. The algorithm creates communities that are densely connected within them and sparsely connected to the other communities (Blondel et al. 2008; Fortunato 2010), yielding an outcome of network modularity along with the modularity classification scores (the score of a node is the community label that it belongs to). Due to the heuristic setting of the community detection algorithm (Fortunato 2010), we run the analysis several times (30 iterations), and we keep the mode (the most frequent value) of the modularity score and its corresponding modularity classification (community membership). In terms of interpretation, nodes belonging to the same community are more relevant in terms of connectivity than those of other communities as nodes in the same community are more connected to each other. Further, in the empirical context of the KFRN, communities that are detected by this procedure are expected to define condensed zones in terms of their functionality (as separately approximated by the component variables shown in Table 1), and conceiving the functional profiles of each zone is a further task that would be expected to contribute to the spatial planning of forest resources.

Table 1 Statistics and network measures of the Koupa Forest Road Network’s communities

Empirical analysis and policy implementation

Following the community detection analysis, we use empirical methods based on tabulation, descriptive, and inferential statistics (Norusis 2008; Walpole et al. 2012) to configure the functional profiles of the previously detected communities. This procedure allows communities to be defined by their total environment attributes (functional characteristics) and thus the geographical space of the forest road network to be organized according to its functional consistency. It also enables us to identify the specific challenges and opportunities that each community presents, allowing to develop tailored policies and best practices for each one. By analyzing the functional profiles of each community, we can identify the most efficient and effective ways to address issues such as road maintenance, resource extraction, and environmental conservation. This approach ensures that the policies and practices implemented are appropriate for the specific context and characteristics of each community, resulting in a more sustainable and environmentally responsible forest road network. The overall approach aims at conceiving the best practices and proper planning and development policies for each community (based on their specific profile), and thus configuring a mix of policies and practices that is customized to the specific total environment attributes of the forest network’s area.

Results and discussion

Community detection

In the community detection analysis, we apply the modularity optimization algorithm of Blondel et al. (2008) to divide the total environment network (G) into communities of functional consistency. The results of the analysis are shown in Fig. 3, where it can be observed that the aggregate network G(n = 245, m = 260) of the KFRN is divided into 21 communities (from #0 to #20). An extra structural grouping is shown in the layouts of Fig. 3: (i) “dipoles,” aggregating the “one edge” (i.e., a dipole) communities; (ii) “joints,” including nodes from different communities that are connected by an edge; and (iii) “triplets,” aggregating communities including three nodes. As far as the spatial distribution of the communities is concerned, we observe that it appears geographically configured, interpreting that membership within each community depends on spatial proximity to the community (thus defining communities as convex geographical areas including neighbor nodes). This outcome is in line with the existing empirical knowledge about spatial networks (Barthelemy 2011), as community detection in networks embedded in geographical space is usually geographically defined and driven by forces of proximity. Further interpretation and research on communities that break this spatial consistency (including non-neighboring members) are needed. Based on this perspective, none of the typical communities present exceptional cases to geographically study, whereas the thematic groups do. In particular, the spatial distribution of the joints group distinguishes two major concentrations of joints in the northern and southern parts of the KFRN, illustrating regions of more diversified functionality in the Koupa Forest Complex. Next, the spatial distribution of the triplets points out the locations of two road parts of secondary functionality in the Koupa Forest Complex: (i) a road within a quarry located in the northern part of the Koupa Forest Complex and (ii) road part which is located in the southern part and is disconnected to the KFRN but connected with a neighboring road network (although it belongs to the defined area of Koupa Forest Complex). Finally, the spatial distribution of the dipoles group distinguishes points of hypsometric importance and, especially for the southern component, a place operating as a bottleneck in the KFRN.

Fig. 3
figure 3

Source: own elaboration

The 20 communities of the Koupa Forest Road Network’s total environment network (dipoles, joints, and triplets suggest a further thematic grouping).

Based on their definition (Blondel et al. 2008; Fortunato 2010), the produced communities are dense (in terms of weighted connectivity, i.e., strength) within and sparse between. This conceptualization allows the generated network communities to be conceived as concise functional areas in the total environment network of the Koupa Forest Complex, and prompts the detection of their specialized attribute profile. Within this context, to configure the specialized functional (total environment) profile of each community, we construct Table 1, which showcases the ordinal classification of each community’s scores by the total environment attributes (as described in “Graph modeling and data”). To quantify the information in Table 1, we further construct the stacked bar charts shown in Fig. 4, which are computed after normalizing the ordinal scores of Table 1. As can be observed, the most functional parts of the Koupa Forest Complex within a total environment context are communities #0 (located in the southeast; vegetated with Fagus), #7 (located in the west; characterized by high forest vegetation), #10 (located in the west; geomorphologically characterized by fields), #13 (located in the northeast; has high vegetation diversity), #14 (located in the north; has high vegetation diversity and high accessibility to the Koupa settlement), and #21 (the joints group). Further, breaking down the total environment stacked bar chart into the five stacked sphere bar charts allows the communities (areas in the Koupa Forest Complex) that either lag or outperform per case to be distingushed, providing structured and organized information for sustainable forest transportation development.

