Dryad DOI: https://doi.org/10.5061/dryad.s1rn8pkp4

Associated publication:
Herlan JJ, McTague SA, Staniczenko PPA, Loh JM, Santos RO, James WR, Gleason ACR, Edwards CB, Gosnell JS, McDonald KC, Friedlander AM, Gaymer CF. Spatial ecology of coral reefs at Rapa Nui (Easter Island): drivers of coral composition and dispersion along a wave exposure gradient. Marine Ecology Progress Series (in press). https://doi.org/10.3354/meps15300

---

Overview

This repository contains the data and R analysis scripts for a study of how wave exposure shapes the spatial organization of reef-building corals at Rapa Nui (Easter Island), Chile. The study integrates variance-to-mean ratio (VMR) dispersion analysis, spatial point-pattern model comparison (Diggle–Cressie–Loosmore–Ford goodness-of-fit), and a descriptive, null-free analysis of Pocillopora neighborhood composition across three reef sites along a wave-exposure gradient:

• North — Anakena (ana): sheltered, low wave exposure, high structural complexity
• West — Manavai (man): moderate wave exposure and complexity
• Southeast — Vaihú (vhu): exposed, high wave exposure, low structural complexity

Surveys were conducted in March–April 2015 from 250 m² photomosaic plots (10 m × 25 m) at approximately 15 m depth at each site. File names and data values write Vaihú without the accent (vhu, Vaihu, VHU).

---

Data accessibility — classification rasters

The primary inputs for the raster-based analyses are large binary PNG files (one per taxon per site: *_poci.png, *_plob.png, *_bare.png, *_dead_poci.png), derived from manual segmentation of orthorectified photomosaics. These rasters are not included in this deposit because of their size; they are available from the corresponding author (James J. Herlan, james.herlan@baruch.cuny.edu) on reasonable request. To allow the published tables and figures to be reproduced without the rasters, the pre-computed intermediate and result CSVs derived from them are provided here (see the File manifest and Data dictionary below). Researchers who obtain the rasters can reproduce every analysis from scratch.

---

Archive structure

The files are organized into the sub-folders listed under File manifest below, and the archive is provided as a single ZIP that preserves this folder structure (RapaNui_coral_dispersion_Dryad_20261108_v1.zip). Unzip it before use and keep the folder layout intact: each R script expects to be run from the top level of the unzipped folder and reads from / writes to the named sub-folders.

---

How to use

Each script is self-contained and begins with a header describing its inputs, outputs, and required packages. The single configuration step is working_dir, which is set to "." — run each script with the top level of this folder as the working directory. Scripts read inputs from the named sub-folders below and create their own output folders.

Tested with R ≥ 4.4 (analyses run under R 4.6.0) and packages: tidyverse, terra, sf, raster, spatstat, landscapemetrics, scales, png.

---

File manifest

Every file in the archive is listed below, grouped by folder. Site prefixes are ana = Anakena (north), man = Manavai (west), vhu = Vaihú (southeast).

Top-level data

• wave.csv — WaveWatch III hindcast; 57,415 three-hourly records across the 4 virtual stations combined, Oct 2012–Oct 2017 (57,287 retained for the directional wave-power summaries after excluding records without a valid direction).
• Rapa_Nui_ww3_coords.csv — coordinates of the 4 WWIII virtual stations.
• ana_rug_100_transects.csv, man_rug_100_transects.csv, vhu_rug_100_transects.csv — Virtual Profile Gauge rod-by-rod transect output (no header row; see Data dictionary).

Top-level scripts

• rpn_wave_exposure_analysis.R — wave-exposure summaries and site exposure (Table 1, Fig. 2A) from wave.csv.
• rpn_wave_rose_dual.R — dual wave-rose figure (Fig. 2A) from wave.csv.
• rpn_rugosity_analysis.R — linear rugosity, Kruskal–Wallis, and pairwise Wilcoxon (Table 1, Fig. 2B); writes output_rugosity/.
• rpn_percent_cover_pointrange.R — percent-cover point-range figure (Fig. 4) from output_percent_cover/.
• rpn_patch_utilities.R — percent-cover bootstrap pipeline (needs the omitted rasters; results provided in output_percent_cover/).
• rpn_poci_abundance_size_bootstrap.R — Pocillopora abundance and colony-size bootstrap (needs rasters; results in output_abundance_size/).
• rpn_poci_vmr_dispersion_bootstrap.R — VMR dispersion and chi-square CSR (needs rasters; results in output_vmr/).
• rpn_poci_pcf_model_comparison.R — DCLF four-model pair-correlation comparison (Table 5, Fig. 7) from output_patch_utilities/*_poci_sizefiltered.csv.

