Testing and integrating CoTS connectivity matrices into ReefMod-GBR (CCIP-RT-08, UQ, JCU, and QUT)

Full description

This dataset documents the evaluation and selection of crown-of-thorns starfish (CoTS) larval connectivity matrices for use in ReefMod-GBR, along with the subsequent calibration of CoTS population parameters. The project evaluated which pre-existing connectivity matrix set produces the most accurate CoTS density predictions when integrated into the model, and determined the optimal parameter values for that matrix set. Note: This metadata record is published for discovery purposes only; the dataset files are not directly available for download. To obtain the data, contact the dataset author. The primary outcomes of this work were: 1. The selection of the GBRLUP connectivity matrix set (a shallow-water subset of the GBRL set, derived from the CCIP-R-05 project) for integration into ReefMod-GBR. 2. Calibrated CoTS Beverton-Holt recruitment parameters (alpha, beta) and mortality coefficients, both GBR-wide and region-specific, optimised against manta tow observation data. These parameter values have been integrated into ReefMod-GBR for ongoing CoTS outbreak prediction. The dataset contains the scripts, input data, configuration files, and reports that document how these decisions were reached. It is not a standalone reproducible workflow, as the core modelling software (ReefMod Engine, RME) is not publicly available. The scripts and configuration files serve primarily as methodology documentation. For further information, refer to the main document: Framework for integrating CoTS matrices.pdf Methods: Three connectivity matrix sets were considered candidates to be integrated, including GBR1, GBRL, and SLIM (Choukroun et al., 2025. Larval dispersal predictions are highly sensitive to hydrodynamic modeling choices). These sets were analysed to confirm compatibility with ReefMod-GBR and to test their prediction accuracy, assessing their suitability for integration. Candidate matrix sets were compared with the default set (modelled using GBR4, 4km resolution, set of seven matrices, each corresponds to summer periods 2008-2009, 2010-2011, 2011-2012, 2012-2013, 2014-2015, 2015-2016, and 2016-2017) in terms of number and strength of links (see report ComparingMatricesCont and ComparingMatricesInitialR04R05_SummaryReport). To validate the prediction accuracy of CoTS density after integrating matrix sets into the model, repeated hindcast runs were performed in the ReefMod Engine (RME) as a proxy for ReefMod-GBR. RME is a C++ re-implementation of the MATLAB ReefMod source code. It produces comparable outputs while being significantly faster, making it suitable for repeated simulations. Prediction accuracy was defined here as the difference between predicted and observed categories of CoTS density per manta tow (Skinner et al., 2024). The prediction accuracy results informed the final selection of GBRLUP (a subset of GBRL representative of the shallow water environment) for ongoing modelling and CoTS outbreak prediction (see report PredictionAccuracyComparison and PredictionAccuracyComparisonCont). Five yearly CoTS connectivity matrices were created by averaging GBRLUP matrices across different settlement radii and pre-competency periods. The GBRLUP set contained 45 matrices, with one matrix for each combination of settlement radius (three values) and pre-competency period (two values), covering spawning seasons from 2018-19 to 2022-23. After integrating the selected matrices into ReefMod-GBR, CoTS parameters were calibrated to refine model accuracy and minimise the number of underpredictions. Calibration was carried out using a genetic algorithm that solved for the minimum value of the criterion of fit, defined here as the accuracy of CoTS density category predictions, expressed through a customised weighted kappa loss function with asymmetrical weighting. Ideally, underpredictions should be avoided, as they increase the risk of failing to detect an outbreak. Parameters were calibrated either for the entire GBR or by region, and the finalised parameters were then integrated into ReefMod for future development. For further information, refer to the main document in the dataset: Framework for integrating CoTS matrices.pdf Reproducibility