Data from: Unexpected system-wide resilience of Australian coral reefs to recent heatwaves and cyclones
Data files
Apr 27, 2026 version files 443.69 KB
-
Edgar_et_al_Coral_Change_Dryad.csv
393.80 KB
-
Edgar_et_al_Coral_Change.R
25.28 KB
-
original_with_focus_on_time_manual_v2.neta
19.12 KB
-
README.md
5.49 KB
Abstract
This data examines our standardized monitoring of 476 reef sites spanning Australian waters (>5,000 km) from 2008 to 2024 despite local reef volatility. Sites were censused using the Reef Life Survey methodology within the period 21 March 2008 to 17 Dec 2024. Site initial and final sampling dates spanned 7.2 years on average, with 91% of sites assessed both before and after the extreme March 2016 heatwave. A total of 48 sites were surveyed across the southern Great Barrier Reef and Coral Sea in November and December 2024, thus capturing mortality from the severe March 2024 bleaching event. Data was analysed in R using Random Forests methodology and based on the CSV file. Complementary to this was a Baysian Network model (developed in Netica, www.norsys.com) which showed similar results but highlights that the partition of factors is informative. Both models are included. The compiled Bayesian Network model can be examined in the freely downloaded software from Norsys.
Dataset DOI: 10.5061/dryad.xd2547dwf
Files and variables
The CSV file shows the field sites that were censused using the Reef Life Survey methodology within the period 21 March 2008 to 17 Dec 2024.
File: Edgar_et_al_Coral_Change.R
Description: R code to run the analysis in R
The R code shows the relationships between absolute change in coral cover and potential drivers of change were assessed using random forest models.
File: original_with_focus_on_time_manual_v2.neta
Description: A compiled Bayesian Network model developed in Netica by Norsys.
File: Edgar_et_al_Coral_Change_Dryad.csv
Description: A table of sites with the survey results and covariates. Variables include:
- case: integer
- X: Integer
- site_code: Catagorical such as CS1
- site_code2: Catagorical such as CS1b
- start_date: dd/mm/yyyy
- end_date: dd/mm/yyyy
- time_diff: Time in days between t1 and t2 (integer)
- start_cover: Total cover (%) of hard coral at t1.
- end_cover: Total cover (%) of hard coral at t2.
- cover_diff: % coral cover difference between t1 and t2
- acropora_cover_diff: % coral cover difference between start and end for Acropora only
- porites_cover_diff: % coral cover difference between start and end for Porites only
- massives_cover_diff: % coral cover difference between start and end for Massives only
- latitude: Site Latitude in degrees
- longitude: Site Longitude in degrees
- depth: Depth (m) of surveyed transect with correction to low tide benchmark. When not identical, nearest depths were matched in t1 and t2 (e.g. 4 m and 5 m at t1 and 5 m and 6 m at t2 would match up as 4-5 m and 5-6 m comparisons) and the mean of the two depths applied.
- exposure: Site wave exposure assessed on site using four categories: 1= sheltered, wind waves <1 m; 2= waves 1-3 m; 3= ocean swell <3 m; open to ocean swell from prevailing direction.
- visibility: Mean underwater visibility measured on the seabed along the transect line by divers, averaged from all surveys at the site (m).
- total_fish_biomass: Total estimated biomass of fishes (kg/ha) observed within 500 m2 transect blocks at each site on the day of survey (Edgar & Stuart-Smith 2014).
- total_fish_richness: a measure of diversity see [63] in paper
- COTS: number of Crown of Thorns Starfish per Hectare, see [54]
- max_cycl_intensity: Magnitude of the largest cyclone impacting site during the interval from t1 to t2. Measured as number of hours that the site was impacted by significant wave heights (Hs) ≥4 m, termed ‘damaging seas’ (see Puotinen* et al.* 2016). Significant wave height represents the average wave height of the top one-third wave heights estimated for a given area. An Hs = 4 m implies maximum wave heights up to about 10 m. Field studies show such waves are sufficient to cause major damage to structurally vulnerable corals. Values were averaged for each site and month.
