News

Development updates

2026

Bugfix
  • 4.3-4-7 :
    • fix set_new_dirname function
    • add check when CV.do.full.models = NULL in BIOMOD_Modeling function
  • 4.3-4-6 :
    • update user related messages :
      • remove test argument from .fun_test[...] functions and use them whenever possible
      • add .fun_testIfNULL, .fun_testIfLength, .fun_testIfInOnlyOne, .fun_testIfSameSize functions, and change .fun_testIf01 and .fun_testMetric to .fun_testIf0X and .fun_testIfSize
      • add .bm_cat2 function and update all cat messages
      • add distinction between cat and message, and switch most warning to message
      • standardize / uniformize all cat, message, stop and warning messages
    • update default parameters in functions (set to NULL if possible)
    • add seed.val parameter in bm_CrossValidation
    • add check in BIOMOD_Projection and BIOMOD_EnsembleForecasting to deal with single model projection failing
    • remove .BIOMOD_EnsembleForecasting.prepare.workdir function
    • fix kfold selection for non-binary data in bm_CrossValidation
    • add check in bm_PseudoAbsences to select presences in PA.user.table when user.defined strategy
    • fix bm_Tuning for DNN : multiple sets (PA x RUN), hidden width and depth parameters
    • fix EMwmean for abundance models using RMSE, MSE, MAE or Max_error as selection metrics
    • add new video tutorial on YouTube
  • 4.3-4-5 : pass CRAN check

4.3-4-4 Clean version (2026-01-22)

  • Upgrade xgboost package from 1.7.10.1 to 3.1.3.1 :
    • change modelling function from xgboost to xgb_train in ModelsTable
    • update default parameters with xgb.params (for objective and verbosity)
    • update OptionsBigboss data
    • deactivate tuning for XGBOOST until bug is fixed
    • authorize categorical variables for XGBOOST
    • change from reshape to type (“response” or “class” value) parameter in xgboost::predict function
    • give data to xgb.train function through xgb.DMatrix
  • New ModelsTable and OptionsBigBoss
Bugfix
  • Remove obsolete parameter (transpose) in pROC::coords function in bm_FindOptimStat
  • Fix bm_PseudoAbsences with strategy = 'user.defined' and true absences in resp.var
  • Update vignettes

2025

Bugfix
  • 4.3-4-3 :
    • split Main functions example in 2 : binary and abund
    • add DataSTOC data set
    • fix tuning function for DNN
    • fix biomod2_template_report.Rmd to adapt to last plot changes and abundance data
    • fix plot function for BIOMOD.formated.data with plot.type == "raster" & plot.valid for _allData_allRun case
    • fix metric.eval in plot.Rscore in bm_ModelAnalysis function
  • 4.3-4-2 :
    • merge BIOMOD.formated.data plot functions to deal with all data types, add plot.valid parameter
    • reformat bm_ModelAnalysis function
  • 4.3-4-1 : change color to fill, and add filtered.by in bm_Plot[...]Boxplot functions (Eval and VarImp)

4.3-4 DNN model (2025-06-04)

  • Add DNN : Deep Neural Network with cito package
  • New ModelsTable and OptionsBigBoss
  • Tuning made with the tune function of cito package
Bugfix
  • Fix EMwmean when only one model selected
  • Fix bm_RunModelsLoop output list
  • Fix plot function of BIOMOD.projection.out object

4.3-3 Multiclass datatype (2025-05-20)

Major Changes
  • In biomod2_classes_0 file :
    • add new data type : “multiclass”
    • adaptation of .BIOMOD.options.default.correct function
  • In biomod2_classes_1 file :
    • adaptation of plot method for BIOMOD.projection.out object
  • In biomod2_classes_3 file :
    • adaptation of .BIOMOD.formated.data.check_data, .plot.BIOMOD.formated.data.abundance, summary and show function
  • In biomod2_classes_4 file :
    • factor_levels slot for biomod2_model class
    • update all predict2 functions for single models to deal with the different data types
  • In bm_RunModelsLoop, BIOMOD_Projection, BIOMOD_EnsembleModeling and BIOMOD_EnsembleForecasting functions :
    • management of model predictions format, in particular for “ordinal” and “multiclass” cases
  • In biomod2_classes_5 file :
    • definition of EMmode and EMfreq with predict methods
Bugfix
  • 4.3-2-3 : fix data.type in bm_ModelingOptions
  • 4.3-2-2 :
    • change the name of the metric ROC to AUCroc. It will switch automatically if you use ROC
    • add the metric AUCprg : Area Under Curve of the Precision-Recall-Gain curve
  • 4.3-2-1 :
    • add togglelayerselected, maximumbackground, maximumiterations, convergencethreshold, autofeature, jackknife, writeclampgrid, writemess, logfile and verbose parameters in MAXENT parameters
    • fix .check_calib.lines_names for BIOMOD.formated.data.PA object in bm_CrossValidation_user.defined
    • merge binary and nonbinary models into OptionsBigboss
    • set data.type = 'binary' by default in bm_ModelingOptions
    • correct .BIOMOD_Modeling.summary
    • remove gam:: in bm_MakeFormula

