Candidate SDM model pool for priority woody invasive plant species (Ailanthus altissima, Acer negundo) in the Danubian riparian corridor

Published: 15 September 2026| Version 2 | DOI: 10.17632/9hfzb63k2w.2
Contributors:
Andrej Halabuk, Petra Gasparovicova, Tomas Rusnak, Jakub Tomes, Lubos Halada

Description

This dataset supports the SDM–MCDA framework developed for two woody invasive alien plant species, Ailanthus altissima and Acer negundo, in the Danubian riparian corridor. Analysis-ready data are provided at a common 1-km modelling resolution for the Danube citizen-science calibration domain and two professional external evaluation domains (Romania and Serbia). The calibration data use a target-group background design, whereas the external datasets contain survey-based presences and non-detections/absences. Eleven transformed and z-standardised environmental predictors are used across species and domains. Candidate models are penalised logistic Generalised Additive Models (pyGAM). Latin Hypercube Sampling is used to vary model complexity and the smoothing parameter λ. Candidates are evaluated by spatially blocked cross-validation in the Danube domain and by external AUC in Romania and Serbia. The candidate pool is then evaluated using six MCDA criteria and four decision scenarios. Scenario weights are perturbed with a Dirichlet distribution to quantify ranking stability. The top-10 models per scenario are used to produce median suitability maps, scenario hotspot masks, and the Core, Union, and Union-minus-Core priority zones. Version 2 adds a self-contained reproducibility package (IAS_Dunaj.zip) with analysis-ready input tables, the aligned predictor stack, configuration files, Python modules, pseudocode, requirements, and an executable Jupyter notebook. After extraction to C:\IAS_Dunaj, users can select either species, fit a demonstration pool (e.g. 50 models) or the full 1,000-model pool, modify MCDA weights, and regenerate the raster priority indicators. The original fitted candidate models and performance tables are retained for direct reuse without refitting. Associated manuscript: Halabuk et al., “From citizen science records to invasive plant indicators: multi-domain evaluation and scenario-based model selection for the Danube corridor”. Acknowledgement: Funded by the EU NextGenerationEU through the Recovery and Resilience Plan for Slovakia under project No. 09I01-03-V04-00005.

Files

Steps to reproduce

Version 2 provides an executable workflow in IAS_Dunaj.zip. Download IAS_Dunaj.zip and extract its contents so that the root directory is: C:\\IAS_Dunaj\\ Create a Python environment and install the supplied dependencies: python -m venv C:\\IAS_Dunaj\\venv C:\\IAS_Dunaj\\venv\\Scripts\\activate python -m pip install --upgrade pip pip install -r C:\\IAS_Dunaj\\requirements.txt python -m ipykernel install --user --name IAS_Dunaj --display-name "IAS Dunaj" Start Jupyter and open: C:\\IAS_Dunaj\\SDM_MCDA_workflow.ipynb In the first notebook cell, select the focal species: SPECIES = "Negundo" for Acer negundo, or: SPECIES = "Ailanthus" for Ailanthus altissima. Set the size of the candidate-model search: N_MODELS = 50 provides a practical demonstration run. To run the full candidate-search scale used in the study, set: N_MODELS = 1000 Each value generates a Latin Hypercube candidate pool using the supplied random seed and the model-generation settings documented in the package. Run the notebook from top to bottom. The model-generation section reads the supplied species-specific Danube, Romania, and Serbia 1-km analysis-ready tables, generates candidate GAM specifications, fits and evaluates the models, and stores the fitted candidates and performance table under: C:\\IAS_Dunaj\\outputs\\<SPECIES>\\run_nXXXX_seed42\\models\\ The MCDA section calculates criteria F1–F6, reads the S1–S4 criterion weights from: C:\\IAS_Dunaj\\config\\scenario_weights.csv performs the Monte Carlo weight perturbation, ranks the candidate models, and retains the scenario-specific top-10 models. To explore another decision context, edit the F1–F6 weights in scenario_weights.csv; each scenario row must sum to 1. Changing only the decision weights does not require the candidate GAMs to be refitted. The spatial section applies each scenario-specific top-10 shortlist to the supplied aligned 1-km predictor rasters. It generates the scenario suitability surfaces, species-specific hotspot masks, the four-scenario overlay, and the Core, Union, and Union-minus-Core priority rasters. The principal outputs are written to: C:\\IAS_Dunaj\\outputs\\<SPECIES>\\run_nXXXX_seed42\\ Additional implementation details, criterion definitions, parameter settings, and the complete algorithmic sequence are provided in README.txt and PSEUDOCODE.txt. The fitted model archives supplied separately with the dataset can be used when users wish to perform post-hoc model selection or inference without rerunning candidate-model generation.

Categories

Biological Invasion, Invasive Species Management

Licence