BiodiversityR versions
This ChangeLog provides information on new functions integrated in the package and some
required updates. Updates related to error messages for new versions of the base and
loaded packages are typically not included here (as these were addressed).
BiodiversityR details on installation are available from:
http://www.worldagroforestry.org/output/tree-diversity-analysis
The manual accompanying the original version of BiodiversityR is available from:
http://www.worldagroforestry.org/downloads/Publications/PDFS/B13695.pdf
An article that documents the ensemble suitability modelling options of BiodiversityR
is available from https://doi.org/10.1016/j.envsoft.2017.11.009
(You could also be interested in https://www.cifor-icraf.org/knowledge/publication/8977/).
Various scripts for creating publication-ready graphs with ggplot2 are available from:
https://rpubs.com/Roeland-KINDT
Version 2.17-2 (April 2025)
* Modified nested.npmanova() based on suggestions by Jari Oksanen about the adonis2()
function (see also: https://github.com/vegandevs/vegan/issues/677).
* Modified nested.anova.dbrda() based on suggestions by Jari Oksanen about suppressing
messages (see also: https://github.com/vegandevs/vegan/issues/682).
* Modified documentation for caprescale() to use scores() instead of summary(), as also
done internally in the function (see version 2.17-1).
Version 2.17-1.1 (January 2025 correction)
* Inclusion of the modified caprescale() function created for version 2.17-1, something that
was not achieved during the previous submission (see also vegan announcement #644)
Version 2.17-1 (January 2025; King Gizzard & the Lizard Wizard US Tour 2024 live captures
by Chris Dempsey version)
* New function treegoer.position() that calculates the position of the planting site in
respect to the median of the species range
* New function treegoer.map() that creates a suitability map with the same methods as
used by treegoer.score().
* Updated treegoer.widen() to be compatible with treegoer.position().
* Expanded functions treegoer.score() and treegoer.filter() to accommodate a climate score
of 0.5 for one-sided checks whether the planting site is outside the species' ranges.
* Modified functions caprescale(), to use the vegan::scores() function instead of
vegan:::summary.cca() (see also vegan announcement #644)
* Compatible with Rcmdr version 2.9-5
version 2.16-1 (March 2024; King Gizzard & the Lizard Wizard at the PALP Rocklette
Festival in August 2023 version)
* New functions treegoer.score(), treegoer.filter() and treegoer.widen() that calculate
a climate score similar to the algorithms that are used internally in the
GlobalUsefulNativeTrees database with environmental ranges documented by the Tree
Globally Observed Environmental Ranges database.
version 2.15-4 (October 2023)
* Removed references to rgeos in the documentation for function ensemble.chull.create()
that caused messages of 'unknown package'.
* Include pvclust among the suggested packages (See also documentation for function
pvclust.long()).
* Made changes to be compatible with Rcmdr 2.9-1.
version 2.15-3 (September 2023)
* Modified function ensemble.envirem.run to use envirem::generateEnvirem()
* Included reference to TreeGOER: A database with globally observed environmental ranges
for 48,129 tree species. Global Change Biology https://doi.org/10.1111/gcb.16914
version 2.15-2 (May 2023)
* New function ensemble.Tjur() to calculate Tjur's Coefficient of Discrimination for
logistic regression models.
* Modified function ensemble.evaluate() to include results from ensemble.Tjur().
* Corrected function ensemble.envirem.run() to work with newer versions of terra.
* Updated citation of van Zonneveld et al. 2023 in the documentation of function
ensemble.concave.hull().
version 2.15-1 (January 2023; King Gizzard & the Lizard Wizard Omnium Gatherum version)
* Removed dependencies on rgdal, rgeos and maptools (requested by Roger Bivand on 19-DEC)
* New function ensemble.outliers() to identify outliers in multivariate space based on
user-defined number of variables and multiples of interquartile range distance.
* Modified function ensemble.chull.create() to remove use of rgeos::gBuffer.
* Modified function ensemble.plot() to remove use of maptools 'wrld_simpl' data set.
* Modified function ensemble.chull.apply() to save the masked suitability rasters in the
'chull' subdirectory rather than overwriting the input files.
* The default output format for ensemble suitability methods that use the raster package
was modified to GTiff for functions ensemble.analogue(), ensemble.batch(),
ensemble.bioclim(), ensemble.chull.create(), ensemble.chull.apply(),
ensemble.ecocrop(), ensemble.habitat.change(), ensemble.mean(), ensemble.novel(),
ensemble.raster() and ensemble.zones().
* Function ensemble.batch() now uses the 'GTiff' output format.
* KML output is no longer supported, modified for functions ensemble.analogue(),
ensemble.batch(), ensemble.bioclim(), ensemble.habitat.change(), ensemble.mean(),
ensemble.novel(), ensemble.raster() and ensemble.zones().
* New option of 'hellinger' for argument 'dist' in function dist.eval().
* New option of by="onedf" for anova.cca() implemented in the Constrained Ordination
Graphical User Interface.