4.3-2 Report (2025-03-11)

  • Add call slot for BIOMOD.formated.data and BIOMOD.formated.data.PA, BIOMOD.models.out, BIOMOD.projection.out, and BIOMOD.ensemble.models.out classes
  • Add BIOMOD_Report function
  • Add .Rmd templates in inst/rmd/ folder (for report, ODMAP and code)

4.3-1 RangeSize (2025-03-03)

  • Change BIOMOD_RangeSize to bm_RangeSize
  • Change BIOMOD_RangeSize inputs to BIOMOD.projection.out objects
  • Add new class BIOMOD.rangesize.out
  • Change bm_PlotRangeSize input to BIOMOD.rangesize.out objects, and project maps with coordinates of BIOMOD.projection.out object
  • Fix show outputs of BIOMOD.[...].out objects

4.3-0 Abundance (2025-01-30)

Major Changes
  • In biomod2_classes_0 file :
    • add new data types : “binary”, “abundance”, “count”, “ordinal”, “relative”, “nonbinary”
    • add data.type, has.filter.raster and biomod2.version slot for BIOMOD.formated.data and BIOMOD.formated.data.PA classes
    • add .BIOMOD.options.default.correct function (default changes and corrections depending on the data type (mainly for type, method, family and distribution options) to BIOMOD.options.dataset)
  • In biomod2_classes_1 file :
    • add .BIOMOD.formated.data.check_data function (routine that can be applied to both original and evaluation datasets)
    • add .plot.BIOMOD.formated.data.abundance function (to plot formated data when data type is not binary)
    • update BIOMOD.formated.data summary to the different data types
  • In biomod2_classes_3 file :
    • add data.type slot for BIOMOD.models.out class
    • add set_new_dirname and set_new_dirname.models functions (to recursively modify the dir.name slot in all biomod2 objects of an existing simulation directory)
  • In biomod2_classes_4 file :
    • add model_type and thresholds_ordinal slot for biomod2_model class
    • update all predict2 functions for single models to deal with the different data types (mainly between binary, ordinal and the others)
  • In bm_CrossValidation function :
    • split .sample_mat function into .sample_num and .sample_class functions
    • update all functions to deal with the different data types (mainly between ordinal and the others)
  • In bm_FindOptimStat function :
    • add new evaluation metrics : “RMSE”, “MAE”, “MSE”, “Rsquared”, “Rsquared_aj”, “Max_error” (for abundance/count/relative data) and “Accuracy”, “Recall”, “Precision”, “F1” (for ordinal data)
    • add k parameter
    • add .contingency_table_ordinal function
    • split bm_CalculateStat function into bm_CalculateStatBin and bm_CalculateStatAbun functions
  • In bm_VariablesImportance function :
    • use Spearman correlation for ordinal, Pearson otherwise
  • Use either on_0_1000 or on_1_1000 when making projections
  • Correct and adapt tuning
  • Add bm_ModelAnalysis function (to analyse the residuals of single models)
Minor Changes
  • Update ModelsTable data with nonbinary data type
  • Update OptionsBigboss data with all nonbinary single models options
  • Change dir.name slot within BIOMOD.formated.data into absolute one instead of relative by default
  • Change aes_string to [aes + .data] in bm_Plot[...] functions
  • Add digits and overwrite parameters in BIOMOD_Projection function
  • Change get_var_type and get_var_range to internal functions (.get_var_type and .get_var_range)
  • Remove deprecated functions and corresponding documentation
  • Add vignette for abundance (to help you navigate through the new datatypes available)
  • Update examples and documentation
Bugfix
  • Fix stratified cross-validation strategy in bm_CrossValidation (partitions not being balanced correctly)
  • Fix selection of metrics in bm_PlotEvalMean
  • Fix formula when given by user in bm_Tuning (to be able to run the step AIC with the user formula)
  • Change scale.models = FALSE by default in BIOMOD_Modeling and bm_RunModelsLoop

