Changes in version 2.3.2 (2012-05-15)

New features or changed behaviour of old functions

    o   new function randomHabitat() generates random binary landscape
    that may be used with sim.popn()

    o   secr.fit() 'biasLimit' argument sets threshold of bias for
    buffer-width check, or turns off bias checking (biasLimit = NA)

    o   mash() now works with signal detectors

    o   secr.design.MS() trial with 3-D usage

    o   read.traps() now accepts multiple covariate fields after '/'

    o   plot.traps() new argument `labelclusters' to label points at
    level of cluster rather separate traps

    o   detector covariates may vary within a session. This uses a new
    traps attribute 'timevaryingcov', and entails changes to
    secr.design.MS()

    o   subset.capthist() new argument 'droplowsignals' to apply a
    new signal threshold at the level of whole sounds rather than specific
    microphones, and verify() check relaxed to allow such data

    o   new detector type 'signalnoise' for paired measurements of
    signals and background noise (not fully documented)

    o   capthist objects whose detector is of type 'signalnoise' include
    a 'noise' attribute with extraction and replacement methods like
    'signal'. make.capthist() has been adapted to accept noise
    measurements in the field after signal (i.e. 6 or 7) and to accept
    further signal covariates in other columns (not fully documented)

    o   derived.nj() gives model-based variance estimates when nj is
    scalar (length 1); new argument 'area' for binomial-n variance
    calculation

    o   sim.popn() argument 'D' may now specify a covariate in the
    'core' mask, or a constant, to use for cell-specific density with
    model2D = 'IHP'

    o   make.mask() acquires new argument 'ny' used only with new type
    'rectangular'

    o   pointsInPolygon() accepts mask as polygon input, and acquires
    new argument 'logical'

    o   print.capthist() adjusted to allow NULL detector
    (i.e. nonspatial data)

    o   unRMarkInput() converts RMark dataframe to single-session
    capthist object.

Bugs fixed

    o   plot.capthist() 'ncap' argument did not work with some detectors

    o   make.capthist() scrambled the order of ancillary data (signal
    strength, x-y coordinates) when input not sorted by animalID; this
    may have prevented the fitting of models for signal-strength or
    polygon detectors (also affects read.capthist(), read.DA(),
    read.SPACECAP() and mash() as these call make.capthist()).

    o   subset.capthist() lost mash-related attributes of the input
    (n.mash, centres)

    o   secr.fit() failed with fixed real parameters in version 2.3.1

    o   make.mask() 'boundingbox' attribute sometimes out by spacing/2

    o   region.N() 'nlowerbound' and 'RN.method' argements ignored when nsess>1

    o   autoini() could fail when all detections at site of first capture

    o   predict.secr failed with multi-session user-supplied 'newdata' not
    containing 'session' (still required for mashed data)

Known bugs

    o   secr.fit() with signal detector: it is not effective to vary
    cutval only via the 'details' argument - use subset.capthist on the
    input data

Other

    o the 'possum' dataset was found to include animals that had been
    re-tagged after previously losing a tag (see ?possum). 

Changes in version 2.3.1 (2011-12-20)

New features or changed behaviour of old functions

    o   secr.fit() new behavioural response models bk, Bk, k, K (e.g. g0
    ~ bk for trap-specific learned response); sim.secr() modified for
    these models

    o   join() new function for making single-session capthist from
    multiple inputs, joining them 'side by side' (number of new
    occasions = sum of old); unjoin() reverses this.

    o   RMarkInput() new function to create object for input to RMark

    o   reduce.capthist() argument 'by' may take the character value
    "all" indicating that all columns (occasions) should be collapsed
    to one.

    o   secr.fit() "backdoor" entry of predictors via the 'dframe'
    argument is now enabled (not for groups)

    o   ip.secr() new argument 'maxtries' for greater stability with
    sparse datasets, and related changes, but see Warning on ip.secr
    help page.

Bugs fixed

    o   secr.fit() failed when detector usage was specified for some
    sessions, but not all, and possibly with other multi-session data

Other
    o   C code tweaked to avoid warnings


Changes in version 2.3.0 (2011-11-18)

New features or changed behaviour of old functions

    o   predictDsurface() predicts height of fitted density surface
    across a mask

    o   `Dsurface' new S3 class with methods head, tail, plot, print and
    summary

    o   head() and tail() new methods for mask, traps and capthist objects

    o   spotHeight() new function for interrogating the density or
    other covariates of a plotted Dsurface or mask

    o   secr.fit() fits multi-session spatial trend models using mask
    covariates

    o   distancetotrap() no longer requires second argument to have
    'detector' attribute (i.e. may be matrix of coordinates other than
    a 'traps' object)

    o   region.N() argument RN.method = 'poisson' uses faster variance
    computation, and parts of the function have been re-coded; E(n) is
    no longer computed by region.N - see expected.n() instead.

