data2oscr: Convert standard SCR data file(s) to oSCR data objects

In jaroyle/oSCR: Multi-Session Sex-Structured Spatial Capture-Recapture Models

Convert standard SCR data file(s) to oSCR data objects

Description

This will take input files similar to those used for the R package 'secr' and convert them to list objects used by oSCR.

Usage

data2oscr(edf, sess.col = 1, id.col = 2, occ.col = 3, trap.col = 4, sex.col = NULL, tdf = NULL, K = NULL, ntraps = NULL, remove.zeros = TRUE, trapcov.names=NULL,remove.extracaps = TRUE, sex.nacode = NULL, tdf.sep = "/")

Arguments

edf	MATRIX – NOT A LIST. Encounter data file having basic information about session, individual, occasion, trap and (optional) sex. Note: if the session column indicates multiple sessions then the result will be to produce a list of appropriate length. e.g., if session has values (1,2) then the output lists will all be of length 2. If you have session indicating "sex" you probably don't want this to happen so add a new column for "sex" and keep that information independent of "session".
sess.col	Column index (integer) of session info
id.col	Column index (integer) of individual ID
occ.col	Column index (integer) of occasion
trap.col	Column index (integer) of trap
sex.col	Optional column index (integer) of sex
tdf	LIST: one for each session. Each element should be a trap deployment file. Necessary format: Trap ID, Xcoord, Ycoord, trap deployment columns, character symbol, trap covariates Optional columns: character symbol is any character symbol which indicates subsequent columns represent covariates. IMPORTANT NOTE: The column names for the trap deployment columns (cols >= 4) must have the same NAMES as the traps referenced in the encounter data file (EDF).
K	Vector number of occasions per session. K = c(8) for a single season study with 8 occasions.
ntraps	Vector number of traps per session. ntraps = c(38) for a single season study with 38 traps.
remove.zeros	If TRUE then individuals that were never captured in a session are removed from the session encounter history data. If FALSE then the resulting data structure preserves individual ID across sesions. For example in a 2 session data set then output$y3d[[1]][7,,] is the same individual as output$y3d[[2]][7,,]. For fitting models in oSCR you should have remove.zeros=TRUE but if you want to fit open SCR models in JAGS you might use the remove.zeros = FALSE option so that the rows are alighned properly.
trapcov.names
remove.extracaps	If TRUE then individuals captured > 1 time in a trap at an occasion have their encounter frequency reduced to 1.
sex.nacode	Which character codes represent unknown sex?
tdf.sep	Which character indicates that subsequent columns should be understood as trap covariates?

Details

This function takes one or more encounter data files and one or more trap deployment files and creates list objects suitable for use by oSCR.fit.

It is best to make sure that you have a proper TDF for EACH EDF that is provided.

The EDF should be a single EDF with sessions differentiated by a column in the EDF.

The TDF should be provided in list form. One element per session.

Value

Returns a list with elements

edf	the encounter data file
y3d	a list of 3-d encounter history matrices, one for each session
sex	a list of sex information suitable for inputting into oSCR.fit
traplocs	a list of trap locations suitable for inputting into oSCR.fit
trapopp	a list of trap operation information suitable for inputting into oSCR.fit
trapcovs	a list of trap covariates suitable for inputting into oSCR.fit
scrFrame	the scrFrame suitable for use in oSCR.fit.

Author(s)

Andy Royle

Examples

# Use data2oscr to process the black bear data (no covariates, no trap operation information)
library(oSCR)
data(beardata)
data<- data2oscr(beardata$edf,
                 sess.col = 1,
                 id.col = 2,
                 occ.col = 3,
                 trap.col = 4,
                 tdf = list(beardata$tdf[,1:3]),
                 K = c(8),
                 ntraps = c(38))
sf <- data$scrFrame
ssDF <- make.ssDF(sf, buffer=3, res = 0.5)
out1 <- oSCR.fit(model=list(D~1,p0~1,sig~1), scrFrame=sf, ssDF=ssDF,plotit=FALSE , trimS=8)
####pred<- predict.oSCR(out1)


# Approach 2
data <- data2oscr( edf = beardata$edf,
tdf = list(beardata$tdf),
sess.col = which(colnames(beardata$edf) 
id.col = which(colnames(beardata$edf) 
occ.col = which(colnames(beardata$edf) 
trap.col = which(colnames(beardata$edf) 
sex.col = which(colnames(beardata$edf) 
sex.nacode = "NA",
K = 8,
ntraps = nrow(beardata$tdf))