Version 2.14-4 (September 2002)
* Made compatible with Rcmdr version 2.8-0 as the Graphical User Interface could not be
launched otherwise with function BiodiversityRGUI()
Version 2.14-3 (August 2022)
* Removed suggested package akima (email from Prof. Ripley 4-JUN-2022). Function
residualssurface() was removed.
* New argument of 'aitchison_pseudocount' in function CAPdiscrim() to allow the new
methods of 'aitchison' and 'robust.aitchison' from vegan::vegdist().
* Added options of 'aitchison' and 'robust.aitchison' to documentation options of
distdisplayed().
* Function nested.npmanova() now uses vegan::adonis2().
* Functions diversityresult(), diversitycomp() and diversityvariables() now include
options for the unbiased Simpson calculated via function vegan::simpson.unb()
* GUI now includes 'aitchison' and 'robust.aitchison' among options for calculations of
distances matrices. Note, however, that such methods may only work when calculating the
distance matrix first before using other methods such as ordination methods.
* Diversity calculation GUI now includes the options of 'simpson.unb' and
'simpson.unb.inverse'.
* BiodiversityR now depends on vegan version 2.6-2.
Version 2.14-2 (April 2022)
* New functions ensemble.concave.hull(), ensemble.concave.venn() and
ensemble.concave.union() that enable analysis of niche overlaps in environmental space
for different combinations of climates and geographical origins of accessions via
concave or convex hulls, building on methods used by Pironon et al. 2019.
https://doi.org/10.1038/s41558-019-0585-7.
* New functions ensemble.envirem.masterstack(), ensemble.envirem.solradstack() and
ensemble.envirem.run() that allow to calculate an expanded set of bioclimatic variables
via teh envirem package with data.frames as input.
* Function ensemble.environmentalThin() was modified to result in exactly the
target number of environmentally-thinned locations.
* The speed of function ensemble.ecocrop() was increased. The function now also works
with the SpatRaster objects of the terra package. Note that the setting of argument
temp.multiply was changed from 10 to 1 to work with later versions of WorldClim.
Version 2.14-1 (January 2022; Ty Segall's Harmonizer version)
* New function ensemble.terra() that creates raster files via the terra package with
similar functionality as ensemble.raster.
* ensemble.calibrate.models() and ensemble.calibrate.weights() functions now work with
package terra's SpatRaster objects (and still with package raster's rasterStack
objects).
* ensemble.PET.season(), ensemble.PET.seasons(), ensemble.prec.season(),
ensemble.tmean.season() and ensemble.season.suitability() functions now work with
package terra's SpatRaster objects (and still with package raster's rasterStack and
rasterLayerobjects).
* Removed dependency on the orphaned 'maptree' package. Used previously for an option
of 'kgs' in the clustering GUI.
* Compatible with Rcmdr version 2.7-2 (January 2022)
Version 2.13-1 (April 2021)
* New functions ensemble.spatialBlock and ensemble.envBlock that allow for
stratification methods available via the blockCV package. These functions were modified
from functions available previously from the ensemble.blockCV package.
* New function pvclust.long that allows plotting of results from pvclust onto
ordination diagrams.
* New functions ensemble.PET.seasons, ensemble.prec.season and ensemble.tmean.season
that return raster layers with the number of growing periods, their starts and lengths,
and the precipitation and mean temperatures of a selected (growing) period.
* New function ensemble.season.suitability that enables the calculation of crop
suitability from ecocrop-like thresholds and growing season temperature and
precipitation.
* Readme file
Version 2.12-3 (December 2020)
* Moved package rgl to Suggests as requested by prof. Brian D Ripley (13-DEC-2020)
* Moved package vegan3d to Suggests
* Removed calls to rgls and vegan3d in function BiodiversityRGUI
* New function ordiellipse.long that can be used to plot ordination diagrams via
ggplot2.
* New plot options of 'ggplot (ordiellipse)' in the constrained and unconstrained
ordination GUIs.
* Included references to https://rpubs.com/Roeland-KINDT in the documentation where
options for ggplot2 graphics were shown.
Version 2.12-2 (November 2020; Fuzz III version )
* New functions sites.long, species.long, centroids.long, vectorfit.long,
ordisurfgrid.long and axis.long that extract data in the 'long' (tidy) format and that
can subsequentially be used to plot ordination diagrams via ggplot2. Various examples
are provided in the documentation, including cases where concave hulls and
superellipses are drawn via the ggforce package.
* New function accumcomp.long that extracts data in the 'long' (tidy) format from a
accumcomp result and that can subsequentially be used by ggplot2.
* New function renyicomp.long that extracts data in the 'long' (tidy) format from a
renyicomp result and that can subsequentially be used by ggplot2.
* New function renyi.long that extracts data in the 'long' (tidy) format from a
renyiresult and that can subsequentially be used by ggplot2.
* New arguments of 'return.data' and 'specnames' that result in function rankabuncomp
to return data that can be used to plot rank abundance curves via ggplot2. Examples were
included in the documentation for rankabuncomp.
* New argument of 'extracted.data' in function 'ensemble.environmentalThin' that avoids
the step of extracting environmental data at the provided locations.
* New argument of 'poly.only' in function ensemble.chull.create that results in only
returning the convex polygon.