    o   region.N() population size is now adjusted automatically for the
    number of clusters in `mashed' models; it relates to a single
    cluster unless 'regionmask' is specified

    o   make.mask() has new logical argument 'poly.habitat' to determine
    whether polygons represent habitat or non-habitat, allowing
    exclusion of designated areas.

    o   read.mask() now saves covariates from the input file or
    dataframe, and has new argument 'columns' to select the columns to
    save.

    o   plot.mask() acquires argument 'meshcol' for colour of pixel
    borders when dots = FALSE (replaces manipulation of par()$fg)

    o   rectangularMask() new function expands an irregular mask to fill
    its bounding box (useful for plotting covariate contours; used
    internally by plot.Dsurface)

    o   sim.popn() and subset.popn() acquire 'poly' argument to
    constrain points to a lie within or outside polygon(s); may be
    matrix of coordinates or SpatialPolygonsDataFrame object from
    package sp. The subset.popn() method also acquires arguments
    'poly.habitat' and 'keep.poly'.

    o   pointsInPolygon() new function (previously internal as
    'insidepoly'). Polygon may be matrix of x-y coordinates or
    SpatialPolygonsDataFrame object from package 'sp'.

    o   secr.fit() allows the user to specify a spatial model for
    density (i.e. the intensity surface of an inhomogeneous Poisson
    distribution) via an external R function instead of the usual
    linear model. The function is passed as component 'userDfn' in the
    'details' argument of secr.fit(). See vignette
    `secr-densitysurfaces.pdf'.

    o   an secr object created by secr.fit() no longer includes D, the
    predicted density at each mask point. Instead it has a component N
    for the summed density of each group and session. deviance.secr()
    has been changed to use this for the N of binomial models. D no
    longer in default drop vector of trim().

    o   secr.make.newdata() now uses values from covariates(mask) to
    construct default newdata for predict.secr (previously set to
    zero, which could cause an error with factor covariates)

    o   tile() replicates popn object on 3 x 3 grid

    o   addCovariates() new function to add spatial covariates to a mask
    or traps object from an ESRI shapefile or other spatial data
    source

    o   subset.capthist() argument 'dropnull' replaced with separate
    arguments 'dropnullCH' and dropnullocc'; 'traps' argument may be
    character vector; other recoding for greater clarity and reliability


Bugs fixed

    o   region.N() sometimes computed SE even when se.N = FALSE

    o   subset.mask() output bounding box was too small by spacing/2

    o   model.average() did not work with covar = TRUE

    o   secr.fit() models with detector-level covariates of detection
    parameters were not fitted. Fix required substantial re-write of C
    code for log likelihood.

Other
    o   the mask and fitted models packaged with possum data have
    changed

    o   new vignette 'secr-densitysurfaces.pdf' describes fitting,
    evaluation and plotting of density surfaces, including the use of
    user-defined model functions.

    o   many small changes to documentation

Changes in version 2.2.0 (2011-10-01)

New features or changed behaviour of old functions

    o   rbind.capthist() has new logical argument 'verify', and along
    with MS.capthist(), has been considerably revised with assistance
    from Mike Meredith. Its behaviour has changed with respect to
    pooling of covariates, coordinates (xy) and signal strength.

    o   experimental 'presence' detector type (undocumented)

    o   secr.fit() allows more than one parameter to be fixed

    o   sim.capthist() has new argument 'p.available' for incomplete
    availability (random or Markovian between sessions) (applies only
    to multi-session sampling of a single population)

    o   levels of polyID and transectID factors retain order of input
    dataframe in make.poly() and make.transect()

    o   sim.capthist() for transects now handles uniform detection
    function (detectfn = 4) separately (without integration, which is
    faster)

    o   simulate.secr(), sim.secr() and sim.capthist() have new argument
    'maxperpoly' for the user to control memory allocation when
    sampling areas and transects

    o   the lognormal confidence interval for realised N returned by
    region.N() is by default now based only on unobserved fraction of
    population, so the lower limit cannot be less than n (the number
    observed). The original behaviour may be restored by setting the
    new argument 'nlowerbound'.

    o   expected.n() default for bycluster argument changed to FALSE,
    and other minor changes to match help page

    o   a subset method has been provided for popn objects

    o   verify.capthist() checks that rownames are not duplicated

    o   bad parameter values* (detection parameters or density) during
    likelihood maximisation trigger warnings but should not abort
    estimation. *These can result from the nlm optimiser 'trying'
    invalid parts of the parameter space, and can safely be ignored if
    the convergence code of the ultimate fit is zero.

