SpatialPosition (version 1.2.0)

mcStewart: Stewart Potentials Parallel

Description

This function computes Stewart potentials using parallel computation.

Usage

mcStewart(knownpts, unknownpts = NULL, varname, typefct = "exponential",
  span, beta, resolution = NULL, mask = NULL, cl = NULL, size = 1000,
  longlat = TRUE)

Arguments

knownpts

sp object (SpatialPointsDataFrame or SpatialPolygonsDataFrame); this is the set of known observations to estimate the potentials from.

unknownpts

sp object (SpatialPointsDataFrame or SpatialPolygonsDataFrame); this is the set of unknown units for which the function computes the estimates. Not used when resolution is set up. (optional)

varname

character; name of the variable in the knownpts dataframe from which potentials are computed. Quantitative variable with no negative values.

typefct

character; spatial interaction function. Options are "pareto" (means power law) or "exponential". If "pareto" the interaction is defined as: (1 + alpha * mDistance) ^ (-beta). If "exponential" the interaction is defined as: exp(- alpha * mDistance ^ beta). The alpha parameter is computed from parameters given by the user (beta and span).

span

numeric; distance where the density of probability of the spatial interaction function equals 0.5.

beta

numeric; impedance factor for the spatial interaction function.

resolution

numeric; resolution of the output SpatialPointsDataFrame (in map units). If resolution is not set, the grid will contain around 7250 points. (optional)

mask

sp object; the spatial extent of this object is used to create the regularly spaced SpatialPointsDataFrame output. (optional)

cl

numeric; number of clusters. By default cl is determined using parallel::detectCores().

size

numeric; mcStewart splits unknownpts in chunks, size indicates the size of each chunks.

longlat

logical; if FALSE, Euclidean distance, if TRUE Great Circle (WGS84 ellipsoid) distance.

Value

A Spatial*DataFrame with the computed potentials in a new field named OUTPUT is returned.

Details

The parallel implementation splits potentials computations along chunks of unknownpts (or chunks of the grid defined using resolution). It only uses Great Cercle distances (with CreateDistMatrix).

See Also

stewart.

Examples

# NOT RUN {
if(require(cartography)){
  nuts3.spdf@data <- nuts3.df
  t1 <- system.time(
    s1 <- stewart(knownpts = nuts3.spdf,resolution = 40000, 
                  varname = "pop2008",
                  typefct = "exponential", span = 100000,
                  beta = 3, mask = nuts3.spdf)
  )
  
  t2 <- system.time(
    s2 <- mcStewart(knownpts = nuts3.spdf, resolution = 40000, 
                     varname = "pop2008",
                     typefct = "exponential", span = 100000,
                     beta = 3, mask = nuts3.spdf, cl = 3, size = 500)
  )
  identical(s1, s2)
  cat("Elapsed time\n", "stewart:", t1[3], "\n parStewart:",t2[3])
  
  r2 <- rasterStewart(s2)
  c2 <- rasterToContourPoly(r = r2, breaks = c(0,1000000,2000000, 5000000,
                                               10000000, 20000000, 200004342), 
                            mask = nuts3.spdf)
  # cartography
  opar <- par(mar = c(0,0,1.2,0))
  bks <- sort(unique(c(c2$min, c2$max)))
  choroLayer(spdf = c2, var = "center", breaks = bks, border = NA, 
             legend.title.txt = "pop")
  layoutLayer("potential population", "","", scale = NULL)
  par(opar)
}
# }