Appendix B - Updating Valavi et al. 2021 benchmark study with biomod2 v4.3-4

In 2021, Valavi et al. published a benchmark study of different algorithms computing species distribution models over the same data set. Model predictions were compared according to three threshold-independent measures of predictive performance. Two ensemble models were included within the study: 1) one built with biomod2 package; 2) one combining single-models pre-selected by the authors. Among the concluding points, it was pointed out that the ensemble model built with biomod2 performed badly compared to simpler algorithms, being not tuned and using default algorithm parameters. Since then, biomod2 package has evolved, particularly in terms of the selection of modeling parameters. Broadly, one user can choose between 3 parameterization choices: 1) default parameters; 2) parameters generally optimized by the developing team (so called bigboss); 3) parameters tuned or self-chosen. We take advantage of the reproducible code made available by Valavi et al. 2021 to try and test biomod2 current version for any performance improvement. Knowing that few users take the time to fully explore each algorithms parameters and possibilities (which we are not in favor of), we decided to test for what we believe are the most common cases: selection of default or bigboss parameter sets. All data and code used for this review are presented here. Original data Valavi_2021_Fig3.png, Valavi_2021_Fig5.png, Valavi_2021_Fig7.png, Valavi_2021_Fig10.png Original figures of Valavi et al. 2021 paper. background_50k/ folder Original background data used in Valavi et al. 2021 paper and not directly available within disdat package.Available in Valavi et al. 2021 supplementary material. Computation of results SuppMat_biomod2_Valavi_SCRIPT1_modelRun_202509.R SCRIPT R used to run biomod2 v4.3-4.The script is based on the one provided in supplementary material by Valavi et al. 2021 and adapted to new biomod2 version. SuppMat_biomod2_Valavi_SCRIPT2_modelEvaluation_202509.R SCRIPT R used to compute evaluation metrics over biomod2 models.The script is based on the one provided in supplementary material by Valavi et al. 2021 and adapted to new biomod2 version.It produces output files contained in models_output.7z archive, as well as the EVAL_COMPLETE_evalSin_ALL.csv and EVAL_COMPLETE_evalEns_ALL.csv files. EVAL_COMPLETE_evalSin_ALL.csv Single models' evaluation table.group, region and spid come from the disdat dataset.time from 1 to 5 are indicative biomod2 computation times (see SCRIPT1).validation_valadi is produced by SCRIPT2, while other columns are directly produced by biomod2 functions. Columns: group region spid time1 time2 time3 time4 time5 full.name PA run algo SET metric.eval cutoff sensitivity specificity calibration_biomod validation_biomod evaluation validation_valavi EVAL_COMPLETE_evalEns_ALL.csv Ensemble models' evaluation table.group, region and spid come from the disdat dataset.time from 1 to 5 are indicative biomod2 computation times (see SCRIPT1).validation_valadi is produced by SCRIPT2, while other columns are directly produced by biomod2 functions. Columns: group region spid time1 time2 time3 time4 time5 full.name merged.by.PA merged.by.run merged.by.algo filtered.by algo SET metric.eval cutoff sensitivity specificity calibration_biomod validation_biomod evaluation validation_valavi Analysis of results SuppMat_biomod2_Valavi_SCRIPT3_results_202511.Rmd SCRIPT MARKDOWN used to analyse the output values and compare to previous results obtained by Valavi et al. 2021.The script is provided for the purposes of transparency, sharing, and reproducibility of analyses. Results AppendixB_biomod2_Valavi_20251211.html (or SuppMat_biomod2_Valavi_SCRIPT3_results_202510.html file) HTML OUTPUT file containing the analysis results. Main values or summary statistics are highlighted for each graphic.For some, parallel is made with results of previous analysis from Valavi et al. 2021.