Other

    o   'unmarked' is no longer an extraction function because unmarked
    data are now stored in a 3-D array, like proximity detection
    histories

    o   the external C code has been substantially re-arranged

Bugs fixed

    o   transectlength() and searcharea unreliable when multiple
    transects or polygons, and did not handle multiple sessions (now
    return a list in this case)

    o   make.mask() trivial bug checking class of polygon

    o   expected.n() bycluster failed with multisession capthist

    o   sim.capthist() multiple sessions not assigned distinct names

    o   rbind.capthist() and MS.capthist() failed with some inputs
    (concatenating single and multiple sessions, pooling single and
    multiple sessions)

    o   region.N() failed for some multi-session inputs

    o   plot.popn() with 'circles' option gave error when no animals in
    population; now plots a blank map


Known bugs

    o   confint.secr (profile likelihood intervals) not working for
    multi-session inputs, and maybe other complex models

Changes in version 2.1.0 (2011-06-17)

New functions

    o   circular.r() reports the radius (quantile) containing a
    specified fraction of a bivariate distribution; circular.p() is
    its inverse (fraction within specified radius)

    o   trap.builder() builds complex detector arrays. Combined with
    make.grid() etc., it provides capability similar to the Trap
    Builder in Density. make.systematic() generates rectangular arrays
    of detector clusters with a random origin.

    o   region.N() calculates the expected number of individuals in a
    user-defined region from a fitted secr model

    o   expected.n() calculates the expected number of detected
    individuals given a fitted secr model; may be stratified by
    cluster.

    o   writeGPS() uploads detector coordinates to a connected GPS
    (assumes GPSBabel software installed).

    o   clusterID(), clusterID<-, clustertrap(), and clustertrap<- for
    extracting and replacing cluster attributes of a compound traps
    object (clusterID() and clustertrap() also work with capthist
    objects)

    o   derived.cluster(), derived.mash(), derived.external(),
    derived.nj(), derived.session() for empirical variance estimation
    from various designs (replacing empirical.varD())

    o   mash() to collapse multi-cluster capthist objects. Mashed input
    is 'unmashed' by the predict method for secr objects, derived()
    and derived.mash()

    o   cluster.counts() returns the number of individuals detected per
    cluster in a single-session multi-cluster capthist object

    o   cluster.centres() returns the x- and y-coordinates of cluster
    centres in a single-session multi-cluster traps object

    o   alongtransect() returns the distance of each detection along the
    transect with which it is associated (transect data only)

New features or changed behaviour of old functions

    o   reduce.capthist() has new argument 'by' for grouping occasions

    o   sim.capthist() has a new argument 'nsessions' for simulating a
    multiple session capthist object, and the existing option of
    providing in argument 'popn' a list of populations to sample has
    been documented (this also generates multi-session output)

    o   sim.popn() argument 'model2D' has a new options 'coastal' and
    'hills'. 'coastal' simulates a density gradient (Fewster and
    Buckland 2004 ADS Ch 10)

    o   subset.traps() has new argument 'occasions' to select
    columns from the usage matrix; also default NULL value of 'subset'
    argument causes all rows to be selected.

    o   plot.popn() has new argument 'circles' for plotting a circle of
    given radius (in metres) around each animal location

    o   plot.capthist() new argument 'type' is used to specify new
    colour-coded frequency plots 'n.per.detector', 'n.per.cluster'

    o   make.mask() argument 'type' has new options 'clusterrect' and
    'clusterbuffer' for efficient masks around dispersed clusters of
    detectors; also new argument 'check.poly' for whther to warn if
    traps lie outside polygon

    o   esa() new argument 'noccasions' to compute effective sampling
    area when the number of occasions differs from the fitted model
    (assuming constant detection parameters)

    o   derived() output fields 'varcomp1' and 'varcomp2' have been
    replaced with their more familiar equivalents 'CVn' and 'CVa', and
    'CVD' is added

    o   animalID() returns a zero-length character value instead of ''
    when argument has no detections, for consistency with similar
    functions

    o   polyarea() extended for polygon classes from package 'sp', using
    package 'rgeos'

    o   read.mask() has 'data' argument (like read.traps) for mask input
    directly from a dataframe

    o   read.capthist() and make.capthist() have new argument 'noncapt'
    that allows a session to be recorded in which there were no
    detections

    o   logmultinom() returns 0 if there were no detections

    o   secr.fit() recalculates the multinomial coefficient for each
    likelihood evaluation. This is more flexible, but less efficient,
    and it is possible that reported log likelihoods will differ from
    previous versions

    o   summary.capthist() has new argument 'terse' for summary table
    from multi-session capthist objects

    o   plot.traps() has new argument 'markused' to show detectors used on
    at least one occasion

