SSN (version 1.1.7)

EmpiricalSemivariogram: Empirical Semivariogram Based on Euclidean Distance

Description

EmpiricalSemivariogram computes the empirical semivariogram from the data based on Euclidean distance.

Usage

EmpiricalSemivariogram(ssn.object, varName, nlag = 20, directions = c(0, 45, 90, 135),  tolerance = 22.5, inc = 0, maxlag = 1e+32, nlagcutoff = 1, EmpVarMeth = "MethMoment")

Arguments

ssn.object

an object of class SpatialStreamNetwork-class or influenceSSN-class

varName

a response or residual variable name in the data.frame of observed data in the SpatialStreamNetwork or influenceSSN object.

nlag

the number of lag bins to create, by direction if directions are specified. The distance between endpoints that define a bin will have equal lengths for all bins. The bin sizes are then determined from the minimum lag in the data, and the specification of maxlag.

directions

directions in degrees clockwise from north that allow lag binning to be directional. Default is c(0, 45, 90, 135). Values should be between 0 and 180, as there is radial symmetry in orientation between two points.

tolerance

the angle on either side of the directions to determine if a pair of points falls in that direction class. Note, a pair of points may be in more than one lag bin if tolerances for different directions overlap.

inc

the distance increment for each bin class. Default is 0, in which case maxlag and nclasses determine the distance increments.

maxlag

the maximum lag distance to consider when binning pairs of locations by the hydrologic distance that separates them. If the specified maxlag is larger than the maximum distance among pairs of points, then maxlag is set to the maximum distance among pairs. If inc is greater than 0, then maxlag is disregarded.

nlagcutoff

the minimum number of pairs needed to estimate the semivariance for a bin. If the sample size is less than this value, the semivariance for the bin is not calculated.

EmpVarMeth

method for computing semivariances. The default is "MethMoment", the classical method of moments, which is just the average difference-squared within bin classes. "Covariance" computes covariance rather than semivariance, but may be more biased because it subtracts off the simple mean of the response variable. "RobustMedian" and "RobustMean" are robust estimators proposed by Cressie and Hawkins (1980). If v is a vector of all pairwise square-roots of absolute differences within a bin class, then RobustMedian computes median(v)^4/.457. "RobustMean" computes mean(v)^4/(.457 + .494/length(v)).

Value

A list of six vectors. The lengths of all vectors are equal, which is equal to nlag*(number of directions) - (any missing lags due to nlagcutoff).

distance

the mean Euclidean distance separating pairs of sites used to calculate the semivariance for each bin

gamma

the estimated semivariance for each bin, based on EmpVarMeth

np

the number of pairs of sites used to calculate the semivariance for each bin

azimuth

the azimuth, equivalent to the direction, used for the bin class

hx

the x-coordinate of the center of the bin lag.

hy

the y-coordinate of the center of the bin lag.

Examples

library(SSN)
# NOT RUN 
# mf04 <- importSSN(system.file("lsndata/MiddleFork04.ssn", 
#	        package = "SSN"), o.write = TRUE)
# use SpatialStreamNetwork object mf04 that was already created
data(mf04)
#make sure mf04 has the correct path, will vary for each users installation
mf04 <- updatePath(mf04, system.file("lsndata/MiddleFork04.ssn", package = "SSN"))
    
# Compute the empirical semivariogram for the raw data. 
# the number of pairs used to estimate the semivariance
EVout <- EmpiricalSemivariogram(mf04, "Summer_mn", directions = 0, tolerance = 180,
    nlag = 10)
# Plot it and set the point size relative to the number of pairs used to estimate
# the semivariance
plot(EVout$distance, EVout$gamma, pch = 19, cex = EVout$np/100, ylim = c(0,6),
    ylab = "Gamma", xlab = "Distance", col = "blue", 
    main = "Empirical Semivariogram - Raw Data")

# generate and plot an empirical semivariogram based on model residuals
data(modelFits)
#make sure fitSP has the correct path, will vary for each users installation
fitSp$ssn@path <- system.file("lsndata/MiddleFork04.ssn", package = "SSN")
fitSpRes <- residuals(fitSp)
names(getSSNdata.frame(fitSpRes))
EVresid <- EmpiricalSemivariogram(fitSpRes, "_resid_", directions = 0, 
    tolerance = 180, nlag = 7,	maxlag = 15000,)
plot(EVresid$distance, EVresid$gamma, pch = 19, cex = EVresid$np/50, ylim = c(0,6),
    ylab = "Gamma", xlab = "Distance", main = "Empirical Semivariogram - Residuals")