kernelbb: Estimation of Kernel Brownian Bridge Home-Range
In adehabitat: Analysis of Habitat Selection by Animals

Description Usage Arguments Details Value Author(s) References See Also Examples

kernelbb is used to estimate the utilization distribution of an animal using the brownian bridge approach of the kernel method (for autocorrelated relocations; Bullard 1991, Horne et al. 2007).
liker can be used to find the maximum likelihood estimation of the parameter sig1, using the approach defined in Horne et al. 2007 (see Details).

kernelbb(tr, sig1, sig2, grid = 40, same4all = FALSE, byburst = FALSE,
         extent = 0.5, nalpha = 25)

liker(tr, rangesig1, sig2, le = 1000,
      byburst = FALSE, plotit = TRUE)

## S3 method for class 'liker'
print(x, ...)

`tr`	an object of class `ltraj` of type II (time recorded), regular or not (see `help(as.ltraj)`).
`sig1`	first smoothing parameter for the brownian bridge method (related to the speed of the animals; it can be estimated by the function `liker`).
`sig2`	second smoothing parameter for the brownian bridge method (related to the imprecision of the relocations, supposed known).
`grid`	a number giving the size of the grid on which the UD should be estimated. Alternatively, this parameter may be an object of class `asc`, or a list of objects of class `asc`, with named elements corresponding to each level of the factor id
`same4all`	logical. If `TRUE`, the same grid is used for all animals. If `FALSE`, one grid per animal is used
`byburst`	logical. Whether the brownian bridge estimation should be done by burst.
`extent`	a value indicating the extent of the grid used for the estimation (the extent of the grid on the abscissa is equal to `(min(xy[,1]) + extent * diff(range(xy[,1])))`).
`nalpha`	a parameter used internally to compute the integral of the Brownian bridge. The integral is computed by cutting each step built by two relocations into `nalpha` sub-intervals.
`rangesig1`	the range of possible values of sig1 within which the likelihood should be maximized.
`le`	The number of values of sig1 tested within the specified range.
`plotit`	logical. Whether the results of the function should be plotted.
`x`	an object of class `khr` returned by `kernelbb`.
`...`	additionnal parameters to be passed to the generic functions `print`

The function kernelbb uses the brownian bridge approach to estimate the Utilization Distribution of an animal with serial autocorrelation of the relocations (Bullard 1991, Horne et al. 2007). Instead of simply smoothing the relocation pattern (which is the case for the function kernelUD), it takes into account the fact that between two successive relocations r1 and r2, the animal has moved through a continuous path, which is not necessarily linear. A brownian bridge estimates the density of probability that this path passed through any point of the study area, given that the animal was located at the point r1 at time t1 and at the point r2 at time t2, with a certain amount of inaccuracy (controled by the parameter sig2, see Examples). Brownian bridges are placed over the different sections of the trajectory, and these functions are then summed over the area. The brownian bridge approach therefore smoothes a trajectory.

The brownian bridge estimation relies on two smoothing parameters, sig1 and sig2. The parameter sig1 is related to the speed of the animal, and describes how far from the line joining two successive relocations the animal can go during one time unit (here the time is measured in second). The function liker can be used to estimate this value using the maximum likelihood approach described in Horne et al. (2007). The larger this parameter is, and the more wiggly the trajectory is likely to be. The parameter sig2 is equivalent to the parameter h of the classical kernel method: it is related to the inaccuracy of the relocations, and is supposed known (See examples for an illustration of the smoothing parameters).

The functions getvolumeUD and getverticeshr can then be used to conpute the home ranges (see kernelbb). More generally, more details on the generic parameters of kernelbb can be found on the help page of kernelUD.

An object of class khr, subclass kbbhrud. This list has one component per animal (or per burst, depending on the value of the parameter perburst). Each component is itself a list, with the following sub-components:

`UD`	an object of class `asc`, with the values of density probability in each cell of the grid
`h`	a vector containing the values of sig1 and sig2.
`locs`	The relocations used in the estimation procedure.
`hmeth`	a character string "bb" (not used)

liker returns an object of class liker, with one component per animal (or per burst, depending on the value of the parameter perburst), containing the value of (i) optimized sig1, (ii) sig2, and (iii) a data frame named "cv" with the tested values of sig1 and the corresponding log-likelihood.

Clement Calenge clement.calenge@oncfs.gouv.fr

Bullard, F. (1991) Estimating the home range of an animal: a Brownian bridge approach. Master of Science, University of North Carolina, Chapel Hill.