Other

    o   LazyDataCompression: xz

    o   New detector type 'unmarked' partially implemented (counts of
    unmarked animals per detector); functions unmarked() and unmarked()<-

    o   Allow zero detections per session

    o   flip, shift and rotate methods for popn objects

    o   ovenbird dataset includes additional fitted model
    ovenbird.model.D for trend in density across years

    o   'Not run' code in Examples checked for errors

Bugs fixed

    o   ip.secr() not working

    o   mask.check() sometimes failed when asked to compute likelihoods

    o   write.traps() header in output file faulty when covariates
    present

    o   sim.popn() failed when population was empty; now returns popn
    with zero rows

    o   plot.traps() 'markvarying' argument showed all used detectors

    o   plot.mask() 'add' argument was ignored

    o   make.grid() with hollow = TRUE and nx<3 or ny<3 caused an error

    o   subset.capthist() could fail with trapping data (single, multi)
    when unused trap sites were dropped, because trap numbers were not
    reassigned

    o   sim.capthist() with renumber = FALSE mislabelled rows in the
    capthist object

    o   closedN() failed when used with non-ML estimators alone

    o   distancetotrap() had ceased to work with non-trap objects

    o   rbind.capthist() ignored xy and signal, so failed with polygon,
    transect or acoustic data

    o   mask.check() failed in the unusual case that a detector was on
    the edge of the fitted mask and 'buffers' defaulted to a multiple
    of the buffer inferred from that mask

Changes in version 2.0.0 (2011-02-21)

New functions

    o   sort method for capthist objects (sort.capthist)

    o   alive() extracts the status of an animal at each detection

    o   fxi.secr(), fxi.contour(), fxi.mode() for estimating the
    location of an individual with a given detection history (these
    require a fitted secr model)

    o   read.DA(), write.DA() convert polygon data to and from BUGS
    format for data augmentation & MCMC

New features or changed behaviour of old functions

    o   new likelihood implementation and many improvements to the
    functionality of 'polygon' and 'transect' detectors; also new
    vignette 'secr-polygondetectors.pdf'

    o   new detector types 'polygonX' and 'transectX' for
    non-independent (exclusive) polygon and transect searches. Animals
    can be detected on at most one exclusive polygon or transect on
    any occasion

    o   in secr.fit() and sim.capthist() binomN has been elevated to a
    named argument from a component of details (secr.fit) or detectpar
    (sim.capthist). It is used to specify the discrete distribution
    for counts. The default value of binomN is Poisson for count-type
    detectors ('count', 'polygon', 'transect')

    o   secr.fit() logical values TRUE and FALSE for argument
    details$hessian may be used instead of 'auto' and 'none'

    o   secr.fit() check on buffer width not performed when 'verify = FALSE'

    o   secr.fit() allows user to choose Gardner & Royle
    parameterisation of detection function for multi-catch detectors
    (details = list(param = 1))

    o   secr.fit() triggers warning when maximization fails

    o   make.poly() and make.transect() have new argument 'exclusive' to
    create traps objects with detector types polygonX and transectX

    o   verify.traps() now checks that polygons are not concave in
    east-west direction

    o   make.poly() has new argument 'verify' to enforce shape check

    o   log likelihood reported for fitted SECR models includes the
    multinomial coefficient only for detector types 'multi' (including
    'single'), 'proximity' and 'count'. The reported log likelihood
    for 'signal' detectors will differ from previous versions by a
    data-specific constant (given by logmultinom())

    o   reduce.capthist() now retains row names from input (function has
    been re-written and may show other subtle changes)

    o   write.capthist(), write.traps() and write.captures() now export
    covariates if requested. Also note changed position of dots argument.

    o   secr.fit() the component 'distribution' of argument 'details'
    may take a numeric value larger than nrow(capthist), rather than
    'binomial' or 'poisson'. The likelihood then treats n as a
    binomial draw from a superpopulation of this size, with
    consequences for the variance of density estimates (cf data
    augmentation)

    o   make.capthist() and read.capthist() now coerce only character
    covariates to factors; numeric covariates are left as numeric.

    o   print.traps no longer adds a column for polygon or transect polyID

New dataset

    o   flat-tailed horned lizard polygon search from Royle & Young
    (2008) (thanks to Andy Royle and Kevin Young for assistance with
    this)

