Ecosystem-based management places a strong emphasis on habitat, but little work has been done to examine how water column properties may influence the distribution, abundances and structure of groundfish assemblages. We identified and described oceanographic habitats in the northern California Current based on temperature, salinity, chlorophyll-a and the inherent variability in these factors. We then examined the distribution and the abundance of groundfishes in relation to these oceanographic habitats and conditions with the long-term goal of improving science for ecosystem-based management of the groundfish fishery of the west coast of the USA. Five summertime oceanographic habitats with distinct physical and biological characteristics were identified in the northeast Pacific Ocean off the northwest Coast of the USA: Offshore Habitat, Upwelling Habitat, Highly Variable Upwelling Habitat, River Plume Habitat, and Highly Variable Habitat. Overall, the species composition differed among the five oceanographic habitats, with certain groundfish species being highly indicative of some habitats; however, the majority of the associations were weak due to overlap of species distributions in the nearshore oceanographic habitats. In contrast, groundfish species showed strong associations with individual oceanographic factors, primarily depth, surface chlorophyll-a, and bottom salinity and temperature. In addition, latitudinal variations in upwelling intensity, river discharge and productivity led to the identification of three regions where high chlorophyll-a concentrations were associated with large abundances of specific groundfish species. The combined oceanographic datasets and data products that we produced have the potential to be a powerful tool for improving our knowledge of the west coast ecosystem.
This study is part of an ongoing strategy to integrate the oceanographic knowledge obtained off the west coast of the USA into the science and management of the groundfish fishery. To date, oceanographic data and information on habitats have not been used to their full potential to examine the distributions and structure of groundfish assemblages due to the difficulty of obtaining relevant datasets commensurate with fisheries data. Consequently, groundfish stocks are managed with incomplete knowledge about their oceanographic habitats, which limits use of an ecosystem-based approach. An ecosystem-based management (EBM) approach creates a framework in which the knowledge of oceanographic processes and climate variability is seen as an essential part of understanding fish populations, their habitats and thus their management.
The ocean circulation off the Washington, Oregon and northern California coasts covers the northern part of the California Current Large Marine Ecosystem, which is one of five eastern boundary current upwelling regions in the world, supporting a major portion of the world’s fisheries (Pauly and Christensen, 1995). The main ocean circulation is driven by seasonal winds creating upwelling conditions during the summer months and downwelling conditions during the winter months (Huyer, 1983). This seasonal circulation is under the influence of two long-term atmospheric processes: El Niño–Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO). Additionally, there is good understanding of the influence of local fresh water inputs (the most important are Fraser River outflow through the Strait of Juan de Fuca, and the Columbia River), local wind forcing and local topographic features such as submarine canyons, banks, headlands and capes on the local water column properties and circulation (Hickey and Banas, 2003; Castelao and Barth, 2005; and Huyer et al., 2005). These complex interactions among ocean processes, acting at different temporal and spatial scales, make the study area an ideal place to investigate how regional oceanographic processes influence groundfish species.
We examined how water column properties are associated with the distributions and abundances of west coast groundfishes. We first assembled and merged dispersed and disparate oceanographic datasets for temperature, salinity and chlorophyll-a in the northeast Pacific Ocean off the northwest coast of the USA from the 1930s to the year 2004. We then generated relevant oceanographic data products for fisheries research, consisting of climatological means, standard deviations and coefficients of variation for each ocean variable at different depths. Finally, we used these data to: (1) examine spatial distributions and abundances of groundfishes for the year 2004; (2) identify cold-regime oceanographic habitats in the water column; and (3) investigate whether the abundance and distribution of groundfish species is associated with a specific oceanographic habitat or individual oceanographic variables, using a combination of univariate, classification and ordination techniques.
The oceanographic parameters selected for analysis were temperature, salinity and chlorophyll-a concentration. These parameters were selected for their relevance to the ecology of groundfishes and their potential as descriptors of oceanographic habitats. Accessible data (Table 1) for each oceanographic variable were assembled from a variety of sources including remotely sensed data from September 1997 to August 2003 and in situ data from the 1930s to 2004 within the study region. The study region covered the northern California Current System (CCS) from the Strait of Juan de Fuca in northern Washington (49°N) to northern California (41°N) and extended from the coastline to 127°W (Fig. 1). After all the in situ data were assembled, a chlorophyll-a dataset (where chlorophyll-a measurements from Niskin bottles and fluorometers were merged) and a conductivity-temperature-depth (CTD) dataset (where all the CTD casts were merged) were created. These datasets, together with chlorophyll-a satellite data from SeaWiFS, were used to compute climatological monthly means, standard deviations (STD) and coefficients of variations (CV) for each variable at different depths to characterize the oceanographic conditions in the northeast Pacific Ocean off the northwest coast of the USA. For a complete description of the assembled datasets and the methodology used in the computation of the climatologies, refer to Juan-Jordá (2006). The historical datasets and the climatologies are available at http://pacoos.coas.oregonstate.edu.
