Getting started

In prioritizr: Systematic Conservation Prioritization in R

# define dummy variables so that vignette passes package checks
tas_features <- raster::raster(matrix(1))

# define variables for vignette figures and code execution
h <- 3.5
w <- 3.5
is_check <-
  ("CheckExEnv" %in% search()) ||
  any(c("_R_CHECK_TIMINGS_", "_R_CHECK_LICENSE_") %in% names(Sys.getenv())) ||
  !identical(Sys.getenv("MY_UNIVERSE"), "")
knitr::opts_chunk$set(
  fig.align = "center",
  eval = !is_check,
  purl = !is_check,
  R.options = list(width = 80)
)

# set up print method
print <- function(x, ...) {
  if (inherits(x, "ConservationProblem")) {
    prioritizr::knit_print.ConservationProblem(x)
  } else if (inherits(x, "OptimizationProblem")) {
    prioritizr::knit_print.OptimizationProblem(x)
  } else {
    base::print(x)
  }
}

Introduction

The aim of this tutorial is to provide a short introduction to the prioritizr R package. It is also intended to help conservation planners familiar the Marxan decision support tool [@r3] start using the package for their work.

Data

Let's load the packages and data used in this tutorial. Since this tutorial uses data from the prioritizrdata R package, please ensure that it is installed. The data used in this tutorial were obtained from the Introduction to Marxan course and the Australian Government's National Vegetation Information System.

# load packages
library(prioritizrdata)
library(prioritizr)
library(sf)
library(terra)
library(vegan)
library(cluster)

# load planning unit data
tas_pu <- get_tas_pu()

# load feature data
tas_features <- get_tas_features()

Let's have a look at the planning unit data. The tas_pu object contains planning units represented as spatial polygons (i.e., a sf::st_sf() object). This object has three columns that denote the following information for each planning unit: a unique identifier (id), unimproved land value (cost), and current conservation status (locked_in). Planning units that have at least half of their area overlapping with existing protected areas are denoted with a locked in TRUE value, otherwise they are denoted with a value of FALSE.

# print planning unit data
print(tas_pu)

# plot map of planning unit costs
plot(st_as_sf(tas_pu[, "cost"]), main = "Planning unit costs")

# plot map of planning unit coverage by protected areas
plot(st_as_sf(tas_pu[, "locked_in"]), main = "Protected area coverage")

Now, let's look at the conservation feature data. The tas_features object describes the spatial distribution of the features. Specifically, the feature data are a multi-layer raster (i.e., a terra::rast() object). Each layer corresponds to a different vegetation community. Within each layer, cells values denote the presence (using value of 1) or absence (using value of 0) of the vegetation community across the study area.

# print planning unit data
print(tas_features)

# plot map of the first four vegetation classes
plot(tas_features[[1:4]])

Problem formulation

Now we will formulate a conservation planing problem. To achieve this, we first specify which objects contain the planning unit and feature data (using the problem() function). Next, we specify that we want to use the minimum set objective function (using the add_min_set_objective() function). This objective function indicates that we wish to minimize the total cost of planning units selected by the prioritization. We then specify boundary penalties to reduce spatial fragmentation in the resulting prioritization (using the add_boundary_penalties() function; see the Calibrating trade-offs vignette for details on calibrating the penalty value). We also specify representation targets to ensure the resulting prioritization provides adequate coverage of each vegetation community (using the add_relative_targets() function). Specifically, we specify targets to ensure at least 17% of the spatial extent of each vegetation community (based on the Aichi Target 11). Additionally, we set constraints to ensure that planning units predominately covered by existing protected areas are selected by the prioritization (using the add_locked_in_constraints() function). Finally, we specify that the prioritization should either select -- or not select -- planning units for prioritization (using the add_binary_decisions() function).

# build problem
p1 <-
  problem(tas_pu, tas_features, cost_column = "cost") %>%
  add_min_set_objective() %>%
  add_boundary_penalties(penalty = 0.005) %>%
  add_relative_targets(0.17) %>%
  add_locked_in_constraints("locked_in") %>%
  add_binary_decisions()

# print problem
print(p1)

Prioritization

We can now solve the problem formulation (p1) to generate a prioritization (using the solve() function). The prioritizr R package supports a range of different exact algorithm solvers, including Gurobi, IBM CPLEX, CBC, HiGHS, Rsymphony, and lpsymphony. Although there are benefits and limitations associated with each of these different solvers, they should return similar results. Note that you will need at least one solver installed on your system to generate prioritizations. Since we did not specify a solver when building the problem, the prioritizr R package will automatically select the best available solver installed. We recommend using the Gurobi solver if possible, and have used it for this tutorial (see the Gurobi Installation Guide vignette for installation instructions). After solving the problem, the prioritization will be stored in the solution_1 column of the s1 object.

