Data from: Using community composition and successional theory to guide site-specific coral reef management
Data files
Jan 24, 2025 version files 483.33 KB
-
Environmental_data_2017-2023.csv
7.04 KB
-
LDM_fun.R
324.10 KB
-
Long_term_environmental_averages_Ocean_Tipping_Points.csv
2.26 KB
-
Maui_Nui_analysis_final.Rmd
73.45 KB
-
Maui_Nui_benthic_data_final.csv
36.56 KB
-
Maui_Nui_monitoring_program_metadata.csv
13.42 KB
-
Percent_cover_confidence_estimate.Rmd
14.54 KB
-
README.md
11.96 KB
Abstract
High spatial or temporal variability in community composition makes it challenging for natural resource managers to predict ecosystem trajectories at scales relevant to management. This is commonly the case in nearshore marine environments, where the frequency and intensity of disturbance events varies at the sub-kilometer to meter scale, creating a patchwork of successional stages within a single ecosystem. The successional stage of a community impacts its stability, recovery potential, and trajectory over time in predictable ways. Here we demonstrate the value of successional theory for interpreting fine-scale community heterogeneity using Hawaiian coral reefs as a case study. We tracked benthic community dynamics on 36 forereefs over a six-year period (2017–2023) that captures impacts from high surf events, a marine heatwave, and unprecedented shifts in human behavior due to the COVID-19 pandemic. We document high spatial variation in benthic community composition that was only partially explained by island and environmental regime. Through hierarchical clustering, we identify three distinct community types that appear to represent different successional stages of reef development. Reefs belonging to the same community type exhibited similar rates of change in coral cover and structural complexity over time, more so than reefs located on the same island. Importantly, communities that were indicative of early succession (low coral cover reefs dominated by stress-tolerant corals) were most likely to experience an increase in coral cover over time, while later stage successional communities were more likely to experience coral decline. Our findings highlight the influence of life-history and successional stage on community trajectories. Accounting for these factors, not simply overall coral cover, is essential for designing effective management interventions. Site-specific management that accounts for a community’s unique composition and history of disturbance is needed to effectively conserve these important ecosystems.
README: Data from: Using community composition and successional theory to guide site-specific coral reef management
https://doi.org/10.5061/dryad.15dv41p6r
Description of the data and file structure
Benthic and environmental data are contained in four .csv files:
- Maui Nui monitoring program metadata.csv lists the island, site name, and coordinates of all sites in our large-area imagery Maui Nui monitoring program, including 16 sites not included in McCarthy et al. 2025. The year_month that each site was established and resurveyed is clearly indicated. This data covers 2014-2024.
- Maui Nui benthic data final.csv includes the following data for each of the 36 sites included in McCarthy et al. 2025 in each year they were surveyed:
- ind: A unique identifying string combining the island code, year surveyed, and site abbreviation (e.g., KAH_2016_01 for Kaho'olawe site 1 surveyed in 2016)
- Latitude: decimal coordinates for latitude
- Longitude: decimal coordinates for longitude
- Depth: Depth of center left calibration tile, in meters
- Hex: Color code used for making figures in R
- Site_abrr: Abbreviation for each site, used for making figures in R
- Island: Abbreviation for each island (KAH = Kaho'olawe, LAN = Lana'i, MAI = Maui, MOL = Moloka'i)
- Year: Survey year
- Date: Survey date
- Site: Site code, combines the island abbreviation and site abbreviation
- Por.com: Percent cover of Porites compressa
- Por.lob: Percent cover of Porites lobata
- Por.evr: Percent cover of Porites evermanni. This rare taxa could not be confidently distinguished from Porites lobata in all cases, so for the purposes of analysis we group Por.evr with Por.lob (see R code)
- Mon.cap: Percent cover of Montipora capitata
- Mon.pat: Percent cover of Montipora patula
- Mon.flb: Percent cover of Montipora flabellata. *This rare taxa could not be confidently distinguished from *Montipora patula in all cases, so for the purposes of analysis we group Mon.flb with Mon.pat (see R code)
- Pav: Percent cover of *Pavona *spp
- Poc: Percent cover of *Pocillopora *spp
- Oth.cor: Percent cover of all other coral taxa detected (rare, less than or equal to 1% cover). Includes Porites rus, Fungia spp., Leptastrea spp., Psammocora spp., and Gardineroseris planulata.