Table 1. Oceanographic data sources.
Temperature and salinity
National Oceanographic Data Center (NODC)
From 1934 to 1997
Not gridded Random sampling
Northeast Pacific GLOBEC Program
From1997 to 2004
Not gridded CTD transects off Oregon
Summer 2001 and Winter 2003
Not gridded CTD transects off Oregon
Not gridded CTD transects off Washington and Oregon
Newport Hydrographic Line, (HL) courtesy of W. Peterson, Oregon State University
Not gridded CTD transects along the Newport HL
Fluorometer and Niskin Bottles
National Oceanographic Data Center (NODC)
Niskin Bottle samples
From 1958 to 1993
Not gridded CTD transects off Oregon
Niskin Bottle samples and Fluorometers
From1997 to 2004
Not gridded CTD transects off Oregon
Niskin Bottle samples and Fluorometers
Summer 2001 and Winter 2003
Not gridded CTD transects off Washington and Oregon
SeaWiFS (Sea-viewing Wide Field-of-view Sensor)
Courtesy of A. Thomas(University of Maine)
Monthly means for the Northeast Pacific
September 1997 to August 2003
Data gridded ∼4 km resolution
Data on the distribution and abundances of west coast groundfish used in the current analysis were obtained from the NMFS bottom trawl survey database. The 2004 survey was identified as the most appropriate for this study as the data were collected using a depth-stratified and random sampling design and it was the first survey that covered the shelf and slope off the west coast of the USA (Keller et al., 2007). The survey started in May and ended in October, extended from approximately 48°N to 32°N along the west coast of North America and covered a depth range of 50–1280 m. Catches were sorted to species or the lowest taxonomic level and then weighed in aggregate and enumerated. The 2004 groundfish bottom trawl survey identified 153 species in the 252 trawls within the study area (Fig. 1) between May and October of 2004. The analysis focused on the 28 most abundant species, which made up 95% of the catch biomass (Table 2). Twenty-five of the 28 species are classified as groundfish species by the Pacific Coast Groundfish Fisheries Management Plan. The three exceptions were giant grenadier Albatrossia pectoralis, slender sole Lyopsetta exilis, and sandpaper skate Bathyraja interrupta and Bathyraja kincaidii, but because they were included in 95% of the catch biomass, we retained them in the analysis. For the purpose of this study, we refer to all 28 species as groundfish. A matrix for 28 species was created with 252 rows (the number of hauls within the study area) and 28 columns (the number of species), and incorporated into the statistical analysis. Prior to the analysis, densities (kg km−2) were computed from the raw species data. The densities were log transformed (natural log) to reduce the effects of the abundant species. The species densities were not standardized to allow differences in abundances of groundfish species to be expressed in the analysis. Thus, abundances will reveal some aspects of the habitat quality.
Table 2. List of 28 groundfish species making up 95% of the total biomass from trawl survey conducted in 2004.
Biomass (kg km−2)
Biomass (kg km−2)
Species with an asterisk are not classified as groundfish species in the Pacific Coast Groundfish Fisheries Management Plan.
†Skates identified during the trawl survey as B. interrupta are now considered to be sandpaper skate, B. kincaidii (Ebert, 2003).