2024

Bugfix
  • 4.2-6-2 :
    • fix get_evaluations and get_variables_importance message when data is not available (extended to bm_Plot[...] functions)
    • add cluster management code through parallel::stopCluster and foreach:::.foreachGlobals
      finally rather through .errorhandling in foreach loop in bm_RunModelsLoop (for MAXENT on Windows)
    • merge request improving balanced partitions for stratification strategy in bm_CrossValidation
    • fix metric selection in bm_PlotEvalMean
    • add new tutorial video (for version 4.2-6)
    • add MAXENT.partitions, MAXENT.kfolds, MAXENT.user.grp and MAXENT.tune.args parameters in bm_Tuning function
  • 4.2-6-1 : fixed path failure within species name (e.g. data format like tif)

4.2-6 RFd and some tuning (2024-07-22)

  • Warning ! Some options for OptionsBigboss have been modified (concerns only ANN, CTA and RF models)
  • Some changes for the tuning ranges.
  • Add RFd : Random Forest with a down-sampling method.
  • Add seed.val for bm_pseudoAbsences and BIOMOD_FormatingData.
  • Add fact.aggr argument for pseudo-absence selection with the random and disk methods. It allows to reduce the resolution of the environment.
  • Possibility to give the same options to all datasets with “for_all_datasets” in bm_ModelingOptions.
Bugfix
  • 4.2-5-2 :
    • fixed BIOMOD_EnsembleModeling when multiple PA datasets (obs and fit not matching when calling bm_FindOptimStat)
    • always keep same order for variable names for new.env in BIOMOD_Projection and BIOMOD_EnsembleForecasting
  • 4.2-5-1 :
    • removed tests/ folder (unused)
    • fixed check for variable names for new.env in BIOMOD_Projection and BIOMOD_EnsembleForecasting
    • fixed connection issue with MAXENT tuning

4.2-5 Modeling options and Tuning (2024-05-02)

Major Changes
  • Add print function for BIOMOD.options.dataset and BIOMOD.models.options classes
  • Add GAM.binary.gam.gam and GAM.binary.mgcv.bam in OptionsBigboss dataset
  • Add checks for obs and fit parameters in bm_FindOptimStat
  • In bm_Tuning function :
    • add MAXENT.algorithm and MAXENT.parallel parameters
    • be sure to update default tuning parameters if not all provided
    • fix tuning for XGBOOST and SRE
    • fix optimization of formula
    • switch from gamSpline to gamLoess method to tune GAM.gam.gam model, and add GAM.span and GAM.degree parameters
  • Fix multiple cases when using different settings of PA / cross-validation / modeling options through all the different secundary functions and BIOMOD_Modeling
Minor Changes
  • Add vignette for cross-validation (to prepare your own calibration / validation datasets)
  • Add vignette for modeling options (to help you navigate through the new way of parameterizing single models)
  • Update examples and documentation
Bugfix
  • Add try and error message in call for models in bm_RunModelsLoop
  • Check for predictions >1000
  • Correct BIOMOD_EnsembleForecasting when using data.frame
  • Correct predictions for EMci (removing round)

2023

4.2-5 Modeling options and Tuning (2023-09-12)

Major Changes
  • Reformat modeling options, how they are created and how they are given to models :
    • Create BIOMOD.options.default and BIOMOD.options.dataset classes, retrieving default parameters and values with formalArgs function
    • Modeling options can now be different for PA x cross-validation datasets x models
    • Add ModelsTable and OptionsBigboss datasets containing single models informations and pre-defined modeling options
    • Move BIOMOD_ModelingOptions to bm_ModelingOptions
    • Call bm_ModelingOptions directly in BIOMOD_Modeling and add related OPT.[...] parameters
  • Move BIOMOD_Tuning to bm_Tuning and adapt it to match with new modeling options
  • Modify the call to single models in bm_RunModelsLoop in a more generalized way dealing with new modeling options
Minor Changes
  • Remove BIOMOD_PresenceOnly function and add BOYCE and MPA indices into bm_FindOptimStat function
  • Save ensemble projections with FLT4S datatype only when EMcv is activated, otherwise use INT2S
  • Add vignette for data preparation (questions you should ask yourself before modeling)
  • Correct getters (variable importance, built models)
  • Update examples and documentation
Bugfix
  • 4.2-4-10 : fixed wrap applied to a data.frame in BIOMOD_Projection
  • 4.2-4-9 : fixed predict method for RF with do.classif = FALSE
  • 4.2-4-8 : improved tests in bm_PlotEvalMean
  • 4.2-4-7 : fixed do.classif ignored in BIOMOD_ModelingOptions
  • 4.2-4-6 : fixed parallelization in BIOMOD_Projection
  • 4.2-4-5 : fixed summary method for BIOMOD.formated.data and BIOMOD.formated.data.PA
  • 4.2-4-4 : fixed bm_PlotResponseCurves for ensemble models merged by algo (for Maxent)
  • 4.2-4-3 :
    • added point.size argument to plot.BIOMOD.Formated.data
    • added maxcell argument to plot.BIOMOD.projected.out
  • 4.2-4-2 : set XGBOOST verbose = 0 (from verbose = 1)
  • 4.2-4-1 : fixed BIOMOD_FormatingData checks for resp.xy