- CCA: Percent cover of crustose coralline algae
- Calf.mac: Percent cover of erect calcifying macroalgae (primarily *Halimeda *spp.)
- Turf: Percent cover of turf algae
- Cyano: Percent cover of fleshy cyanobacteria or cyanobacterial mats (where visually distinguishable from surrounding turf algae)
- Fl.Mac: Percent cover of fleshy macroalgae (non-cyanobacteria)
- Lob.Peys: Percent cover of encrusting macroalgae belonging to *Lobophora *spp. or *Peyssonnelia*spp.
- Other: Percent cover of other sessile taxa (non coral and non algae). Primarily comprised of sponges, zoanthids, and tunicates. This category comprises less than 1% cover for all but one site (Kaho'olawe 6 in 2021)
- Sand: Percent cover of sand or sediment
- cm.1: Linear rugosity (site level mean) measured at 1 cm resolution
- cm.50: Linear rugosity (site level mean) measured at 50 cm resolution
- X1.to.50: Fractal dimension, calculated as D = 1 - S, where S is the slope between each successive rugosity value on a log−log plot of linear rugosity vs. profile gauge resolution (see McCarthy et al. 2022; https://www.int-res.com/abstracts/meps/v702/p71-86)
- heterogeneity: the proportion of unlike adjacencies within a Voronoi tessellation of benthic data
- Long term environmental averages Ocean Tipping Points.csv includes long-term averages of environmental conditions at our sites. These data represent the long-term (mean) environmental conditions prior to our timeseries, rather than punctuated (max) conditions associated with disturbance during our timeseries. Other than depth data, which we obtained in situ using dive computers, all data in this tab were sourced from the Ocean Tipping Points project. This data is freely and publicly available for use, although the Ocean Tipping Points project requests that you cite the appropriate sources (which can be found here: https://www.pacioos.hawaii.edu/projects/oceantippingpoints/#datasources). Columns are as follows:
- site: the name of each site. Format is an abbreviation for island followed by the site code. Island abbreviations are as follows: KAH = Kaho'olawe, LAN = Lana'i, MAI = Maui, MOL = Moloka'i
- lat: decimal coordinates for latitude
- long: decimal coordinates for longitude
- depth: the depth of the site at the GPS coordinate, in meters, measured in situ using a dive computer.
- sst_sd: the standard deviation of sea surface temperature (°C) from 2000-2012
- wave_power: the maximum monthly mean of wave power (kW/m) from 1979-2013
- effluent_total: modeled data representing the total effluent from onsite sewage disposal systems (gal/kg2/day).
- sedimentation: modeled data representing the annual amount of sediment reaching the coast (tons/ha/yr).
- chl_a: long-term mean of chlorophyll-a (mg/m3) from 2002-2013.
- irradiance: mean surface irradiance (mol m-2 d-1) from 2002-2013.
- Environmental data 2017-2023.csv includes environmental conditions at our sites during our 2016/17 to 2023 monitoring program. We used this data in our study to conduct our GAMs multiple regression analysis of drivers of coral cover and rugosity change over time. Columns are as follows:
- site: the name of each site. Format is an abbreviation for island followed by the site code. Island abbreviations are as follows: KAH = Kaho'olawe, LAN = Lana'i, MAI = Maui, MOL = Moloka'i
- lat: decimal coordinates for latitude
- long: decimal coordinates for longitude
- timestep: options include 2017-2019, 2019-2021, and 2021-2023. Indicates the time period between large-area imagery surveys that the data corresponds to. The only exception here is dhw_CRW_2015, which corresponds to a time period before monitoring (2015), and was used to test the hypothesis that past thermal stress before monitoring began was a driver of coral cover and rugosity change.
- depth: the depth of the site at the GPS coordinate, in meters, measured in situ using a dive computer.