# Ocelot data

data(ocelot)
 # need to have occasions labeled same as in TDF
colnames(tdf1) <- c("Detector","X","Y", 1:44)
colnames(tdf2) <- c("Detector","X","Y", 1:96)

# This fails because one TDF is given and the data imply DIFFERENT # occasions
data<-data2oscr(rbind(edf1,edf2),sess.col=1, id.col=2, occ.col=3, trap.col=4, sex.col=5,   
             ntraps=c(23,23),
             tdf=list(tdf1),
             remove.zeros=TRUE, remove.extracaps=TRUE,sex.nacode=c("U") )

# This works. Always best to input K and ntraps!
data<-data2oscr(rbind(edf1,edf2),sess.col=1, id.col=2, occ.col=3, trap.col=4,sex.col=5, K=c(44,96),   ntraps=c(23,23),
             tdf=list(tdf1,tdf2),
             remove.zeros=TRUE, remove.extracaps=TRUE,sex.nacode=c("U") )

# There is one covariate "distance to disturbance" which is not defined for session = 1
#    therefore I set it to 0 here. This is probably ok if we define the "exposure" covariate to be 1/dist
#    because then as "dist -> 0" (i.e., close to disturance) then exposure -> BIG whereas if dist -> infinity then
#    exposure -> 0.  Therefore 0 values are like "no exposure to disturbance" which is a good description of
#    pre-exposure.


# The covariate "distance" is in the input TDF file as a trap covariate. The function data2oscr processes that into a matrix
#  but it does NOT exist for session 1!
dist1 =  rep(0,23)
dist2 =  as.vector(data$trapcovs[[2]])/1000   # Convert to km
# Organize the covariates into a matrix that is J x K (traps x occasions)
dist1 <- matrix(dist1, nrow=23,ncol=44,byrow=FALSE)
dist2 <- matrix(dist2, nrow=23,ncol=96, byrow=FALSE)
# Transformation to exposure to disturbance. Scaling is pretty bad, some extreme values
exposure1<- dist1    # This is 0 for all traps
exposure2<- 1/dist2  # This goes to 0 as distance increase

# Package these into a suitable trapCovs format. Example of 2 covariates, defined for each session here:
tc<-  list(expose=list(exposure1, exposure2), expose2 = list(exposure1^2, exposure2^2) )

tc<-  make.trapCovs(tc)
# Example of 1 covariate
# "list of lists" format
tc <-  list(expose=list(exposure1, exposure2))
tc <-  make.trapCovs(tc)
# example of 2 covariates
#tc<-  list(dist=list(dist1, dist2) , dist2=list(dist1,dist2) )
#tc<- make.trapCovs(tc)

# Make the SCRframe. Sex is not included here. Either way is fine....
ocelot.sf  <-  make.scrFrame(caphist=data$y3d, indCovs=NULL,
                    traps=data$traplocs,trapCovs=tc ,
                    trapOperation=data$trapopp )


# Make a state-space
ocelot.ss <- make.ssDF(ocelot.sf, buffer=2000, res = 600)
plot(ocelot.ss[[1]], pch=20)
points(ocelot.sf$traps[[1]],pch=3,col="red",lwd=2)
spiderplot(y = data$y3d[[2]],traplocs = data$traplocs[[1]] )
spiderplot(y = data$y3d[[2]],traplocs = data$traplocs[[1]],add=TRUE)

##  fit a model
out1 <- oSCR.fit(model=list(D~1,p0~1,sig~1), ocelot.sf, ssDF=ocelot.ss,plotit=FALSE , trimS=2000)

## Good starts, improves run-time (can make a huge difference in complex models)
out1c <- oSCR.fit(model=list(D~1,p0~1,sig~1), ocelot.sf, start.vals=c(-3.4, 6.4, -0.58),
                 ssDF=ocelot.ss,plotit=FALSE, trimS=2000 )

# Fit a model with "exposure to disturbance"
## default vals for some models are especially bad. Could be geographic scaling issue.
out5 <- oSCR.fit(model=list(D~1,p0~ expose ,sig~1), ocelot.sf, start.vals=c(-3.5, 6.4, 0.035, -1.5),
                 ssDF=ocelot.ss, plotit=FALSE, trimS=2000  )

jaroyle/oSCR documentation built on Sept. 23, 2023, 12:46 p.m.