4.2-4 XGBOOST (2023-06-21)

Major Changes
  • added XGBOOST as a possible algorithm in BIOMOD_Modeling
Minor Changes
  • changed CV.do.full.models default value to FALSE
Bugfix
  • 4.2-3-5 :
    • fixed some more issues related to categorical variables badly interacting with missing values. Projection and Ensemble Forecasting are now only calculated on cells without any missing values.
    • removed obsolete code for SRE pseudo absences sampling with categorical variables, as SRE do not work with categorical variables
    • added internal function to get mask of data .get_data_mask
  • 4.2-3-4 : fixed bug in BIOMOD_Modeling when using sampsize as a vector. argument strata was badly formatted
  • 4.2-3-3 : fixed bug in BIOMOD_EnsembleModeling for additional projection with only one environmental variables
  • 4.2-3-2 : fixed bugs in BIOMOD_EnsembleForecasting when several projection are running simultaneously and using the same temporary directory
  • 4.2-3-1 : fixed bugs in bm_CrossValidation with user.defined tables badly formatted (TRUE/FALSE for data not in the given PA dataset are now properly transformed into NA)

4.2-3 Cross-Validation and Pseudo-Absences (2023-05-09)

Major Changes
  • Improved pseudo-absence management: it is now possible to have pseudo-absence dataset of different size and algorithm can be setup to run on different pseudo-absence dataset (with models.pa argument in BIOMOD_Modeling).
  • Rework and harmonization of cross-validation function. BIOMOD_CrossValidation have been renamed bm_CrossValidation and cross-validation with k-fold, stratified and environmental strategy now work properly with pseudo-absence dataset. All cross-validation strategy can now be called directly through BIOMOD_Modeling.
Minor Changes
  • improved Documentation (get_evaluations, BIOMOD_EnsembleModeling, bm_RunModelsLoop, bm_RunModel)
  • updated website tutorial to use current biomod2 version
  • removed unused parameter save.output. output are now automatically saved.
  • improved management of categorical raster for both terra and raster
  • CV.perc (formerly data.split.perc) now uses a 0-1 range (instead of 0-100)
  • deprecated arguments for BIOMOD_EnsembleModeling now gives an error.
  • added argument metric.select.dataset to BIOMOD_EnsembleModeling to choose the dataset which evaluation metric should be used to filter and/or weigh the ensemble models. Default value is now ‘validation’ instead of ‘evaluation’.
  • added argument na.rm to BIOMOD_EnsembleModeling to harmonize the management of NA among individual model predictions.
Bugfix
  • validation metric calculation now properly use the calibration threshold (i.e a threshold optimized on calibration data instead of validation data). This can lead to less optimistic threshold-dependent validation metric.
  • fixed SRE projection assuming the same variables ordering in calibration and projection data
  • ensemble model can again be calculated over models without validation
  • correct print for multiple values in RF$sampsize parameter in BIOMOD_ModelingOptions
  • fixed layer name in BIOMOD_Projection and BIOMOD_EnsembleForecasting when terraOption(todisk = TRUE) is activated (for large or numerous raster).
  • fixed Ensemble Models based on models without cross-validation (“allRun”)
  • model is now robust to using data.table object (that are converted into standard data.frame).
  • fixed projection raster name when using do.stack = FALSE and resp.name with . inside.
  • fixed using user-defined pseudo-absences along with filter.raster = TRUE in bm_PseudoAbsence.
  • fixed weights calculation when using only one pseudo-absence dataset
  • fixed summary and show method for BIOMOD.formated.data.PA
Internal
  • add internal function get_species_data and get_eval_data
  • removed .BIOMOD_Modeling.prepare.data
  • reorganised bm_RunModelsLoop to do the PA loop within the function
  • calib.lines and eval.lines variable names are standardised (no more calibLines or eval_lines)
  • removed dependency to data.table (removed use of rbindlist)
  • added .get_env_class to reduce code redundancy
  • renamed categorical_stack_to_terra into .categorical_stack_to_terra
  • dispatched some of BIOMOD_FormatingData checks into bm_PseudoAbsences