- dhw_CRW_2019: max degree heating weeks (°C-weeks), a measure of integrated heat stress over time, for the 2019 bleaching event. Sourced from Coral Bleaching Watch; http://pacioos.org/metadata/dhw_5km.html
- dhw_CRW_2015: max degree heating weeks (°C-weeks), a measure of integrated heat stress over time, for the 2015 bleaching event. Sourced from Coral Bleaching Watch; http://pacioos.org/metadata/dhw_5km.html
- wH_90: the 90th percentile of wave height (in meters), sourced from the WaveWatch III (WW3) Global Wave Model; http://pacioos.org/metadata/ww3_global.html
- turb_qmax_mean: the mean quarterly max of turbidity (in FNU), sourced from the Allen Coral Atlas "Ocean Water Turbidity" layer, which is based on Li et al. 2022; https://zslpublications.onlinelibrary.wiley.com/doi/full/10.1002/rse2.259. The turbidity data has been omitted in this Dryad upload due to the copyright constraints of Li et al. 2022. You can download turbidity data for our sites directly from Allen Coral Atlas, provided you cite Li et al. 2022 and do not use the data for commercial purposes. Instructions for downloading data from Allen Coral Atlas can be found at https://storage.googleapis.com/coral-atlas-static-files/resources-page-materials/Data_Download_Instructions.pdf. Alternatively, you can contact omccarth@ucsd.edu and directly request turbidity data for the sites in this study.
- stream_dist: the distance from each site to the nearest stream (in kilometers), sourced from the Hawai'i Statewide GIS Program; http://pacioos.org/metadata/hi_hcgg_all_darstreams.html
Code/software
Code for analyses is included in three files:
- Maui Nui analysis final.Rmd performs the following analyses:
- How does benthic community composition vary spatially and temporally in the Maui Nui region?
- Creates an NMDS to visualize benthic community composition trends in multivariate space
- Conducts a cluster analysis to identify sites with similar community composition, then conducts an ANOSIM to check if there are significant differences in community composition based on cluster and/or island
- Runs a PERMANOVA, with matrix of community data (percent cover, structural complexity, and landscape heterogeneity) as the response variable and both island and longterm averages of environmental variables as explanatory variables to see if community composition varies significantly between island and across environmental gradients
- Does coral community composition (cluster) or location (island) better explain patterns of benthic change over time?
- Uses a linear mixed effect model to test whether island, cluster, or both explain patterns of change over time across sites
- Looks for evidence of spatial autocorrelation in community composition and change in coral cover and rugosity over time
- What environmental factors are driving change in coral cover and structural complexity on Maui Nui’s reefs, and are the drivers of benthic change consistent over time?
- Tests various hypotheses of environmental, spatial, and ecological drivers for each timestep via linear and GAM multiple regression, selects the best model for each timestep
- How does benthic community composition vary spatially and temporally in the Maui Nui region?
- LDM_fun.R is dependency file that should be downloaded and stored in the working directory used in Maui Nui analysis final.Rmd. This dependency is required to install the LDM function (Hu & Satten 2023; https://cran.r-project.org/web/packages/LDM/LDM.pdf) that we use to perform a repeated measures PERMANOVA to account for the non-independence of resampling fixed sites.
- Percent cover confidence estimate.Rmd conducts stratified random sampling on a simulated coral reef to test the variation of percent cover estimates using a single 10x10 m plot with 2,500 stratified random points. The code creates an empty 10,000 x 10,000 raster, where each cell represents 1 x 1mm. Next it generates a population of coral colonies (represented as circles) where colony area was selected from a log normal distribution. it places a random number of coral colonies in the raster and then calculates the true percent cover of corals in this raster. Then, to simulate stratified random sampling with 2,500 points, the original raster is aggregated into a 50 x 50 cell raster, so that each cell would correspond to 20 x 20 cm. The code randomly samples one 1 x 1mm cell from within each aggregated cell to mimic Viscore’s stratified random sampling approach, and each aggregated cell is coded as either “coral” or “empty”. Next the code calculates the estimated percent coral cover of the simulated reef based on these aggregated cells, and calculates the absolute difference between true and estimated percent cover. The code is set up to repeat this process 10,000 times, each time simulating a new coral population. It allows users to calculate the 95% quantile of the absolute difference in percent cover estimates, and repeat this entire process for coral communities with different mean colony sizes.