Bathyraja interrupta and Bathyraja kincaidii
Pacific ocean perch
Combined ocean dataset
Oceanographic data (with the exception of bottom temperature) were not collected as part of the 2004 groundfish survey, and the oceanographic data acquired from other sources in the year 2004 in the study region were also scarce. Therefore, we relied on the monthly climatologies computed in this study to attempt to reproduce what oceanographic conditions were like during the year 2004. However, there are three things to consider before using the monthly climatologies: (1) the climatologies should represent the best estimate for the average oceanographic conditions at a given month from the earliest time available (e.g., CTDs go back to 1930s) to 2004; (2) the period of time (from 1930s to 2004) is characterized by changing oceanographic regimes in the North Pacific Ocean (e.g., Mantua et al., 1997; Bond et al., 2003); and (3) the 2004 trawl survey is only 1 yr within the time range of the climatologies. For these reasons, the oceanographic conditions off the northwest coast of the USA during 2004 were examined to see if they were similar to either a cold or a warm regime. The assembled dataset for temperature was divided according to the regime periods identified in the North Pacific: cold-regime (1947–76), warm regime (1977–98), cold-regime (1999–2004). Climatological monthly means of temperature for each regime were computed and compared to monthly means of temperature for 2004 at the Newport Hydrographic (NH) line off central Oregon at 45 m depth. The comparison indicated that 2004 monthly means at the NH line were more similar to the climatological monthly means for both cold-regime periods than for the warm-regime period (Fig. 2). In addition, climatological monthly means from both cold-regime periods showed similar temperatures (Fig. 2; Goericke et al., 2005). Consequently, we computed cold-regime summer climatologies utilizing the following data: (1) ocean data from both cold-regime periods to maximize coverage in our study region and (2) ocean data between May and October, when the groundfish survey was conducted. Cold-regime summer climatological means and CVs were computed for each variable (temperature, salinity and chlorophyll-a) at three depths (surface, 50 m, and near bottom) to characterize the water column. The CVs were used to identify regions in the study area that experience greater variability. The near-bottom characteristics were defined as all measurements within 10 m of the bottom.
The cold-regime summer climatologies were gridded and mapped to a common geographic extent, uniform spatial grid, and resolution of 0.16° longitude (∼12 km) and latitude (∼17 km) for all depths, using matlab version 7.1 (Fig. 3 MathWorks, Natick, MA, USA). The climatologies were also interpolated using a pseudo-spline method with a UNIX-based program called ZG, developed by Ian Crain in 1978 (unpublished) and further developed by Stephen Pierce (Oregon State University, Corvallis, OR, USA). In addition, noisy data were removed by applying two types of filters: (1) a standard deviation filter which considered an observation to be an outlier when its value was bigger (smaller) than the mean value of the grid box plus (minus) two times the standard deviation of the value in the grid box; (2) a spatial filter called medfilt2 from the statistical matlab toolbox which used a two-dimensional median filter to spatially smooth the climatologies. A subjective data screening resulted in the removal of subsurface chlorophyll-a from the analysis due to insufficient spatial coverage. For the statistical analysis, each oceanographic climatological value was extracted at each trawl location. This information created an environmental matrix composed of 252 rows (number of trawl stations) and 17 columns (14 oceanographic variables, longitude, latitude and depth at each of the trawl stations).
Several ordination and classification techniques were used to identify oceanographic habitats representative of cold regime periods and to examine patterns in the distribution and abundance of groundfish species during the year 2004 in relation to distinct oceanographic habitats and variables. We assumed the 2004 groundfish distributions and abundances were stable and representative of a cold-regime period. Past work from Tolimieri and Levin (2006) have shown no significant interannual variability in an assemblage of slope-dwelling groundfishes during the period 1999–2002 along the west coast of the USA, supporting our assumption of stable groundfish distributions during the cold-regime period. In addition, off the coast of Namibia, groundfish assemblages have also been shown not to be affected by interannual variations in temperature (Hamukuaya et al., 2001). Therefore, groundfish assemblages might be adapted to variable environments.
Two different approaches were used to identify distinct oceanographic habitats in the study region. The analyses were applied to the environmental matrix composed only of the oceanographic variables (depth, longitude and latitude were not included in the analysis). First, a hierarchical agglomerative clustering analysis was performed using Euclidean distances and Ward’s linkage method to partition the distance matrix into a determinate number of groups (McCune and Grace, 2002), which in this case were called ‘oceanographic habitats’. The distance matrix was successively partitioned into three, four, five and six groups, to evaluate the best partitioning. Each partitioning was evaluated based on the current knowledge of the oceanographic processes of the northern California Current to identify how many oceanographic habitats typify the region. Once the number of habitats was determined, each of the ocean variables was averaged within each habitat to describe the oceanographic habitats in detail. Prior to the cluster analyses, the columns in the environmental matrix were standardized by subtracting the mean and dividing by the standard deviation. The second approach consisted of a principal components analysis (PCA) to describe the direct effects of physical and biological forcing on the system and to separate discrete groups (‘oceanographic habitats’) with similar ocean characteristics. We used s-plus® statistical software for this analysis (MathSoft, 2000).