Requirements: The workflow documented in this dataset cannot be independently reproduced without access to the ReefMod Engine (RME). The scripts and configuration files are provided as methodology documentation rather than as executable code. Required components (not included): - ReefMod-GBR: Available on GitHub: https://github.com/ymbozec/REEFMOD.7.3_GBR - ReefMod Engine (RME): Contact the developer John Hedley or Shu Kiu Leung. - Custom configuration file: An example is provided in "2. RunningRME/data/config and init/config_hindcast_GBRLUP_cycle_0SR.xml". - Reef ID files for CoTS control: Needs to be manually created, as described in "Framework for integrating CoTS matrices.pdf". Limitations of the data: - This dataset documents the decision-making process and cannot be used to independently reproduce the analysis without access to the ReefMod Engine (RME). - Data from manta tow surveys from AIMS Long Term Monitoring Program, Field Management Program, and the GBR CoTS Control Program were used in this analysis. Manta tow surveys are less effective for detecting juveniles and solitary adults so might underestimate CoTS outbreaks. However, the data remains the most comprehensive and widely used CoTS dataset for the GBR to date. - While the model hindcast spans across 2007 to 2024, connectivity matrix data were only available for five years from 2018 to 2022. To address this, we cycled through the available matrices, but we acknowledge that this approach does not fully represent real-world conditions and the decision to cycle the matrices may have introduced some inaccuracies. - The analysis was conducted using RME version 4.01, which is still under development and does not yet include a built-in hindcast function. As such, the hindcast was simulated using customised configurations, which do not exactly replicate the base ReefMod-GBR conditions. RME is only a proxy of the source code and does not produce 100% identical results even with the same random seed. Other differences might arise from the fact that the source code can simulate bleaching despite cyclone events during the hindcast (while RME can’t), which affects coral cover and, in turn, CoTS dynamics. Format of the data: This dataset consists of 1 main document (PDF), 30 MATLAB scripts, 3 R scripts, 28 data files (MAT, CSV, XML, and XLSX), 58 connectivity matrices (CSV, MAT), 1 RME configuration file (XML), 6 relevant reports (DOCX and PDF). Main document: Framework for integrating CoTS matrices.pdf MATLAB scripts: - RME setup and analysis of connectivity matrices - AveragingmatricesR05.m creates a mean matrix from multiple matrices in the dataset. This script was used to average GBRLUP matrices selected for integration. - random_init_coral_rme.m stores the code that generates the initialisation files for coral cover and rubble cover. - connectivity_hindcast_run_new.m Runs RME with the customised configuration. Produces CoTS per manta tow per reef for yearly time points 2007 to 2024. - Predicted_COTS_SL_RME.m converts RME outputs (.mat file) into .csv file that is readable by R. This script is run before simple_analysis_RME.R - simple_analysis_RME.R - processes RME outputs and compares them with observations to produce plots and statistical summary tables of prediction accuracy. - Sensitivity analysis of parameters for calibration: - analyze_sensitivity_parallel.m - script for performing simple sensitivity analysis by varying parameters one at a time: alpha, beta, and mortality values are adjusted from -20% to +20%, and outbreak duration is varied from 1 to 10 years in fixed increments. - lossfun_kappa_org.m - function to calculate criterion of fit (weighted Kappa loss), which was fed into the script for sensitivity analysis. Includes all five parameters - alpha, beta, CoTS mortality, and outbreak duration. - runRME_R05S_cat_org.m - wrapper function to run RME to produce predictions, which was passed to the criterion of fit function. Includes all five parameters - alpha, beta, CoTS mortality, and outbreak duration. - Scripts for optimising CoTS parameters: - scaled_run_optimization_cat_ga.m - optimisation script that uses genetic algorithm (GA) in MATLAB to search for the lowest value of loss within bounds of parameter values - lossfun_kappa.m - function to calculate