BioModelling Pro

Biogeography and species distribution modelling, in your browser. Download presence records from GBIF, clean them, extract bioclimatic, climate and soil variables, explore them with statistics and multivariate methods, and model the species distribution with ten algorithms and an ensemble — with maps, validation, thresholds, variable importance and projections to climate scenarios. Nothing to install, no account, no server of our own: your files and analyses stay on your computer.

The interface is bilingual (Spanish / English) and has light and dark themes. Both switches are in the top bar; the choice is remembered.

Use it online

👉 https://luisangelbg.github.io/BioModellingPro/

You only need a modern browser and an internet connection for GBIF and the base maps. The Python engine ships inside the app (vendor/pyodide/): served by the local server it works offline; opened by double-click, the browser cannot read it and it is fetched online the first time.

Use it locally

Download or clone the repository and double-click index.html. Everything works from file://, including the example records (Pinus cembroides, 2,400 records from GBIF). The climate layers you download once: WorldClim does not allow them to be redistributed, so they do not travel with the program — datos/LEE-ME.md says where to get them and step 5 reads them from wherever you keep them. Köppen-Geiger is bundled. To serve the folder on your network, run Open BioModelling Pro.bat (Windows) or server.ps1 and open http://localhost:9400.

The workflow, in ten steps

#	Step	What it does
1	GBIF or your own data	Resolves the scientific name and downloads every record with coordinates —optionally restricted to the countries you tick, with one-click shortcuts for Mexico, Central America and South America— or loads your own file (CSV, TXT or Excel with species, longitude and latitude; DMS accepted), alone or added to the GBIF records. A bundled example is available.
2	Filter	By taxon, country, year, record type, a bounding box you can draw, and state or region: the 32 states of Mexico come bundled and the state of each record is assigned from its coordinates (which completes the records GBIF delivers without one), with shortcuts by group of states; you can also load your own region layer (municipalities, watersheds, protected areas) as GeoJSON.
3	Clean	Nine coordinate-quality rules (duplicates, impossible values, country centroids, low precision, outliers…) with a before/after report.
4	Map	Publication-ready map: base layers, legend, north arrow, scale bar, graticule, PNG export.
5	Variables	19 bioclimatic variables and elevation (WorldClim 2.1), Köppen-Geiger climate and WRB soil type (SoilGrids) at every point; thematic map.
6	Correlation	Interactive Pearson / Spearman correlation matrix with dendrogram, live VIF of the variables you tick, a recommended low-collinearity selection that respects the variables you pin (for example BIO1 and BIO12) and a pair viewer. Your choice becomes the default of the next steps.
7	Statistics	Descriptive statistics, histograms, box plots and normality tests with suggested transformations.
8	ML, ecology and diversity	PCA, clustering (hierarchical, k-means, PAM, fuzzy, Gaussian mixture, DBSCAN), correspondence analysis, metrics of range area, spatial pattern, niche breadth and overlap, and species diversity: richness and its estimators, Shannon, Simpson, Hill numbers, evenness, rarefaction at equal effort and coverage, and beta diversity with its turnover / nestedness partition and ordination of sites.
9	SDM modelling	Ten algorithms plus an ensemble, spatial cross-validation, thresholds, variable importance, response curves, maps, and a future-projection workbench (many scenarios, model ensemble and agreement, range-shift metrics, dispersal assumptions, refugia).

In steps 7, 8, 9 and 10 you can still change which variables enter the analysis — even collinear ones that step 6 flagged, when you consider them ecologically important.

The home page includes a virtual-species lab: a synthetic landscape with a niche you know, sampled presences (with or without sampling bias) and three real algorithms whose estimate is compared with the truth.

Interpretation help

Every number in steps 6 to 10 carries a help button. It opens a card that says, in plain words, what the value measures, how to read it, on which scale (a bar of coloured bands shows where the value falls: |r|, VIF, AUC, TSS, Boyce, silhouette, Hopkins, Shannon, the red-list area thresholds, the land suitability classes…), the mistake most often made with it, and the reference it comes from. When the scale is a reading convention rather than a statistical law the card says so, and when no accepted scale exists it says that too. Each step ends with a collapsible interpretation guide holding all of its cards, and opens with a box explaining what the step is for and what you decide in it.

Map export

The maps of steps 4, 5 and 9 are edited in a studio with tabs (base layers, text and fonts per element, legend, scale bar, north arrow, graticule, colours, saved styles). On export you can see a preview of the exact image that will be written, clip the map to the country or to the states chosen in step 2 (everything outside becomes transparent, with the outline of the country drawn), choose the framing and the paper size with its resolution, and save as PNG, JPEG, WebP, SVG, TIFF and georeferenced GeoTIFF for further editing in QGIS, ArcGIS or Illustrator.

