a, Location of the Commonwealth of the Northern Mariana Islands (CNMI). b, Location of Maug. c, The three main islands of Maug, with 100 m isobaths and the location of both the vent and control sites. d, Detail of the vent with the high-pCO2 and mid-pCO2 study sites, together with 2 m isobaths.
Study site.
This study was conducted at Maug Island (20° 1′ N, 145° 13 E), in the northernmost region of the CNMI (Fig. 1). Initial investigation of the pH/CO2 gradient was conducted with a pH probe (ROSS Ultra pH, Orion) and non-dispersive infrared CO2 analyser (LI-820, LI-COR Biosciences) paired with a Global Positioning System (GPS). These data were used to inform subsequent chemical, environmental and biological sampling. For the purposes of this study, three sites were established along a gradient of vent influence. A high-pCO2 site was located along the vent field/reef margin. An intermediate, mid-pCO2 site was located roughly 50 m south of the vent, in an area dominated by reef framework and coral. Finally, an unaffected control site was located on the southern end of the island, roughly 1 km south of the research site. All were located at approximately nine metres depth, to control the influence of extraneous sources of variance during comparison.
Environmental data.
To characterize the extent of carbonate chemistry alteration, 33 discrete water samples were collected in a grid pattern over the area influenced by the vent, covering both the high-pCO2 and mid-pCO2 sites. Water was collected from 20 cm below the water’s surface using borosilicate glass bottles, which were immediately fixed with HgCl2 and sealed. Temperature and salinity were recorded at the same depth using a handheld meter (EC300A, YSI) and sites were marked with a handheld GPS (GPSMAP 78S, Garmin). Water samples were collected in the same manner at the control site.
Samples were transported to NOAA’s Atlantic Oceanographic and Meteorological Laboratories (AOML), where they were analysed for dissolved inorganic carbon (DIC) and total alkalinity (TA) using autotitrators (AS-C3 and AS-ALK2 respectively, Apollo SciTech). The carbonic acid system was solved using CO2SYS (ref. 36) with the dissociation constants of ref. 37 as refitted by ref. 38 and ref. 39 for boric acid. Carbonate chemistry parameters were plotted over the extent of the vent using ArcGIS (ESRI). An interpolated raster map was created from these points using the Spatial Analyst Toolbox and the inverse distance weighted (IDW) technique.
SeaFET pH loggers were deployed and recorded data every half hour at each of the three sites (control, mid-pCO2, high-pCO2). Data were collected from 19 May to 10 August 2014. Shorter-term diel oscillation in carbonate chemistry was investigated using discrete water samples collected every 6 h over a 48 h period from 11 August to 13 August 2014. Water was collected at each of the three study sites immediately above the benthos using a Niskin bottle, and then immediately transferred to borosilicate bottles while minimizing bubble formation and gas exchange. Samples were analysed with the same methodology used for spatial characterization.
Temperature loggers (HOBO Water Temp Pro v2, Onset) were deployed over the same period as the SeaFETs and were attached to stable platform bases approximately 10 cm above the benthos at the control, mid-pCO2 and high-pCO2 sites.
Light loggers (ECO-PAR, Wet Labs) were placed at each of the three sites and were programmed to record photosynthetically active radiation (PAR, 400–700 nm) every 30 min from 19 May to 9 August 2014. The instrument at the high-pCO2 site failed immediately on deployment. The mid-pCO2 and control site instruments were subsequently redeployed at the high-pCO2 and control sites, collecting every 10 min from 10–13 August, to measure relative PAR levels. ECO-PAR instruments contain wipers that clean the sensor after each reading, and no drift was observed over the deployment period. We report daily PAR dose following ref. 40, where mean PAR over the period 10 am to 3 pm is multiplied by the total time of that period (5 h).
Two acoustic Doppler current profilers (ADCPs, Nortek Aquadopp) were deployed at the high-pCO2 and control sites to measure current. The upward facing devices were turned over during a storm and stopped recording useable data on 4 July.
Vent gas was collected underwater using a conical collection cup connected to gas impermeable 1 l Tedlar sampling bags. Sealed bags were transported to Miami and subsequently analysed using gas chromatography (Varian CP3800 and HP 5890).
Biological data.
Changes in benthic cover were investigated using photo quadrats and was conducted across two spatial scales: large-scale differences between the three instrumented high-pCO2, medium-pCO2 and control sites, and fine-scale community shifts occurring outside the zone of active bubbling, expressed as a function of proximity to the high-pCO2 site. For quantification of benthic cover among sites, high-resolution photomosaics were constructed following ref. 41. Mosaics were subsequently subsampled into 100 images per site and the benthic cover under 30 randomly located points were identified using the CPCe software package42. To examine changes in benthic cover with increasing distance from the area of active bubbling, East–West oriented transects, perpendicular to the CO2 gradient, were placed at increasing distance from the vent. Photos were taken every 2 m along the 20-metre transect, and were subsequently analysed using CPCe (ref. 42), whereby 40 random points were overlaid over each image and identified.
Finer-level taxonomic identification of coral and algae can be difficult from photographs and community richness data were collected in situ using SCUBA. As with benthic cover, analysis was conducted across both large and small spatial scales. Immediately outside of the vent, six 15 m transects were placed, starting at ~5 m depth, and arranged perpendicular to the shore (East–West). Transects were spaced 10–20 m apart, incrementally further away from the vent site (North–South). Five 0.25 m2 quadrats were placed haphazardly along each transect. All algae species within each quadrat were identified to the lowest reliable taxonomic level. Analysis of algae richness was conducted on all identified algae taxa (excluding turf algae), as well as on calcifying algae species, as shown in Supplementary Table 8. Turf algae is defined as the low-lying (<2 cm) community of small and juvenile algae species that are not taxonomically distinguishable in situ. Five additional 0.25 m2 quadrats were used to quantify coral richness, as listed in Supplementary Table 7. Species-specific prevalences at each site were calculated as the proportion of richness quadrats containing each species, and 95% confidence intervals were calculated following ref. 43. For larger-scale site comparisons of community richness, near-vent data were grouped and compared against three additional transects placed at the control site.
Cores (5 cm diam. × 10 cm length) were taken from colonies of massive Porites sp. in close proximity to the instrumented mosaic sites using a pneumatic drill and SCUBA tank rig. Cores were slabbed parallel to the growth axis and scanned using microCT (Skyscan 1174, Bruker). Density was plotted versus distance and Coral XDS+ (ref. 44) was used to delineate yearly banding (peak–peak method), as well as to calculate extension, density and calcification rate.
Statistical analysis.
Light was analysed using a t-test (2-tailed). Current and pH data were analysed using nonparametric Mann–Whitney and Kruskal–Wallis tests, respectively. Percentage cover data were arcsine-transformed45 and analysed using general linear models (GLMs). Species richness data were loge-transformed and were analysed using t-tests (2-tailed). Transformation was unnecessary for coral core and calcification data, which were presented by year. Sample-specific averages over a five-year period (2009–2013) were compared between sites using GLMs. Post hoc pair-wise comparisons were conducted with Tukey’s tests.
To investigate the effects of vent proximity on benthic community composition at the vent site, linear (y = b0 + b1x), parabolic (y = b0 + b1x + b2x2), asymptotic (y = b0 + b1x−1) and Ricker models (b0xe−b1x) were fitted to percentage coral cover data. Linear, parabolic and asymptotic models were fitted to coral, algae and calcifying algae community richness data. Goodness of fit was evaluated on statistical significance (p < 0.05), R2, and Akaike’s information criterion (AIC). Statistical analysis was conducted with the SPSS and GraphPad Prism software packages46, 47.