raster: interpolate – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

interpolate

Interpolate

Description

Make a RasterLayer with interpolated values using a fitted model object of classes such as 'gstat' (gstat package) or 'Krige' (fields package). That is, these are models that have location ('x' and 'y', or 'longitude' and 'latitude') as independent variables. If x and y are the only independent variables provide an empty (no associated data in memory or on file) RasterLayer for which you want predictions. If there are more spatial predictor variables provide these as a Raster* object in the first argument of the function. If you do not have x and y locations as implicit predictors in your model you should use predict instead.

Usage

## S4 method for signature 'Raster'
interpolate(object, model, filename="", fun=predict, xyOnly=TRUE,
   xyNames=c('x', 'y'), ext=NULL, const=NULL, index=1, na.rm=TRUE, debug.level=1, ...)

Arguments

`object`	Raster* object
`model`	model object
`filename`	character. Output filename (optional)
`fun`	function. Default value is 'predict', but can be replaced with e.g. 'predict.se' (depending on the class of the model object)
`xyOnly`	logical. If `TRUE`, values of the Raster* object are not considered as co-variables; and only x and y (longitude and latitude) are used. This should match the model
`xyNames`	character. variable names that the model uses for the spatial coordinates. E.g., `c('longitude', 'latitude')`
`ext`	Extent object to limit the prediction to a sub-region of `x`
`const`	data.frame. Can be used to add a constant for which there is no Raster object for model predictions. This is particulary useful if the constant is a character-like factor value
`index`	integer. To select the column if 'predict.model' returns a matrix with multiple columns
`na.rm`	logical. Remove cells with NA values in the predictors before solving the model (and return `NA` for those cells). In most cases this will not affect the output. This option prevents errors with models that cannot handle `NA` values
`debug.level`	for gstat models only. See ?
`...`	additional arguments passed to the predict.'model' function

Value

Raster* object

Examples

## Thin plate spline interpolation with x and y only
# some example data
r <- raster(system.file("external/test.grd", package="raster"))
ra <- aggregate(r, 10)
xy <- data.frame(xyFromCell(ra, 1:ncell(ra)))
v <- getValues(ra)
# remove NAs
i <- !is.na(v)
xy <- xy[i,]
v <- v[i]

#### Thin plate spline model
library(fields) 
tps <- Tps(xy, v)
p <- raster(r)

# use model to predict values at all locations
p <- interpolate(p, tps)
p <- mask(p, r)

plot(p)
## change the fun from predict to fields::predictSE to get the TPS standard error
se <- interpolate(p, tps, fun=predictSE)
se <- mask(se, r)
plot(se)

## another variable; let's call it elevation
elevation <- (init(r, 'x') * init(r, 'y')) / 100000000
names(elevation) <- 'elev'

z <- extract(elevation, xy)

# add as another independent variable
xyz <- cbind(xy, z)
tps2 <- Tps(xyz, v)
p2 <- interpolate(elevation, tps2, xyOnly=FALSE)

# as a linear coveriate
tps3 <- Tps(xy, v, Z=z)

# Z is a separate argument in Krig.predict, so we need a new function
# Internally (in interpolate) a matrix is formed of x, y, and elev (Z)

pfun <- function(model, x, ...) {
   predict(model, x[,1:2], Z=x[,3], ...)
}
p3 <- interpolate(elevation, tps3, xyOnly=FALSE, fun=pfun)

#### gstat examples
library(gstat)
data(meuse)

## inverse distance weighted (IDW)
r <- raster(system.file("external/test.grd", package="raster"))
data(meuse)
mg <- gstat(id = "zinc", formula = zinc~1, locations = ~x+y, data=meuse, 
            nmax=7, set=list(idp = .5))
z <- interpolate(r, mg)
z <- mask(z, r)

## kriging
coordinates(meuse) <- ~x+y
crs(meuse) <- crs(r)

## ordinary kriging
v <- variogram(log(zinc)~1, meuse)
m <- fit.variogram(v, vgm(1, "Sph", 300, 1))
gOK <- gstat(NULL, "log.zinc", log(zinc)~1, meuse, model=m)
OK <- interpolate(r, gOK)

# examples below provided by Maurizio Marchi
## universial kriging
vu <- variogram(log(zinc)~elev, meuse)
mu <- fit.variogram(vu, vgm(1, "Sph", 300, 1))
gUK <- gstat(NULL, "log.zinc", log(zinc)~elev, meuse, model=mu)
names(r) <- 'elev'
UK <- interpolate(r, gUK, xyOnly=FALSE)

## co-kriging
gCoK <- gstat(NULL, 'log.zinc', log(zinc)~1, meuse)
gCoK <- gstat(gCoK, 'elev', elev~1, meuse)
gCoK <- gstat(gCoK, 'cadmium', cadmium~1, meuse)
gCoK <- gstat(gCoK, 'copper', copper~1, meuse)
coV <- variogram(gCoK)
plot(coV, type='b', main='Co-variogram')
coV.fit <- fit.lmc(coV, gCoK, vgm(model='Sph', range=1000))
coV.fit
plot(coV, coV.fit, main='Fitted Co-variogram')
coK <- interpolate(r, coV.fit)
plot(coK)

raster

Geographic Data Analysis and Modeling

v3.4-10

GPL (>= 3)

Authors

Robert J. Hijmans [cre, aut] (<https://orcid.org/0000-0001-5872-2872>), Jacob van Etten [ctb], Michael Sumner [ctb], Joe Cheng [ctb], Dan Baston [ctb], Andrew Bevan [ctb], Roger Bivand [ctb], Lorenzo Busetto [ctb], Mort Canty [ctb], Ben Fasoli [ctb], David Forrest [ctb], Aniruddha Ghosh [ctb], Duncan Golicher [ctb], Josh Gray [ctb], Jonathan A. Greenberg [ctb], Paul Hiemstra [ctb], Kassel Hingee [ctb], Institute for Mathematics Applied Geosciences [cph], Charles Karney [ctb], Matteo Mattiuzzi [ctb], Steven Mosher [ctb], Babak Naimi [ctb], Jakub Nowosad [ctb], Edzer Pebesma [ctb], Oscar Perpinan Lamigueiro [ctb], Etienne B. Racine [ctb], Barry Rowlingson [ctb], Ashton Shortridge [ctb], Bill Venables [ctb], Rafael Wueest [ctb]

Initial release

2021-05-02