Methods
Data Collection and Processing
We surveyed 36 fixed long-term monitoring sites in the Maui Nui region using large-area imagery, also known as Structure from Motion photogrammetry. We completed our first survey between July 2016 and July 2017, and resurveyed sites in 2019, 2021, and 2023. Most sites were surveyed four times (n = 30), while a minority were surveyed only three times (n = 2) or twice (n = 4). An interactive map of these data is available here. Surveys were completed as part of the 100 Island Challenge, a broader effort to document global patterns of coral reef condition and change over time. Sites were confined to forereef habitat with leeward exposure at approximately 10m depth to decrease environmental differences and enable fine-scale comparisons between sites, with eight to twelve sites selected per island. Several sites on Maui and Moloka'i are positioned close to or overlap with fixed long-term CRAMP transects maintained by the state of Hawai'i.
Our large-area imaging surveys followed the methods of Edwards et al. 2017. A diver recorded the GPS coordinates of a stainless-steel stake or eyebolt embedded in the reef, which served as a permanent reference for site relocation. Each 10 x 10 m site was marked with six calibration tiles and four reference floats to aid in diver navigation. During the survey, one diver measured the depth of each calibration tile and placed four 0.5m long scale bars within the site boundaries. The other diver swam with a custom camera rig 1.5m over the benthos in a gridded pattern to produce high overlap between photos. The rig contained two Nikon DSLR cameras (D7000, except for the 2023 surveys which utilized D780) in Ikelite underwater housings. One camera was equipped with a wide-angle lens (18mm for D7000, 24mm for D780) while the other had a longer focal length (55mm for D7000, 60mm for D780) to capture more magnified images. Lens focal lengths were changed slightly in 2023 to ensure that D780 photos covered the same spatial footprint as the earlier D7000 photos. Both cameras used an automatic one-second interval timer, which resulted in roughly 5,000 photos per site. The diver swam several meters beyond the core 10 x 10 m site to create a buffer zone of additional imagery. Each survey took approximately 50 minutes to one hour to complete.
After each field expedition, we generated a dense-point cloud reconstruction (3D model) of each site using Agisoft Metashape. Imagery from both cameras was aligned, but only imagery from the wide-angle camera was used to generate 3D models. The longer focal length images, which had higher resolution but a smaller spatial footprint than wide-angle images, were used as a reference for species identification. We used Viscore, a custom ecological analysis and visualization software to input depth and scale metadata and align 3D models of the same site from different years. This co-registration of models allowed us to analyze the same 10 x 10 m benthic community over time. One site with incomplete image overlap (Kaho'olawe 6) necessitated the use of a rectangular (12 x 5 m) plot instead.
Using aligned 3D models, we quantified three site-level metrics of benthic community composition for each timepoint: 1) percent cover of benthic taxa, 2) landscape heterogeneity, and 3) structural complexity. A single coral reef ecologist manually identified benthic cover at 2,500 stratified random points for each site (10x10m plot). Each point was identified to the finest taxonomic resolution, and then all point IDs were aggregated into 15 categories chosen to reflect the dominant benthic taxa and functional groups in Hawai'i (Montipora capitata, Montipora patula, Porites lobata, Porites compressa, Pocillopora spp., Pavona spp., other corals, turf algae, crustose coralline algae, calcifying erect macroalgae, fleshy encrusting macroalgae, fleshy erect macroalgae, cyanobacteria, sand, and other). We interpolated this spatially-explicit benthic data using a Voronoi tessellation to quantify landscape heterogeneity, defined here as the proportion of Voronoi polygon boundaries where two different taxa border each other (termed “unlike adjacencies”). Finally, we quantified structural complexity using linear rugosity and fractal dimension, which we derived from depth measurements using Viscore’s Virtual Profile Gauge tool, which assesses the height of the substrate along virtual transects.