# solve problem
s1 <- solve(p1)

# plot map of prioritization
plot(
  st_as_sf(s1[, "solution_1"]), main = "Prioritization",
  pal = c("grey90", "darkgreen")
)

Feature representation

Let's examine how well the vegetation communities are represented by existing protected areas and the prioritization.

# create column with existing protected areas
tas_pu$pa <- round(tas_pu$locked_in)

# calculate feature representation statistics based on existing protected areas
tc_pa <- eval_target_coverage_summary(p1, tas_pu[, "pa"])
print(tc_pa)

# calculate  feature representation statistics based on the prioritization
tc_s1 <- eval_target_coverage_summary(p1, s1[, "solution_1"])
print(tc_s1)

# explore representation by existing protected areas
## calculate number of features adequately represented by existing protected
## areas
sum(tc_pa$met)

## summarize representation (values show percent coverage)
summary(tc_pa$relative_held * 100)

## visualize representation  (values show percent coverage)
hist(tc_pa$relative_held * 100,
     main = "Feature representation by existing protected areas",
     xlim = c(0, 100),
     xlab = "Percent coverage of features (%)")


# explore representation by prioritization
## summarize representation (values show percent coverage)
summary(tc_s1$relative_held * 100)

## calculate number of features adequately represented by the prioritization
sum(tc_s1$met)

## visualize representation  (values show percent coverage)
hist(
  tc_s1$relative_held * 100,
  main = "Feature representation by prioritization",
  xlim = c(0, 100),
  xlab = "Percent coverage of features (%)"
)

We can see that representation of the vegetation communities by existing protected areas is remarkably poor. For example, many of the vegetation communities have nearly zero coverage by existing protected areas. In other words, are almost entirely absent from existing protected areas. We can also see that all vegetation communities have at least 17% coverage by the prioritization -- meaning that it meets the representation targets for all of the features.

Irreplaceability

After generating the prioritization, we can examine the relative importance of planning units selected by the prioritization. This can be useful to identify critically important planning units for conservation -- in other words, places that contain biodiversity features which cannot be represented anywhere else -- and schedule implementation of the prioritization. To achieve this, we will use the Ferrier metric [@r34].

# calculate irreplaceability
irrep_s1 <- eval_ferrier_importance(p1, s1["solution_1"])
print(irrep_s1)

# manually coerce values for planning units not selected in prioritization
# to NA, so that they are shown in white
irrep_s1$plot_total <- irrep_s1$total
irrep_s1$plot_total[s1$solution_1 < 0.5] <- NA_real_

# plot map of overall importance scores
plot(st_as_sf(irrep_s1[, "plot_total"]), main = "Overall importance")

Portfolios

Conservation planning exercises often involve generating multiple different prioritizations. This can help decision makers consider different options, and provide starting points for building consensus among stakeholders. To generate a range of different prioritizations given the same problem formulation, we can use portfolio functions. Here we will use the gap portfolio to generate 1000 solutions that are within 20% of optimality. Please note that you will need to have the Gurobi solver installed to use this specific portfolio. If you don't have access to Gurobi, you could try using the shuffle portfolio instead (using the add_shuffle_portfolio() function).

# create new problem with a portfolio added to it
p2 <-
  p1 %>%
  add_gap_portfolio(number_solutions = 1000, pool_gap = 0.2)

# print problem
print(p2)

# generate prioritizations
prt <- solve(p2)
print(prt)

After generating all these prioritizations, we now want some way to visualize them. Because it would be onerous to look at each and every prioritization individually, we will use statistical analyses to help us. We can visualize the differences between these different prioritizations -- based on which planning units they selected -- using a hierarchical cluster analysis [@r35].

