Locating seamounts: seamount databases
Two seamount databases were used in this study: the validated Pacific database published by Allain et al. in 2008 (referred to in the text as the “Allain database”) and the most up-to-date global database published by Yesson et al. in 2011 (referred to in the text as the “Yesson database”). The primary analyses were conducted on a representative subsample of the Allain seamount database, the most spatially expansive (45°S–32°N and 130°E–120°W), validated and crosschecked published seamount database16. This area covers a large swath of the Pacific, which contains the vast majority of seamount features on our planet. Only “validated” seamounts, whose location and associated data were confirmed by at least one ship-based dataset rather than purely derived from satellite estimations, were used in the analyses. This subset was further reduced to include only features with validated summit depths deeper than 30 m (the optically shallow cutoff used after Gove et al. 2016) and elevations greater than 1,000 m (to follow the classic definition of a seamount as a feature rising more than 1,000 m above the seafloor). The resulting dataset was then subsampled to meet computational restrictions on database size. All seamounts with summit depths shallower than or equal to 300 m (48) were included, and the remaining features were subsampled such that 5 features were selected from each 100 m height bin and 1,500 m elevation bin for a total of 196 seamounts (of 485).
Second, to examine patterns globally, a subsample of the unvalidated Yesson database was analyzed with identical methodology. This database was based on the same global bathymetry used in this paper to derive underlying water depths for each chlorophyll pixel1. Only seamounts with estimated summit depths deeper than 30 m, elevations greater than 1,000 m, and estimated base areas greater than 500 km2 were selected from because the smallest features have the largest position and depth errors associated with them. Eight features were randomly sampled for each 150 m summit depth bin (ranging from − 30 to − 1,050 m) and from each 1,000 m elevation bin. These cutoffs were selected to create a subset of comparable size to the Allain subset and to maximize the chances of selecting from “real” features (those accurately detected via satellite and the Yesson seamount algorithm)1, 2. Because this database is unvalidated, these added precautions were taken in subset selection. The resulting subset (192 of 2,560) was then examined visually, summit depth estimates were corrected where needed, and features which were mistakenly identified as seamounts were removed from the subset. Despite our subsetting process, the manual revision still revealed problems with the published database, especially in estimated summit depth and location; therefore, approximately 19% of the initially selected seamounts had to be excluded from the final global analysis (final included number of seamounts = 166).
Quantifying chlorophyll-a enhancements around seamounts
Chlorophyll-a (mg/m3) data were derived from the August 2015 version of the level 3 monthly composite, scientific quality, 0.0417° squared (~ 4 km) Moderate Resolution Imaging Spectroradiometer (MODIS) data (https://oceancolor.gsfc.nasa.gov/). Data were accessed through the NOAA ERDDAP, griddap site (https://coastwatch.pfeg.noaa.gov/erddap/griddap/erdMH1chlamday.html). A decade’s worth of chlorophyll data (Jan 2006-Jan 2016) were analyzed around each feature for a seamount-centered square with 100 km sides. Though seamounts whose validated summit depths were shallower than 30 m were excluded from the dataset entirely, an additional 30 m pixel depth (data source described below) cutoff was applied to all chlorophyll data to avoid potential bias from optically shallow waters anywhere in the sampling area, following the methods of Gove et al.19. Additionally, to avoid confusing the island mass effect (IME) with SICE, all seamounts whose sample area included one or more pixels with satellite estimated depths were emergent (≥ 0) were labeled “Emergent”. For all reported analyses these features flagged as ‘emergent’ (N = 19) were removed before statistical anlysis. All analyses included temporal predictors to account for seasonality (month predictor) and annual variability (year predictor) in chlorophyll patterns.
Sea surface temperature
To test for the occurrence of seamount uplifted water, monthly daytime SSTs on the same ~ 4 km resolution from the Aqua MODIS platform were also downloaded for each 100 km sided seamount box (https://coastwatch.pfeg.noaa.gov/infog/MH1_sstMask_las.html). This data is science quality data from the August 2015 reprocessing of the global Level 3, 11 km SST data.
