https://github.com/JosiahParry/sfweight

Calculate spatial weights and lags with {sf} objects and the tidyverse.
https://github.com/JosiahParry/sfweight

gis rspatial

Last synced: 3 months ago
JSON representation

Calculate spatial weights and lags with {sf} objects and the tidyverse.

• Host: GitHub
• URL: https://github.com/JosiahParry/sfweight
• Owner: JosiahParry
• Created: 2021-05-24T01:16:51.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-03-29T17:46:55.000Z (over 2 years ago)
• Last Synced: 2024-06-11T16:08:26.916Z (5 months ago)
• Topics: gis, rspatial
• Language: R
• Homepage:
• Size: 7.28 MB
• Stars: 59
• Watchers: 6
• Forks: 2
• Open Issues: 8
• Metadata Files:
  • Readme: README.Rmd

Awesome Lists containing this project

README

        
---

output: github_document

---

```{r, include = FALSE}

knitr::opts_chunk$set(

  collapse = TRUE,

  comment = "#>",

  fig.path = "man/figures/README-",

  out.width = "100%"

)

```

[![CRAN status](https://www.r-pkg.org/badges/version/sfweight)](https://CRAN.R-project.org/package=sfweight)

[![Lifecycle: experimental](https://img.shields.io/badge/lifecycle-experimental-orange.svg)](https://lifecycle.r-lib.org/articles/stages.html#experimental)

NOTE: this package is under active and experimental development. Functions are likely to change. 

sfweight is an opinionated translation of the wonderful spdep package. The goal is to provide a streamlined method of doing spatial statistics that works with sf objects, data frames, and the tidyverse. spdep is more flexible with the types of input objects and a bit more idiosyncratic in the syntax that is used. 

## Installation

You can install the development version from GitHub with 

```{r eval=FALSE}

remotes::install_github("Josiahparry/sfweight")

```

## Motivating examples

### Spatial OLS

We can fit a spatial Durbin model by calculating spatially lagged predictors. 

```{r, warning = FALSE, message=FALSE}

library(sfweight)

library(tidyverse)

acs_lagged <- acs %>% 

  mutate(nb = st_contiguity(geometry),

         wts = st_weights(nb),

         trans_lag = st_lag(by_pub_trans, nb, wts),

         bach_lag = st_lag(bach, nb, wts))

durbin_lm <- lm(med_house_income ~ trans_lag + by_pub_trans + bach_lag + bach, 

   data = acs_lagged)

broom::tidy(durbin_lm)

```

### Local Autocorrelation

We can create a Moran plot by creating a spatially lagged variable. Additionally the function `categorize_lisa()` will categorize high-high, high-low, etc., groupings of these variables.

```{r}

acs_lagged %>% 

  mutate(inc_lag = st_lag(med_house_income, nb, wts),

         lisa_group = categorize_lisa(med_house_income, inc_lag)) %>% 

  ggplot(aes(med_house_income, inc_lag, color = lisa_group)) +

  geom_vline(aes(xintercept = mean(med_house_income)), lty = 2, alpha = 1/3) +

  geom_hline(aes(yintercept = mean(inc_lag)), lty = 2, alpha = 1/3) + 

  geom_point() +

  labs(title = "Moran Plot",

       y = "Med. HH Income Spatial Lag",

       x = "Median Household Income") +

  theme_minimal() +

  scale_x_continuous(labels = scales::dollar) + 

  scale_y_continuous(labels = scales::dollar)

```

We can also calculate the Local  Moran's I for each observation using the function `local_moran()` this will create a dataframe column containing the I, expected I, variance, Z-value, and P-value for each observation. You can extract this using `tidyr::unnest()`.

```{r}

acs %>% 

  mutate(nb = st_contiguity(geometry),

         wt = st_weights(nb),

    lisa = local_moran(med_house_income, nb, wt)) %>% 

  unnest(lisa) %>% 

  ggplot(aes(fill = lisa_category)) + 

  geom_sf(color = "black", lwd = 0.25) +

  scale_fill_manual(values = c("HH" = "#528672","LL" = "#525586", "Insignificant" = NA))

  

```

## Basic usage & contiguities

```{r}

str(acs)

```

We can get neighbors based on Queen contiguities with `st_contiguity()`.

```{r}

nbs <- st_contiguity(acs)

nbs[1:5]

```

If needed, we can also identify the cardinalities from the neighbors list as well. 

```{r}

st_cardinalties(nbs)

```

We can get the weights from the neighbor contiguities as well. By default, `st_weights()` uses row standardization. 

```{r}

wts <- st_weights(nbs)

wts[1:5]

```

We can also calculate the spatial lag with the weights and neighbors. 

```{r}

inc_lag <- st_lag(acs$med_house_income, nbs, wts)

inc_lag[1:5]

```

## K-Nearest Neighbor Distances

If we have point data we can also identify the k-nearest neighbors with `st_knn()`. For an example we can use the `airbnb` dataset that's imported with `sfweight`. 

```{r}

airbnb

```

```{r message=FALSE, warning=FALSE}

airbnb_knn <- st_knn(airbnb)

airbnb_knn[1:5]

```

## Other weights

Point based weights implemented based on Luc Anselin and Grant Morrison's [notes](https://spatialanalysis.github.io/lab_tutorials/Spatial_Weights_as_Distance_Functions.html#kernal-weights). 

Inverse distance band

```{r}

airbnb_idw <- st_inverse_weights(airbnb$geometry, airbnb_knn)

airbnb_idw[1]

```

### Kernel based weights

Available kernels are:

- uniform

- triangular

- epanechnikov

- quartic

- gaussian

```{r}

airbnb_gauss <- st_kernel_weight(airbnb$geometry, airbnb_knn, "gaussian")

airbnb_gauss[1]

```

## Higher order neighbors 

```{r}

acs %>% 

  transmute(nb = st_contiguity(geometry),

            nb_2 = st_nb_lag(nb, 2),

            nb_cumul_2 = st_nb_lag_cumul(nb, 2))

```