Bugs fixed

    o   read.table() sometimes failed when reading covariates -
    mechanism now more robust

    o   summary.traps() lost summary of usage and covariates

    o   make.capthist() and read.capthist() stored signal strengths
    incorrectly.

    o   plot.capthist() labelled first captures incorrectly when used
    with 'polygon' detectors; also several other bugs fixed in the
    course of testing polygon detectors, and now fails gracefully
    when the palette is inadequate for 'varycol'

    o   simulate.secr() ignored 'distribution' of fitted secr model
    (binomial or poisson), with consequences for sim.secr() (always
    Poisson)

    o   moves() did not work with transect data

    o   summary.capthist() and counts() mis-reported the numbers of
    detections and detectors visited for proximity and count detectors
    when losses > 0 (no other known effects)

    o   plot.secr() limits = TRUE not working

    o   secr.fit() non-default settings of 'method' did not work
    (e.g. method = 'BFGS')

Other

    o   check for too few observations when computing AICc (suggested by
    Mike Meredith)

    o   removed incorrect line ZipData: yes from package Description file

    o   removed multinomial term from polygon, transect and signal likelihoods


Changes in version 1.5.1 (2010-12-02)

Changed defaults 

    o   in secr.fit() the distribution fitted to counts now defaults to
    Poisson rather than Bernoulli (i.e. by default details$binomN = 0
    when detector(traps(capthist)) = 'count')
    o   plot.secr() confidence limits are computed only if requested
    explicitly ( default: limits = FALSE)
    o   the argument `buffer' in secr.fit() now defaults to NULL; a
    value of 100 is substituted if it is needed (i.e. neither mask nor
    buffer specified). This allows a warning to be given when the
    user relies on the default; it has no other effect.
    o   in secr.fit() the default detection parameter values when
    detectfn = 9 are now b0 = 2 and b1 = -0.1; this works for the
    dataset 'captdata', but in general you need to set 'start'.
    o   pdot() default 'detectpar' is now list(g0 = 0.2, sigma = 25, z =
    1) (consistent with sim.capthist())

Changed output

    o   the single value of 'esa' returned by derived() is now the
    harmonic mean (reciprocal of arithmetic mean of reciprocals)
    rather than the arithmetic mean, following a suggestion by David
    Borchers. This matters only when esa varies between individuals
    due to the inclusion of individual covariates in the detection
    model. Density estimates are unchanged.

New functions

    o   mask.check() evaluates the effect of varying buffer width and
    mask spacing on either the likelihood or density estimates from
    secr.fit(). This implements and extends the functionality of
    'Tools | ML SECR log likelihood' in Density 4.
    o   esa.plot() displays the effect of mask buffer width on effective
    sampling area (esa) for fixed values of detection
    parameters. Examine the asymptotic curve to choose a buffer width
    for model fitting. 
    o   suggest.buffer() proposes a buffer width to keep the
    buffer-related relative bias of D-hat within a bound
    o   detectpar() extracts the central estimates of detection function
    parameters from a fitted model as a list
    o   pdot.contour() display contours of the net probability of
    detection p.(X)
    o   buffer.contour() adds a conventional 'boundary strip' to a
    detector (trap) array, where the argument 'buffer' represents the
    strip width(s)
    o   utility functions polyarea() to calculate the area of a simple
    polygon and ms() to indicate whether an object has data from
    multiple sessions
    o   methods spacing.default() and 'spacing<-' to extract or replace
    mean detector spacing attribute of a 'traps' or 'mask' object (or compute if 
    NULL)
    o   methods AIC.secrlist(), plot.secrlist() and predict.secrlist()
    for lists of fitted models that have been assigned the class
    c('list','secrlist')
    o   method logLik.secr() for consistency with standard model fitting
    functions