Fig. 4
figure 4

Source: own elaboration

Stacked bar charts showing the community scores (normalized per sphere) for each variable (community #21 corresponds to the joints group; see other communities and their labels in Fig. 3).

To better organize the information provided in Table 1 and Fig. 4, we construct Table 2, which includes a summary of the attributes and the semiology of the communities resulting from community detection analysis. The table contains a list of communities with their corresponding descriptions. Each community is characterized by a set of high and low values for certain features or factors. The high values refer to the factors that have a strong influence on the community, while the low values are less influential. Community #0 has a moderate overall score, with high scores in the atmosphere and biosphere layers, moderate scores in the anthroposphere and hydrosphere layers, and a low score in the lithosphere layer. Next, although community #2 may have the potential for wind energy generation, the infrastructure and site conditions for other types of economic activity, such as agriculture or forestry, may be limited. Next, community #3 appears to be strongly rooted in the local environment, with significant potential for wood production, but it may face challenges in terms of economic connectivity and infrastructure. Next, community #4 appears to have better access to water resources but may face challenges in other areas such as topography, site quality, and distances to production units and settlements. Community #5 has an overall moderate-to-high performance score, with relatively good water availability for vegetation growth but presenting a challenge for wind energy generation (which could impact agriculture and crop production) and a lower potential for wood production and silvicultural activities in the area. The performance of community #6 is relatively low compared to other communities, with some potential for improvement in terms of road infrastructure and access to resources.

Table 2 Summary of the attributes and semiology of the communities resulting from the multilevel network analysis of the Koupa Forest Road Network

Community #7 has a moderately high performance profile, with relatively high scores in the atmosphere, lithosphere, and anthroposphere layers, but a low volume of logging and silvicultural types, moderate distances from the nearest spring, and moderate spring dynamics. Community #8 shows a moderate-to-high profile, but a relatively low performance in terms of infrastructure and forest management, implying that it may benefit from improvements in road infrastructure and forest management practices to increase its economic potential and improve the well-being of its residents. Next, community #9 is characterized by a relatively high score in the atmosphere layer, particularly in aeolic potential (indicating that the road segments in this community experience higher wind speeds compared to other communities) and a high score in the anthroposphere layer (possibly indicating a more rural and less developed area), and road segments that are less influenced by water and vegetation factors. Next, Community #10 appears to have a relatively good potential for wind energy, but its road infrastructure and location are not ideal for timber production and it may have limited access to water resources, although it has good road quality. Community #11 has a mostly moderate-to-high score, particularly in the atmosphere and anthroposphere layers. Its road segments may not be as straight and easy to traverse, and it seems to have good potential for economic activity, especially in agriculture and forestry, due to its favorable environmental and anthropogenic factors. Community #12 has a profile indicating some favorable conditions for forestry development, with moderate-to-high natural resources and moderate infrastructure. Next, community #13 has average to high atmospheric conditions, its road segments have a low slope, and it has also low-to-moderate levels of anthroposphere conditions. Community #14 appears to be a moderate to accessible community, while community #17 has an overall moderate performance, with a balanced distribution of high, medium, and low scores across the different layers. It could benefit from improving the wind potential and water availability as well as enhancing the forest diversity and spatial impedance. The dipole (a term retrieved from graph theory) group of communities (#15, #16, and #18), along with the triplet group of communities (#1 and #19), show relatively low performance compared to the other communities, with a mix of medium and low scores prevailing, suggesting that they may face challenges in terms of connectivity, transportation, and access to resources. Next, community #20 appears to have moderate-to-high potential for wind energy generation, with relatively challenging terrain and access to water resources, but with moderate levels of logging activity and relatively accessible roads. Finally, the joints community (#21) is characterized by high values for network size, average temperature, site quality, geological structure, cherry tree volume, vegetation map, and road type, but low values for road straight length and network clustering.

