Elodea SDM pipeline: an R workflow for modelling the distribution, nutrient response and spread of Elodea nuttallii and Elodea canadensis in Germany

An R workflow that tests whether an invasive aquatic plant has filled its potential range, and translates the answer into catchment-level management zones. It is applied to Nuttall’s waterweed (Elodea nuttallii), an invasive alien species of Union concern, and its long-established congener Elodea canadensis in Germany. A single script runs every step, from raw GBIF downloads to figures, tables and maps, and writes an audit trail as it goes. The workflow first cleans GBIF records of seven aquatic macrophytes with identical filters (location in Germany, coordinate uncertainty of 10 km or less, collection year 1980 or later, one record per species and location) and thins them spatially. A target-group background built from all surveyed locations corrects for uneven survey effort. Species distribution models (GLM, GAM, GBM and MaxNet in biomod2) are fitted with predictors selected by correlation and variance-inflation screening. They are validated by river-basin cross-validation on ECRINS sub-basins and combined into an ensemble weighted by the true skill statistic. The ensemble is projected to 2021–2040 under CMCC-ESM2 (SSP2-4.5 and SSP5-8.5), with checks for extrapolation, a multivariate environmental similarity surface and response-curve diagnostics. Binomial generalized linear mixed models (lme4) relate occurrence in ECRINS catchments to agricultural phosphorus input and the nitrogen-to-phosphorus input ratio from the Land Use Intensity Index for Stream Ecosystems (LUIS), survey effort and region. A colonisation analysis uses dated records of a single source in Bavaria to model recorded colonisation after an early invasion front, with winter temperature, elevation and straight-line distance to early-invaded catchments as predictors, and downstream position and river-network distance (igraph) in robustness checks. Robustness tests examine sensitivity to data sources (the national atlas, dated records only, records without citizen-science sources and curated sources only), species × region contrasts, block-bootstrap tests of within-region signals, a model trained in the west and projected to the east, alternative regional boundaries, river basin districts and steps in predictors at federal state borders. Further analyses compare the former GDR with eastern Bavaria and test for a step at the former inner-German border, compare winter temperatures with the native range, include catchments recorded only by the study species in the colonisation analysis, check residual spatial autocorrelation and alternative suitability thresholds, and summarise all key odds ratios in one robustness figure. Finally, each ECRINS catchment is assigned to a management zone, from harvest candidate to prevention priority, based on its records, river basin district and predicted suitability. Fitted models are stamped with their settings and inputs and reused only when the stamp matches, and a preflight check lists all missing inputs before any long computation starts. Input data are not redistributed. They are available from GBIF (https://doi.org/10.15468/dl.dhvq4g and https://doi.org/10.15468/dl.x4fhp2), WorldClim v2.1, LUIS 2017 (https://doi.org/10.5281/zenodo.17204506), EEA ECRINS v1.1 and GADM v4.1. README.md gives the expected folder layout, requirements, run instructions and the mapping of the article’s figures and tables to the outputs; CHANGELOG.md lists the changes between versions. The workflow was tested with R 4.5.2, terra 1.9-1, biomod2 4.3-4-5, mgcv 1.9-3, gbm 2.2.3, maxnet 0.1.4, caret 7.0-1 and lme4 1.1-38.