New features in old functions

    o   closedN() now computes AICc model weights for models within dmax
    of the best model.
    o   new detection functions (7) 'cumulative lognormal' and (8)
    'cumulative gamma' are alternative 3-parameter functions. (8)
    lacks the heavy tail of (1) 'hazard rate'.
    o   pdot() uses compiled C code for ordinary detectors and is much faster 
    o   pdot() should now work for all detection functions 
    o   pdot() now takes account of incomplete 'usage' of detectors
    o   'covariates<-' now allows replacement of individual covariates in
    a multi-session capthist object (replacement value should be a list
    of dataframes)
    o   plot.mask() now handles multi-session masks
    o   detectfnplot() and attenuationplot() now allow the argument 'pars'
    to be a list from detectpar()
    o   make.mask() now clips a masks constructed by any method to a
    boundary polygon if one is provided (previously required type =
    'polygon')
    o   make.mask() and related functions now accept boundary polygons
    in the form of a SpatialPolygonsDataFrame object from 'sp'. This
    effectively adds support for ESRI shapefiles as readShapePoly() in
    the maptools package reads a shapefile and creates a
    SpatialPolygonsDataFrame object (see Examples in ?make.mask).
    Polygons in a shapefile may have multiple parts and/or holes.
    o   a new argument in make.mask() controls the saving or otherwise 
    of a boundary polygon (keep.mask)
    o   read.traps() has a new argument 'covnames' and can accept more
    than one covariate 
    o   read.capthist() has a new argument 'trapcovnames' that is passed
    to read.traps()
    o   read.capthist() has a new argument 'cutval' and can be used to
    input signal strength data. Detections with signal strength
    missing (NA) or below 'cutval' are discarded. Likewise in
    make.capthist().

Other

    o secr.fit() generates a warning when the 'buffer' appears to be
    too small (predicted RB(D-hat) > 0.01); applies only when 'mask'
    not specified.
    o   datasets now use 'lazy loading' so it is not necessary to call the
    data() function before referring to data objects
    o   the MASS and nlme packages have been shifted from Depends to
    Suggests; they are now loaded as required (for eqscplot() and
    fdHess()) 
    o   the utils package is added under Depends for its function
    packageVersion()
    o   the sp, maptools and gpclib packages are added under Suggests
    o   the warning from secr.fit() when the user relied on the default
    detectfn (0) has been suppressed
    o   substantial speed improvement in distancetotrap(), with benefits
    for several other functions
    o   output from secr.fit() now includes the actual call rather than
    the just call as a character string, and print.secr() has been modified
    to interpret call as a string (users should notice no difference)
    o   secr.fit() now gives an explicit error message when individual
    covariates appear in a full-likelihood model
    o   definition of compound halfnormal detection function changed to
    allow g(0) < 1, and now consistent with other functions
    o	tweaked documentation of sim.secr
    o	revised message for missing cutval in plot.secr and detectfnplot
    o	suppressed `spacing not computed' message in summary.traps
    o	documentation revised for detection functions, and tweaked for 
    secr.fit wrt binomN
    o	additions to vignette 'secr-datainput.pdf'
    o	plot.secr() and detectfnplot() streamlined internally, and now
    handle 'binary signal strength' function
    o   help pages for flip(), rotate(), shiftxy(), flip.traps(),
    rotate.traps() and shiftxy.traps() have been consolidated under
    'transformations'
    o   verify.capthist() no longer rejects negative counts (animal died)
    o   undocumented detectors 'quadratbinary' and 'quadratcount' removed
    o   cosmetic changes to many warning and error messages

Bugs fixed

    o	annular halfnormal detection function not working in sim.capthist()
    o   confusion between detection functions 3 & 4 in ip.secr(),
    plot.secr() and detectfnplot()
    o   read.traps() did not set the 'spacing' attribute    
    o   make.capthist() failed with 'polygon' detector when some points
    outside any polygon; now discards such points, with a warning, as
    originally intended
    o   read.traps() could fail reading usage fields from a trap layout file
    o   polyarea() could suffer from numeric overflow with integer
    coordinates, so coordinates now coerced to floating point (solved
    flow-on problem in summary.traps() with polygon detectors)
    o   esa() sometimes ignored incomplete detector usage
    o   sim.secr() failed with 'count' detector when traps object lacked a
    'binomN' attribute, and did not work with negative 'binomN'; now
    takes 'binomN' from 'details' of fitted secr object, default 0
    (Poisson).
    o   sim.secr() did not handle inhomogeneous Poisson density
    correctly, and could make excessive memory demands

Changes in version 1.4.1 (2010-08-29)

New functions
    o   empirical.varD() computes an estimate of the sampling error of a
    Horvitz-Thompson-like estimate of density for replicated spatial
    samples.  The formulation closely resembles that in conventional
    distance sampling with equal-length transects.

    o   moves() returns a list of the distances between consecutive
    detections for each animal in a capthist object (meaningful only
    for trap-like detectors, for which the sequence of detections is
    unambiguous).

    o   make.tri() generates a detector array based on a triangular grid
    and clip.hex() retains a hexagonal subset.

