Step 1 Preprocessing (Optional):

The input of an edge list is required in most graph theoretic approaches. Fig. 2 represents a typical directed weighted graph. In the edge list, the first two columns utilize the nodes IDs to represent edges and the last column corresponds to the value of the edges. This value actually represents the connectivity value between the corresponding nodes and may represent for instance values resulting from particle drift models, migration probabilities or other. This preprocessing step can be conducted using the function preprocess_graphs() and serves in reading and transforming the initial input data into an edge list. Further details on the initial data formats are given below. In case that an edge list is available, this step can be skipped.

Fig. 2: Weighted directed graph (left): Circles represent nodes and arrows represent weighted edges. The tabular representation of the graph represents the edge list (right).

Function preprocess_graphs() takes as input a list of .txt/.csv objects. Each object represents the connections between a node and all the other nodes. For the model to read the data, it is necessary to have all the .txt/.csv objects in one folder. There are two ways to incorporate connectivity data, based on their linkage to features:

• Case 1: the connectivity data correspond to specific biodiversity features. If a biodiversity feature has its own connectivity dataset then the file including the edge lists needs to have the same name as the corresponding feature. For example, consider having 5 species (f1, f2, f3, f4, f5) and 5 connectivity datasets. Then the connectivity datasets need to be in separate folders named: f1,f2,f3,f4,f5 and the algorithm will understand that they correspond to the species.

• Case 2: the connectivity dataset represents a spatial pattern that is not directly connected with a specific biodiversity feature. Then the connectivity data need to be included in a separate folder named in a different way than the species. For example consider having 5 species (f1,f2,f3,f4,f5) and 1 connectivity dataset. This dataset can be included in a separate folder (e.g. “Langragian_con”).

In our example we use connectivity values that are not directly connected with a specific species, therefore we illustrate Case 2. Fig. 3 represents the structure used in the tutorial. The data need to be stored in this way in order for the algorithm to read them properly.

Fig. 3: Connectivity folder

A typical Lagrangian output is a set of files representing the likelihood of a point moving from an origin (source) to a destination (target). This can be represented using a list of .txt files (as many as the origin points) including information for the destination probability. The .txt files need to be named in an increasing order. The name of the files need to correspond to the numbering of the points, in order for the algorithm to match the coordinates with the points (Fig. 4).

Fig. 4: The 001.txt file contains the following information: Each row represents the probability of movement between point 001 and any other destination points.

As long as the data are set in this way, the preprocessing algorithm can run and transform the format of the inputs to an edge list. If an edge list dataset is available, this preprocessing step can be skipped.

# Import packages
library(priorCON)
library(tmap)
library(terra)

# Read connectivity files from folder and combine them
  combined_edge_list <- preprocess_graphs(
    system.file("external",package="priorCON"),
    header = FALSE, sep =";"
   )

Step 2 Connectivity Metrics Estimation:

Function get_metrics() is used to calculate graph metrics values. The edge lists created from the previous step, or inserted directly from the user are used in this step to create graphs. The directed graphs are transformed to undirected. The function is based on the igraph package (Csárdi and Nepusz 2006; Csárdi et al. 2024) which is used to create clusters using Louvain and Walktrap and calculate the following metrics: Eigenvector Centrality, Betweenness Centrality and Degree. S-core is calculated using the package brainGraph (Watson 2024). The user can choose between these options to create the respective outputs. 's_core', 'louvain', 'walktrap', 'eigen', 'betw' or 'deg'.

Detailed information on the theory and equations of the used graph metrics are provided in Nagkoulis et al.(2024; subm Methods in Ecology and Evolution).

# Set seed for reproducibility
set.seed(42)

# Detect graph communities using the s-core algorithm
pre_graphs <- get_metrics(combined_edge_list, which_community = "s_core")

Step 3 Prioritization:

