Methods
Data collection
To identify where coral restoration is occurring globally and determine variables mediating success, I integrated global datasets of climatic stressors (Liu et al. 2017), local stressors (Andrello et al. 2022), reef habitat quality (>2,000 reefs; Darling et al. 2019; Madin et al. 2019), and coral restoration project design (Boström-Einarsson et al. 2020). Whereas reef habitat and restoration project data were primarily collected using underwater SCUBA surveys along replicate transects or quadrats during the years 2010 to 2016 (Figure 5), the stressor data were extracted and calculated from satellite data.
Figure 5. Under water visual survey of coral reef along transect conducted via SCUBA (NOAA).
Analysis
Geospatial manipulations
To integrate these datasets, I used spatial joins to geospatially match reef sites where climate stressors, local stressors (Figure 6A), and reef habitat quality (Figure 6B) were monitored with restoration sites (Figure 6C). Following a sensitivity analysis, whereby varying buffer sizes were applied to reefs sites, I created a 10km buffer around all sites monitored for reef habitat quality prior to spatially joining the data. This buffer size ensured data were joined for all reefs within the same reef tract. Once data were integrated, I generated two data tables: (1) unique reef sites (n = 2,227) where environmental (climatic and local) stressor and reef habitat quality data were present, and (2) unique reef sites (n = 183) where restoration is currently being implemented and restoration design, environmental stressor, and reef habitat quality data were present. The predictor and response variables associated with each question and associated table are listed below (Figure 7).
Where is coral reef restoration occurring?
To evaluate the conditions where reef restoration is being implemented, as well as where restoration is not being implemented, I distinguished reefs (n = 2,227) globally based on environmental stress and reef habitat quality via k-means clustering, which delineates clusters by identifying high-density regions within the data (Borcard et al. 2018). I used Bray-Curtis distance to cluster reefs based on these variables across both sites targeted by restoration and sites not targeted by restoration. Hereafter, I refer to these clusters as reef type.
To assess spatial variation in variables across reef types, we applied Principal Coordinate Analyses (PCoA) to the dissimilarity matrix of environmental stress and habitat quality at all reefs, correcting for negative eigenvalues using Lingoes correction (Gower and Legendre 1986). To visualize the primary conditions characterizing each reef type, we extracted the top scores of environmental stressor and reef habitat quality variables along the first PCoA axis. We evaluated how restoration sites distribute along these axes to approximate which reef types are currently being targeted by restoration and what conditions are most associated with current restoration sites. I expanded on this by calculating the percentage of sites within each reef type at which restoration is currently being implemented, as well as the mean value of environmental stressor and reef habitat quality variables for sites within each reef type.
To better understand how and where restoration sites are currently being selected by practitioners, I applied a classification and regression tree (CART) analysis, a constrained clustering method that successively partitions the data based on a response variable and threshold values of predictor variables (De'Ath 2002). I selected the final tree based on explanatory and predictive power, minimizing relative and cross-validated relative error (Borcard et al. 2018). This analysis was used to classify (n = 2,227) reefs globally by restoration occurrence based on threshold values of environmental stressor and reef habitat quality variables. These classifications helped identify conditions that best explain restoration presence or absence at a reef.
Figure 6. Spatial distributions of coral reef sites and data globally. Maps display location of reefs for which environmental stress was calculated (A; Andrello et al. 2022), reefs for which reef habitat quality was monitored (B; Darling et al. 2019), and restoration sites (C; Boström-Einarsson et al. 2020). Restoration sites are coloured pink-red.
Figure 7. Explanatory and response variables used to evaluate the current scope of coral reef restoration and determinants of restoration success. Variables were extracted from environmental stressor (Andrello et al. 2022), reef habitat quality (Darling et al. 2019), and restoration methods (Boström-Einarsson et al. 2020) datasets.
What determines coral reef restoration success?
To identify the variables mediating the success of coral restoration projects globally, I conducted a CART analysis. This analysis was used to partition restoration sites (n = 183; for sites with restoration success recoded n = 73) based on climate stressors, local stressors, reef habitat quality, and restoration methods to determine which variables best predict restoration success, defined as the percentage of surviving restored corals (Boström-Einarsson et al. 2020). I selected the smallest trees with the highest explanatory and predictive power, or the lowest relative and cross-validated relative error.
Where will coral reef restoration be successful?
I then used this information to predict restoration success at sites where restoration is not currently being implemented by running a random forest analysis (500 trees) across the remaining (n = 2,044) reefs. From these predictions and actual values, I calculated the mean survival percentage of corals for each reef type to determine how restoration success varies across different levels of environmental stress and reef habitat quality conditions globally.