4.2-2 Improvement patch (2023-01-13)

Major Changes
  • '.tif' is available as an output format for raster projection
  • '.tif' is the new default output format for raster projection
  • Improved plot and summary methods for BIOMOD_FormatingData output. These method now support the use of calib.lines to explore how the cross-validation dataset are structured.
  • Updated plot methods for BIOMOD.projection.out objects so that it uses ggplot2 for nicer plots.
  • Binary and Filtered transformation are now properly stored in BIOMOD.projection.out objects. They can be loaded from the disk with get_predictions or represented through BIOMOD.projection.out plot method.
  • get_predictions now return a proper data.frame (unless projection on spatial data) with many additional information available. Old behavior can be reproduced by using get_predictions(x, model.as.col = TRUE).
  • get_evaluations now return a cleaner data.frame with more consistent information available.
Minor Changes
  • Simplified maxent model names: ‘MAXENT.Phillips’ -> ‘MAXENT’ (based on maxent.jar); ‘MAXENT.Phillips.2’ -> ‘MAXNET’ (based on maxnet package).
  • BIOMOD_FormatingData now gives warning when several input data points are located in the same raster cells
  • Added options filter.raster in BIOMOD_FormatingData to filter data points so that none are located in the same raster cells.
  • BIOMOD_EnsembleModeling now have an argument em.algo to select the ensemble algorithm to be computed. Separate arguments are now deprecated (prob.mean, prob.median, prob.cv, prob.ci, committee.averaging, prob.mean.weight). Building all possible ensemble models can now be done with em.algo = c('EMmean','EMmedian','EMcv','EMci','EMca','EMwmean').
  • Some possible values for em.by have slightly changed: ‘PA_dataset’ -> ‘PA’, ‘PA_dataset+repet’ -> ‘PA+run’ and ‘PA_dataset+algo’ -> ‘PA+algo’
  • Added an appropriate message when all models fail for BIOMOD_Modeling and BIOMOD_EnsembleModeling.
Bugfix
  • Fixed MAXENT.Phillips.2 and single variable models.
  • Fixed ensemble models when several filtering metrics were asked and some combination of ensemble/metrics had no models
  • Bugfix for projection for ensemble models to ensure that the proper set of models was selected
  • Bugfix for BIOMOD_CrossValidation for block-stratified sampling
  • Bugfix for BIOMOD_CrossValidation for pseudo-absences
Internal Changes
  • Array have disappeared from most internal functions
  • Removed rasterVis from Suggests
  • Added tidyterra and ggtext to Suggests
  • Added checks to get_evaluation when models have no evaluations.

2022

4.2-1 Bugfix patch

Major Changes
  • Package sp is back into Imports due to the need to use sp::read.asciigrid
  • Added control for terra version number (>= 1.6-33) as terra 1.6-41 was released on CRAN.
  • With do.stack = TRUE, only stacked projection are now saved to the disk.
Minor Changes
  • Added initial_heap_size and max_heap_size in MAXENT.Phillips modeling options
  • Improved projection efficiency for raster with MAXENT.Phillips.
Bugfix
  • Fixed MAXENT.Phillips predict method for large dataset (require sp::read.asciigrid).
  • Fixed ensemble models using a single PA dataset with em.by = 'all' or 'algo'.
  • Models using repetition dataset cannot be merged anymore with models using Full dataset in BIOMOD_EnsembleModeling.
  • Fixed error in BIOMOD_EnsembleForecasting when a single evaluation metric was available and binary/filtered transformation were asked for.
  • Bugfix for plot method for BIOMOD.formated.data.PA object.
  • Fixed MAXENT.Phillips for Windows.
  • Fixed using do.stack = FALSE with BIOMOD_Projection.
  • Fixed EMcv ensemble modeling for data.frame by removing dependency to raster::cv.
  • Fixed free method with PackedSpatRaster
  • Fixed BIOMOD_FormatingData in case where no coordinates are given
Internal
  • Updated github workflow : removed obsolete ubuntu 18.04 ; added test on r-devel ; added cache for R packages.
  • Updated MAXENT.Phillips predict2 method for SpatRaster so that it saves environmental data as .asc and do not use the data.frame method.
  • Fixed some automatic boolean conversion
  • Cleaned up [email protected] slot. data.mask can now be safely saved and re-opened ; data.mask can now store a different extent for evaluation dataset