* New plot methods of 'ggplot (ordisymbol1)', 'ggplot (ordisymbol2)',
'ggplot (ordispider1)', 'ggplot (ordispider2)', 'ggplot (ordisurf1)',
'ggplot (ordisurf2)', 'ggplot (add species)' and 'ggplot (add vector)' in the
unconstrained ordination and constrained ordination GUIs. Except from the latter two
options, each method generates a ggplot2 graph ('plotgg1') based on information
available in an ordiplot object ('plot1') and ordination result. The latter two options
add information on species or continuous variables to 'plotgg1', resulting in 'plotgg2'.
As for other scripts generated by the GUIs of BiodiversityR, it is expected that users
will further modify these scripts to get exactly the graphs that are wanted.
* New option in the Species accumulation curves, Renyi profiles and Rank abundance
curves GUI of 'ggplot'. These require that first the appropriate object (eg the result
from accumcomp) was generated.
* New feature for the Compare distance matrices GUI with betadisper options to also
provide results for pairwise permutation tests.
* New accumulation method in the Species accumulation curves GUI to explicitly refer
to the rarecurve function (previously this method was available via plot, but data
used were not shown).
Version 2.12-1 (July 2020; Wand's Ganglion Reef + 1000 Days version)
* New function ensemble.PET.season that calculates the maximum climatological water
deficit (accumulated aridity), corresponding to the dry season with the largest
accumulated difference between precipitation and potential evapotranspiration. The
methodology is the same as Aragao et al. 2007 (Spatial patterns and fire response of
recent Amazonian droughts. Geophysical Research Letters 34(7)), but the function uses
monthly PET values instead of fixed values at 100 mm per month.
* New function ensemble.chull.buffer.distances that give details on potential buffer
widths of the ensemble.chull.create function.
* New function ensemble.chull.MSDM that creates input data and a suggested script for
the MSDM_Posteriori function of the MSDM package (Mendes, P.; Velazco S.J.E.; Andrade,
A.F.A.; De Marco, P. (2020) Dealing with overprediction in species distribution models:
how adding distance constraints can improve model accuracy, Ecological Modelling,
in press.)
* New argument 'TrainTestData' for function ensemble.calibrate.weights that allows the
function to calibrate suitability models and weights for the ensemble model based on
a data.frame. When providing k-fold arguments as a list, the user should ensure that
these match the number of presence and absence (background) records of the data.frame.
* New argument 'buffer.maxmins' for function ensemble.chull.create that estimates the
buffer width from the observation that is farthest away from its closest neighbour.
* New argument 'lonlat.dist' for function ensemble.chull.create that calculates the
distance in km for analyses based on longitude-latitude coordinate systems
* Modification in function ensemble.calibrate.models to fit MAXENT and MAXNET models
with data.frames when raster stack is not provided.
* New arguments of 'rainbow_hcl', 'rainbow_hcl.c', and 'rainbow_hcl.l' of ordisymbol to
use the colorspace::rainbow_hcl palette and to modify chroma and luminance.
* Internally modified the 'rainbow' palette of functions accumcomp, rankabuncomp and
renyiplot to use the colorspace::rainbow_hcl palette
* Modified overall colour scheme of the GUI plot options to rainbow_hcl of the
colorspace package
* Expanded options in the presence-absence GUI for various models that are available
also from the ensemble suitability modelling algorithms. These include GBM (gbm),
RF (randomForest), CF (cforest), EARTH (earth), RPART (rpart), NNET (nnet), FDA (fda),
SVM (ksvm) and SVME (svm).
* New option in the clustering GUI to use the fastcluster package.
* New option in the clustering GUI for fuzzy K-means analysis via e1071::cmeans
* New options in the clustering GUI for kernel K-means (kkmeans) and spectral clustering
(specc) via the kernel package
* New options in the clustering GUI to determine the number of clusters via the NbClust
package.
* New options in the clustering GUI to change branch and label colors of dendrograms
via the color_branches and color_labels functions of the dendextend package
* New options in the clustering GUI to create heat maps via the aspectHeatmap of the
ClassDiscovery package
* New option in the clustering GUI to create a heat map based on a Thresher analysis
(available as plot option).
* Corrected an error in function ensemble.batch of assigning absence locations.
Version 2.11-3 (October 2019)
* New function ensemble.SEDI that calculates the Symmetric Extremal Dependence Index
(SEDI), as formulated by Ferro et al. 2011. Wea. Forecasting 26: 699-713 and
recommended as evaluation metric by Wunderlich et al. 2019. Nature Conservation 35:
97-116.