Seamount locations (summit latitude, summit degrees poleward or absolute latitude, summit longitude) and seamount specific information (elevation above the surrounding seafloor and summit depth below sea level) were derived from the published seamount databases described above2,16. Seasonality and annual variability were also included in the model through the incorporation of month and year terms. Each of the predictors was included for their theoretical influence on primary producers around seamounts. Summit location (i.e. latitude, longitude, and degrees poleward—defined as the absolute value of latitude) can influence internal wave dynamics13, mixed layer depth34, and global productivity dynamics including light versus nutrient limitation on production 38. Whether a seamount enhances production may well depend upon the background or long-term average productivity of the area, and this may co-vary with latitude and average SST (oligotrophic gyres are warm) at the summit. Average euphotic layer depth may influence the depth that physical seamount effects would need to reach in order to influence phytoplankton production. Finally, seamount summit depth greatly influences circulation patterns at the feature13 and thus possibly nutrient injection into the euphotic zone. However, seamounts often have complex geomorphologies, and therefore a variety of measures of summit depth were included: the shallowest depth at summit, proportion of pixels with depths shallower than the average euphotic layer depth, and proportion of pixels shallower than 800 m.
Depth data were derived from the Shuttle Radar Topography Mission (SRTM30 PLUS) 30 arc-second global bathymetry grid, which combines high resolution (~ 1 km) ship-based bathymetry data with ~ 9 km satellite-gravity data39 (https://topex.ucsd.edu/WWW_html/srtm30_plus.html). For each selected seamount, bathymetry and chlorophyll data were analyzed from a square region centered on the given summit location measuring 100 km2. Previous research suggested that the island mass effect (IME) extends approximately 30 km from the shore of islands19, and that seamount effects can extend up to 40 km from the summit location18, therefore, a box extending 50 km from the seamount summit was selected in order to ensure that the entire feature and both seamount-influenced waters and the surrounding unmodified open ocean waters were included in the analyses. Depth was extracted for each chlorophyll pixel using the extrapolation methods in the NOAA marmap package (getdepth function)40. In addition, because summit depth uses data from only the single shallowest point on a complex feature, two further depth-based predictors were derived: the proportion of chlorophyll pixels with depths shallower than 800 m (an estimate for the daytime maximum depth of vertical migration) and proportion of pixels shallower than the average euphotic depth at the seamount summit location.
Monthly composite 4 km resolution euphotic depth (in meters) calculated from the Lee algorithm was obtained from the NASA ocean color data product Zeu (e.g.: A200600A20060012006031.L3m_MO_ZLEE_Zeu_lee_4km.nc). The data were downloaded for the same period (2006–2016) as the chlorophyll data for each pixel around each selected seamount feature. The proportion of pixels in the sample region shallower than or equaling the overall average euphotic layer depth was calculated for each seamount.
Decadal average sea surface temperature (SST) at the summit locations were derived from available monthly mean ARGO SST data for each seamount (https://apdrc.soest.hawaii.edu/dods/public_data/Argo_Products/monthly_mean). These are therefore in-situ measured temperatures. Only data from the shallowest depth bin were used to derive these long-term average SSTs.
Statistical models and model selection
All statistical analyses were conducted using the software package R. To identify seamounts characterized by SICE, defined as a statistically significant increase of chlorophyll with shallowing depths, we fit a Gaussian GAM for each seamount in each dataset analyzed. These models use the natural log of chlorophyll as the response and include a spatial predictor (two-dimensional relative latitude and longitude smoother), and a temporal predictor (month) to account for spatial and temporal autocorrelation respectively. Because phytoplankton are naturally patchy throughout the ocean, we included a two-dimensional spatial smoother to detect and account for this natural spatial structure. This approach made it possible to distinguish between depth related chlorophyll enhancements and random patchiness. An alternative approach might be to randomly select a control region away from the seamount for comparison. However, chl-a enhancements are likely to be asymmetrical and background levels are inherently patchy19,41,42,43. Our approach implicitly controls for such patchiness by testing for increases in chl-a with shallowing depth in a seamount-centered region that spreads well beyond the radius of any measured seamount effect, creating a control region that forms a ring around the region of interest instead of a single offset control region whose different position within the larger latitudinal and longitudinal spatial gradients in chlorophyll concentrations could skew the analysis18,19. Gove et al. (2016) took a very similar approach to their analysis of the island mass effect. These GAMs also fit a slope for each seamount between chlorophyll and depth using the decade of chlorophyll data for each corresponding sample area (see Supplementary Information 1 Table 1 for all full model formulas). The seamounts for which the resulting chlorophyll/depth estimate (seamount-specific slopes) were significantly positive (P < 0.05), were identified as SICE seamounts. Identical statistical techniques were applied to the SST analysis such that a GAM was used to describe seamount specific SST-depth slopes. In addition, correlations between SST and depth, SST and chlorophyll, and significantly negative SST slopes and SICE were estimated via Spearman correlations which assess the strength and direction of a monotonic relationship between two variables.