Statistical Analysis
To analyze spatial and temporal trends in benthic community composition, we calculated the Bray-Curtis dissimilarity among all sites in all years. Our multivariate data consisted of 1) percent cover of 15 benthic classes, 2) landscape heterogeneity (the proportion of unlike adjacencies), and 3) structural complexity metrics (linear rugosity collected at 1 cm and 50 cm resolution, and fractal dimension). For comparability with percent cover data, we standardized landscape heterogeneity and structural complexity metrics to range from 0 to 1. We used non-metric multidimensional scaling (NMDS) to visualize benthic community composition for all islands and years together in multidimensional space using the vegan package. Next, we analyzed the same multivariate benthic data with a permutational multivariate analysis of variance (PERMANOVA) to identify significant drivers of benthic community variation. These drivers included island (fixed effect; Kaho'olawe, Lāna'i, Maui, and Moloka'i), depth (m), wave power (max monthly mean, kW/m), sea surface temperature (standard deviation, °C), chlorophyll-a (mean, mg/m3), surface irradiance (mean, mol m-2 d-1), sediment export (tons/ha/yr), and effluent (gal/kg2/day) (Figure S6, Table S1). Environmental variables represent long-term average conditions for our sites and were sourced from the Ocean Tipping Points project. None of the environmental variables had a correlation coefficient more extreme than 0.54. We used the LDM package to perform a repeated measures PERMANOVA to account for the non-independence of resampling fixed sites.
To identify groups of sites with similar benthic community composition, we used the same Bray-Curtis dissimilarity matrix of benthic data from our NMDS to conduct a hierarchical cluster analysis. We used the complete linkage method within the hclust function in R. Our observations from the field led us to believe that three ecologically distinct community types exist, which was supported by the hierarchical clustering dendrogram. To explore the robustness of our results to clustering method, we also implemented clustering using Ward’s method. While a few sites changed position, the same three ecological groupings were achieved regardless of the clustering method employed. In addition, we used analysis of similarities (ANOSIM) to test whether dissimilarity was greater between or within groups, first using island and then using cluster as the grouping factor. We calculated each ANOSIM using 9,999 free permutations with the vegan package.
Our ANOSIM corroborated our observation that these clusters likely represent distinct benthic communities, so next we tested the hypothesis that reefs of the same community type would respond similarly to disturbance events regardless of their location within Maui Nui. We used mixed-effects linear models to test whether the fixed factor of community type or island better described the following response variables: 1) change in coral cover, and 2) change in linear rugosity (1 cm resolution) in each timestep. We calculated each response variable as the monthly rate of change between each survey for each site to account for inconsistent durations between surveys. We included site as a random effect in our models and modeled variance using the varIdent function to account for heteroskedasticity. We used Tukey’s honest significance test to identify significant pairwise differences using the emmeans package, and selected the best model using AIC.
Lastly, we used multiple regression with generalized additive models (GAMs) to identify which environmental variable(s) best explained patterns of change in coral cover and rugosity for each timestep of our monitoring program (2016/17–2019, 2019–2021, and 2021–2023). We hypothesized that wave damage, bleaching related mortality, and sedimentation stress could have driven coral mortality during our timeseries (Figure 1C) and constructed a GAM to explore each hypothesis. We also quantified maximum thermal stress at each site in 2015 to test whether bleaching impacts pre-dating our monitoring could explain coral cover and rugosity change from 2017–2019. For our second and third timestep, we tested the hypothesis that change in coral cover and rugosity in the preceding timestep would explain subsequent patterns of benthic change (via recovery dynamics, succession, or chronic degradation). We obtained site-level data for three environmental variables: thermal stress (max degree heating weeks (°C/weeks) from NOAA Coral Reef Watch, wave height (90th percentile (m) from the WaveWatch III (WW3) Global Wave Model, and turbidity (mean of quarterly max (FNU) using remotely sensed data from Planet Dove. Turbidity data were not available for our first timestep, so for each site we also quantified the distance to the nearest stream as a proxy for sedimentation using the Hawai'i Statewide GIS Program, which we log transformed to reduce the influence of outliers. Finally, we incorporated island and community type to account for the effect of local environmental factors and benthic community composition.