Several community analysis techniques were used to examine groundfish distributions and abundances in relation to oceanographic habitats (as categorical variables) and oceanographic variables (as quantitative variables) in the northeast Pacific Ocean off the northwest coast of the USA. First, a non-metric multi-response permutation procedure (MRPP) (Mielke, 1984; Mielke and Berry, 2001) was used to test the null hypothesis of no differences in groundfish species composition among the oceanographic habitats. The distance measure chosen was the Bray–Curtis distance (Bray and Curtis, 1957), which tends to emphasize differences in relative abundances (Anderson and Willis, 2003). Secondly, indicator species analysis (ISA) (Dufrêne and Legendre, 1997) was used to identify species that are good indicators for each of the oceanographic habitats (McCune and Grace, 2002). The ISA combines information on the concentration of species abundance in a particular habitat and the faithfulness of occurrence of a species in a particular habitat to calculate an indicator value for each species in each habitat. The indicator value ranges from zero (no indication) to 100% (perfect indication). The analysis also assesses the significance of the indicator values with a Monte Carlo randomization technique testing the null hypothesis that the highest indicator value is no larger than would be expected by chance (McCune and Grace, 2002). We used the software package pc-ord version 4.1 for the above analyses (McCune and Mefford, 1999).
Lastly, two different ordination methods were used to explore variation in groundfish assemblage structure and associations with the oceanographic habitats and individual oceanographic variables. First, we used non-metric multidimensional scaling (NMDS) (Kruskal, 1964; Mather, 1976), an unconstrained ordination method, to characterize the main patterns of variation among sample units (trawl stations) in species space without reference to an a priori hypothesis. The ‘slow-and-thorough’ autopilot mode of NMDS in PC-ORD used the best of 40 runs with the real data along with 50 runs with randomized data for a Monte-Carlo test of significance. The resulting ordination was rotated to load the strongest environmental variables on a single axis. Secondly, we used canonical analysis of principal coordinates (CAP), a permutation-based constrained ordination (Anderson, 2003; Anderson and Willis, 2003), to examine variation in assemblage structure in relation to the oceanographic habitats identified (categorical variables, canonical discriminate analysis – CDA) and the individual oceanographic variables (continuous variables, canonical correlation – CCoA). We ran 9999 permutations and chose m based on misclassification error (Anderson, 2001). The combined information of both methods is useful to obtain a more complete understanding of the patterns in the multivariate ecological data (Anderson and Willis, 2003). Both ordination techniques were done with the same distance measure (Bray–Curtis distance) in order for their joint information to be interpretable.
Cold-regime summer climatologies
The cold-regime summer climatologies (Fig. 3) represent the best estimate for oceanographic conditions during summer upwelling in a cold-regime year. This analysis assumes that these ocean conditions are the ones manifested in the summer of 2004 off the northwest coast of the USA. The cold-regime summer climatological means for temperature and salinity (Fig. 3a,c) at the surface and at 50 m delineate the cold, saltier upwelling region. At the surface, the upwelling region is more visible and extends farther off the Oregon coast, while at 50 m the upwelling region is wider off Washington. The cold-regime summer climatological mean for salinity at the surface strongly demarcates the riverine fresh waters from the northern Washington coast and the Columbia River plume. The cold-regime summer climatological means for temperature and salinity near the bottom reflect the general stratification of the ocean. However, the near-shore deep waters off Washington are warmer and fresher than those off Oregon.
The CVs of temperature and salinity (Fig. 3b,d) showed that the surface waters are spatially the most variable. The Columbia River plume at the surface and the near-shore waters off Washington throughout the water column showed the highest levels of variability. The near-shore region off Washington State also had high variability. This could be associated with the annual changes in freshwater water discharge from the Strait of Juan de Fuca, rivers on the Olympic Peninsula and intermittently from the Columbia River plume. The region off Cape Blanco also had high variability in the temperature and salinity field at the surface and at 50 m.
The cold-regime summer climatological mean for chlorophyll-a (Fig. 3e) shows highly productive waters within the upwelling region. Higher chlorophyll-a concentrations are found off the Washington coast and over Heceta Bank and south of Cape Blanco. The CV of chlorophyll-a (Fig. 3f) shows that the offshore area off the Washington coast, the upwelling region south of Cape Blanco and, to a lesser extent, the Heceta Bank region are the most variable areas.
