locationallocation <a href = "https://giacfalk.github.io/locationallocation/"><img src = "man/figures/logo.png" align="right" width="120" /></a>

<!-- quarto render --> <!-- Install the package before rendering this file: `devtools::install()` --> <!-- badges: start -->

<!-- badges: end -->

Overview

Assessing and planning infrastructure and service networks, given a dispersed demand, limited capacity, accessibility targets, and concerns about spatial justice, is a central policy challenge. Problems of this type are commonly referred to as Maximal Coverage Location-Allocation (MCLA) spatial optimization problems.

locationallocation is an R package that provides tools for solving MCLA problems with geospatial data. It builds on widely used spatial libraries in R, follows tidyverse principles, and integrates seamlessly with the broader tidyverse ecosystem. The package can generate travel-time maps and optimize the placement of facilities or infrastructure according to accessibility criteria, which can be weighted by one or more variables or by a user-defined function.

Potential applications of the package extend to the domains of public infrastructure assessment and planning (public services provision, e.g. transport, social services, healthcare, parks), urban environmental and climate risk reduction interventions, logistics and hubs allocation, commercial and strategic decisions.

If you find this project useful, please consider giving it a star!  

Installation

You can install locationallocation using the remotes package:

# install.packages(remotes)
remotes::install_github("giacfalk/locationallocation")

A CRAN version of the package is planned for the near future.

Usage

To use the package, start by loading it to your R session using the library function:

library(locationallocation)

As an example, we demonstrate how the package can address urban-scale climate risk through infrastructure assessment and geospatial planning. For this demonstration, we use the demo datasets included with the package.

These datasets include the coordinates of public drinking water fountains (blue dots) in Naples, Italy (naples_fountains); a gridded population raster from the Global Human Settlement Layer (GHSL) Population Grid (GHS-POP) (naples_population); a 100-meter resolution heat hazard map, representing the number of days with Wet-Bulb Globe Temperature above 25°C during 2008–2017, obtained from the UrbClim model (naples_hot_days); and the city’s administrative boundaries (naples_shape).

We can use the traveltime() function to create a map of current accessibility to the facility points (represented by the sf object naples_fountains) within the specified geographical boundaries. The function allows the user to select a travel mode (walking or fastest route) and an output spatial resolution in meters, achieved through dissevering spatial downscaling techniques.

traveltime_data <-
  naples_fountains |>
  traveltime(
    bb_area = naples_shape,
    dowscaling_model_type = "lm",
    mode = "walk",
    res_output = 100
  )

Once traveltime() has completed, the resulting layer can be visualized using the traveltime_plot() function.

traveltime_data |>
  traveltime_plot(
    bb_area = naples_shape,
    facilities = naples_fountains,
    contour_traveltime = 15
  )

We can also generate a summary plot and compute statistics using the output of the traveltime() function, together with a demand raster (e.g., population density) and a specified time threshold, using the traveltime_stats function:

traveltime_data |>
  traveltime_stats(
    demand = naples_population,
    breaks = c(5, 10, 15, 30),
    objectiveminutes = 15
  )
#> ℹ 85.45148% of coverage within the 15 minutes threshold.

We can now use the allocation() function to optimize the placement of new water fountains, ensuring that (virtually) everyone (i.e., the full extent of the raster layer specified by the demand parameter) can reach one within 15 minutes, as defined by the objectiveminutes parameter:

allocation_data <-
  naples_population |>
  allocation(
    bb_area = naples_shape,
    facilities = naples_fountains,
    traveltime = traveltime_data,
    weights = NULL,
    objectiveminutes = 15,
    objectiveshare = 0.99,
    heur = "max",
    approach = "norm",
    exp_demand = 1,
    exp_weights = 1
  )

#> ✔ Target coverage share of 99% attained with 28 facilities within the 15
#> minutes threshold. The achieved coverage share is 99.03953%.

allocation_data |> allocation_plot(naples_shape)

Note that it is also possible to solve an allocation problem using the weights parameter, which assigns greater relative importance or priority to areas where demand overlaps with weighting factors defined by another raster layer, such as exposure to hot days, as shown in the following example:

allocation_data <-
  naples_population |>
  allocation(
    bb_area = naples_shape,
    facilities = naples_fountains,
    traveltime = traveltime_data,
    weights = naples_hot_days, # <--- Changed
    objectiveminutes = 15,
    objectiveshare = 0.99,
    heur = "max",
    approach = "norm",
    exp_demand = 1,
    exp_weights = 1
  )

#> ✔ Target coverage share of 99% attained with 25 facilities within the 15
#> minutes threshold. The achieved coverage share is 99.16903%.

allocation_data |> allocation_plot(naples_shape)

It is also possible to apply normalization and exponentiation to different demand and weighting layers, enhancing their relative influence on the allocation, using the approach, exp_demand, and exp_weights parameters (see the function documentation for details):

allocation_data <-
  naples_population |>
  allocation(
    bb_area = naples_shape,
    facilities = naples_fountains,
    traveltime = traveltime_data,
    weights = naples_hot_days,
    objectiveminutes = 15,
    objectiveshare = 0.99,
    heur = "max",
    approach = "norm",
    exp_demand = 2, # <--- Changed
    exp_weights = 1
  )

#> ✔ Target coverage share of 99% attained with 9 facilities within the 15
#> minutes threshold. The achieved coverage share is 99.15597%.

allocation_data |> allocation_plot(naples_shape)

A variant of the allocation problem arises when the set of candidate locations for new facilities is discrete, rather than continuous across the study area as in the previous example.

In this case, the user must provide a set of candidate points via the candidate parameter of the allocation_discrete() function, along with the maximum number of facilities that can be selected (n_fac parameter). The function applies a quasi-optimality heuristic based on a randomization approach, where the number of replications (n_samples parameter) gradually approaches the global optimum, although computational time increases linearly. As with the continuous allocation problem, a weight layer and normalization and exponentiation parameters can also be specified.

library(sf)

allocation_data <-
  naples_population |>
  allocation_discrete(
    bb_area = naples_shape,
    candidate = naples_shape |> st_sample(20),
    facilities = naples_fountains,
    n_fac = 5,
    n_samples = 100,
    traveltime = travelt