R’s Geospatial Kaleidoscope:
Exploring Perspectives, Strengths, and Challenges
2024-11-07
Vector data
Source: https://r-tmap.github.io/
Vector data
library(sf)
linestring_sf = read_sf("data/linestring_sf.gpkg")
linestring_sfSimple feature collection with 1 feature and 1 field
Geometry type: LINESTRING
Dimension: XY
Bounding box: xmin: 0.8 ymin: 1 xmax: 1 ymax: 1.2
Geodetic CRS: WGS 84
# A tibble: 1 × 2
id geom
<chr> <LINESTRING [°]>
1 1 (0.8 1, 0.8 1.2, 1 1.2)Raster data
Spatial raster data model: describing the real world as continuous surfaces
Raster maps represent continuous phenomena such as elevation, temperature, population density or spectral data
They also can represent discrete features such as soil or land-cover classes
Raster data
Raster data
library(terra)
nz_elev = rast("data/nz_elev.tif")
nz_elevclass : SpatRaster
dimensions : 1450, 1115, 1 (nrow, ncol, nlyr)
resolution : 1000, 1000 (x, y)
extent : 995456.5, 2110457, 4741961, 6191961 (xmin, xmax, ymin, ymax)
coord. ref. : NZGD2000 / New Zealand Transverse Mercator 2000 (EPSG:2193)
source : nz_elev.tif
name : elevation
min value : 0.000
max value : 4140.333 
Other types of spatial data
Data cubes/Spatiotemporal Arrays: https://r-spatial.github.io/stars/
Meshes: https://rpubs.com/cyclemumner/mdsumner-geomesh-foss4g
Point clouds: https://r-lidar.github.io/lidRbook/
Coordinate Reference Systems (CRSs)
Geographical coordinates
Projected coordinates
Coordinate Reference Systems (CRSs)
CRSs are crucial for spatial data analysis:
- Data integration
- Data visualization (map making)
- Spatial operations: distance, area, etc.
Spatial statistics
Data types conversion exist in many R spatial statistics packages
spatstat:
my_points = read_sf("my_points.gpkg")
my_points_ppp = as.ppp(my_points)sfnetworks:
my_lines = read_sf("my_lines.gpkg")
my_lines_sfn = as_sfnetwork(my_lines)spdep:
my_polygons = read_sf("my_polygons.gpkg")
my_polygons_nb = poly2nb(my_polygons)
Many statistical techniques make the assumption that observations are independent.
This assumption can be easily violated when working with ecological/environmental/geographical data.
Violation of this assumption can risk that the results will be wrong.
Temporal autocorrelation
Spatial autocorrelation
Spatial autocorrelation
Friend or Foe?
What we want to achieve?
- Understand and describe the autocorrelation
- Draw inferences about the spatial processes
- Make predictions in unsampled locations
Describing spatial autocorrelation
library(tmap)
library(gstat)
data("lsl", "study_mask", package = "spDataLarge")
lsl_sf = st_as_sf(lsl, coords = c("x", "y"))
plot(lsl_sf["lslpts"], cex = 2)
plot(variogram(lslpts ~ 1, locations = lsl_sf))
Drawing inferences about spatial processes
fit = glm(lslpts ~ slope + cplan + cprof + elev + log10_carea, family = binomial(), data = lsl)
lsl_sf$residuals = residuals(fit)tm_shape(lsl_sf) + tm_symbols(fill = "residuals")
plot(variogram(residuals ~ 1, locations = lsl_sf))
Drawing inferences about spatial processes
lsl_sf$residuals2 = residuals(model_e)tm_shape(lsl_sf) + tm_symbols(fill = "residuals2")
plot(variogram(residuals2 ~ 1, locations = lsl_sf))
Making predictions in unsampled locations
Machine learning
The results of the hyperparameter tuning, feature selection, and (often) model evaluation depend on the selected cross-validation strategy.
File formats
Spatial data is stored in various file formats
Vector data:
- Shapefile (.shp): avoid if possible
- GeoPackage (.gpkg): better alternative
- GeoJSON (.geojson): good for web applications
- GeoParquet (.parquet): columnar storage format
Raster data:
- GeoTIFF (.tif): the most common format
- NetCDF (.nc): multidimensional raster data
- Zarr (.zarr): cloud-optimized format for multidimensional data
Point clouds:
- LAS (.las): the most common point cloud format
- LAZ (.laz): compressed version of LAS
File formats
A shapefile example
gpkg_file = system.file("shapes/world.gpkg",
package = "spData")
world = read_sf(gpkg_file)[0]
world$the_best_column_ever = 1
names(world)[1] "geom" "the_best_column_ever"write_sf(world, "world.shp")
world_shp = read_sf("world.shp")
names(world_shp)[1] "th_bs__" "geometry"File formats capabilities
spain = read_sf(gpkg_file,
query = 'SELECT * FROM world WHERE name_long = "Spain"')
plot(spain[1])
File formats
myurl = paste0("/vsicurl/https://zenodo.org/record/6211990/files/",
"copernicus_DSM_100m_mood_bbox_epsg3035.tif") #~9.2 GB
dem = rast(myurl)
demclass : SpatRaster
dimensions : 73590, 78430, 1 (nrow, ncol, nlyr)
resolution : 100, 100 (x, y)
extent : 869000, 8712000, -485000, 6874000 (xmin, xmax, ymin, ymax)
coord. ref. : ETRS89-extended / LAEA Europe (EPSG:3035)
source : copernicus_DSM_100m_mood_bbox_epsg3035.tif
name : copernicus_DSM_100m_mood_bbox_epsg3035
min value : -427025
max value : 5606973 
sev = st_as_sf(data.frame(lon = -5.99, lat = 37.39), coords = c("lon", "lat"), crs = "EPSG:4326")
sev_3035 = st_transform(sev, crs = "EPSG:3035")
dem_sev = extract(dem, sev_3035)
dem_sev ID copernicus_DSM_100m_mood_bbox_epsg3035
1 1 20551(True elevation is the output value divided by 1000)
Cartography
That was then…
R cartography
… and this is now
library(mapSpain)
sev_borders = esp_get_munic_siane(region = "Sevilla",
epsg = "3035")
plot(sev_borders)
dem_sevm = crop(dem, sev_borders, mask = TRUE)
plot(dem_sevm)
plot(sev_borders, add = TRUE,
col = NA, border = "white")
R (static) cartography
library(ggplot2); library(ggspatial)
ggplot(data = sev_borders) +
geom_sf() +
annotation_scale(location = "tl") +
labs(title = "The Province of Seville") +
theme_minimal()
library(tmap)
tm_shape(sev_borders) +
tm_graticules() +
tm_polygons() +
tm_scalebar() +
tm_title("The Province of Seville") 
R (interactive) cartography
R cartography
A lot of developments in the last 10 years
At the same time, there seems to be many challenges and opportunities ahead – how to reimagine cartography using programming languages?
(Keep in mind to use the right tool for the right job)
R spatial in various domains
Geospatial methods in R
Different domains use different terminologies
Reinventing the wheel is common
More collaboration accross domains is needed
The whole picture
Learn
A lot of resources are available: books, websites, and courses
Apply
Think if you can use R for your spatial data analysis: your domain expertise is crucial
Collaborate
Conversation is needed: learning from other successes and mistakes; learning from other fields
Extend
More people are welcome to the R spatial community: there are still many things to do
Learn
Apply
Collaborate
Extend
Contact
Mastodon: fosstodon.org/@nowosad
Website: https://jakubnowosad.com
Slides
http://jakubnowosad.com/IIIRqueR/