* New function ensemble.evaluate that calculates the following evaluation statistics:
AUC: Area Under The Curve for the testing ModelEvaluation; TSS: maximum value of the
True Skill Statistic over range of threshold values; SEDI: maximum value of the
Symmetric Extremal Dependence Index over range of threshold values; TSS.fixed: True
Skill Statistic at the fixed threshold value; SEDI.fixed: SEDI at the fixed threshold
value; FNR.fixed: False Negative Rate (= omission rate) at the fixed threshold value;
MCR.fixed: Missclassification Rate at the fixed threshold value; and AUCdiff: Difference
between AUC for calibration and the testing data
* Internally call ensemble.evaluate in function ensemble.calibrate.models
* New function ensemble.VIF.dataframe that provides similar types of Variance Inflation
Factor analysis for a data.frame as function ensemble.VIF
* Decreased the step size for PresenceAbsence threshold calculations in function
ensemble.threshold from 0.005 to 0.0001 (as in dismo::evaluate)
* Corrected an error in function ensemble.mean when RASTER.stack.name=""
* Methods and examples of generating and using spatial folds for presence and absence
points via the blockCV package are available from the BiodiversityRblockCV package
(URL https://github.com/RoelandKindt/BiodiversityRblockCV) through functions
ensemble.spatialBlock and ensemble.envBlock (These functions are in a
separate package as blockCV is not available from CRAN). Use the commands
devtools::install_github("rvalavi/blockCV") and
devtools::install_github("RoelandKindt/BiodiversityRblockCV") to install these packages.
* example of calculating diversity profiles from importance values calculated with
importancevalue.comp (email exchange with Uday Pimple)
Version 2.11-2 (June 2019)
* new function ensemble.environmentalThin.clara that enables environmental thinning for
large data sets. Prior to using function ensemble.environmentalThin internally in the
function, the points are first divided in clusters via the cluster.clara function.
Subsequently points in each cluster are environmentally thinned.
* new option for ensemble.calibrate.weights to import k-fold groupings for presence and
background points as a list with elements groupp (k-fold groupings for presence points)
and groupa (k-fold groupings for background points). The option was inspired on using
outputs from the blockCV package. The option is invoked by using the list as input for
argument k of ensemble.calibrate.weights.
* modification in ensemble.environmentalThin to provide a final data set that matches
the requested number of environmentally thinned points (previously the function stopped
the second algorithm when the identified minimum distance was reached, in analogy to
the ensemble.spatialThin function).
* modification in ensemble.environmentalThin to possibly retain only 1 axis if the
variance on 1 axis exceeds the selected threshold (the previous behaviour was to
select a minimum of 2 PCA axes).
* new argument 'silent' for the ensemble.spatialThin, ensemble.environmentalThin and
ensemble.spatialThin.quant functions, with option to have no reports on the processing
of points.
* new arguments 'cex.lab' and 'cex.axis' for the accumplot and accumcomp functions to
modify sizes of axis labels (requested by Julio Cesar Canales).
* new options for the unconstrained and constrained ordination GUI to create
3-dimensional ordination graphics via the ordiplot3d, ordirgl, orglspider and
orglellipse functions of the vegan3d package.
* additional options of 'ward.D', 'ward.D2' and 'mcquitty' for the cluster GUI.
* new option 'labels (variable)' of the cluster GUI to change labels in dendrograms
to data from the environmental data set.
* new option in the Analysis of species abundance menu to summarize differences in
abundance with the Rmisc::summarySE function.
* corrected an error in the ensemble.raster function where projection ran into problems
with CATCH.OFF=FALSE for the GBM algorithm.
* corrected the error of having no visible global definitions for 'data' and
'wrld_simpl' of the ensemble.plot function (error reported in the CRAN Package Check
results)
* corrected an error for the envfit plotting option of the ordination GUI where results
of permutation tests were not shown
* compatible with Rcmdr 2.5-3 (May 2019)
Version 2.11-1 (January 2019; Ty Segall & White Fence's Joy version)
* new function ensemble.environmentalThin that provides a method similar to function
ensemble.spatialThin, but operates in environmental instead of geographical space. The
environmental space is represented by principal components obtained via vegan::rda.
The user needs to define the target number of environmentally thinned points instead
of the minimum distance.
* new function ensemble.outlierThin that filters out point observations that could be
outliers based on a Local Outlier Factor analysis.
* new function ensemble.spatialThin.bins that enables spatial thinning for large data
sets. With the function, points are first classified in different bins. In the next
step, spatial thinning is done for each bin. Bins are defined by quantile limits of
geographical coordinates.
* new argument of 'MAXNET' in ensemble.calibrate.models, ensemble.calibrate.weights,
ensemble.raster, ensemble.batch, ensemble.drop1, evaluation.strip.data and
evaluation.strip.plot functions, allowing for maximum entropy species distribution
modeling using glmnet for model fitting as implemented in the maxnet package.
* new argument of 'CF' in ensemble.calibrate.models, ensemble.calibrate.weights,
ensemble.raster, ensemble.batch, ensemble.drop1, evaluation.strip.data and
evaluation.strip.plot functions, allowing for random forest and bagging ensemble
algorithms utilizing conditional inference trees as base learners as implemented in the
party package. (Note that integration of cforest required modifications of
raster::predict that were integrated in raster version 2.8-19.)
* new argument of 'target.groups' in ensemble.calibrate.models,
ensemble.calibrate.weights, ensemble.batch and ensemble.drop1 that results in 'target
groups' of species that were all collected or observed using the same methods or
equipment as for the species for which species distribution models are calibrated.
