R pipeline and field dataset characterizing the functional diversity of the flora and fauna of the Las Peñas campus (ESPOL, Guayaquil, Ecuador) as a pre-intervention reference condition for the campus remodeling.
Capstone project (Proyecto Integrador), Biology · Miguel Ángel Morán Piedrahita · ESPOL, 2026. Supervisor: Julián Alfredo Pérez Correa, PhD.
🇪🇸 Resumen en español at the bottom, and a full Spanish version in LEEME.md.
More than half of a 3.25 ha campus — sitting among the last dry-forest fragments inside Guayaquil, a few metres from the Guayas river — is being rebuilt, and no one had recorded what lived there. Ecuador's seasonally dry forest shrank by 2 632 km² between 1990 and 2018; only 13 % of the coastal remnant has any protective status, and roughly 80 % of its biota is endemic to the Tumbesian region.
Counting species does not answer the question a construction project raises. Richness says nothing about what those species do, and it is the doing that gets disrupted. This repository measures the traits instead.
Computes, per zone and per taxonomic group:
| Index | What it captures |
|---|---|
| FRic | Functional richness — volume of trait space occupied |
| FEve | Functional evenness — regularity of abundance across that volume |
| FDiv | Functional divergence — how far the abundant species sit from the centre |
| FDis | Functional dispersion — abundance-weighted mean distance to the centroid |
| Rao's Q | Quadratic entropy — abundance-weighted mean pairwise trait distance |
| FR | Functional redundancy — the share of species diversity that is not functional diversity |
| SES | The above, standardized against a null model, so richness is controlled |
Two patterns dominate, and they belong together: functional richness is low everywhere — the compression urban filters reliably produce, amplified by a flora that is 100 % introduced ornamentals — while functional divergence is high (0.59–1.00), meaning the abundant species sit at the periphery of that small volume.
One result deserves to be read carefully. Bird FRic is NA for Zones 3, 4 and 5
because they contain exactly the same four species, and an unweighted volume
cannot distinguish identical species lists. Zone 1 is the exception — the only
zone with Columba livia and Quiscalus mexicanus. That the car park would
therefore post the campus's highest bird functional richness is the clearest
available argument against using any single index as a conservation criterion.
The gap between the two bars is functional redundancy: species whose ecological role is already covered by another species present. It is the measurement behind the resilience argument — a low FRic says the community occupies little trait space, but it does not say different species do the same job. That is a separate claim and it needs its own number.
docs/verification.md, item V1),
not a settled result. Both use FR = 1 − Q/D; the gap is systematic and most
likely comes down to how the dissimilarity inside Q was scaled. Do not quote a
redundancy figure from this repository without reading that entry first.
git clone https://github.com/mamoran-bio/FD_functional_diversity_indices.git
cd FD_functional_diversity_indices
# Figures and the redundancy / taxonomic analysis (needs only ggplot2, reshape2,
# cluster, vegan — no FD, so this runs almost anywhere)
Rscript R/figures_reference.R
# Full pipeline: all indices, null models, sensitivity analysis (needs FD)
Rscript R/functional_diversity_analysis.RRun both from the repository root, not from R/. Outputs go to results/,
which is git-ignored and regenerated on every run.
To restore the exact package versions:
install.packages("renv")
renv::restore()
renv.lockis arenv::snapshot()lockfile: it pins the full dependency tree (~140 packages), not just top-level declarations. Frozen 2026-09-11 under R 4.5.3.