Identification of oceanographic habitats
Five oceanographic habitats were chosen from the cluster dendrogram as habitats that most closely represented the oceanographic conditions during cold-regime summer months off the Washington and Oregon coasts (Fig. 4). Five habitats were chosen because four groups did not discriminate the Columbia River plume region from the Highly Variable Habitat, and six groups divided the Highly Variable Habitat into two, which was counter to expectations and expert knowledge. In addition, the PCA on the environmental matrix supported the cluster analysis in identifying five oceanographic habitats. For a detailed description of the PCA results, refer to Juan-Jordá (2006). The first group (red circles, N = 14) from the cluster analysis was characterized by the highest CVs for temperature and salinity among all oceanographic habitats, indicating that it is the habitat with the highest natural variability (Table 3). In addition, this habitat had the highest temperature and lowest salinity near the bottom of all the habitats. This group was named the Highly Variable Habitat for its high variability. The second cluster (blue circles, N = 108) was characterized by being the largest habitat with cold temperatures and high salinities at the surface (12.7°C, 31.84) and at 50 m (7.8°C, 33.13). In addition, this habitat had relatively high concentrations of surface chlorophyll-a (3.6 mg m−3). These conditions describe a typical upwelling region and therefore it was named the Upwelling Habitat. The third cluster (green circles, N = 52) covered the southern region of the study area, south of Cape Blanco. This group was characterized by the lowest temperatures at the surface (11.5°C) and the highest salinity at the surface (32.95) and at 50 m (33.54). In addition, it had the highest CV for temperature at the surface (CV ∼ 0.13) and the highest CV for chlorophyll-a at the surface (CV ∼ 0.67). These characteristics describe a very intense and variable upwelling region with high mesoscale activity. For these reasons, this group was named the Highly Variable Upwelling Habitat. The fourth cluster (black circles, N = 65), covering the deepest waters of the study region, was characterized by the warmest surface temperatures (14.3°C), the lowest surface chlorophyll-a (1.6 mg m−3) and the highest bottom salinities (34.17). The offshore area was not affected by the cold, saline waters found in the coastal upwelling region and therefore was named the Offshore Habitat. The last cluster (yellow circles, N = 13), confined to the area influenced by the Columbia River plume, was characterized by the lowest surface salinities (28.04) and the second highest surface temperatures (13.7°C), similar to the Offshore Habitat. In addition, it had the highest CV for salinity at the surface (CV ∼ 0.14) and the highest surface chlorophyll-a concentrations (5.9 mg m−3) of all the other habitats, making it the most productive. This group was called the River Plume Habitat.
Table 3. Average oceanographic conditions and groundfish biomass within each cold-regime summer ocean habitat. The first 14 oceanographic variables were included in the statistical analysis.
Averages within each habitat
Highly Variable Upwelling Habitat
River Plume Habitat
Highly Variable Habitat
Mean temperature at 0 m (°C)
Mean temperature at 50 m (°C)
Mean temperature near the bottom (°C)
Mean salinity at 0 m
Mean salinity at 50 m
Mean salinity near bottom
Mean chlorophyll-a at 0 m (mg m−3)
CV of temperature at 0 m
CV of temperature at 50 m
CV of temperature near the bottom
CV of salinity at 0 m
CV of salinity at 50 m
CV of salinity near the bottom
CV of chlorophyll-a at 0 m
Number of samples (N)
Fish biomass (kg km−2)
2 177 897.90
The MRPP indicated that the groundfish community composition was significantly different among oceanographic habitats (P-value ≤ 0.001). In addition, MRPP found moderate within-group agreement (A ∼ 0.17). When A is equal to its maximal value of 1, all items within the group are identical. This is not realistic for community data. In community ecology, usually the value of A is below 0.1 and when A is greater than 0.3 it is recognized as a high value (McCune and Grace, 2002). Therefore, the community composition differed strongly among the oceanographic habitats.
The indicator species analysis (ISA) showed significant species associations for all the oceanographic habitats except for the Upwelling Habitat (Fig. 4). In terms of species composition, the Highly Variable Habitat (the smallest habitat) had the highest number of indicator species. The Upwelling Habitat (the largest habitat) had only two indicator species, with indicator values that were not statistically significant. Overall, only three species showed indicator values around 40% or higher: spiny dogfish Squalus acanthias in the Highly Variable Habitat, shortspine thornyhead Sebastolobus alascanus in the Offshore Habitat, and Pacific sanddab Citharichthys sordidus in the River Plume Habitat. These were considered high indicator species. The next set of species showing statistically significant indicator values of around 15–40% were considered to be moderate indicator species. Species with indicator values of less than 15% were considered weak indicator species.
