fit.sscr: Fitting an SCR model with second-order spatial dependence

Description Usage Arguments

View source: R/fit-sscr.r


Fits an SSCR model. Estimation is by maximum likelihood. The second-order spatial dependence is modelled via trap-level random effects for each detected individual. The likelihood function is calculated by integrating over these random effects using the Laplace approximation.


fit.sscr(capt, traps, mask, resp = "binom", = NULL,
  detfn = "hn", detfn.scale = "er", cov.structure = "none",
  re.scale = "er", start = NULL, toa = NULL, trace = FALSE,
  test = FALSE, test.conditional.n = TRUE, hess = FALSE, = "nlminb")



A capture history object. It should be a matrix, where the jth element of the ith row should provide a detection record of the ith individual at the jth detector.


A matrix with two columns, providing the Cartesian coordinates of the detector locations.


A mask object for integration over the survey area.


Response distribution for capture history elements. Either "binom" for a binomial distribution, or "pois" for a Poisson distribution.

A named vector of known, fixed parameters for the response distribution. If resp is "binom", then this must have a single element named "size" giving the fixed number of trials; if this argument is not provided, then the default is 1.


Detection function, given by a character string. Use "hn" for halfnormal and "hr" for hazard rate.


A character string, either "er" or "prob". This indicates whether the detection function should provide the encounter rate (expected number of detections) or the probability of detection.


Covariance structure of the random effects. The current options are (1) "none" for no random effects (regular SCR), (2) "independent", for independent random effects (equivalent to counts of detections being overdispersed), (3) "exponential", for random effects with an exponential covariance structure, (4) "sq_exponential" for random effects with a squared exponential covariance structure, (5) "matern", for random effects with a Matern covariance structure, (5) "individual", for random effects that are restricted to being the same at all traps (equivalent to having an independent random effect on lambda0 for each individual), or (6) "lc_exponential" for a linear combination of exponential covariance functions.


A character string, either "er" or "prob". This indicates whether the Gaussian random effects effect the encounter rate (expected number of detections) or the probability of detection.


A named list of parameter start values.


A matrix with the same dimensions as capt that provides time-of-arrival information for acoustic detections.


Logical. If TRUE, parameter values for each step of the optimisation algorithm are printed.


Logical. If TRUE, the negative log-likelihood is calculated at parameter start values. If FALSE, a model is fitted. Alternatively, a character string. If "nll", then the negative log-likelihood is calculated. If "gr", then the partial derivatives of the negative log-likelihood function with respect to the parameters is also calculated. If "hess" then the Hessian if also calculated.


Logical. If TRUE, tests are carried out conditioning in the number of detections.


Logical. If TRUE, a Hessian is computed and a variance-covariance matrix is returned.

A character string representing the R function to maximise the likelihood. This can either be "nlminb" or "nlm".

b-steve/sscr documentation built on March 23, 2018, 1:29 a.m.