See Phillips et al. 2009 https://doi.org/10.1890/07-2153.1 for details. When selecting
the option, users should provide the presence locations of the other species as
background. For model calibrations and evaluations, pseudo-absence locations will
correspond to centres of cells with presence locations of the other species, possibly
excluding cells where the focal species occurs (a similar option is available from
dismo::randomPoints). Feature requested by Tobias Fremout.
* new argument of 'block.default' in ensemble.calibrate.weights and ensemble.batch that
results in different options of blocking point data sets (first by latitude as only
option available from ENMeval::get.block [i.e. the default], or first by longitude).
Feature suggested by Evert Thomas.
* new argument of 'get.subblocks' in ensemble.calibrate.weights and ensemble.batch that
results in a double implementation of ENMeval::get.block, whereby point data sets
assigned to one block in the first step are again assigned to four blocks ('subblocks')
in a second step.
* corrected error in ensemble.spatialThin to allow retaining only 1 point in case that
the random elimination process retained two points above the minimum distance in the
previous elimination step.
* modified ensemble.spatialThin to round geographical distances to 2 decimals (10 m)
* ensured that ensemble.raster function only retained numeric variables for maxlike and
glmnet algorithms in situations where categorical variables were used by other
algorithms.
* corrected error for GBMSTEP and probit links where the gbm.x argument did not select
the correct explanatory variables
* compatible with Rcmdr 2.5-1 (September 2018)
* new option in the GUI to import Excel data sets via readxl::read_excel
* modified edit of community and environmental data sets in the GUI to be more
similar to edit options of the R-Commander
* new GUI for spatial and environmental thinning of point observation data
Version 2.10-1 (July 2018)
* new function ensemble.red that calculates statistics of Area of Occupancy (AOO) and
Extent of Occurrence (EOO) that are used as thresholds for the IUCN Red List (URL
http://www.iucnredlist.org/static/categories_criteria_3_1). AOO and EOO are expected
to be calculated for 'count' suitability layers that represent different levels of
consensus among ensembles or algorithms. AOO and EOO are calculated via the red package.
* new functions ensemble.chull.create and ensemble.chull.apply that restrict suitable
habitat to polygons that are included in a convex hull around known presence locations.
The convex hull is expanded by a user-defined buffer based on the maximum distance among
the known presence locations.
* new function ensemble.pairs that creates scatterplots similar to graphics::pairs
* new function BiodiversityR.changeLog
* corrected error in ensemble.VIF where keeping explanatory variables resulted in
never-ending loops if VIF of some kept explanatory variables was above VIF.max
* compatible with Rcmdr 2.4-4 (April 2018)
* Species accumulation GUI plots horizontal error bars (via argument ci.length for
plotting of specaccum objects)
* Distance matrix calculation GUI now includes the possibily to directly calculate the
Hellinger and Chord distances.
* Distance matrix calculation GUI now includes the scaled Hellinder and Chord distances
(Conde and Dominguez 2018. https://doi.org/10.1016/j.japb.2018.01.001).
* Clark's 1952 coefficient of divergence (appropriate for the analysis of species
abundance data; Legendre and Legendre 2012 p. 306) was included among the options
in distance, ordination and clustering GUIs.
* vegan::avgdist function was included as option in the Distance matrix calculation GUI
* vegan::chaodist function was included as option in the Distance matrix calculation GUI
* Ensemble suitability modeling GUI now allows for the option that absence locations
are NULL - i.e. allowing for random selection of background locations during each run.
* Corrected error generated by constrained ordination GUI by calling vegan::goodness for
capscale results (this option is no longer available for vegan >= 2.5-1).
* Corrected errors generated by constrained ordination GUI by calling vegan::inertcomp
in vegan >= 2.5-1
Version 2.9-2 (27 February 2018)
* modified gam (version 1.13 of 2018-02-25) functions to gam::predict.Gam and
gam::step.Gam in ensemble.calibrate.models, ensemble.raster and evaluation.strip.data
to safely retain BiodiversityR on CRAN (problems reported on 2018-02-26).
* corrected error in for R-devel for rankabuncomp
* new argument 'srt' for rankabunplot to allow rotation of species labels in rank-
abundance graphs
* new method 'max' for diversityresult and diversitycomp functions
Version 2.9-1 (February 2018; Ty Segall's Freedom's Goblin version)
* new function ensemble.spatialThin that creates a spatially thinned point location
data set where all pairwise distances are above a threshold distance. Although similar
to functions in the spThin or red packages, the geodesic is calculated more accurately
via the geosphere package. The function was documented separately and also was
integrated in the ensemble.batch function.
* an article that documents the ensemble suitability modelling options of BiodiversityR
has been published (Kindt R. 2018. Ensemble species distribution modelling with
transformed suitability values. Environmental Modelling & Software 100: 136-145. URL
https://doi.org/10.1016/j.envsoft.2017.11.009. References to this article has been
included in the documentation of various functions.
* a new argument of 'keep' in ensemble.VIF allows to always keep a subset of
explanatory variables within the final subset of variables
* in the output of the ensemble.mean function, niche overlap between input suitability
layers and the mean (consensus) layer is calculated via Warren's I index that is based
on the Hellinger distance (dismo::nicheOverlap; Warren et al. 2008. Evolution
62-11: 2868?2883)
* in the ensemble.batch function, random selection of background locations (if an
option) is repeated in each run
* included options in ensemble.plot to specify colours for absence and presence
suitabilities, including a new option only to plot areas where the species is predicted
to be present. (Also included an example how to generate the colour scheme used in the
http://www.worldagroforestry.org/atlas-central-america climate change atlas.)