The NMDS analysis further indicated that groundfish communities differed among the oceanographic habitats (Fig. 5). The NMDS results provided a three-dimensional representation with a statistically significant reduction in stress, as compared with the randomized data (final stress = 9.53, P-value =0.0196). The first three axes explained 95% of the community variation (axis 1 captured 3.4% of the variance, axis 2, 77% and axis 3, 14%). Of the environmental parameters measured, depth, mean salinity near the bottom, mean temperature at 50 m and CV of chlorophyll-a at the surface had strong positive correlations with axis 2 (black lines in Fig. 5a) (r = 0.91, 0.63, 0.61 and 0.49, respectively). Mean temperature near the bottom and mean chlorophyll-a concentration at the surface had strong negative correlations with axis 2 (r = −0.74 and r = −0.69, respectively). The environmental variables with strong correlations with axis 2 explained primarily the natural stratification of the water column and upwelling characteristics during the summer, suggesting that depth gradients in the physical properties of the water column are the main factors defining groundfish distributions and abundances in the study area.
Abundances of 21 groundfish species had intermediate to high correlations (|r| ≥ 0.3–0.91) with ordination scores from axis 2 (Table 4). Seventeen of the 21 species negatively correlated with axis 2 indicating that these species were more abundant in shallower, colder and richer chlorophyll-a waters that were characteristic of the Highly Variable Habitat, River Plume Habitat and Upwelling Habitat. In contrast, five species positively correlated with axis 2 (Table 4), indicating that these species were more abundant in the deeper offshore water that was characteristic of the Offshore Habitat and to some extent the Highly Variable Upwelling Habitat.
Table 4. Pearson correlation of groundfish abundances with ordination axis 2 for the 28 most abundant species from the NMDS analysis.
Correlation with axis 2
Pacific ocean perch
A second pattern arose when examining how environmental variables correlated with axis 3 (Fig. 5a). Mean chlorophyll-a at the surface correlated negatively with axis 3 (r = −0.209) and to a lesser extent surface salinity (r = −0.1), whereas surface temperature correlated positively with axis 3 (r =0.207). This suggested that coastal productivity and upwelling intensity varies along the coast, making some areas more productive than others. The ordination plot in Fig. 5a showed an arch effect. We believe that the arch effect in Fig. 5a is created by the high abundances of particular species in the trawl stations located at the far end of both sides of axis 2. For example, stations 141, 144 and 55 had high abundances of lingcod and station 142 had high abundances of petrale sole. Thus, the high abundances of particular species at some stations pull the ordination at each side of axis 2 creating the arch shape (B. McCune, Oregon State University, Corvallis, OR, USA, personal communication). In addition, it is not surprising that the first axis separates the most different habitats and that the second axis separates the reminder of the habitats (Levin and Tolimieri 2006;Legendre and Legendre, 1998). We investigated the trawl stations that showed strong association with high chlorophyll-a concentrations at the surface (these are the stations on the negative side of axes 2 and 3) and found an unexpected pattern. The stations that showed strong associations with high chlorophyll-a concentrations were located in three discontinuous regions (with a few exceptions) within the study area (Fig. 6): the Juan de Fuca Canyon region, the Columbia River plume region and the Heceta Bank region.
A new pattern arises when examining the ordination of stations along axis 1 and 2 (Fig. 5b). Shallower stations (less than 500 m depth) exhibit greater spread along axis 1 than deeper stations, indicating that species composition in shallower waters is more diverse than in deep waters (Fig. 5b). This suggests a major shift in groundfish community structure at around 500–600 m depth. This agrees with Tolimieri and Levin (2006), who found an abrupt change in groundfish assemblage structure at approximately 500–600 m depth along the west coast of the USA. In addition, mean surface chlorophyll-a was the environmental variable which correlated highest with axis 1 (r = −0.21). The species Pacific sanddab (r =−0.47), English sole (r = −0.25) and big skate (r = −0.2) showed moderate negative correlations with axis 1. This suggests that these species were more abundant at stations with higher chlorophyll-a concentrations and shallower waters, which was characteristic of the Plume River Habitat, the Highly Variable Habitat and to some extent the Upwelling Habitat.
The canonical discriminant analysis (CDA) allowed the classification of the groundfish species matrix in multidimensional space to maximize differences between the five oceanographic habitats (Fig. 7, m = 6, ∂21 = 0.47, ∂22 = 0.1, P-value < 0.001). The CDA yielded two canonical axes based on differences in species composition between shallow and deep habitats, which is consistent with the NMDS analysis. However, a secondary north-south pattern emerged, but more weakly, separating stations in the Offshore Habitat from the stations in the Highly Variable Upwelling Habitat. In this case, the CDA method uncovered some of the group differences that were not easily seen in NMDS ordination (Fig. 5).