Horne, J.S., Garton, E.O., Krone, S.M. and Lewis, J.S. (2007) Analyzing animal movements using brownian bridge. Ecology, in press.

as.ltraj for further information concerning objects of class ltraj. kernelUD for the classical kernel estimation. asc for additionnal informations on objects of class asc, mcp for estimation of home ranges using the minimum convex polygon, and for help on the function plot.hrsize. kver for information on objects of class kver

## Not run: 

#########################################################
#########################################################
#########################################################
###
###        Example of a typical case study
###       with the brownian bridge approach
###

## Load the data
data(puechcirc)
x <- puechcirc[1]

## Field studies have shown that the mean standard deviation (relocations
## as a sample of the actual position of the animal) is equal to 58
## meters on these data (Maillard, 1996, p. 63). Therefore
sig2 <- 58

## Find the maximum likelihood estimation of the parameter sig1
## First, try to find it between 10 and 100 meters
liker(x, sig2 = 58, rangesig1 = c(10, 100))

## Wow! we expected a too large standard deviation! Try again between
## 1 and 10 meters:
liker(x, sig2 = 58, rangesig1 = c(1, 10))

## So that sig1 = 6.23

## Now, estimate the brownian bridge
tata <- kernelbb(x, sig1 = 6.23, sig2 = 58, grid = 50)
image(tata, addpoints=FALSE)

## OK, now look at the home range
image(tata[[1]]$UD)
contour(getvolumeUD(tata)[[1]]$UD, level=95,
        add=TRUE, col="red", lwd=2)





#########################################################
#########################################################
#########################################################
###
###      Comparison of the brownian bridge approach
###            with the classical approach
###


## Take an illustrative example: we simulate a trajectory
set.seed(2098)
pts1 <- data.frame(x = rnorm(25, mean = 4.5, sd = 0.05),
                   y = rnorm(25, mean = 4.5, sd = 0.05))
pts1b <- data.frame(x = rnorm(25, mean = 4.5, sd = 0.05),
                    y = rnorm(25, mean = 4.5, sd = 0.05))
pts2 <- data.frame(x = rnorm(25, mean = 4, sd = 0.05),
                   y = rnorm(25, mean = 4, sd = 0.05))
pts3 <- data.frame(x = rnorm(25, mean = 5, sd = 0.05),
                   y = rnorm(25, mean = 4, sd = 0.05))
pts3b <- data.frame(x = rnorm(25, mean = 5, sd = 0.05),
                    y = rnorm(25, mean = 4, sd = 0.05))
pts2b <- data.frame(x = rnorm(25, mean = 4, sd = 0.05),
                    y = rnorm(25, mean = 4, sd = 0.05))
pts <- do.call("rbind", lapply(1:25, function(i) {
          rbind(pts1[i,], pts1b[i,], pts2[i,], pts3[i,],
                pts3b[i,], pts2b[i,])
}))
dat <- 1:150
class(dat) <- c("POSIXct","POSIXt")
x <- as.ltraj(pts, date=dat, id = rep("A", 150))

## See the trajectory:
plot(x)


## Now, we suppose that there is a precision of 0.05
## on the relocations
sig2 <- 0.05
## and that sig1=0.1
sig1 <- 0.1

## Now fits the brownian bridge home range
(kbb <- kernelbb(x, sig1 = sig1,
                 sig2 = sig2))


## Now fits the classical kernel home range
(kud <- kernelUD(pts))


###### The results

opar <- par(mfrow=c(2,2), mar=c(0.1,0.1,2,0.1))
plot(pts, pch=16, axes=FALSE, main="The relocation pattern", asp=1)
box()
plot(x, axes=FALSE, main="The trajectory")
box()
image(kud[[1]]$UD, axes=FALSE, main="Classical kernel home range")
contour(getvolumeUD(kud)[[1]]$UD, lev=95, col="red", add=TRUE)
box()
image(kbb[[1]]$UD, axes=FALSE, main="Brownian bridge home range")
contour(getvolumeUD(kbb)[[1]]$UD, lev=95, col="red", add=TRUE)
box()
par(opar)



###############################################
###############################################
###############################################
###
###         Image of a brownian bridge.
###          Fit with two relocations
###               


xx <- c(0,1)
yy <- c(0,1)
date <- c(0,1)
class(date) <- c("POSIXt", "POSIXct")
tr <- as.ltraj(data.frame(x = xx,y = yy), date, id="a")

## Use of different smoothing parameters
sig1 <- c(0.05, 0.1, 0.2, 0.4, 0.6)
sig2 <- c(0.05, 0.1, 0.2, 0.5, 0.7)

y <- list()
for (i in 1:5) {
  for (j in 1:5) {
     k <- paste("s1=", sig1[i], ", s2=", sig2[j], sep = "")
     y[[k]]<-kernelbb(tr, sig1[i], sig2[j])[[1]]$UD
   }
 }

## Displays the results
opar <- par(mar = c(0,0,2,0), mfrow = c(5,5))
foo <- function(x)
   {
     image(y[[x]], main = names(y)[x], axes = F)
     points(tr[[1]][,c("x","y")], pch = 16, col = "red")
   }
lapply(1:length(y), foo)

par(opar)



## End(Not run)