dcrw: Correlated Random Walk Filter
In SWotherspoon/Marseille: Telemetry Track Filtering and Simulation

Description Usage Arguments Details Value

Correlated Random Walk Filter

dcrw(data, subset = rep(TRUE, nrow(data)), tstep = 6/24, nu = 10,
  gamma = 0.5, span = 0.1, verbose = FALSE, optim = c("nlminb",
  "optim"), extrap = FALSE, parameters = NULL, esf = aesfCRAWL(),
  theta.zero = FALSE, common.tau = FALSE)

`data`	A dataframe representing the track (see details)
`subset`	Logical vector indicating which rows of the data frame be kept in the filtering
`tstep`	The time step to predict to (in days)
`nu`	The degrees of freedom parameter for the location errors
`gamma`	The autocorrelation parameter used to estimate initial parameters.
`span`	The span parameter for the loess fits used to estimate initial locations.
`verbose`	Enable tracing information
`optim`	The function used to minimize the negative log likelihood.
`extrap`	If TRUE, the final predicted state occurs immediately before the last observation, otherwise the final predicted state occurs immediately after the last observation.
`parameters`	The TMB parameter list.
`esf`	The error scale factors for the location classes.
`theta.zero`	Should theta be fixed at zero.
`common.tau`	Should a common tau parameter be fitted for longitude and latitude.

Fit a correlated random walk to filter a track and predict locations on a regular time step.

The input track is given as a dataframe where each row is an observed location, with columns

'date': observation time (as GMT POSIXct),
'lon': observed longitude,
'lat': observed latitude,
'lc': location class.

The TMB parameter list can be specified directly with the parameters argument. Otherwise suitable parameter values are estimated from predicted locations estimated by fitting loess smooths to the raw observations.

If extrap is TRUE, the last observations occur after the last predicted location and the last fitted locations are extrapolations, otherwise the final observations occur before the final predicted locations and all fitted locations are interpolated.

The filtering model assumes the errors in longitude and latitude are proportional to scale factors determined by the location class. The scale factors are speficied through the aes argument. By default the function uses the same scaling factors for location accuracy as used in crawl for ARGOS data.

a list with components

`predicted`	a dataframe of predicted locations
`fitted`	a dataframe of fitted locations
`par`	model parameter summary
`data`	the input dataframe
`subset`	the input subset vector
`tstep`	the prediction time step
`theta.zero`	has theta been fixed at zero
`common.tau`	has a common tau been fitted for lon and lat
`opt`	the object returned by the optimizer
`tmb`	the TMB object