4.2-0 Terra Update

Major Changes
  • Package now rely only on terra (> 1.6.33) and do not automatically import raster and sp.
  • Moved raster and sp package into SUGGESTS rather than DEPENDS.
  • raster and sp input data type are still supported.
  • Package dataset now are now documented and loaded with data() .
Minor Changes
  • bm_BinaryTransformation now always returns 0/1 and never TRUE/FALSE
  • Added a check to bm_PlotResponseCurves for new.env possible data types.
  • BIOMOD_Projection and BIOMOD_EnsembleForecasting now properly support matrix as new.env
  • get_prediction on biomod.projection.out generated from BIOMOD_Projection based on SpatRaster with arg as.data.frame = TRUE are now possible.
  • bm_BinaryTransformation now return same type of object as its input
  • Improved communication for BIOMOD_RangeSize, indicating how comparison are done depending on the number of models in current vs future.
  • Added argument check for BIOMOD_CrossValidation
Bugfix
  • MAXENT.Phillips models can now properly fail
  • bugfix for bm_BinaryTransformation with data.frame/matrix and do.filtering = TRUE
  • Removed obsolete warning about CTA and categorical variables when using raster
  • bm_PlotResponseCurves now work with factors in univariate representation
  • bm_PlotResponseCurves properly handles SpatRaster and Raster as new.env
  • Bugfix for predictions with MAXENT.Phillips and a single environmental variable
  • Bugfix for BIOMOD_EnsembleForecasting so that it properly accounts for new.env.xy when projecting on matrix or data.frame.
  • BIOMOD_EnsembleModeling now works when called for a single ensemble model
  • Improved argument check for BIOMOD_RangeSize. Comparisons with non-binary values throw errors.
  • bugfix for BIOMOD_RangeSize and data.frame method
  • BIOMOD_RangeSize data.frame method now handles 1 current vs n future projection
  • bugfix for BIOMOD_PresenceOnly that can now work when evaluation data are provided
  • bugfix for BIOMOD_PresenceOnly that can now work when only the EM have been provided
  • expanded support for BIOMOD_PresenceOnly to SpatRaster and SpatVector.
  • build_clamping_mask now support categorical variables
  • fixed ensemble model EMcv based on a single environmental variable
Internal Changes
  • New internal function .categorical2numeric to transform categorical variables into numeric within a data.frame.
  • New internal function .get_categorical_names to retrieve categorical variable names from a data.frame.
  • Split load_stored_object method into a method for BIOMOD.stored.SpatRaster and a method for all other BIOMOD.stored.data.
  • BIOMOD.stored.SpatRaster stores PackedSpatraster and not SpatRaster.
  • New internal function .CompteurSp based on old function CompteurSp that was defined within a function.
  • Removed obsolete function check_data_range().

4.1-3

Bugfix
  • Ensemble models that fails (e.g. EMcv with only one models) will not crash the full ensemble run. Instead a warning is generally displayed at the beginning and the resulting object will list failed models.
  • Fixed CTA raster prediction for categorical variables.
  • Fixed binary transformation in BIOMOD_EnsembleForecasting.

Internal Changes

  • New internal function .get_kept_models to generate list of models kept by ensemble modeling depending on metric.select.
  • Improved checks for BIOMOD_EnsembleModeling to generate warnings when ensemble models are expected to be run with <= 1 models.
  • Repaired support for cross-validation table given as data.frame instead of matrix.