criterion of fit (weighted Kappa loss), which was then fed into the optimisation script. Only includes alpha, beta, and CoTS mortality. - runRME_R05S_cat.m - wrapper function to run RME to produce predictions, which was passed to the criterion of fit function. Only includes alpha, beta, and CoTS mortality. - optimize_region_(1-4).m - refined optimisation scripts that use genetic algorithm (GA) in MATLAB to search for the lowest value of loss within bounds of parameter values for each region. - lossfun_kappa_regions.m - function to calculate criterion of fit (weighted Kappa loss) for each region, which was then fed into the optimisation scripts. - runRME_R05S_cat_regions.m - wrapper function to run RME to produce predictions, which was passed to the criterion of fit function. Same as GBR-wide. - generate_region_scripts.m - script to quickly generate individual optimisation scripts for all 4 regions based on optimize_region_4.m. - Validation of calibrated parameters: - Predicted_COTS_SL_RM.m converts RM (source code) outputs (.mat file) into .csv file that is readable by R. This script is run before simple_analysis_RM.R. - simple_analysis_RM.R processes ReefMod-GBR (source code) outputs and compares them with observations to produce plots and statistical summary tables of prediction accuracy. - f_runmodel_SL.m - script to run 1 simulation in the source code. Updated to call new settings for COTS and GBR-wide/region-specific parameters. The region-specific version controls how individual reefs can have different alpha and beta values depending on which region they belong to. - settings_CONNECTIVITY_NEW_SL.m - example script to load GBRLUP_CONNECT.mat - settings_GBR_NEW.m - settings for initial values of coral cover, rubble cover, and CoTS density across the GBR. Comment or uncomment the specified lines to turn on or off observation forcing. - f_apply_COTS_predation.m - updates the calculation of the CoTS population and coral predation. - MULTIPLE_REEF_SETUP.m - updates how individual reefs can have different mortality rates depending on which region they belong to. - setting_COTS.m - settings of CoTS parameters, including alpha, beta, mortality coefficients. - INITIALISATION.m - script to generate and load initial parameters for ReefMod. Updated to load new connectivity matrix settings (settings_CONNECTIVITY_NEW_SL.m) - plot_param_regions.m - plot showing changes in the shape of the Beverton-Holt recruitment curve and the mortality function using the newly calibrated GBR-wide and region-specific parameter sets (see report CalibrationResults.pdf) R scripts: - ComparingMatrices.R calculates multiple metrics for each connectivity matrix set, including number and strength of links per reef, reef rankings as sources or sinks at different spatial scales, self-retention, and the proportion of source and sink numbers or strengths between different sets of matrices. - simple_analysis_RME.R processes RME outputs and compares them with observations to produce plots and statistical summary tables of prediction accuracy. - simple_analysis_RM.R processes outputs from the source code and compares them with observations to produce plots and statistical summary tables of prediction accuracy. Reports: - ComparingMatricesInitialR04R05_SummaryReport.pdf compares the number of reefs, as well as the ranking of source/sink reefs of GBR4 and R05 connectivity matrices. - ComparingMatricesCont.pdf compares the number of links, strength of links, proportion of number/strength of links of GBR4 and R05 connectivity matrices. - PredictionAccuracyComparison.pdf compares the prediction accuracy of RME models incorporating different connectivity matrices. - PredictionAccuracyComparisonCont.pdf continues this analysis using updated R05 matrices, providing further evaluation of model performance. - SelfRetentionSummary.pdf compares self-retention values between GBR4 and R05 connectivity matrices. - CalibrationResults.pdf details the results of preliminary sensitivity, calibration, and validation of both GBR-wide and region-specific parameters calibrated to the CoTS control dataset. Configuration xml: - config_hindcast_GBR4Cots.xml - example custom configuration file. Data files: - Summary_Correlations.xlsx summarises the output regarding Kendall's tau correlation coefficients