* It is now possible to disable calls to the tryCatch function in the ensemble
suitability (parameter CATCH.OFF). This option was implemented to facilitate error
checking (such as problems when not all packages were installed).
* An error was encountered when attempting to plot evaluation results. This error seems
to be caused by interpretation of the ModelEvaluation signature. The error was
circumvened by disabling plotting in the ensemble functions, but giving an example of
getting plot results in the examples. Error reported by Thomas Fremout 22-JAN-2018.
* the location of the dismo examples folder is provided explicitly in the GUI for
creation of raster stacks
Version 2.8-4 (September 2017)
* compatible with Rcmdr 2.4-0 (August 2017)
Version 2.8-3 (June 2017)
* corrected error that prevented loading of the Graphical User Interface after previous
workspace was saved
* corrected error in function multiconstrained
* option added of method="dbrda" in function multiconstrained
Version 2.8-2 (May 2017)
* full integration of the MAXLIKE algorithm in ensemble suitability modelling
* new argument of 'get.block' in ensemble.calibrate.weights and ensemble.batch.
The argument splits 4-fold cross-validation presence and background point
location data sets along longitudinaland latidudinal lines through function
get.block of the ENMeval package.
* new argument of 'mmax' in the CAPdiscrim function. This argument limits the search for
the optimal m (number of PCoA axes) to a a maximum number of axes equal to mmax.
The argument was introduced to deal with the lda error of a variable appearing
to be constant within groups.
* function CAPdiscrim saves correct predictions per factor level
* function ordisymbol now allows different palettes
* errors corrected in the 'diversity indices' menu in saving results in the
environmental data set
* function evaluation.strip.data includes suitability values prior to the probit
transformation. Errors obtained for MAHAL, MAHAL01 and DOMAIN algorithms were
corrected for the same function.
Version 2.8-1 (January 2017, malaria recovery version)
* compatible with Rcmdr 2.3-1 (October 2016)
* BiodiversityRGUI function gives information on backward compatibility with prior
versions of BiodiversityR
* new function ensemble.calibrate.models that replaces function ensemble.test.
The original name of ensemble.test originated from the first (2012) versions of
ensemble suitability modelling functions in BiodiversityR where both
ensemble.raster and ensemble test internally calibrated and evaluated (tested)
models, but only ensemble.raster went ahead with creating suitability raster
layers. For possible backwards compatibility, assign
ensemble.test <- ensemble.calibrate.models
* new function ensemble.calibrate.weights that replaces function ensemble.test.splits
* slot ensemble.calibrate.weights$AUC.table (ensemble.calibrate.weights) replaces
ensemble.calibrate.weights$table
* argument SSB.reduce (ensemble.calibrate.models, ensemble.batch) replaces argument
CIRCLES.at
* ensemble suitability layers are now create in different 'ensemble' subfolders
subfolder 'suitability' contains suitability layers for each ensemble
subfolder 'presence' contains presence layers for each ensemble
subfolder 'count' contains number of algorithms predicting presence
subfolder 'consensussuitability' contains consensus suitability layers
subfolder 'consensuspresence' contains consensus presence layers
subfolder 'consensuscount' contains count of ensembles predicting presence
subfolder 'consenssd' contains consensus standard deviation
* new function ensemble.VIF that selects subsets of explanatory data layers until all
Variance Inflation Factors are below a user-defined threshold (default: 10).
By an iterative procedure, the data layer with largest VIF is removed in each
step. The function is called from within the ensemble.batch function.
* new function ensemble.bioclim that provides an algorithm that more closely represents
the original BIOCLIM algorithm than function dismo::bioclim
* new function ensemble.bioclim.graph that graph function makes graphs that show mean,
median, minimum, maximum and lower and upper limits for species or climates
* new function stackcommunitydataset that creates a data.frame with separate columns
for sites, species and abundance values (therefore the inverse function of
makecommunitydataset). Function is useful to combine information from
different community data sets into one community data set.
* new option for including the glmnet algorithm in ensemble suitability modelling.
Option available via argument of GLMNET in ensemble.calibrate.models,
ensemble.batch, GUI interface and other functions.
* new option for including the maxlike algorithm (Royle et al. 2012. Methods in Ecology
and Evolution) in ensemble suitability modelling. However, because the maxlike
function does not accept data frames for model calibration and the predict
function does not have a newdata argument, inclusion of the algorithm is
problematic and may often not be available (as in evaluation.strip.data)
or fail (Roeland will write a new email to the authors of maxlike indicating
the need to expand maxlike to handle data frames).