4.1-2 (2022-09-29)

Major changes
  • dir.name can now be provided as project argument so that results may be saved in a custom folder.
  • predict with CTA algorithm and categorical variables on raster is now possible.
  • Changed evaluation for EM models merging PA datasets (em.by = "algo" or em.by = "all") so that evaluation uses the union of PA data sets instead of the whole environmental space supplied.
Minor changes
  • Individual EM models projected as raster are now saved with INT2S data format when on_0_1000 is set to TRUE.
  • Homogenize the use of load functions (use get_[...], load_stored_object and BIOMOD_LoadModels, instead of get(load(...))) and the workflow within get_[...] functions (use load_stored_object and similar arguments such as as.data.frame, full.name, …).
  • Adapting predict workflow to properly use S4 class.
  • Homogenize BIOMOD.ensemble.models.out and BIOMOD.models.out objects
    • Add slots in BIOMOD.ensemble.models.out object for evaluations, variables importance and predictions.
    • Moved .Models.save.objects in BIOMOD_modeling to .fill_BIOMOD.models.out in biomod2_internal.R.
    • Save slots within BIOMOD.ensemble.models.out and use load_stored_object to directly get them within get_[...] functions.
Bug Fix
  • Validation data can now be properly combined with Pseudo-Absence in BIOMOD_FormatingData, instead of throwing an error linked to data.mask.
  • Argument on_0_1000 can now be passed without errors so that projection may either be on a range from 0 to 1 or from 0 to 1000. The latter option being more effective memory-wise.
  • Completed argument check for function BIOMOD_EnsembleModeling so that em.by can not be of length > 1.
  • Corrected function .get_models_assembling so that it did not confound MAXENT.Phillips2 with MAXENT.Phillips when grouping models by algorithm in BIOMOD_EnsembleModeling.
  • get_predictions method for BIOMOD.ensemble.models.out now accepts an evaluation arg. Evaluation values, variables’ importance and Calibration/Evaluation predictions for ensemble models are now properly saved by BIOMOD_EnsembleModeling().
  • Evaluation metrics are no longer calculated for models prob.ci.inf et prob.ci.sup.
  • Package now properly pass R CMD check.
  • BIOMOD_PresenceOnly now properly manage NA.
  • Corrected bm_PlotResponseCurves to only plot show.variables.
  • get_predictions.BIOMOD.projection.out now properly works when asked for a subset of model.
Miscellaneous

4.1-1 (2022-08-30)

  • add do.progress parameter (to render or not progress bar) and dir.name parameter in BIOMOD_FormatingData and biomod2 objects (Mathieu B. request)
  • fix BIOMOD_PresenceOnly function by removing ecospat dependency

4.1 (2022-07-12)

  • fix bugs following major release 4.0

4.0 (2022-03-01)

  • MAJOR RELEASE
  • clean all functions, reorganize files, remove old / unused functions
  • standardize function names and parameter names
  • update roxygen2 documentation for all functions, including examples
  • create github website to host documentation, examples, vignettes, news

2021

3.5-3 (2021-11-02)

  • clean BIOMOD classes definitions and functions (biomod2_classes files)
  • clean BIOMOD_FormatingData function
  • clean BIOMOD_ModelingOptions function
  • fix BIOMOD_FormatingData : test class condition only a first element (to deal with matrix / array objects)
  • fix BIOMOD_EnsembleForecasting for EMcv model when only one single model was kept

3.5-2 (2021-10-18)

  • fix BIOMOD_PresenceOnly function (previously BIOMOD_presenceonly)
  • fix BIOMOD_CrossValidation function (previously BIOMOD_cv)
  • fix internal function to find MinMax values, when factor included : should get clamping mask to work

2018-2019

3.3-20 (2019-03-05)

  • Remove maxent Tsurukoa because not maintained anymore (required by CRAN team)

3.3-18 (2018-07-04)

  • fix the gbm multicore issue

3.3-17 (2018-04-23)

  • correct the single presence pseudo-absences generation bug (Matthias G.)

2016

3.3-6 (2016-01-14)

  • add get_predictions function for ensemble models

3.3-5 (2016-01-04)

  • MARS models are now computed throw earth package (was mda in previous versions)
  • MARS now supports factorial explanatory variables
  • MARS now supports formula

2015

3.3-4 (2015-11-04)

  • update BIOMOD_tuning function (Frank B.)

3.3-3 (2015-10-27)

  • force sampling of each level of factorial variables
  • add betamultiplier parameter to tune MAXENT.Phillips (Frank B. request)

3.3-00 (2015-10-05)

  • MAJOR RELEASE
  • optimize the memory consumption of projections and ensemble projections procedure
  • add the possibility to run MAXENT.Phillips with proper background data
  • classical version of MAXENT has been renamed MAXENT.Phillips
  • add a new version of MAXENT MAXENT.Tsuruoka

3.2-00 (2015-07-28)

  • add 3 new functions in biomod2 (Frank B. contribution)
  • BIOMOD_cv to control models cross validation procedure
  • BIOMOD_presenceonly to evaluate biomod models using boyce and mpa indices
  • BIOMOD_tuning to automatically tune BIOMOD_ModelingOptions parameters

2014

3.1-59 (2014-10-23)

  • add model evaluation scores plotting function
  • dependence to ggplot2

3.1-53 (2014-08-06)

  • new ensemble models names to be more coherent with formal models names

3.1-44 (2014-05-20)

  • possibility to use user defined function to influence the way models are weighted in weighted mean ensemble models (thanks to Frank B.)

