Geospatial Distribution Dynamics for Simple Features

Tools for exploratory geospatial distribution dynamics with 'sf' objects and tidy data. Provides pooled and time-specific classification of longitudinal spatial values, classic discrete Markov transition matrices, spatial Markov matrices conditioned on spatial-lag classes, endpoint and adjacent rank mobility summaries, and 'ggplot2' visualizations. Methods follow Rey (2001) and Rey et al. (2016) ; design is inspired by the Python 'PySAL' 'giddy' package < https://pysal.org/giddy/>.

griddy

griddy is an R package for geospatial distribution dynamics with sf and tidy data. It is inspired by PySAL giddy and by established spatial distribution dynamics work, especially Rey (2001). Prior R implementations exist, including estdaR and spdyn, but neither is on CRAN or designed along tidy principles. This package is designed to work best with long data and sf native spatial workflows.

Current scope:

classify longitudinal spatial values into comparable classes
estimate classic Markov transition matrices
estimate spatial Markov transition matrices conditioned on spatial lag class
compute simple rank mobility
return tidy tables and ggplot2 plots

The initial CRAN release focuses on the core classification, Markov, spatial Markov, rank mobility, and plotting workflows.

Installation

# install.packages("pak")
pak::pak("dshkol/griddy")

The minimal example below also uses spData for state geometry:

install.packages("spData")

What it does

griddy keeps the workflow keyed by explicit id, time, and value columns instead of matrix row position. The analytical outputs preserve transition labels, class intervals, and spatial-lag intervals so they can be inspected, joined, and plotted without reverse-engineering array dimensions.

Minimal example

The bundled usjoin panel is the canonical PySAL giddy reference dataset: 48 contiguous US states, per-capita personal income, 1929 to 2009.

library(griddy)
library(dplyr)
library(sf)
library(sfdep)
library(spData)

data(usjoin)

geom <- us_states |>
  filter(NAME %in% usjoin$name) |>
  arrange(NAME) |>
  mutate(
    nb = st_contiguity(geometry),
    wt = st_weights(nb)
  )

panel <- usjoin |>
  filter(name %in% geom$NAME) |>
  arrange(name, year)

classes <- classify_dynamics(panel, name, year, income, k = 5)
classic <- markov_dynamics(classes, name, year, class)
spatial <- spatial_markov(panel, name, year, income, geometry = geom, k = 5)

classic$transitions |> select(id, from_time, to_time, transition) |> head()
lag_intervals(spatial)

spatial_markov() takes a geometry argument: an sf tibble with one row per spatial unit and nb / wt list-columns produced by sfdep. This keeps the spatial frame, neighbor structure, and row-standardization choice in one tidy object. listw and nb arguments remain accepted for compatibility with existing workflows.

Documentation

The pkgdown site is organized around:

core workflow and concepts
tidycensus and cancensus examples showing applications with US and Canadian census data
prior-art comparison against estdaR and spdyn
performance benchmarking notes