output_wave/ — wave-exposure results (from rpn_wave_exposure_analysis.R and rpn_wave_rose_dual.R)

• wave_exposure_by_site.csv — site-specific wave-exposure summary (Table 1).
• wave_power_sector_summary.csv — directional wave power by 16 compass sectors.
• wave_power_station_summary.csv — per-station wave-power summaries.
• Figure_1B_wave_power_rose.png, Figure_1B_wave_rose_dual.png, wave_direction_histogram.png, wave_site_exposure_comparison.png — wave-climate figures (Fig. 2A components).

output_percent_cover/ — bootstrapped percent-cover summaries (B = 1000, seed = 42; Table 2, Fig. 4)

• rpn_poci_bootstrap_results.csv — live Pocillopora percent cover.
• rpn_plob_bootstrap_results.csv — massive Porites lobata percent cover.
• rpn_bare_bootstrap_results.csv — bare-substrate percent cover.
• rpn_dead_poci_bootstrap_results.csv — dead-Pocillopora percent cover.

output_patch_utilities/ — size-filtered Pocillopora colony centroids (inputs to the DCLF comparison)

• ana_poci_sizefiltered.csv, man_poci_sizefiltered.csv, vhu_poci_sizefiltered.csv — colony centroids after the 9 cm² size filter (124, 368 and 3,106 colonies).

output_dispersion_from_csv/ — DCLF model-comparison results (Table 5)

• Table_dclf_4model_manuscript.csv — DCLF four-model p-values as reported in the manuscript.
• Table_dclf_4model_results.csv — full DCLF four-model results with adequacy/interpretation.
• Table_model_parameters.csv — fitted point-process model parameters.

output_vmr/ — dispersion and CSR results (Table 4, Fig. 6C)

• Table_vmr_dispersion_results.csv — variance-to-mean ratio dispersion results.
• Table_chisq_csr_results.csv — chi-square goodness-of-fit test for complete spatial randomness.

output_abundance_size/ — abundance and colony-size results (Table 3, Fig. 6A,B)

• rpn_poci_abundance_size_results.csv — bootstrap Pocillopora abundance (count per 250 m² plot) and mean colony size.

output_rugosity/ — rugosity results (Table 1; from rpn_rugosity_analysis.R)

• rugosity_site_summary.csv — per-site linear-rugosity summary.
• rugosity_transect_data.csv — per-transect linear-rugosity values.
• rugosity_kruskal_wallis.csv — Kruskal–Wallis test result.
• rugosity_pairwise_wilcoxon.csv — Bonferroni-corrected pairwise Wilcoxon results.

neighborhood_composition/ — null-free neighborhood-composition workflow (Table 2, Fig. 5)

Numbered pipeline; run 00_run_all.R to execute 01–04 in order.

• 00_run_all.R — master script that sources 01–04.
• 01_load_neighborhood_data.R — loads the *_interactions_full.csv neighbor data.
• 02_descriptive_bootstrap.R — observed mean composition with bootstrap CIs.
• 03_cross_site_comparison.R — cross-site bootstrap difference intervals.
• 04_figures.R — neighborhood-composition figures (Fig. 5).
• config.R — configuration (paths, site labels, bootstrap settings).
• functions.R — helper functions used by the pipeline.
• ana_interactions_full.csv, man_interactions_full.csv, vhu_interactions_full.csv — per-focal-colony neighbor proportions.

neighborhood_composition/outputs/ — neighborhood-composition results

• neighborhood_observed_means_CI.csv — observed mean neighborhood composition with 95% bootstrap CIs (Table 2, Fig. 5).
• cross_site_pairwise_differences.csv — cross-site pairwise difference intervals.

---

Data dictionary

Unless noted otherwise: wave power is in kW m⁻¹ (kilowatts per meter of wave crest); wave height in m; wave period in s; directions in degrees (compass bearing, the direction waves propagate from); coordinates and distances in m; colony size / area in cm²; percentages and proportions as indicated; p-values, VMR, and rugosity are dimensionless.

