Quantile Random Forests for Freshwater Typology Type Coherence Prediction

This repository contains fitted Quantile Random Forest models and associated training data for benchmarking freshwater ecological typologies across four taxonomic groups: diatoms, fishes, macroinvertebrates, and macrophytes. The models predict the expected distribution of typology evaluation metrics (e.g., ANOSIM R, Classification Strength) conditional on the typology's structural, spatial, and taxonomic properties. The models were fitted as part of the PULSE project, a pan-European freshwater ecotypology initiative . They enable users to place a candidate typology's performance in the context of comparable typologies from across Europe, providing quantile-based benchmarks rather than single point estimates. Details of the training data, modelling approach, and validation are reported in the accompanying manuscript which soon to be submitted. Content Each .rds file contains a fitted workflows::workflow object, bundled with the bundle package for portable serialisation. Files are named <taxon>__<metric>.rds for models and <taxon>__training_data.rds for the associated training sets. Models can be loaded in R with readRDS() followed by bundle::unbundle(). Intended use These models power the companion Shiny application [LInK], which allows interactive prediction of benchmark distributions for user-specified typology configurations. The files can also be used directly in R for custom analyses. See here for the github repositroy with all code related to the project.