New features in old functions
    o   MS.capthist() now accepts any combination of single-session
    capthist objects, multi-session capthist objects, and lists of
    single-session capthist objects.

    o   sim.popn() now optionally generates a spatiotemporal population
    i.e. a list whose components form a time series of simple popn
    objects with mortality and recruitment and movement between
    times. Two new arguments are used (number.from and nsession);
    turnover parameters are passed as presently undocumented
    components of 'details'.

    o   sim.capthist now generates a multi-session capthist object when
    passed a list of populations (see preceding).

    o   sim.capthist now has an option to truncate detections beyond a
    fixed radius, regardless of the shape of the detection
    function. The radius is specified as component `truncate` in the
    `detectpar` argument. The default is no truncation (truncate =
    NULL).

    o   secr.fit() new automatic model effect 'Session' coded as 0:(R-1)
    where R is the number of sessions. Contrast with lower-case
    'session' which is a factor. Equivalent to manual sessioncov =
    data.frame(0:(R-1)); useful for modelling trend.

    o   secr.fit() new automatic covariates 'x2', 'y2','xy' for density
    trend models (defined for all points on habitat mask).

    o   new detection function = 6 'annular normal' has peak at distance
    +w where parameter w >= 0.

    o   Various functions with the argument 'detectfn' (secr.fit(),
    sim.capthist(), detectfnplot() etc.) now allow the name of the
    function to be supplied as a character value (e.g. 'halfnormal';
    see ?detectfn for names). Values are still stored as the integer
    codes.

    o	rbind.mask() now handles covariates rather than discarding them
    with a warning; rbind.traps() has been revised internally.

    o   read.capthist() and make.capthist() accept negative occasion
    numbers for death on capture even when detector type is not
    'single' or 'multi'.  Previously this coding was ignored because
    modelling of such data poses problems; responsibility now lies
    with the user.

Bugs fixed
    o   write.captures() and write.capthist() used row (animal) number
    rather than row names (animal identifiers) when detector='single'
    or detector='multi'. This has been rectified.

    o   write.captures() and write.capthist() when used with
    multi-session data lost the session names and exported all
    detections as session 1. Actual session names are now used to
    distinguish sessions in the output.

    o   sim.capthist() became confused when detectfn=3 or
    detectfn=4. Simulation with a compound halfnormal detection
    function (3) or uniform detection function (4) should now work.

    o   rbind.popn() returned with meaningless 'Ndist' and 'model2D'
    attributes. Now Ndist is set to 'user' and 'model2D' to the value
    for the first contributing population.

    o   secr.fit() with distribution = 'binomial' failed to calculate
    the asymptotic variance of D because it called the R function
    dbinom that takes only integer values for the size parameter
    (N). This call has been replaced by a local function that accepts
    non-integer N. The associated warning has been suppressed.

    o   rbind.traps() should now combine covariates and usage correctly.

    o   secr.make.newdata() failed with trend-over-occasion models (T),
    causing errors in predict.secr(), collate() and model.average(). 

Other
    o	Displays startup message

    o   Minor improvements to documentation (alias for detectfn, all
    data objects documented etc.)

    o	details$binomN argument of secr.fit has been documented. binomN is
    used to specify the discrete distribution fitted to count data
    (binomial, Poisson etc.), and the size parameter of the binomial.

    o   print.mask() has been dropped as print.dataframe() works
    perfectly well with masks.


Changes in version 1.4.0 (2010-05-07)

New functions
    o	write.capthist() exports capthist object as DENSITY text files.
    o	read.capthist() imports capthist object from DENSITY text files.
    o	write.SPACECAP() exports capthist object and mask as SPACECAP
    text files.
    o	read.SPACECAP() imports capthist object and mask from SPACECAP
    text files.
    o	write.captures() has been renamed from write.capthist.
    o	closedN() computes various closed population estimators.
    o	closure.test() tests for population closure of Otis et al.
    (1978) and Stanley & Burnham (1999).
    o	MMDM() computes observed (trap-revealed) range length (Otis et
    al. 1978).
    o	ARL() computes asympototic range length (Jett & Nichols 1987).
    o	logmultinom() computes the multinomial coefficient of the SECR
    likelihood (internal function, now exported).

New datasets
    o	'skink' lizard pitfall trapping (Efford, Thomas & Spencer unpubl)
    o	'stoatDNA' stoat microsatellites (Efford, Borchers & Byrom 2009)
    o	'housemouse' Coulombe Mus musculus (Otis et al. 1978)
    o	'deermouse' Reid Peromyscus (Otis et al. 1978)

New features in old functions
    o	make.capthist() now sorts rows numerically if all values of
    animalID are valid numbers.
    o	make.capthist() now checks for missing values.
    o	make.capthist() places multi-session capture data in order by
    session before matching 'traps' objects.
    o	print.capthist() now sorts rows numerically if all values of
    animalID are valid numbers.
    o	make.grid() has a new option for labeling detectors "xy".
    o	secr.fit() has a new 'details' option to centre trap and mask
    coordinates; this can solve some numerical problems fitting
    coordinates with large absolute values.
    o	plot.mask() produces more attractive images and does not call
    image(), but is somewhat slower.
    o	read.mask() has been tweaked to allow import from csv format as
    used by SPACECAP.

