Modeling the response of tropical coral reefs to back-to-back coral bleaching events based on 20 years of ocean data from the Pacific Islands Fisheries Science Center
Researchers analysed 20 years of reef data from a 200-kilometre stretch of coastline on Hawai’i Island. The data included an unprecedented marine heatwave in 2015, providing the researchers with a unique opportunity to explore how factors such as wastewater runoff and fishing intensity shaped the coral reef’s response. The heatwave was the strongest in the last 120 years and saw ocean temperatures jump to 2.2 C above normal, peaking at 29.4 C.
In the 12 years leading up to the heatwave, reefs with more fish — particularly herbivorous scrapers — and lower levels of wastewater pollution experienced coral growth.
Scrapers “literally scrape the reef”, says Jamison Gove, an oceanographer at the Pacific Islands Fisheries Science Center at the US National Oceanic and Atmospheric Administration in Honolulu, Hawaii. He says that they clear space to allow for the settlement of coralline algae, which is a step in the process of coral growth.
Wastewater pollution from sewage disposal systems and pollution from urban environments can damage coral health. He said that the soup of bacteria and other things can cause coral disease. “You also have whatever everybody’s flushing down or putting down their sink. You have household chemicals, you have pharmaceuticals, you have all sorts of toxins.
Hughes gives the example of back-to-back bleaching events that struck the Great Barrier Reef in 2016 and 2017, and says that climate modelling suggests that most coral reefs will experience bleaching events every year by 2050 under a business-as-usual emissions scenario.
The same process was used to remove outliers in our predictor variables. We then square-root transformed the following predictors to down-weight the influence of values at the extreme ends of their distributions: total fish biomass, wastewater pollution, sediment input and nutrient loading. Pearson’s correlation coefficients were calculated among all predictors (Supplementary Fig. 8), removing highly correlated (r > 0.7) predictors. Human population density was excluded from the model-fitting process due to the reasons outlined in the GAMM analysis. This resulted in the following predictors included in the models (correlated predictors in parentheses were removed): total fish biomass, biomass of scrapers, biomass of grazers (total herbivore biomass), wastewater pollution, nutrient input, sediment input, urban runoff (phytoplankton biomass), wave exposure, fishing gear restrictions and depth. The logic behind the decision which correlated predictors to retain was the same as the logic behind our analysis. The mean and variability were not taken into account because there was a small range of values among the reefs. All possible candidate models were computed while limiting the total number of predictors in any given candidate model to four (to reduce overfitting and to account for the lower response variable replication compared to our GAMM analysis). The models were computed using a function in a computer program. The models that are used in the comparison are in the Extended Data Table 3. For the highest ranked models, the pseudo-R2 ranged from 0.21 to 0.22. This is an excellent fit due to the pseudo-R2 values of more than 0.2. Models within ΔAICc ≤ 2 of model 1 in Extended Data Table 3 demonstrated comparable levels of goodness of fit and parsimony99,100. Many of the parameter coefficients within these models were sensitive to the underlying variability in the data and their estimates did not differ significantly from zero (P < 0.05). The top model contained parameters that were different from the zero ones. We looked at the change in the probability of a given reef being high, moderate or low. Probability curves for high, moderate and low were calculated on the basis of changing scraper biomass and wastewater pollution and holding all other predictors at their mean.
There are various resource management scenarios presented. The rationale that led to 4b being selected was the same that led to them being chosen. The management target for scrapers was 250 ha1, which is the same value as the long-term mean for scrapers in Kealakekua Bay. Kealakekua Bay is also exposed to numerous land-based stressors, including high levels of wastewater pollution (258,000 l h−1 in 2019). The value of 250 kilo ha1 is an estimate of scraper abundance on a reef with strong fishing protection but with land-based stressors present. In addition, we compared our upper (250 kg ha−1) and lower (30 kg ha−1) scraper biomass values to the distribution of scraper biomass among all reefs (n = 80) in 2019, the most recent time point in which all reefs were surveyed within the same year (Supplementary Fig. 10). The upper and lower limits represent the 92nd and 36th percentiles, respectively. For wastewater pollution, we used our 2019, 100 m grid cell values that fell along the 10 m isobath (same as Fig. 1d) The latitudinal extent was constrained so that it was consistent with the north and south of the survey. This approach gave us much better replication and an assessment of the pollution along the coastline, which we can use for our management scenarios. The 95th and 36th percentiles of the distribution were represented by the upper (600,000 l ha 1) and Lower (2,500 l h 1) values for wastewater pollution.