3.1-43 (2014-05-20)

  • add of as.data.frame argument for get_evaluations() function to enable formal and ensemble models evaluation scores merging

3.1-42 (2014-05-19)

  • enable ensemble forecasting models selection (thanks to Robin E.)

2013

3.1-17 (2013-10-23)

  • add parameter to control amount of memory reserved for MAXENT calculations (via java) (thanks to Burke G.)
  • optimization of memory consumption in models projections when do.stack argument is set to FALSE
  • binary and filtering projections output re-activated

3.1-1 (2013-09-04)

  • limitation of package dependencies
  • fairely definition of package namespace
  • add functions to update biomod2 objects from a version to the current one

3.0.2 (2013-07-23)

  • new functions to evaluate a-posteriori models quality
  • remove weights for models scaling and set it FALSE by default

3.0.0 (2013-07-01)

  • MAJOR RELEASES
  • ensemble models are now biomod2 models objects (should be predicted, evaluated, and you can do variables importance) the same way than all formal biomod2 models
  • possibility to produce ensemble models response plot
  • ensemble forecasting output is now a biomod2_projection object: should be plotted…
  • ensemble forecasting is now doable without doing previous projections (even if it is still advised). Can work with raw explanatory variables
  • getter and setter function have been renamed
  • new variable_importance function
  • …

2.1.37 (2013-06-12)

  • change (temporally?) gam default package from mgcv to gam to deal with memory (cache) over-consuming (thanks to Burke G.)
  • update of response.plot2 function (optimization + deal with factorial variables)

2.1.32 (2013-05-30)

  • weights for user defined pseudo-absences are now supported (thanks to Rui F.)
  • deal with unknown factors predictions (thanks to Denis M.)

2.1.13 (2013-03-06)

  • Add ProbDensFunc() function to package to produce nice plots that show inter-models variability

2.1.12 (2013-03-04)

  • add rasterVis dependency for nicer biomod2 plots
  • PA.dist.min and PA.dist.max are now defined in meters when you work with unprojected raster in disk pseudo-absence selection

2.1.9 (2013-02-28)

  • possibility to indicate manually which data should be used for calibration (resp. for validation) of models within BIOMOD_Modeling

2.1.9 (2013-02-27)

  • one var modeling supported (thanks Anne O.)
  • new options for response curves plotting (col, lty, data_species…)

2.1.8 (2013-02-25)

  • response plot supports now formal models

2.1.0 (2013-02-21)

  • MAJOR RELEASE
  • CRAN SUBMISION
  • add of a modeling.id arg (BIOMOD_Modeling) for prevent from no wanted models overwriting and facilitate models tests and comparisons (thanks Frank B.)
  • change of biomod2 dataset
  • vignettes and help files update (thanks Sam P. & Signe N.)
  • save link between modeling and projection objects
  • add pROC package dependency
  • add a modeling cleaner that remove modeling objects from both memory and hard drive: RemoveProperly()

2.0.11 (2013-02-18)

  • possibility to consider a user.defined pseudo-absence selection (thanks to Signe N.)
  • possibility to switch off stepwise glm selection (*thanks Frank B.)

2.0.9 (2013-02-15)

  • automatic save on hard drive of BIOMOD_Projection outputs

2.0.8 (2013-02-14)

  • BIOMOD_LoadModels supports multiple models input
  • deal with NA in evaluation table issue (*thanks Frank B.)

2.0.7 (2013-02-12)

  • bug on weights corrected (thanks to Lugi M.)

2.0.3 (2013-01-18)

  • deal with MAXENT categorical variables and categorical raster input

2.0.0 (2013-01-17)

  • MAJOR RELEASE
  • CRAN SUBMISION
  • models built within biomod2 are now defined as “biomod2 models objects” (own scaling models, own predict function, …)
  • full paths are replaced by relative paths to favor portability
  • harmonization of names of objects that are saved on hard drive (more coherence between functions)
  • possibility to save projections directly in raster format (.grd or .img)

Year 0

1.x.x

  • development phase