Defunct
    o	read.captures() has been ditched - use read.table() instead.

Bugs fixed
    o	A bug in 'traps<-' has been fixed; replacement of traps for
    multiple sessions should now work.

Other
    o	Dataset 'secrdemo' now incorporates 'rawdata' and 'captdata'.
    o	Example text files 'capt.txt,'trap.txt','stoatcapt.txt',and
    'stoattrap.txt' are provided in the package installation folder
    'extdata' (e.g. setwd(system.file('extdata', package = 'secr'))).
    o	New vignette: secr-datainput.pdf
    o	Revised vignette: secr-overview.pdf


Changes in version 1.3.0 (2010-03-11)

New functions
    o	ellipse.secr() computes and plots confidence ellipses of
    fitted parameters.
    o	LLsurface.secr() plots log-likelihood surface of SECR model
    for 2 dimensions at a time.
    o	attenuationplot() plots the sound attenuation function from
    an acoustic SECR model.

New datasets
    o	ovensong - new acoustic dataset (Dawson & Efford J Appl Ecol;
    see also 'secr-sound.pdf')

New features in old functions
    o	The handling of structural relationships in secr.fit() with
    scaleg0 and scalesigma has been improved.
    o	The 'start' argument of secr.fit() may now be a previously
    fitted nested model.
    o	The default algorithm for initial values in secr.fit() now
    uses 'RPSV' instead of 'dbar' for spatial scale.
    o	secr.fit() has a new pre-defined covariate term T for time
    trend in detection parameter model formulae.

Other
    o	New detector types 'count', 'signal' ('polygon' and
    'transect' are present but should be treated as 'under
     development')
    o	New detection functions 'binary signal strength', 'signal
    strength', 'signal strength spherical'
    o	Finite mixture models for heterogeneity in detection
    parameters (use 'h2' in model formula)
    o	New vignettes: secr-overview.pdf, secr-sound.pdf,
    secr-finitemixtures.pdf

Version 1.2.11 2009-12-22

Bug fix for session names not in alphanumeric sequence

Version 1.2.10 2009-10-21

large models -- more robust memory allocation in external C code (malloc)
new example datasets -- 'possum' and 'ovenbird'
confint.secr -- improved profile likelihood confidence intervals, both beta and real parameters
secr.fit -- detects non-finite values of real parameters during maximization & terminates (cf stepmax in nlm)
sim.capthist -- time-varying detection parameters
sim.resight -- new function
secr.fit -- experimental and undocumented mark-resight

Version 1.2.9 2009-10-05

split.traps -- new method
print.traps -- now displays component name if list
secr.fit -- new argument details$minprob; new default 1e-50 may affect likelihood
print.capthist -- new argument 'condense' for condensed printing of proximity data
verify.mask -- fixed bug with mask lists
verify.capthist -- additional checks for missing or badly coded data

Version 1.2.8 2009-09-30 and later

Further changes to external C code for non-Windows platforms (now expected to work)
Minor editting of documentation files
Improved handling of multi-session capthist objects (2009 10 01)
Restore output of dAICc in AIC.secr (2009 10 02)

Version 1.2.7 2009-09-29

External C code re-written to conform to ISO C99 for use on non-Windows platforms
External C code now relies on R math functions dpois, dbinom
make.capthist -- new argument 'sortrows'
print.secr -- minor changes
secr.fit -- secr version number recorded in secr object
secr.fit -- optional check of input data with 'verify'
verify (new) -- function to check data in capthist objects
ip.secr (new) -- density estimated by simulation and inverse prediction
collate -- minor bug fixed
plot.secr -- confidence limits of detection function by delta method
vcov.secr -- new option for covariances between real parameters

Version 1.2.6 2009-09-10

Simulation functions (sim.popn, sim.capthist, simulate, sim.secr) use R random number generators and seeds.
New S3 method simulate.secr and function sim.secr for simulating from a fitted secr model
New S3 methods deviance and df.residual
secr.make.newdata is now exported (previously internal)
Additional components in secr object returned by secr.fit: design, D, start
Bugs fixed in secr.make.newdata, secr.fit and make.capthist