and p-values from ComparingMatrices.R. - averageConnectivityMatrix(R05Up, R05Dp, R05all, UpNormed, DpNormed, Normed).csv - averaged connectivity matrices generated during the CCIP-R-05 project. Under CCIP-R-05, new CoTS connectivity matrices span 2018-2022: 2018 (10), 2019-2021 (20 each), and 2022 (10), with half for shallow and half for deep water. Normed = with normalised values, Up = shallow, Dp = deep. - coral_(sp1-sp6)_2007_pct.csv - initial coral cover files - rubble_2007_pct.csv - initial rubble cover file - cots_2007.csv - initial CoTS density file - GBR_past_DHW_CRW_5km_2007_2024.csv - yearly DHW file from 2007 to 2024 - GBR_PAST_CYCLONES_2007_2024.csv - cyclone categories file from 2007 to 2024 - Observations_COTS_2025.csv - CoTS observations data up to 2025 - GBR_REEFS.csv - GBR reef reference data - region_reefs.csv - List of reef IDs and corresponding region IDs. - GBR_PAST_COTS_NEW_2025.csv - CoTS density data from manta tow. - IB_reefs.csv - list of reef IDs within the initiation box. Connectivity matrices: - 'Coral_GBR1_matrices' - folder containing 49 GBR1 matrices for coral connectivity - 'CoTS_GBRLUP_matrices' - folder containing 5 averaged GBRLUP matrices for CoTS connectivity - GBRLUP_CONNECT.mat and R05S_CONNECT.mat - example sets of connectivity matrices that can be loaded into the ReefMod-GBR source code Data dictionary: GBR_REEFS_2023.csv - RM_ID: Reef ID in ReefMod (1-3806). - AIMS_sector: 11 AIMS sectors for Long Term Monitoring Program (LTMP). - Sector: 1, Name: Cape Grenville - Sector: 2, Name: Princess Charlotte Bay - Sector: 3, Name: Cooktown / Lizard Island - Sector: 4, Name: Cairns - Sector: 5, Name: Innisfail - Sector: 6, Name: Townsville - Sector: 7, Name: Cape Upstart - Sector: 8, Name: Whitsunday - Sector: 9, Name: Pompey - Sector: 10, Name: Swain - Sector: 11, Name: Capricorn Bunker - Region: Management regions of Great Barrier Reef Marine Park (GBRMP). - FN = Far northern - N = Northern - C = Central - S = Southern - Shelf_position: Position of the reef on the shelf. - I = Inner - M = Mid - O = Outer Summary_Correlations.xlsx - SourceCoEff: source correlation coefficient between GBR4 and R05 matrices. - SourceP: p-value of the source correlation coefficient. - SinkCoEff: sink correlation coefficient between GBR4 and R05 matrices. - SinkP: p-value of the sink correlation coefficient. GBR_REEFS.csv - Reef_ID: Reef ID in ReefMod (1-3806). - LAT: latitude - LON: longitude - Shelf position: Position of the reef on the shelf. - 1 = Inner - 2 = Mid - 3 = Outer - AIMS_sector: 11 AIMS sectors for LTMP (see GBR_REEFS_2023). - GreenZone: Indicates whether the reef is within the green zone (GZ). - 0 = outside GZ - 1 = inside GZ - CullSites: Number of cull sites at the reef. - GBRMPARegion: Reclassified management regions of the Great Barrier Reef Marine Park. - FN = Far northern - N = Northern - C = Central - S = Southern - Reef_Type: Type of reef for CoTS control (i.e., whether the reef is a priority reef for CoTS control). - N = not priority - P = priority Please refer to Great Barrier Reef (GBR) Features (https://catalogue.eatlas.org.au/geonetwork/srv/eng/catalog.search#/metadata/ac8e8e4f-fc0e-4a01-9c3d-f27e4a8fac3c) for UNIQUE_ID, LABEL_ID, and GBR_NAME. Observations_COTS_2025.csv - ReefModID: Reef ID in ReefMod (1-3806). - ReefID: same as LABEL_ID. - ReefName: Name of the reef. - Source: Source of observation data. - LTMP = Long Term Monitoring Program - FMP = Field Management Program - CP = GBR CoTS Control Program - meanCoTS = mean number of CoTS per manta tow. - n_obs = number of observations Data access: The data files are not published with this metadata record. To obtain the dataset, contact Shu Kiu (Suki) Leung. A listing of the files contained in the dataset is provided in dataset-files.txt. Location of the data: This dataset is filed in the eAtlas enduring data repository at: data\custodian\2020-2029-other\GBR_CCIP-RT-08_Reef-connectivity-matrices_2025

Lineage

Maintenance and Update Frequency: asNeeded

Notes

Credit
The development of the analytical method described in this record is funded by The University of Queensland (UQ), James Cook University (JCU), and Queensland University of Technology (QUT). The connectivity data were provided by Severine Choukroun, Michael Bode, and Owen Stewart.