* new option for including a modification of the dismo::mahal algorithm to ensure
that predictions are within the 0 to 1 range similar to probabilities. Option
available via argument of MAHAL01 and MAHAL.shape in ensemble.calibrate.models,
ensemble.batch, GUI interface and other functions with
* new output of the ensemble.batch function to report which ensemble had the best
overall AUC. (This option can be used to plot count models
* new options for ensemble.threshold
threshold.method == "threshold2013.mean" provides the mean threshold of
optimal thresholds identified by Liu et al. 2013. Journal of
Biogeography 40: 778-789
threshold.method == "threshold2013.min" provides the minimum threshold of
optimal thresholds identified by Liu et al. 2013. Journal of
Biogeography 40: 778-789
threshold.method == "threshold2005.mean" provides the mean threshold of
optimal thresholds identified by Liu et al. 2005. Ecography 28:
385-393
threshold.method == "threshold2005.min" provides the minimum threshold of
optimal thresholds identified by Liu et al. 2005. Ecography 28:
385-393
* change of handling of data extraction from raster stacks with only 1 layer
* internal change of raster extent no longer possible with main ensemble functions.
Hence rasterstack objects should be modified prior to using them in
ensemble functions.
* change of implementation in ensemble.calibrate.models whereby dummy variables
are no longer excluded for BIOCLIM and BIOCLIM.O algorithms
* change of implementatin in ensemble.calibrate.models whereby only dummy variables
with no variation are excluded for the DOMAIN algorithm
* new option for ensemble.drop1 to also fit models with only the focal variable.
Option available via argument of variables.alone = TRUE.
* new option for ensemble.novel.object to exclude factor variables. The documentation
was updated to show dealing with factor variable through dummy variables.
* new option for ensemble.dummy.variables to create dummy variables for levels not
observed in the input factor layer
* increased output of ensemble.calibrate.models and ensemble.calibrate.weights to
show how ensemble weights are determined (input weights, parameters and
output weights)
* warning message in ensemble GUI that title of calibration stack is expected to be
the same as object name of the calibration stack
* ensemble suitability calibration modelling allows to set VIF.max (maximum Variance
Inflaction Factors of explanatory variables) and CIRCLES.d (the distance in m
of circular neighbourhoods for testing background locations to reduce spatial
sorting bias).
* added example of superimposing GBIF locations in ensemble.ecocrop documentation
* GUI option of showing calibration stack also provides names of factor variables,
names of dummy variables and makes a plot of the calibration stack
* Diversity indices menu now also accepts two factor variables to create a 2-way
table with diversity statistics for each combination of factor levels
* Added links from the GUI help to vignettes on diversity analysis in vegan and
ordination analysis in vegan. Also added link to the community analysis
website of Jari Oksanen from the GUI help.
Version 2.7-2 (August 2016)
* compatible with Rcmdr 2.3-0 (August 2016) and vegan 2.4-0 (June 2016)
* new function ensemble.threshold to calculate threshold values to differentiate
between absence and presence suitabilities. This function was used earlier
only internally in the ensemble.test, ensemble.test.splits and ensemble.batch
functions, whereas it is now available and documented as separarte function.
* corrected error in makecommunitydataset (error reported by Pascal Hirsch)
* Graphical User Interface (GUI) now supports following functions or options from
vegan:
o dispweight function (via Transform community matrix menu)
o fitspecaccum function (via options of "arrhenius", "gleason", "gitay",
"lomolino", "asymp", "gompertz", "michaelis-menten", "logis" and
"weibull" in Species accumulation curves menu)
o specslope function (via options in Species accumulation curves menu)
o rareslope function (via options in Species accumulation curves menu)
o rarefy function (via options in Species accumulation curves menu)
o tsallis function (via options in Renyi profile menu)
o tsallisaccum function (via options in Renyi profile menu)
o dbrda function (via Constrained ordination menu)
o monoMDS function (via Unconstrained ordination menu)
o stressplot function (via unconstrained and constrained ordination menus)
o betadisper function (via Compare distance matrices menu)
o designdist function (in Distance matrix calculation menu)
o Lingoes and Cailliez adjustments for wcmdscale, capscale, dbrda and
betadisper
o sqrt.dist argument for wcmdscale, capscale, dbrda and betadisper
o descriptive string of "species" in scaling of ordination axes (via
unconstrained and constrained ordination menus)
o RsquareAdj function for cca, dbrda and capscale (in summaries of constrained
ordination)
o ordibar function (via plotting functions for unconstrained and constrained
ordination)
o plotting factor levels in different colours with ordihull, ordiarrows,
ordisegments, ordispider, ordibar and ordiellipse (via plotting
functions for unconstrained and contrained ordination)
o summary and ordiareatest for ordihull and ordiellipse (via plotting
functions for unconstrained and contrained ordination)
o ehull option of ordiellipse ((via plotting functions for unconstrained
and contrained ordination))
* change in calculation for ensemble.threshold function when using thresholds from the
PresenceAbsence package with ensemble.threshold options of "threshold.mean"
and "threshold.min": five thresholds from the PresenceAbsence package will
be used (see Details)
* new parameter ENSEMBLE.weight.min to set minimum output weight for ensemble.test,
ensemble.test.splits and ensemble.batch. Note that parameter ENSEMBLE.min
refers to the minimum input weight (and typically used to exclude algorithms
with low AUC values).