The difference in local human impacts and environmental factors between positive and negative trajectory reefs were then calculated as the difference in the mean drop-one jackknife values for each impact or factor88. Upper and lower bars in Fig. 2d represent the respective maximum and minimum differences in drop-one jackknife values between positive and negative trajectory reefs. Before calculating the drop-one jackknife values, we identified and removed outliers that fell outside a threshold of ±2 standard deviations of the median. We formally tested for a difference in the local conditions of positive versus negative trajectory reefs using a multivariate permutational analysis of variance (PERMANOVA)89 based on a Euclidean distance similarity matrix, type III (partial) sums-of-squares and unrestricted permutations of the normalized data. We visualized the results in Fig. 2c using a constrained analysis of principal coordinates90 and calculated the cross-validation allocation success (a measure of group distinctness) from the leave-one-out procedure of the constrained analysis of principal coordinates analysis.
The maximum value of the 7-day moving window averaged for each SSTpixel was used to calculate the mean and variability in summertime sea surface temperature. The mean regional temperature was averaged out across the coastal areas within the study region. coral reef scientists use a widely used metric to assess heat stress on their corals during a marine heatwave. All data was available daily at 5 km resolution35.
Satellite derived chlorophyll-a was used, along with irradiance from two sources. The long-term mean (2002–2013) in 8-day, 4 km data were obtained from ref. 80 And shown in a picture. The data was 2d and extended. 3. All subsequent analysis used the visible-infrared imaging/radiometer suite, which has high spatial (750 m) and temporal (daily) resolution data starting in 2014 (provided by NOAA’s Coral Reef Watch). All data was masked for cloud cover and shallow waters after ref. 83 (Supplementary Fig. 27).
We created a categorical value using the regulation information and marine managed area boundary designations. 80. The rules for fishing for reef finfish species over time were evaluated to see if they should be changed. There are 6 ranked fishing gear categories, including full no-take, no lay net, spear or aquarium, no lay net or aquarium, and open to all gear types.
The Integrated Valuation of Ecosystem Services and Tradeoffs sediment delivery model was used to derive long-term annual average sediment input (kg ha−1) reaching the coast77,78,79,80 at 100 m resolution. Peak rainfall data is used to calculate the long-term annual average stream flow by Watershed. Following ref, discharge by Watershed was calculated. 81. Sediment load was assumed to scale with discharge according to a approximate ratings curve following ref. 82 (Supplementary Figs. 23 and 24).
We were able to calculate the annual rainfall and peak total at 100 m. Daily rainfall data were generated following refs. 75,76. Rainfall from each rain station was used to derive interpolated surfaces at annual time steps using Empirical Bayesian Kriging in ArcGIS. Subwatershed catchment data74 were clipped to 0–10 km from the coast and used to calculate rainfall per drainage area (Supplementary Figs. 21 and 22).
We quantified the total area of impervious surfaces (that is, paved roads, parking lots, sidewalks and roofs) within 10 km of the coastline at 100 m resolution for each year from 2000 to 2017 (Supplementary Figs. 19 and 20). Data was derived from land-use data from 1992 to 2010. We also digitized 2017 impervious surface cover from a single cloud-free Landsat 8 image (courtesy of the United States Geological Survey, USGS) (15 m resolution pan-sharpened). Linear interpolation filled in years between data availability.
We calculated wastewater effluent (l ha−1 yr−1) and nitrogen input (kg ha−1 yr−1) from onsite sewage disposal systems (for example, cesspools and septic tanks) and injection wells (collectively OSDS) in coastal waters at 100 m resolution. Only OSDS located within a modelled one-year groundwater travel time of the coast were included in the analysis and nutrients from OSDS were assumed to flow to the nearest point on the shoreline. Wastewater effluent and nutrient input were estimated on the basis of ref. 67 and discharge rates and nutrient loading according to ref. 67. dispersal offshore is estimated using a Gaussian decay function to be at zero at 2 km. This same dispersal function was also used for nutrient input, urban runoff, sediment input and rainfall, which are each described below.
We quantified human population density using NASA Gridded Population of the World v.4 (ref. 66). At a 5 year intervals, the data is at 1 km resolution. Linear interpolation was used to fill in the missing years and produce annual time steps of human population within 15 km of each 100 m grid cell across our study region (Supplementary Fig. 12).
The Hawaiian Islands’ Seashore: A Reseditive System with a Large Ocean-Scale Refining Reef Density
The methods for calculating the fish population were established. The allometric length–weight conversion, where parameters a and b are species-specific constants, was used to estimate the food supply of individual fishes. Length–weight fitting parameters were obtained from a comprehensive assessment of Hawai‘i specific parameters56 and FishBase65. Fish species were excluded from fish biomass calculations according to life history characteristics that are not well captured with visual surveys, including cryptic benthic species, nocturnal species, pelagic schooling species and manta rays.
The most populous island of the Hawaiian Islands is Hawai ‘‘i Island, which is located in the northern central Pacific. The majority of the western section’s coastline is to the north and south. The main Hawaiian Islands’ coastline contains the longest contiguous reef system in the world and has a large range in human population, land–sea impacts, and environmental factors that are similar to reef systems around the world. The region represents an ideal study location for resolving the land–sea human impacts driving reef ecosystem change and coral trajectories following acute climate-driven disturbance.