Shared categorical keys (used across many files):

• Site: ana / north / North (Anakena) / ANA = Anakena (north coast); man / west / West (Manavai) / MAN = Manavai (west); vhu / southeast / Southeast (Vaihu) / VHU = Vaihú (southeast).
• Wave exposure: Low (north), Moderate (west), High (southeast). Some files use the equivalent exposure labels Sheltered / Intermediate / Exposed.
• Complexity class: High (north), Moderate (west), Low (southeast).
• Neighbor / cover class: poci / pocilloporid = live Pocillopora (conspecific); plob / porites = Porites (massive Porites lobata); dead_poci = dead Pocillopora skeleton; bare = bare substrate.
• Direction sector: the 16 compass points N, NNE, NE, ENE, E, ESE, SE, SSE, S, SSW, SW, WSW, W, WNW, NW, NNW.

Top-level data

wave.csv — WaveWatch III three-hourly hindcast, stacked across the 4 virtual stations (57,415 rows).

Rapa_Nui_ww3_coords.csv — one row per virtual station (4 rows).

ana_rug_100_transects.csv, man_rug_100_transects.csv, vhu_rug_100_transects.csv — Virtual Profile Gauge (VPG) rod-by-rod output; no header row. Six comma-separated columns, in order:

output_wave/

wave_exposure_by_site.csv — one row per site (3 rows).

wave_power_sector_summary.csv — one row per compass sector (16 rows). Columns: direction_sector (sector code); n_obs (record count); freq_pct (% of all records); mean_power, sd_power, median_power, max_power, se_power (wave power, kW m⁻¹); mean_height, sd_height, max_height (m); mean_period, sd_period (s); mean_swell_height (m); mean_swell_period (s); sector_center (sector mid-point bearing, degrees).

wave_power_station_summary.csv — one row per virtual station (4 rows). Columns: latitude, longitude (station coordinates); n_obs (record count); mean_power, sd_power (kW m⁻¹); dominant_direction (modal sector); mean_height, max_height (m); mean_period (s).

output_percent_cover/

All four files share one row per site (the ana/man/vhu key) and the following columns; rpn_bare_* and rpn_dead_poci_* add a final substrate column. Values are percent cover (%).

output_patch_utilities/

{ana,man,vhu}_poci_sizefiltered.csv — one row per Pocillopora colony centroid retained after the 9 cm² size filter.

output_dispersion_from_csv/

Table_dclf_4model_manuscript.csv — one row per site; columns site plus the four model columns Homogeneous Poisson, Inhomogeneous Poisson, Homogeneous Cauchy, Inhomogeneous Cauchy, each the DCLF p-value for that model.

Table_dclf_4model_results.csv — long form, one row per site × model (12 rows). Columns: site (3-letter code), site_name, model (candidate model name), p_value (DCLF p-value), adequate (Yes/No, i.e., not rejected at α = 0.05), interpretation (plain-text verdict).

Table_model_parameters.csv — fitted point-process parameters; one row per site × kernel-bandwidth candidate. Columns: site, site_name, n_colonies (colonies used), bandwidth.sigma (kernel bandwidth σ, m), bandwidth.cv (cross-validation score at that σ).

output_vmr/

Table_vmr_dispersion_results.csv — one row per site. Columns: Wave Exposure, Site, Location (site name); N Colonies (colony count); Median VMR, Mean VMR, SD (variance-to-mean ratio of quadrat counts, bootstrap distribution); 95% CI (text interval); Dispersion Pattern (e.g. Clustered); Significant Departure (Yes/No, VMR 95% CI excludes 1).

Table_chisq_csr_results.csv — one row per site. Columns: site, wave (exposure), n_colonies, chi_sq (χ² statistic), df (degrees of freedom, 249), p_value, reject_CSR (TRUE/FALSE).

output_abundance_size/

rpn_poci_abundance_size_results.csv — long form, one row per metric × site (6 rows). Columns: Metric (Abundance = colonies per 250 m² plot, or Colony Size (cm^2)); Wave Exposure; Site; Mean (bootstrap mean); 95% CI (text interval, lower - upper).