* ensemble.plot allows to add country boundaries (via maptools package)
* ensemble.plot exports breaks and colours if only one suitability map is plotted
* breaks are forced to be unique in ensemble.plot
* ensemble.analogue gives warning is variables are very strongly correlated with
method 'mahal' (as algorithm my fail)
* corrected problem in ensemble.novel function when x only has one layer
* avoid error when launching the GUI when dismo package was not installed
* corrected error in labeling the y-axis in reyniplot function (error reported by
Mauricio Carrasquilla)
Version 2.7-1 (April 2016)
* ensemble.batch: procedure stops when no algorithms retained in the k-fold
cross-validation procedure
* new Graphical User Interface (GUI) for renaming layer names of calibration
and prediction stacks
* included link to Hijmans RJ and Elith J. 2016. Species distribution modelling with R.
* options to set width and height of new graphical device for ensemble.plot,
evaluation.strip.plot and ensemble.centroids functions. Setting any of these parameters
to a value < 0 results in no opening of a new device .
Version 2.7-0 (March 2016)
* new Graphical User Interface (GUI) for ensemble suitability modelling.
Guidelines for getting started are available from the GUI and in the 'doc'
folder of BiodiversityR
* included the ensemble.mean function within ensemble.batch, so that
ensemble.batch also produces the mean ('_MEAN_') ensemble files when several
ensembles are calibrated via ensemble.batch
* new function ensemble.plot, especially to apply a colourscheme that
differentiates between suitabilities below and above a threshold that
discriminates between predicted absence-presence of the organism
* new feature in the evaluation.strip.plot to include a boxplot for the
presence locations of a data set used for model calibrations
* included help buttons in various GUI windows
Version 2.6-1 (January 2016)
* ensemble.centroid and ensemble.zones: New functions to determine centroid
locations in principal components space of presence locations of an organism.
Mahalanobis distance from centroid locations to map zones based on the
minimum distance from centroid locations.
* report information on spatial sorting bias in ensemble.test.
* CIRLES.at and CIRCLES.d options in ensemble.test function to reduce spatial
sorting bias between evaluation presence and absence locations
* threshold.PresenceAbsence option in ensemble.test to use the PresenceAbsence
package to calculate absence-presence thresholds.
* ensemble.novel: New function to map novel environmental conditions.
* ensemble.area: New function to calculate areas of suitability raster maps.
* ensemble.analogue: New function to map environmental (climate) analogues.
* ensemble.ecocrop: New function to map suitability based on temperature
and rainfall optimal and absolute ranges.
* corrected problem in handling probit-transformations in ensemble.raster function
* started ChangeLog
Version 2.5-4 (September 2015)
* compatibility with Rcmdr restored (eg qqPlot)
* menu interface scripts updated
* new menu option of Analysis of diversity > Add diversity variables to data set
* new menu option of silhouette in the Cluster analysis plot options
Version 2.5-3 (July 2015)
* changes in the CAPdiscrim function to result in same results as PRIMER (problem
handled by including as.dist() function prior to cmdscale() function).
* new function importancevalue to calculate Importance Value Index, and new
ifri data set (collaboration with the International Forestry Resources and Institutions
[IFRI] research network; http://www.ifriresearch.net)
* new menu option of replacing NA with 0 in the community data set
* corrected problem in ensemble.test.splits function in obtain AUC-derived
weights for submodels when weights are not tuned
* ensured that subfolder to save models is created (ensemble.test function)
* included package information for function visibility
Version 2-5.2 (April 2015)
* included droplevels() function in function ensemble.test
* avoiding loading of libraries in ensemble functions (including gam::s()
for gam formula)
* included topography-derived variables in BCI.env data (provided by P. Legendre)
Version 2-5.1 (November 2014)
* menu option to remove species with zero total abundance
* handle new 8-character species codes in vegan in examples
* include options to plot as.phylo() cluster dendrograms
* option to reorder cluster dendrograms
* option to plot symbols in cluster diagrams
* BiodiversityR start up message
* expand vegan functions from menu interface (vegemite, tabasco, coverscale,
screeplot, bioenv, betadisper, contribdiv, beals, betadiverr, wcmscale, pcnm,
polaccum, estaccumR, eventstar, indpower, isomap, meandist, nestedchecker,
nestedtemp, nestednodf, nestedbetasor, nestedbetajac, RsquareAdj, drarefy,
simper, treeheight)
* corrected problems with jackknife of diversity analysis (diversityresult
function with method='jackknife')
Version 2-4.4 (May 2014)
* new PROBIT option in ensemble.test function to transform outputs of
ensemble submodels with probit transformation (also supported in
ensemble.raster function)
* new ENSEMBLE.exponent option to modify weights of ensemble submodels
in ensemble model by using exponent
* corrected use of persp() for accumulated renyi profiles in menu interface
Version 2-4.1 (January 2014)
* new functions of ensemble.simplified.categories and ensemble.dummy.variables
* new function of ensemble.evaluation.strip
* only allow ensemble calibration in ensemble.test and ensemble.test.splits
functions
* include legends in menu-interface generated plots no longer via locator()
Older versions
* mainly changes related to newer version of R base and loaded packages,
including changes required to maintain compatility with R-commander