Step 9 · Species distribution modelling

• Algorithms (10): MaxEnt-style penalised logistic model (cloglog output), GLM, GAM, random forest, boosted regression trees, support vector machine, neural network, Bioclim envelope, Domain and Mahalanobis distance, plus an ensemble (weighted mean, median or vote committee; weights by AUC, TSS or Boyce).
• MaxEnt-style model: L1-penalised logistic regression on linear, quadratic, product and hinge features, with cloglog output. It is statistically equivalent to maximum-entropy modelling (Renner & Warton 2013).
• Data: one presence per cell, background sampled from the accessible area, predictors you choose. The map extent can be the box of the records plus a margin, a whole country or region (for example all of Mexico, Yucatán peninsula included) or custom coordinates; the model is projected over all of it.
• Validation: spatial blocks, random folds or a 70/30 hold-out; AUC, TSS, continuous Boyce index, sensitivity, specificity and omission; thresholds (max TSS, 10th percentile, minimum presence, sensitivity = specificity). Random forest and Domain use out-of-bag / leave-one-out predictions so the training fit is not overstated.
• Variables: permutation importance on the test data and response curves.
• Maps: continuous suitability, threshold presence/absence and between-model uncertainty, with suitable area in km²; export to PNG, ASCII grid and GeoTIFF (Float32, EPSG:4326).
• Projection to other scenarios (e.g. WorldClim CMIP6 future, one 19-band file or separate bio_N layers): gain/loss of suitable area and an extrapolation map (MESS, Elith et al. 2010).

Data and credits

• GBIF — species occurrence records (public API).
• WorldClim 2.1 (Fick & Hijmans 2017) — bioclimatic variables and elevation.
• Köppen-Geiger (Beck et al. 2018, Scientific Data, CC BY 4.0) — climate classification.
• SoilGrids (ISRIC; Poggio et al. 2021) — WRB soil classification.

No WorldClim data travels with the program. Their terms do not allow it to be redistributed without prior permission, so since version 1.1.0 the program ships none — not even the small cropped subset the example used to carry. You download the layers once (datos/LEE-ME.md says how — 10 arc-minutes to learn, 30 arc-seconds for a real study) and step 5 reads them with the folder picker, the file picker, or from the program's own datos/ folder. Köppen-Geiger is bundled, because CC BY 4.0 allows it.

Who is behind the records of the example: datos/GBIF-FUENTES.md, 70 institutions and 91 datasets. What is ours and what is not, source by source: NOTICE.md. Third-party libraries and their licences: vendor/THIRD-PARTY-NOTICES.txt.

Privacy and architecture

All processing happens in your browser: GeoTIFF files are read locally, and the statistics, machine learning and modelling run in Python compiled to WebAssembly. The only network calls are to the public GBIF and SoilGrids APIs, to the map-tile servers and to a public content-delivery network that provides the Python engine. There is no back end and no database; nothing is stored or shared beyond your computer (only your language, theme and a soil-query cache are kept in the browser).

Repository layout

index.html                 the application (home + 10 steps)
css/style.css              styles, light and dark themes
js/                        one file per block (i18n, state, ui, maps, GBIF, filters, cleaning, rasters, statistics,
                           ML, ecology, SDM, export) and the home page (art, home, lab)
js/example-records.js      the bundled GBIF example (Pinus cembroides); no climate layers are bundled
vendor/                    Leaflet, geotiff.js, html2canvas and their licences
datos/                     the Köppen-Geiger layer, where to download the climate layers, and who published
                           the records of the example
manual/                    the user manual in Spanish, in HTML by parts and as a PDF
NOTICE.md                  what is ours and what belongs to someone else, source by source
server.ps1, *.bat          optional local web server (Windows)

How to cite

Barrera-Guzmán, L. Á. (2026). BioModelling Pro: biogeography and species distribution modelling in the browser (Version 1.1.1) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.22907825

That DOI is the concept DOI: it always resolves to the latest version. Version 1.0.0 also has its own, 10.5281/zenodo.22907826, if you need to point at exactly the code you used.

See CITATION.cff. Please also cite the data sources and the methods you use.

Licence, and what it covers

The program is free software under the GNU General Public License v3.0 or later (see LICENSE). That licence covers the source code written by its author.

The libraries the program bundles or loads, and the data it ships as an example, belong to other people and keep their own terms — and one of them has a restriction worth knowing before you redistribute the program: WorldClim does not allow redistribution of its layers without prior permission, which is why the global layers are not in this repository and neither is any subset of them. NOTICE.md says exactly what is whose, source by source, and datos/GBIF-FUENTES.md lists the 70 institutions whose records make up the example.