The CDA showed little overlap of species between the shallower and deeper habitats, with most of the species either having high correlation with the shallower habitats or high correlation with the deep habitats (Fig. 7). The stations within the shallow habitats seem to be mixed, indicating that the three shallow oceanographic habitats possess similar fish assemblages (Fig. 7). The exceptions were longnose skate (r = 0.31) and Pacific cod (r = −0.4) since they correlated moderately with canonical axis 2. It seems that these species are associated with different types of habitats but the particular habitat is not obvious in Fig. 7. On the other hand, sablefish (r = 0.43) and giant grenadier (r = −0.31) showed a strong association with the Highly Variable Upwelling Habitat. Pacific ocean perch Sebastes alutus (r = −0.25) was the only species that weakly correlated with the Offshore Habitat.
Whereas the CDA suggested a weak relationship between groundfish species and individual oceanographic habitats (as categorical variables), the results from the CCoA showed that groundfish abundances and distribution were highly correlated with the environmental variables (as quantitative variables) (Fig. 8, m = 7, ∂21 = 0.89, ∂22 = 0.49, P-value <0.001). The first canonical axis explained primarily trends in depth (r = −0.93) and trends in salinity and temperature near the bottom (r = 0.76 and r = −0.64, respectively). In addition, temperature at the surface (r = −0.3) and at 50 m (r = −0.59), salinity at 50 m (r = 0.38) and chlorophyll-a (r = 0.69) and CV of chlorophyll-a at the surface (r = 0.49) also loaded on the first canonical axis, characterizing the highly productive upwelling waters. Therefore, the first canonical axis separates the shallower highly productive habitats, which show more variation in terms of species distributions and environmental conditions (the stations are spread out). This contrasts with the deeper less productive habitats, which have more uniform variations in species distribution (stations are clumped together) and less environmental variability. The second canonical axis was associated primarily with latitude (r = 0.43), chlorophyll-a at the surface (r = 0.38) and longitude (r = −0.3). The association of latitude and chlorophyll-a arises because the Washington coast is more productive than the Oregon coast (Table 3). Interestingly, the CVs of salinity and temperature at different depths loaded weakly and positively with canonical axis 2, indicating that oceanographic conditions at northern latitudes are more variable, which is characteristic of the Highly Variable Habitat and the River Plume Habitat. In contrast, the CV of chlorophyll-a (r = −0.27) loaded negatively with canonical axis 2 depicting the more variable southern upwelling region off the Oregon coast.
The same groundfish species that correlated with ordination scores from axis 1 in the CDA analysis correlated with canonical axis 1 in the CCoA, suggesting that the majority of species were associated with the shallower, highly productive coastal habitats and a few deep species were associated with the deeper offshore habitats (Fig. 8). Within the group of shallower species, six species – spiny dogfish (r = 0.59), Pacific cod (r = 0.43), English sole (r = 0.36), petrale sole (r = 0.34), Pacific sanddab (r = 0.26) and yellowtail rockfish Sebastes flavidus (r = 0.25) – correlated positively with canonical axis 2 (and positive along axis 1), suggesting that they were more abundant in the richer and variable northern latitudes.
The cold-regime summer climatologies proved to be good descriptors to identify oceanographic habitats in the northeast Pacific Ocean off the northwest coast of the USA. A cluster analysis of trawl stations with their associated ocean characteristics at three different depths was able to classify the region into five cold-regime summer oceanographic habitats. Past work delineating regional oceanographic habitats using oceanographic data in the study region does not exist for comparison. At larger spatial scales, the California Current System has been divided into four regions, each forced by different physical processes and each presumed to maintain a different ecosystem structure (Parrish et al., 1981; U.S. GLOBEC., 1994). The boundary regions were located at Cape Blanco, Oregon (43°N), Point Conception (35°N) and Punta Baja in northern Baja California (30°N). At smaller scales, there is also ample literature indicating that the upwelling regimes (Barth et al., 2000; Huyer et al., 2005), alongshore currents (Smith, 1995), groundfish assemblage structure (Tolimieri and Levin, 2006) and composition of pelagic communities (Weitkamp, 1997; Peterson and Keister, 2002) north and south of Cape Blanco differ in many aspects. This makes our study the first regional attempt to identify distinct oceanographic habitats off the northwest coast of the USA and to examine whether groundfish distributions and abundances are associated with any of the oceanographic habitats. We expect that the identification of five regional oceanographic habitats off the northwest coast of the USA will spur studies to examine ecological patterns in relation to local ocean conditions.