- date_max_cycl_intensity: dd/mm/yyyy of cyclone impact for maximum
- days_since_max_cycl: Time in days from the most damaging (longest exposure to damaging seas) cyclone identified in max_cycl_intensity to t2.
- SST_mean: Degrees Celcius Sea Surface Temperature
- SST_max: Degrees Celcius Sea Surface Temperature maximum
- max_DHW: Degree Heating Week, Magnitude of the largest heatwave recorded during the interval from t1 to t2, measured as cumulative degree heating weeks (DHW) for the 12 week period prior to day; determined as the maximum monthly values for the period of t1 to t2. Zero values were applied if no DHW. Data were obtained from NOAA Coral Reef Watch at a 5 km-resolution (NOAA Coral Reef Watch & Potemra 2021); extracted using the rerddap package within R statistical software language (Chamberlain 2025; R Core Team 2025).
- no_DHW_8: number of DHW more than 8 (see max_DHW)
- date_max_DHW: dd/mm/yyyy The date of the maximum DHW
- days_since_max_DHW: Days, Time in days from the most impactful heatwave, as identified in max_DHW above, to t2.
- MPA: Marine Protected Area status in text format
- prop_diff: Proportional difference in change in Coral Cover base on %
- months_since_max_cycl: Time in months from the most damaging (longest exposure to damaging seas) cyclone identified in 4 to t2.
- log_months_since_max_cycl: log scale of months since cyclone max
- months_since_max_DHW: Time in months from the most impactful heatwave, as identified in Max_DHW, to t2.
- log_months_since_max_DHW: log scale of months of max DHW
- log_total_fish_biomass: Log scale of total fish biomass
- months_since_last_survey: Time in months between t1 and t2
Code/software
The R code requires the free R software from www.cran.org
R version 4.4.2 (2024-10-31 ucrt) based on the libraries cowplot, randomForestSRC and tidyverse
Code input requires "Edgar_et_al_2025_Coral_Change.csv" only
The neta file requires the Netica software downloadable free from www.norsys.com
Access information
Other publicly accessible locations of the data:
- The CSV contains data that can be partially access from the Reef Life Survey website at https://reeflifesurvey.com/
Data was derived from the following sources:
- The Degree Heating Weeks data was obtained from NOAA Coral Reef Watch program.
Site initial and final sampling dates spanned 7.2 years on average, with 91% of sites assessed both before and after the extreme March 2016 heatwave. A total of 48 sites were surveyed across the southern Great Barrier Reef and Coral Sea in November and December 2024. Divers photographed 50-m transect lines at 2.5 m intervals to assess coral cover (20 0.4-x-0.3 m photoquadrats per transect), with two transects generally assessed at each site. Sites were each surveyed 2.9 times on average. Corals in photoquadrats were identified to highest taxonomic level (generally species) by a single expert (ET) for consistency. A total of 505 coral taxa were categorized in images, of which 351 were known species (73% of total cover identified to species) and 154 recorded as a higher taxonomic and morphological combination (98% of total cover identified to genus).
For the Bayesian Network Model, the causal network formed the basis of a naive Bayesian Network (BN) within the Netica V6.04 software environment (Norsys Software Corp 2016). Each factor was broken into five approximately equal sample number bins (or states). At this stage the model is considered naïve, with the joint probabilities required to be updated based on the case file (Marcot et al. 2006). For this model each of the 1668 time couplets was considered an independent case. The conditional probability tables (CPTs) were updated by importing a randomly selected 90% survey case file using a counting process (Marcot 2012). The BN model was compiled, and each node contained the marginal probabilities for each factor. The BN was then tested for predictive accuracy focused on the Coral Cover Difference factor using the associated 10% isolated data set. The testing compared the observations of coral cover change with the BN predictions given the environmental data. This generated a number of indices such as correlation matrix error (Pearl 1978), which provide a measure of accuracy of the model structure and parameterization (Marcot 2012). To assess the extent that the target node could be influenced by a single finding at each of the other nodes the variance reduction was calculated (Marcot 2012).