Empirical analysis and policy implementation

In the final part of the analysis, we use the information provided in Table 1, Fig. 4, and Table 2 to get insights into the semiology of the communities resulting from community detection analysis. The overall approach can provide a useful tool for planners and policymakers as it incorporates all critical variables that should be taken into account in forest management and planning. Each cell of the table provides customized information based on the specialized total environment profile that each community detected by the previous analysis has. In general, the network of forest roads built in an area together with other wood transport facilities is closely related to the intensity of forestry in that area. The longer and more intensive the forestry operation, the higher the quantity and better the quality of wood products, the more favorable the soil conditions, the lower the construction and maintenance costs of the forest roads, and the higher the costs of transporting the wood, the denser and more technically sound the network of forest roads must be. On the contrary, less-dense forest road networks are constructed when the soil conditions are less favorable, the costs of construction and maintenance of forest roads are higher, the cost per unit volume of wood displacement is cheaper, the production capacity in terms of both quantity and quality is lower, and the intensity of forest exploitation is extensive. The quantity of wood to be moved is a decisive element for the opening of the forest, as (i) the quantity of wood decisively determines and substantially influences the size of the road network because means and methods of shifting and transporting the wood that are adapted to the various geomorphological conditions, accommodate the various types of wood transported, and do not damage the remaining stand are needed; and (ii) it influences the capital invested for the construction of a forest road and the costs of road construction. The spatial allocation of timber capital allows the most economical and least damaging displacement of wood in the forest. The solution to this problem influences the mode of displacement (modern means of displacement) and the road density of the road network (denser). Further, timber transportation involves costs that affect the choice of the drilling used. The road network chosen should satisfy, from both economic and technical points of view, both forestry and wood transport, based on the costs of transport and the construction of forest roads, as well as the transport possibilities of the network, e.g., the transport of valuable species and long lengths of wood, minimizing damage to the stand and to the wood transported caused by dragging it along the ground, increasing the duration of wood harvesting, etc. The costs of moving wood are also reduced by reducing the number of intermediate storage facilities from the logging site to the place in which the wood is finally consumed or processed. This requires, on the one hand, a dense network of forest roads and the use of transport vehicles with a wood-loading mechanism, and, on the other hand, the provision of small wood collection points along the forest roads upstream or downstream from where the logs will be collected by mechanical loading. In this way, the timber relocation phase is economically affected, since the costs of the transport route and the costs of stacking and loading the timber are reduced. Within this context, based on the previous analysis, we construct Table 3, which includes some good policies and practices that are proposed as measures to recover, upgrade, and manage the specialized features describing the profile of each community.

Table 3 List of suggested policies per sphere attribute

As can be observed by combining Tables 2 and 3, suggested policies can be tailored to each community. In some cases they should be preservative, ensuring that their road network is well maintained and provides access to water sources and settlements, while also considering the environment and the economic benefits of logging and vegetation growth and focusing on environmental conservation. On the other hand, they can be more decisive in places, with the aims being to increase forestation and logging activities, improve cultivation practices to enhance the overall economic and ecological sustainability, promote sustainable forest management practices that take into account the needs of both the community and the environment (including protecting biodiversity and managing forest resources in a way that supports the community’s long-term economic and social goals), as well as to apply forest road infrastructure works (to improve road network clustering and other topological features) and upgrade road types (to improve road safety and reduce transportation costs).

Conclusions

This paper has built on the total environment conceptualization and used the network paradigm to develop a multilayer network model that serves the demand for integration for sustainable forest transportation development. The multilayer network model was composed of five layers, each representing a sphere (atmosphere, lithosphere, hydrosphere, biosphere, and anthroposphere) in the total environment context, along with an aggregated (overlaid) one including overall information on the sum of the spheres. In methodological terms, the total environment (multilayer) network was proposed as an approach to enclose information in a single model (in the aggregated form of a collection of graph layers) incorporating attributes, land uses, and conditions of a forest road network’s total environment (which are currently being studied separately in the relevant literature) to promote forest road transportation management and planning through the quantitative study of one model. To do so, the methodological framework was composed of several steps, including (i) the collection of relevant variables (expressing land uses and similar attributes of the forest road network); (ii) the conversion of the available variables into edge variables; (iii) the classification of edge variables into total environment sphere groups; (iv) the construction of the multilayer graph model (where a layer corresponded to one variable); (v) the normalization of edge weights per layer; (vi) the averaging of edge weights per sphere group; and (vii) the summation of the weights across the five spheres to construct the aggregate layer of the total environment graph model.

Within this context, we applied community detection analysis, which resulted in the emergence of 21 communities in the Koupa Forest Road Network (KFRN). Areas (sub-networks) of dense interior weighted connectivity were defined, thus dividing the Koupa Forest Complex into areas of functional consistency. Further elaboration of these network communities by their total environment attributes resulted in community functional (total environment) profiles revealing some of the most functional parts of the KFRN, which were located in the southeast (with a high amount of Fagus vegetation), western (with a high amount of forest vegetation, high geomorphological diversity, and geomorphologically characterized by fields), northeastern (with a high vegetation diversity), and northern (with a high vegetation diversity and high accessibility to the Koupa settlement). The analysis also revealed the importance of places that bridge different communities as points of interest for retaining “market integration” in the Koupa Forest Complex. The community detection analysis contributed to the decomposition of the forest network into sub-areas (communities) and to the achievement of the quantitatively driven spatial organization of a forest complex, facilitating the collection of best practices and proper planning and development policies for each community with respect to achieving sustainable forest management. Within this context, we constructed a table collecting a list of measures that fitted each total environment variable and were combined according to each community profile to configure specialized policies for sustainable forest transportation development.

Overall, this paper has introduced a novel quantitative methodological framework that builds on the conceptual framework of the total environment and can guarantee sustainability in road forest management and planning. The proposed method enjoys broad applicability (at any geographical scale, place, variable set, and level of resolution) and promotes the total environment conceptualization by incorporating all five spheres into a common quantitative context. Finally, this paper has provided insights into the case study of the KFRN, and has provided a list of good practices and policy measures according to the specialized profile of the communities resulting from the analysis.