System dependencies. rcdd and nloptr — needed only by the optional
mFD cross-check — require GNU MP and NLopt headers:
# R installed from apt
sudo apt install libgmp-dev libnlopt-cxx-dev cmake
# R installed from conda / miniforge: the conda toolchain does not look in
# /usr/include, so the libraries must live inside the conda prefix instead
conda install -c conda-forge gmp nlopt├── R/
│ ├── functional_diversity_analysis.R main pipeline
│ ├── redundancy.R Ricotta et al. (2016)
│ ├── null_models.R 999-permutation SES
│ ├── taxonomic_diversity.R Shannon, Simpson, Pielou, rarefaction
│ └── figures_reference.R README figures (no FD needed)
├── data/
│ ├── README.md ← full data dictionary, units, caveats
│ ├── traits_flora.csv 42 species × 7 traits
│ ├── abundance_flora.csv crown cover by zone
│ ├── traits_fauna_survey{1,2,3}.csv
│ ├── abundance_fauna_survey{1,2,3}.csv
│ ├── species_metadata.csv origin, group, provenance notes
│ ├── zones.csv areas and sampling effort
│ └── reference_results_thesis.csv published Tables 8–9, for regression checks
├── docs/
│ ├── analysis.Rmd narrative notebook
│ └── verification.md ← every code/thesis discrepancy found
├── figures/
├── Tablas_muestreos.ods raw field record (9 sheets)
├── renv.lock
└── CITATION.cff
| Decision | Choice |
|---|---|
| Flora abundance | Crown-projection cover, fraction 0–1 |
| Fauna abundance | Detection counts per zone and campaign |
| Distance | Gower (mixed traits, internal range rescaling) |
| Transformation | log10 on skewed continuous traits: mass, height, SLA |
| Ordination | PCoA, Cailliez correction for negative eigenvalues |
| Engine | FD::dbFD, with mFD as an optional cross-check |
| Redundancy | Ricotta et al. (2016), FR = D − Q, reported as 1 − Q/D |
| Null model | 999 trait-label permutations → SES + exact two-tailed p |
| Sensitivity | Indices recomputed under raw, log(x+1) and presence/absence weighting |
On the correction. Gower distances on mixed traits are generally not Euclidean, which produces negative PCoA eigenvalues and leaves the convex hull behind FRic ill-defined. Lingoes adds a smaller constant and distorts less, so it is arguably the better choice on the merits. Cailliez is nevertheless the default here, because it is what produced the published tables, and a repository whose job is to reproduce them must default to what produced them. Switching is a one-line change and belongs in a reported sensitivity check, not in a silent edit.
Apis mellifera accounted for 90.1–95.7 % of the Zone 4 fauna records. Because FDis weights distances by relative abundance, pooling it with the vertebrates drags the community centroid onto one functional type and depresses the index — an artefact, not an ecological finding. Analysed separately, the effect disappears.
Two further reasons make the separation obligatory rather than convenient. The honeybee counts are extrapolated, not observed: four 2.5 m units in Ixora coccinea planters, scaled by a factor of 30.4 to the 304 m the plant occupies. And those units sat in the four longest planters, not random ones — since honeybees concentrate foragers on the richest patches through dance recruitment, the number is an upper bound, an index of visitation intensity per unit of floral resource rather than a population size.
Every index here scales with richness by construction, and the zones span 2 to 26 plant species. Comparing zones directly therefore confounds how many species with how they are arranged in trait space — "Zone 4 is functionally richer" risks meaning no more than "Zone 4 has more species".
R/null_models.R shuffles the trait matrix's species labels 999 times while
holding the community matrix fixed:
- SES < 0 — co-occurring species are more similar than an equally rich random draw: functional clustering, the expected urban-filtering signature.
- SES > 0 — more different than expected: overdispersion.
- SES ≈ 0 — the zone is a random subset, and the between-zone difference is a richness difference and nothing more.
The pool is the campus species list, so the test cannot ask whether the campus itself is filtered relative to the regional dry-forest flora. That needs a regional pool this study did not sample.
Conserve Zone 4 and its large-crowned trees (Tabebuia aurea, Gliricidia sepium, Mangifera indica, Ficus benjamina). The case does not rest on the functional indices — for birds they do not discriminate between zones — but on convergent evidence: highest plant richness (26 species), highest accumulated crown cover (101.3 %), the only canopy stratum the frugivorous birds use, five of the six native animal species, and the campus's only Ixora coccinea, the sole floral resource on which pollinator activity was ever observed.
Plant for the missing traits, not for the pretty species. The trait matrix names three concrete gaps: 86 % of species are evergreen and all six deciduous species are introduced trees, so no deciduous shrub exists on campus; 26 % are ornamental palms, all evergreen and zoochorous, the most redundant block in the assemblage; and only 8 of 42 species sit at the conservative end of the leaf-economics spectrum. Native dry-forest species that fill them: Handroanthus chrysanthus, Cordia lutea, Bonellia sprucei, Pithecellobium excelsum.
Diversify the floral resource, so the pollinator community is not one introduced species visiting one introduced shrub.