Finally, to test which predictors influence the likelihood of finding a significant seamount-induced chlorophyll enhancement as defined above, a binomial (or logit) GLM was fit. This model uses the presence/absence of SICE (determined by the results of the previously described GAM) as the binomial response variable and all available seamount-specific terms as predictors. These predictors are listed and fully described in Table S1: degrees poleward, summit latitude, summit longitude, summit depth, summit elevation, long-term average SST at the summit, decadal average chlorophyll, the standard deviation of the average chlorophyll, average euphotic depth, proportion of chlorophyll pixels with depths shallower than 800 m and shallower than the average euphotic depth, and a categorical variable identifying whether or not there is an emergent feature within the seamount sampling area (Table S1). Collinearity amongst the predictors was assessed via the variance inflation factor (VIF via the car package)44. Following the method by Zuur et al. 2010, collinearity was reduced using an iterative procedure whereby the predictor with the largest VIF was removed until all VIF estimates were below three45. Models of all possible combinations of the remaining predictors were then compared and ranked using the corrected Akaike’s information criterion (AICc in the MuMIn packag)46. The ‘best’, equivalent models, (those with AICc scores within 2 of the lowest AICc model), were evaluated and the frequency of significance of each predictor was tallied. Those predictors, which were significant across all ‘best’ models, were reported (Supplementary Information 1 Tables 1 and 2). Equivalent methods were also applied to the subset of seamounts from the global seamount database (N = 166).
The magnitudes of SICE were evaluated using similar techniques to those described previously for the logit model. However, the response was the magnitudes of the significant regression slopes. Models had low explanatory powers, non-uniform residuals for all applicable error distribution families, and no consistently superior model performance relative to null model. Therefore, instead of a statistical analysis of the magnitudes, those seamounts with the most positive slopes (the most extreme cases of SICE) were investigated in detail for the percent increase of chlorophyll over the seamount relative to surrounding surface waters over at least 4,000 m depth, over the entire 10-year period, as well as by year and by month. This gave a maximum enhancement magnitude and an evaluation of the seasonality and persistence of seamount-induced chlorophyll enhancements.
Seamount fisheries catch analyses
To evaluate possible bottom up subsidies from SICE, historical seamount fisheries yields between the years of 1950 and 2015 were compared between those seamounts with significant chlorophyll enhancements and those without using data from Watson et al. 2018, a compilation of 15 different fisheries databases20,47. This database provides taxon-specific catches (ranging from class to species in taxonomic certainty) in tonnes mapped into half degree resolution cells by year, fishing country, and gear type broken into reported, illegal/unreported, and discards for both industrial and non-industrial fishing20,47. Only reported catch was analyzed here, but both industrial and non-industrial catch were included. Total reported catch was extracted for all fishing cells that overlapped partially or totally with any of the 177 seamount areas in the Allain database. Catch was prorated by the fraction of overlap between each fishing cell and the seamount area in question (e.g. if only 50% of the spatial area of a given cell overlapped with a given seamount area, catch from that fishing cell was multiplied by 0.5). Historical total reported fisheries catch (tonnes) was then calculated for each of the 177 seamounts (only those with no emergent pixels within their area) and compared between SICE and non-SICE seamounts using a generalized linear model with a Gamma distribution. In addition, maximum annual catch was calculated for each seamount because seamount fisheries are often characterized by boom-bust cycles, where fishing once commencing quickly ramps up, overexploits and decimates vulnerable stocks, and soon after returns to low levels as fishers move on to another fishing ground7,9,48. We define maximum annual catch as the total reported catch for a given seamount for its best fishing year on record. Again, maximum annual catch was compared between SICE and non-SICE seamounts using a generalized linear model with a Gamma distribution. All full model equations and model results are given in Supplementary Information 1 Table 3. Normality of residuals was evaluated visually using quantile–quantile plots, and residual variance was checked for similarity across groups. To further explore differences in catch between SICE and non-SICE seamounts, we also compared family-level catch between the two seamount categories by placing each taxa (309 unique taxa) in the database into its appropriate family and summing catch by family (see Supplementary Information 1 Tables 4 and 5 for a breakdown by SICE/non-SICE for families making up more than 1% of the catch).
Source: Ecology - nature.com