Phylogenetic Structure of Plant Communities Along Elevational Gradients: Context-Dependent Patterns from a Global Mountain Network

This repository contains the code and data pipeline for analyzing phylogenetic structure of plant communitites along elevational gradients. The project leverages the targets R package for reproducible workflows.

Abstract

Phylogenetic structure of communities helps to improve our understanding of biodiversity maintenance by providing insights about how species respond to constraining and changing conditions . Mountain gradients provide a great opportunity to study community assembly processes as they act as environmental filters, shifting from benign to increasingly stressful within relatively short distances. Because tolerance to harsh conditions is often phylogenetically conserved, the phylogenetic structure of communities is expected to change from overdispersed to clustered with increasing elevations. However, other biotic and abiotic pressures might change the expected patterns of the stress gradient model.

In this study, we asked how community phylogenetic structure changed along elevational gradients, and how this relationship varied with fire history and topography. Across eight field campaigns in South America, North America, Europe, Asia, and Africa, we measured community compositional changes along elevational gradients that collectively span from 0 to 5000 m a.s.l., although each individual gradient covers a narrower range. We quantified community phylogenetic structure using abundance weighted and non-abundance-weighted mean phylogenetic distance and mean nearest taxon distance (MNTD), as well as phylogenetic beta diversity. We then tested multiple predictors to identify which best predicted community phylogenetic structure, and whether “harsher” environmental conditions resulted in the hypothesized increase in clustering due to filtering.

We find that the relationships between phylogenetic structure and elevation were highly heterogeneous, with different gradients showing responses that differed in slope, magnitude, and shape (unimodal vs linear). Site identity explained a large portion of beta-MNTD variation, underscoring that the evolutionary distance between regional floras far exceeds any within-mountain gradient effect. The interaction between fire regime and elevation was consistent with a post-disturbance weakening of environmental filtering: fire reset community composition and resulted in a less taxonomically-filtered set of species. When comparing topographic effects, we found that aspect structures community phylogenetic structure above and beyond elevation.

Our results suggest that community phylogenetic assembly varies strongly with elevation, but this relationship is not uniform and instead modulated by additional biotic and abiotic factors. Together, these dynamics highlight how the interplay between evolutionary history and ecological processes shapes biodiversity patterns across elevational gradients.

Data

The pipeline downloads, imports, cleans, and analyzes data from multiple sources and ecosystems, including:

• Community composition
• Climate data (downscaled extract, summarised per plot)

Most inputs download automatically from OSF and Zenodo. Three small files are tracked in git instead, so a fresh clone has everything it needs:

File	Used by	Notes
data/metaCH.csv	download_meta_ch	China site metadata
data/downscaled_climate.csv	download_climate	Downscaled climate extract, summarised per plot
data/taxonomy_tnrs.csv	taxonomy_cache	TNRS name resolutions (see below)

Taxonomy lookup

data/taxonomy_tnrs.csv maps each submitted name to its accepted name and family, as resolved by TNRS against WCVP/WFO. It is committed on purpose, for two reasons:

• The phylogeny builds offline. TNRS has been intermittently unavailable, and without the lookup a clone cannot build taxonomy at all.
• The taxonomy is pinned. Results do not shift when WCVP/WFO publish revisions, so a rebuild months from now reproduces the same tree.

The pipeline only ever queries TNRS for names that are missing from the file, so adding one species costs one lookup. To re-resolve everything against current TNRS — for instance to pick up genuine upstream corrections — set refresh_taxonomy <- TRUE in R/phylogeny_plan.R, run tar_make(), then set it back and commit the updated file.

taxonomy_cache is a format = "file" target, so pulling a colleague's updated lookup, or correcting a row by hand, invalidates the phylogeny downstream of it rather than being silently ignored.

Cover vs relative cover

Community abundance is recorded differently across regions. Most sites use percent cover (roughly 0–100 per species). Colorado records individual abundance counts (RMBL_2022_abundance.xlsx), which are not percent cover and can exceed 100 for a single species.

To make abundance-weighted analyses comparable, the pipeline follows the same approach used for Arizona in TransPlantNetwork/transplant_network_data_cleaning: keep the raw measurement and add within-plot relative cover.

When community_raw is built, add_relative_cover() adds:

Column	Meaning
cover	Raw recorded value (percent cover, or abundance counts for Colorado)
cover_unit	"percent" or "abundance"
cover_sum	Sum of cover within plot × year (country, gradient, site, plot_id, year)
rel_cover	cover / cover_sum (proportions that sum to 1 within each plot × year)

Use rel_cover for abundance-weighted diversity metrics (e.g. abundance-weighted MPD / MNTD). Presence/absence metrics can ignore cover. After tip labels are attached, rel_cover is recomputed so collapsed synonyms still sum to 1 per plot.

Pipeline Structure

The analysis is organized using the targets package, with plans for:

• Download: Automated retrieval of raw data files
• Import: Reading and initial formatting of data
• Cleaning: Standardizing and filtering datasets
• Transformation: Merging community data and joining downscaled climate
• Phylogeny: Taxonomic harmonization (TNRS, against WCVP/WFO) and grafting the pooled species list onto the GBOTB megatree with rtrees
• Diversity: Phylogenetic diversity indices (MPD, MNTD, beta-MNTD)
• Analysis: Statistical modeling and visualization

Custom functions for cleaning and processing are located in R/Functions/.

Getting Started

Prerequisites

• R (>= 4.0)
• R packages: targets, tarchetypes, dataDownloader, tidyverse, DBI, RSQLite, janitor, vegan, ggvegan, readxl, broom, broom.mixed, glue, geodata, terra, MetBrewer, maps, performance, quarto, see, rgee, sf, lmerTest, gt, ggridges, patchwork, betapart, plotbiomes, ape, picante, TNRS, and rtrees (as specified in _targets.R).

Running the Pipeline

1. Install dependencies (in R): The project uses renv. From a fresh clone, restore the recorded library rather than installing by hand:

   renv::restore()

   .Rprofile sources renv/activate.R, so starting R from the project root will automatically activate the project environment (and may bootstrap renv if needed).

2. Run the pipeline (from the project root):

   source("run.R")

   Or, in the shell:

   Rscript run.R

3. Inspect results:

   • Main results and figures are generated in results.qmd and results_files/.

Project Structure

pftc_phylogeny/
├── _targets.R           # Main targets pipeline definition
├── run.R                # Script to run the pipeline
├── R/                   # R scripts for plans and functions
│   ├── functions/       # Custom function scripts
│   └── ...              # Plan scripts (analysis, cleaning, etc.)
├── data/                # Raw and processed data (mostly gitignored)
├── results_files/       # Output files (gitignored)
├── results.qmd          # Quarto/Markdown results document
└── ...                  # Other supporting files

Reproducibility

• The pipeline is fully reproducible using the targets workflow.
• Data files are managed via automated download scripts.
• Intermediate and final results are cached for efficiency.