Environmental data, zooplankton abundance, and community structure
With the aim of evaluating the correlations between environmental parameters and the structure of the zooplankton community, revealing the spatial structure within each cruise, and revealing seasonal and inter-annual environmental changes, we analyzed the relationships between the 8 environmental predictors and the zooplankton distribution. Those predictors included the average (0-200 m) environmental conditions (i.e., temperature, oxygen, salinity, fluorescence, and density) as well as geographic and bathymetric predictors (e.g., latitude, longitude, and depth). Since the cruises reflect late spring (XIXIMI-05) and summer (XIXIMI-04 and XIXIMI-06) conditions, season was used as a constraining factor in subsequent analyses.
The effect of environmental parameters was tested using the proxy of abundance for each taxa at the family level. A distance-based linear modeling (DistLM) analysis was applied with a multivariate multiple regression analysis using Primer 6+P (K. R. Clarke & Warwick, 2001). We conducted these analysis to estimate the independent ordination of all predictors (Marginal test), which determines the proportion of zooplankton variance explained by each environmental variable independently, and to obtain an optimal ordination model (Sequential Test), in which the model partitions the variation in the data based on a multiple regression model selected by the user (e.g., forward, stepwise, or best fit; K. R. Clarke & Warwick, 2001). The latter outcome may be considered to be the best statistical combination of all abiotic predictors. In this study, we implemented the Akaike Information Corrected Criterion (AICc) and ‘stepwise’ options for model selection. We selected this approach because stepwise multiple regression adds or subtracts predictor variables from a model until most of the variation is explained; variables are excluded if they behave like random variables in terms of the additional variation explained. A significance evaluation of the multidimensional model was conducted with a permutation method with 9999 permutations (Usov, Khaitov, Smirnov, & Sukhotin, 2019) implemented in Primer 6+P (K. R. Clarke & Warwick, 2001).
A distance-based redundancy analysis (dbRDA) was performed using Primer 6+P (K. R. Clarke & Warwick, 2001) to visualize the relative importance of all predictor variables (McArdle & Anderson, 2001). The potential relationships were evaluated using normalized environmental data and the fourth root transformed read abundance of zooplankton at the family level. The fourth root transformation was deemed to be the most appropriate transformation since it reduces the weight of highly abundant taxa and facilitates comparisons among different datasets (Howson, Buchanan, & Nickels, 2017; Vause et al., 2019). The resultant data were converted to a resemblance matrix using the Bray-Curtis similarity index.
According to the multivariate multiple regression results (reported below), stations were categorized into discrete categories according to their environmental profiles. These categories were: low/high oxygen concentration, warm/cold temperatures, and east/west longitude. With regard to the latter, east stations were located at longitudes < 86.30 °W, while the remaining stations (> 86.30 °W) were attributed to the west group. Likewise, stations with lower/higher values than the average mean values for oxygen and temperature (specific average threshold values of oxygen and temperature were calculated according to the stations in each analysis) were grouped as “Low/High_O2” and “Cold/Warm”, respectively. These oceanographic variables (averaged for the first 200 m of the water column) were visualized in contour plots generated with Ocean Data View software (Schlitzer, 2018). This approach was used to compare spatial (within each cruise) and temporal patterns (among spring and summer cruises) and to test for similarities/differences in zooplankton community structure. In addition, all data for the three cruises were also included in a comprehensive analysis.
Potential spatial and seasonal segregation of the environmental parameters after normalizing the data was tested using a Cluster/SIMPROF analysis in Primer 6+P software. Clustering was carried out with a Euclidean distance matrix using the group average method. Statistical differences among groups of stations were evaluated by a PERMANOVA with 4999 permutations using Primer 6+P software, while comparisons of the average abiotic parameters were conducted with ANOVAs in Statistica software.