# extract solutions
prt_results <- sf::st_drop_geometry(prt)
prt_results <- prt_results[, startsWith(names(prt_results), "solution_")]

# calculate pair-wise distances between different prioritizations for analysis
prt_dists <- vegan::vegdist(t(prt_results), method = "jaccard", binary = TRUE)

# run cluster analysis
prt_clust <- hclust(as.dist(prt_dists), method = "average")

# visualize clusters
opar <- par()
par(oma = c(0, 0, 0, 0), mar= c(0, 4.1, 1.5, 2.1))
plot(
  prt_clust, labels = FALSE, sub = NA, xlab = "",
  main = "Different prioritizations in portfolio"
)
suppressWarnings(par(opar))

We can see that there are approximately six main groups of prioritizations in the portfolio. To explore these different groups, let's conduct another cluster analysis (i.e., a k-medoids analysis) to extract the most representative prioritization from each of these groups. In other words, we will run another statistical analysis to find the most central prioritization within each group.

# run k-medoids analysis
prt_med <- pam(prt_dists, k = 6)

# extract names of prioritizations that are most central for each group.
prt_med_names <- prt_med$medoids
print(prt_med_names)

# create a copy of prt and set values for locked in planning units to -1
# so we can easily visualize differences between prioritizations
prt2 <- prt[, prt_med_names]
prt2[which(tas_pu$locked_in > 0.5), prt_med_names] <- -1

# plot a map showing main different prioritizations
# dark grey: locked in planning units
# grey: planning units not selected
# green: selected planning units
plot(st_as_sf(prt2), pal = c("grey60", "grey90", "darkgreen"))

Marxan compatibility

The prioritizr R package provides functionality to help Marxan users generate prioritizations. Specifically, it can import conservation planning data prepared for Marxan, and can generate prioritizations using a similar problem formulation as Marxan [based on @r1]. Indeed, the problem formulation presented earlier in this vignette is very similar to that used by Marxan. The key difference is that the problem formulation we specified earlier uses "hard constraints" for feature representation, and Marxan uses "soft constraints" for feature representation. This means that prioritization we generated earlier was mathematically guaranteed to reach the targets for all features. However, if we used Marxan to generate the prioritization, then we could have produced a prioritization that would fail to reach targets (depending the Species Penalty Factors used to generate the prioritization). In addition to these differences in terms problem formulation, the prioritizr R package uses exact algorithms -- instead of the simulated annealing algorithm -- which ensures that we obtain prioritizations that are near optimal.

Here we will show the prioritizr R package can import Marxan data and generate a prioritization. To begin with, let's import a conservation planning data prepared for Marxan.

# import data
## planning unit data
pu_path <- system.file("extdata/marxan/input/pu.dat", package = "prioritizr")
pu_data <- read.csv(pu_path, header = TRUE, stringsAsFactors = FALSE)
print(head(pu_data))

## feature data
spec_path <- system.file(
  "extdata/marxan/input/spec.dat", package = "prioritizr"
)
spec_data <- read.csv(spec_path, header = TRUE, stringsAsFactors = FALSE)
print(head(spec_data))

## amount of each feature within each planning unit data
puvspr_path <- system.file(
  "extdata/marxan/input/puvspr.dat", package = "prioritizr"
)
puvspr_data <- read.csv(puvspr_path, header = TRUE, stringsAsFactors = FALSE)
print(head(puvspr_data))

## boundary data
bound_path <- system.file(
  "extdata/marxan/input/bound.dat", package = "prioritizr"
)
bound_data <- read.table(bound_path, header = TRUE, stringsAsFactors = FALSE)
print(head(bound_data))

After importing the data, we can now generate a prioritization based on the Marxan problem formulation (using the marxan_problem() function). Please note that this function does not generate prioritizations using Marxan. Instead, it uses the data to create an optimization problem formulation similar to Marxan -- using hard constraints instead of soft constraints -- and uses an exact algorithm solver to generate a prioritization.

# create problem
p3 <- marxan_problem(
  pu_data, spec_data, puvspr_data, bound_data, blm = 0.0005
)

# print problem
print(p3)

# solve problem
s3 <- solve(p3)

# print first six rows of solution object
print(head(s3))

Conclusion

This tutorial shows how the prioritizr R package can be used to build a conservation problem, generate a prioritization, and evaluate it. Although we explored just a few functions, the package provides many different functions so that you can build and custom-tailor conservation planning problems to suit your needs. To learn more about the package, please see the package vignettes for an overview of the package, instructions for installing the Gurobi optimization suite, benchmarks comparing the performance of different solvers, and a record of publications that have cited the package. In addition to this tutorial, the package also provides tutorials on incorporating connectivity into prioritizations, calibrating trade-offs between different criteria (e.g., total cost and spatial fragmentation), and creating prioritizations that have multiple management zones or management actions.

References

prioritizr documentation built on Aug. 9, 2023, 1:06 a.m.