output_rugosity/

rugosity_site_summary.csv — one row per site. Columns: site_code, site_name, site_short, wave_exposure, complexity_class; n_transects, n_valid (transects total / with valid data); mean, median, sd, se, min, max, iqr, q25, q75, ci_lower, ci_upper (linear-rugosity summary statistics, dimensionless ratio ≥ 1); mean_depth, mean_depth_sd, mean_depth_range (mean z-coordinate of valid points — relative model depth in m, not absolute survey depth); mean_pct_valid (mean % of valid rods per transect); wave_order (ordered factor for plotting).

rugosity_transect_data.csv — one row per transect. Columns: site identifiers as above; transect (index); n_total, n_valid, n_segments, n_gap_points (rod counts and gap structure); contour_distance (summed 3-D surface distance, m); profile_length (regression-line base length, m); linear_rugosity (contour ÷ profile length, dimensionless ≥ 1); depth_mean, depth_sd, depth_range (relative model depth, m); pct_valid (% valid rods).

rugosity_kruskal_wallis.csv — Columns: test, response (linear_rugosity), group (grouping variable), statistic_chisq (Kruskal–Wallis χ²), df, p_value.

rugosity_pairwise_wilcoxon.csv — one row per site pair. Columns: test, comparison (e.g., West vs North), p_value_adj (Bonferroni-adjusted p).

neighborhood_composition/

ana_interactions_full.csv, man_interactions_full.csv, vhu_interactions_full.csv — one row per focal Pocillopora colony only (no Porites or bare-substrate foci): 117, 351 and 2,902 colonies at Anakena, Manavai and Vaihú. These are the colonies remaining after the 9 cm² size filter (124, 368 and 3,106) and after excluding colonies whose 10 cm buffer crossed the plot boundary (7, 17 and 204 colonies, respectively). Columns: focal_taxa (focal class, pocilloporid); buffer_id (unique focal-colony identifier, e.g. pocilloporid_8); bare, dead_poci, pocilloporid, porites (proportion of the focal colony's 10 cm annular buffer, 0–1, occupied by bare substrate, dead Pocillopora skeleton, conspecific live Pocillopora, and Porites, respectively). The four proportions sum to at most 1 (0.95–0.98 on average); the remainder is unsegmented area.

neighborhood_composition/outputs/

neighborhood_observed_means_CI.csv — one row per site × neighbor class. Columns: site, site_label, exposure; neighbor (class key), neighbor_label (readable class); n_colonies (focal colonies at the site); observed_pct (observed mean neighbor proportion, %); ci_lower_pct, ci_upper_pct (95% bootstrap CI, %); boot_se_pct (bootstrap standard error, %).

cross_site_pairwise_differences.csv — one row per neighbor class × site pair. Columns: neighbor, neighbor_label; site_a, site_b, site_a_label, site_b_label (the two sites compared); mean_a_pct, mean_b_pct (site means, %); diff_pct (difference, %); diff_lower_pct, diff_upper_pct (95% bootstrap CI of the difference, %); distinct_diffCI (TRUE if the difference CI excludes 0); ci_overlap (TRUE if the two sites' own CIs overlap).

---

Notes

• The neighborhood analysis is null-free and descriptive: it reports observed mean composition with bootstrap CIs and compares sites via bootstrap difference intervals, rather than testing against a pooled-community null.
• The DCLF model comparison is run from size-filtered centroids (rpn_poci_pcf_model_comparison.R), with set.seed and 200 Monte Carlo simulations per model at Anakena and Manavai and 99 at Vaihú (patterns with more than 1,000 colonies), so the smallest attainable p is 0.005 and 0.01, respectively; all four candidate models are rejected at the exposed southeast reef.
• Percent-cover values are reported as bootstrapped means with 95% quantile ranges.

---

Funding

This work was funded by the Chilean Millennium Initiative, Center for Ecology and Sustainable Management of Oceanic Islands (ESMOI). Idea Wild provided field equipment, and the cameras used for large-area imaging were funded through a crowdfunding campaign on Experiment.com (https://doi.org/10.18258/1657).

---

Version history

• 8 Nov 2026: the three *_interactions_full.csv files now include the dead_poci column used for Table 2 and Fig. 5 (the previous upload lacked it, so the neighborhood pipeline could not run); they contain the focal Pocillopora colonies only, after edge exclusion of 7, 17 and 204 colonies. DCLF simulation counts documented (99 at Vaihú). The Electronic Supplement figure was removed (it is published with the article). README expanded with a data dictionary, raster-access statement, archive-structure note and funding, and updated to the article's final title.