Two alternative functions can be used for the prioritization step: i) the connectivity_scenario() function, which includes connectivity into the optimization procedure and ii) the basic_scenario() function, which does not include connectivity. The two functions can be run separately. Users may use both functions, if they wish to compare the results obtained from the two scenarios, i.e. with and without connectivity. Alternatively, only the connectivity_scenario() function can be run to obtain the prioritization outputs of the connectivity scenario. Both functions are based on the priorititizr package (Hanson et al. 2024). The connectivity metrics are first transformed to rasters, following an approach similar to Marxan Connect (Daigle et al. 2020). Then priorititizr maximizes the utility obtained from protecting both features and connections. Mathematically, the target of the optimization is to maximize U under the budget (B) limitations (eq. 2 and 3 of Nagkoulis et al.(2024)):

\[ U = \sum_{\boldsymbol{PU_i} \, \in \,SA} -\lambda c_i + \sum_{\boldsymbol{PU_i} \, \in \,SA} \sum_{j=1}^J \mu_j f_j + \sum_{\boldsymbol{PU_i} \, \in \,SA} \sum_{j=1}^J \mu_j M_j, \]

\[ \sum_{\boldsymbol{PU_i} \, \in \,SA} c_i \leq B \]

A set of planning units \(\boldsymbol{A}=(\boldsymbol{PU}_1, \boldsymbol{PU}_2, ..., \boldsymbol{PU_i}, ..., \boldsymbol{PU_i})\) is considered to be protected when \(\boldsymbol{PU_i} \, \in \, SA\). The protection of any \(\boldsymbol{PU_i}\) results in a cost \(c_i\). A finite set of features \(\boldsymbol{F}=(F_1, F_2, ... , F_j, ..., F_J)\) is also distributed in \(A\), suggesting that these features can be spatially mapped within the PUs. Each \(PU\) is thus defined as a spatial object containing the following properties: \(\boldsymbol{PU_i}=(f_1, f_2, ..., f_j, ..., f_J, c_i)\), where \(f_j\) indicates the quantity of \(F_j\) located in \(PU_i\). We use the annotation \(M_j\) for the values of the metric for every feature \(j\). By inserting metrics into the analysis, each PU’s (\(\boldsymbol{PU_i}\)) properties are extended and can be expressed as follows: \(\boldsymbol{PU_i} = (f_1, f_2, ..., f_j, ..., f_J, M_1, M_2, ..., M_j, ..., M_J, c_i)\).

The first input that can be given to the algorithm is a typical cost layer representing the cost of protecting each single planning unit. This cost layer is inserted to prioritizr. If such a cost layer is not available, a cost layer with all planning units having equal cost can be inserted (Fig. 5. The algorithm needs a cost layer, in order to determine the planning units.

# Set tmap to “view” mode
tmap_mode("view")

# Read the cost raster
cost_raster <- get_cost_raster()

# Plot the cost raster with tmap
tm_shape(cost_raster) +
  tm_raster(title = "cost")

Fig. 5: Cost raster

The second input are the features rasters. In case such rasters are not available, a raster with equal values can be given to the algorithm. In this case we have used such a pseudo-raster, adding some noise to improve the performance of the algorithm.

# Read the features raster
features <- get_features_raster()

# Plot the features raster with tmap
tm_shape(features) +
  tm_raster(title = "f1")

Fig. 6: Features raster

# Solve an ordinary prioritizr prioritization problem
basic_solution <- basic_scenario(
                    cost_raster = cost_raster,
                    features    = features,
                    budget_perc = 0.1
                  )

# Solve a prioritizr prioritization problem,
# by incorporating graph connectivity of the features
connectivity_solution <- connectivity_scenario(
                           cost_raster = cost_raster,
                           features    = features,
                           budget_perc = 0.1,
                           pre_graphs  = pre_graphs
                         )

Step 4: Post-processing:

The results obtained from prioritizr are presented using matrices and plots, allowing the user to compare the outcomes of incorporating connectivity metrics in the analysis. Function get_outputs() takes as input a prioritization solution. The user can also get the outputs as shapefiles and raster to use them in their analysis.

 # Get outputs from basic_scenario function for feature “f1”
basic_outputs <- get_outputs(solution   = basic_solution, 
                             feature    = "f1",
                             pre_graphs = pre_graphs)
basic_outputs$tmap

Fig. 7: Basic solution with connections shown

# Print summary of features and connections held
# percentages for basic scenario
print(basic_outputs$connectivity_table)

##  feature relative_held connections(%)
## 1     f1      0.178563              0

# Get outputs from connectivity_scenario function for feature “f1”
connectivity_outputs <- get_outputs(solution   = connectivity_solution,
                                    feature    = "f1",
                                    pre_graphs = pre_graphs)
connectivity_outputs$tmap

Fig. 8: Connectivity solution with connections shown

# Print summary of features and connections held percentages
# for connectivity scenario
print(connectivity_outputs$connectivity_table)

##  feature relative_held connections(%)
## 1     f1     0.1637209      0.3339886