Re-run this pipeline after the works, with the same methods and this study as the reference.
- Three campaigns, one seasonal window (wet-to-dry transition). Full seasonal turnover is not captured.
- Sampling effort varies 4.4-fold between zones (2.9 to 12.9 points ha⁻¹; effective coverage 20.5 % to 90.9 %). Rarefaction helps for counts; it does not fully correct this.
- No before–after design and no unintervened control. Future change cannot be attributed unambiguously to the remodeling. The reference condition supports description, not causal inference.
- Small n throughout — four zones, 2–26 species each. Little room for inferential statistics; the null models are the right tool at this scale.
- LDMC is systematically overestimated (no rehydration before fresh-mass
determination) and leaf subsamples came from one individual per species.
Relative ordering along the leaf-economics axis holds — which is what Gower
uses — but absolute values are not comparable to TRY or GLOPNET. Four further
documented protocol deviations are in
data/README.md. - Zone 2, 57.3 % of the campus, was never assessed — it was already under construction. The reference condition covers 42.7 % of the site.
- Open verification items, including the unresolved redundancy discrepancy:
docs/verification.md.
See CITATION.cff — GitHub renders it as a "Cite this
repository" button.
Code (R/, docs/): MIT. Data (data/, Tablas_muestreos.ods):
CC BY 4.0. See LICENSE.
- Villéger, S., Mason, N. W. H., & Mouillot, D. (2008). Ecology, 89(8), 2290–2301.
- Laliberté, E., & Legendre, P. (2010). Ecology, 91(1), 299–305.
- Ricotta, C., et al. (2016). Methods in Ecology and Evolution, 7(11), 1386–1395.
- Mouchet, M. A., et al. (2010). Functional Ecology, 24(4), 867–876.
- Maire, E., et al. (2015). Global Ecology and Biogeography, 24(6), 728–740.
- Magneville, C., et al. (2022). mFD. Ecography, 2022(1).
- Pérez-Harguindeguy, N., et al. (2013). Australian Journal of Botany, 61(3), 167–234.
- Wilman, H., et al. (2014). EltonTraits 1.0. Ecology, 95(7), 2027.
- Sol, D., et al. (2020). Urbanisation and the loss of avian functional diversity.
- Cruz-García, et al. (2026). Urban herpetofauna of Guayaquil.
Este repositorio contiene el pipeline de R y los datos de campo que caracterizan la diversidad funcional de la flora y fauna del campus Las Peñas (ESPOL, Guayaquil), como línea de referencia previa a su remodelación.
Más de la mitad de un campus de 3.25 ha —ubicado entre los últimos relictos de bosque seco dentro de Guayaquil— está en obra, y no existía ningún diagnóstico de la biodiversidad presente. Contar especies no responde la pregunta: la riqueza no dice qué hacen esas especies, y es justamente eso lo que una obra civil interrumpe.
Qué calcula: FRic, FEve, FDiv, FDis, Q de Rao, redundancia funcional (Ricotta et al. 2016) y tamaños de efecto estandarizados frente a modelos nulos, por zona y grupo taxonómico.
Hallazgos principales. La riqueza funcional es baja en todas las zonas —compresión típica del filtrado urbano, acentuada porque las 42 especies de flora son introducidas ornamentales— mientras la divergencia funcional es alta (0.59–1.00). En aves, FRic no discrimina: las Zonas 3, 4 y 5 comparten exactamente las mismas cuatro especies, y la Zona 1 (el estacionamiento) sería la de mayor riqueza funcional por albergar Columba livia y Quiscalus mexicanus. Ese resultado es el mejor argumento contra usar un solo índice como criterio de conservación.
Recomendación central: conservar la Zona 4 y su arbolado de gran porte, y sembrar por los rasgos que faltan (no existe ningún arbusto caducifolio en el campus; el 26 % de las especies son palmas ornamentales, el bloque más redundante) con nativas del bosque seco: Handroanthus chrysanthus, Cordia lutea, Bonellia sprucei, Pithecellobium excelsum.
docs/verification.md, ítem V1, y afecta a la cifra
del "≈87 %" en la que se apoyan las conclusiones. No cite un valor de
redundancia sin leer esa entrada.
Diccionario de variables completo, unidades y desviaciones de protocolo:
data/README.md. Versión completa en español:
LEEME.md.