Restoration cannot be scaled up globally to save reefs from loss and degradation

Abstract

Coral restoration is gaining popularity as part of a continuum of approaches addressing the widespread, recurring mass mortality events of corals that-together with elevated and chronic mortality, slower growth and recruitment failure-threaten the persistence of coral reefs worldwide. However, the monetary costs associated with broad-scale coral restoration are massive, making widespread implementation challenging, especially with the lack of coordinated and ecologically informed planning. By combining a comprehensive dataset documenting the success of coral restoration with current and forecasted environmental, ecological and climate data, we highlight how such a coordinated and ecologically informed approach is not forthcoming, despite the extent of previous and ongoing efforts. We show that: (1) restoration sites tend to be disproportionally close to human settlements and therefore more vulnerable to local anthropogenic impacts; (2) the immediate outcomes of restoration do not appear to be influenced by relevant ecological and environmental predictors such as cumulative impact; and (3) most restored localities have a high and severe bleaching risk by the middle of this century, with more than half of recently restored sites already affected. Our findings highlight the need for the coral reef community to reinforce joint development of restoration guidelines that go beyond local objectives, with attention to ocean warming trends and their long-term impacts on coral resilience and restoration success.

Fig. 1. Quantifying the hypothetical cost of broad-scale coral restoration.

The plots show the upper and lower bounds (shaded areas) and median values (solid lines) of estimated restoration costs of coral reefs based on published values as a function of the area of degraded reef that could be rehabilitated according to different techniques. The data used to estimate the upper, median and lower boundaries are from Table 1 of Bayraktarov et al.. We report these data together with estimates of the costs to restore 10% of the 11,700 km² (that is, 1,170 km² = 117,000 ha) area of damaged reef (dashed lines in each plot). Source data

Fig. 2. Factors affecting the choice of restoration sites.

Partial dependency plots showing the marginal effects of the independent variables included in our model on the predicted probability of a site being the target of restoration. We report both the fitted functions (solid orange lines; n = 10,000) and fitted values (blue dots; n = 3,324). The rug plots on top of the panels show the density of observed values in the target independent variables. The percentage values in parentheses indicate the relative influence of each variable. Gravity was obtained from ref. and represents the summed gravity of locations within a radius of 500 km from a target reef location, with individual gravity of each location within the radius computed as the ratio between the population of that location, and the squared distance (measured as travel time, in minutes) from the location and the target reef location. In the plot, we report the gravity values quoted by 1,000 to ease visualization of labels in the x axis. For the quantification of cumulative impacts, refer to ref. . Remoteness is reported as log-transformed travel time from a target reef location to the closest large human settlement, as in ref. . Bleaching alert levels refer to NOAA Coral Reef Watch data and we considered as ‘severe bleaching alert’ level events, those with alert levels I or II (Methods). Source data

Fig. 3. Observed success of coral restoration projects compared with the success predicted by 1,000 boosted regression tree models based on a large set of independent variables identifying coral restoration techniques, ecological factors and disturbances.

The data are from a cross-validation exercise in which we trained and tested 1,000 models on spatially independent sets of observations (including 80% of data in the training sets and 20% in the testing sets). The mean R² of observed versus predicted restoration success in the 1,000 replicates was 0.05 ± 0.06 s.d. and the overall R² for all of the points in the plot was 0.00002. See Methods for details on how ‘success’ was quantified. Source data

Fig. 4. Fraction of recently restored sites exposed to bleaching alert level I and/or II per year compared with control (non-restored) sites.

For each target year, we identified all of the localities restored in the preceding five years and identified as a control all of the remaining reef localities. We then computed the fraction of restored and control localities that were exposed to bleaching alert levels I and/or II. The data are aggregated in 5-year intervals, with values on the x axis indicating the upper boundary of the interval (that is, 1990 = 1986–1990). We shifted the time series for the restored and control data horizontally by ±1 unit to ease visualization. The data are presented as means ± s.d. (n = 5). Source data

Fig. 5. Number of predicted end-of-century coral mass mortality years compared with the timing of the first mass mortality year under an intermediate emissions scenario (SSP2-4.5).

The black dots indicate restored localities and the grey dots represent the other reef localities that have not undergone restoration (0.5° × 0.5° grid cells). Mass mortality years for a given locality are identified as those when the maximum projected DHW value is ≥20 °C-weeks. This analysis was based on projected DHW data from ref. . The vertical and horizontal solid lines indicate the means of the x and y values. Source data

Fig. 6. First occurrence and frequency of years when reef localities (n = 3,929) experienced a maximum DHW value of ≥20 °C-weeks under an intermediate emissions scenario (SSP2-4.5).

Restored sites (n = 256) are identified in the maps by small black dots within the coloured pixels. The projected DHW data are from ref. . The boxplots summarize the values shown in the maps, with boxes indicating first and third quartiles, horizontal lines indicating median values, whiskers indicating the largest and lowest points inside the range defined by the first or third quartile +1.5 times the interquartile range and circles indicating outliers. Source data

Extended Data Fig. 1. Global distribution of restored sites from the dataset.

We aggregated data from Boström-Einarsson et al. at a resolution of 4° × 4° latitude/longitude to ease visualization. Source data

Extended Data Fig. 2. Frequency distribution of the independent variables used to predict the choice of restoration sites.

Total n observations = 3324. Source data

Extended Data Fig. 3. Frequency distribution of the independent variables used to predict coral restoration success.

Total n observations = 134. Source data

Extended Data Fig. 4. Measure of standardized success in coral restoration actions (S_r).

S_r is computed as the deviation of the observed survival (S_o) from expected survival (S_e) (after a given number of post restoration months) as $S_r={\log }_e\left(1+100-\frac{(S_{\mathrm{e}}-S_{\mathrm{o}})}{S_{\mathrm{e}}}100\right)$. S_e is computed based on an empirical curve of coral survival % versus time obtained from literature. Source data

Extended Data Fig. 5. Mean number of sites experiencing a maximum yearly degree heating weeks ≥ 20 per decade.

The plot shows the mean number of sites with predicted maximum yearly DHW ≥ 20 globally within a moving time window of 10 years. We mapped reef localities at a resolution of 0.5° × 0.5° latitude/longitude (n = 3780). The moving time window started at year 2015, so the first reported year is 2025. Predictions are based on an intermediate CMIP6 climate projection (SSP2-4.5, ‘middle of the road’; see Methods for details), with DHW obtained from ref. . Source data