Overall, the statistical analyses have given insight into the associations between groundfish communities and oceanographic habitats and individual oceanographic variables during cold-regime summertime. First, the MRPP analysis indicated that groundfish species composition differed among the five oceanographic habitats. Secondly, the ISA showed that all the oceanographic habitats except the Upwelling Habitat had statistically significant indicator species. The fact that there were no statistically significant species indicators in the Upwelling Habitat could be the result of: (1) species were less abundant within the habitat; (2) species fidelity to the habitat was lower; and (3) the number of stations was relatively large within the habitat. The Upwelling Habitat was the largest, composed of 103 trawl stations, and therefore it was expected to be more heterogeneous in species composition, making it more difficult for one particular species to arise from the analysis.
The ISA showed that the majority of the species had statistically significant indicator values of around 20–40% and thus we concluded that, although species might be indicators of one particular habitat, they might utilize other habitats, too. In addition, the ISA illustrated that the River Plume Habitat and the Highly Variable Habitat, which are the smallest habitats, had the largest number of indicator species. A potential explanation for these findings is the high chlorophyll-a concentrations found in both of these habitats, previously noted by Ware and Thomson (2005). The freshwater flux from the Columbia River and the numerous rivers found in Washington create a stable and highly productive upper layer. The upwelling intensity is also weaker off the Washington coast and the continental shelf is wider than off Oregon. In addition, a large anti-clockwise (cyclonic) eddy develops over the Juan de Fuca Canyon at the mouth of the strait during the summer which upwells deep nutrient-rich water into the surface (Freeland and Denman, 1982). All these characteristics may favor growth and allow the retention of high chlorophyll-a concentrations over the Washington continental shelf, leading to the differences in chlorophyll-a concentrations between the Oregon and Washington coasts, and thus playing a role in explaining the high number of indicator species found in this region.
The examinations of groundfish distributions and abundances in relation to individual oceanographic habitats consistently showed that shallow species have some degree of overlap among the nearshore oceanographic habitats. This was unexpected since the three oceanographic habitats have different physical and biological characteristics. It may be possible that the oceanographic habitats have not been defined by the necessary combination of critical parameters and / or that the environmental variables used to define the oceanographic habitat did not capture the dominant gradients influencing groundfish distributions and abundances in the study region. In addition, the confounding effects of depth, bottom temperature, and bottom salinity may also explain the overlap in species distributions and abundances in the nearshore oceanographic habitats. In the future, when new data become available, such as subsurface chlorophyll-a, water velocity, oxygen and nutrient distributions among others, their incorporation will be a valuable addition to the identification of oceanographic habitats. Nevertheless, the results suggest that the interpretation of the oceanographic habitats as biogeographical boundaries is not possible at the scale of the current study. The study area only covers the waters off the northwest coast of the USA, whereas a number of the groundfish species examined in this study are distributed from Alaska to Baja California. To properly understand habitat preferences we encourage essential fish habitat (EFH) studies to incorporate distributions and habitat data from a species’ entire range.
The examination of groundfish distributions and abundances in relation to individual oceanographic variables (instead of from a habitat perspective) revealed a different picture. There was a high consistency among the analyses suggesting a strong association between groundfish communities and environmental variables (depth, latitude, bottom temperature and salinity and surface chlorophyll-a). The analysis suggested that latitudinal changes affecting fish assemblages are probably caused by variations in upwelling intensity, coastal productivity and fresh water river discharge along the coast. Studies of demersal assemblages in other regions of the world (e.g., Hamukuaya et al., 2001; Biagi et al., 2002) and off the west coast of the USA (e.g., Tolimieri and Levin, 2006) had also identified assemblages along depth and latitudinal gradients, suggesting that changes in temperature and oxygen levels with depth and latitudinal variations in temperature and upwelling play a role in assemblage structure. However, these types of studies used depth and latitude as surrogates of temperature and productivity. In contrast, our study shows possible associations between water columns properties and groundfish distributions and their abundances.
The associations between groundfish distributions and abundances and individual oceanographic variables also led to the identification of three highly productive regions – Juan de Fuca Canyon, the Columbia River Plume region and Heceta Bank. The Heceta Bank region has already been identified as a very productive area and a hotspot (Pearcy et al., 1989; Reese and Brodeur, 2006). The River Plume Habitat was characterized as the most productive habitat in this study, as it showed the highest surface chlorophyll-a concentrations among all the habitats. Finally, the region off the Strait of Juan de Fuca also has been identified as a highly productive area as well as a highly productive foraging grounds for birds, fish and whales and an important commercial fishing ground (Healy et al., 1990; Ware and Thomson, 2005). Based on these findings we suggest that surface chlorophyll-a concentration is an important factor in determining fish distributions and abundances on regional scales. Surface chlorophyll-a proxies have the advantage of being readily available from satellite data.
In addition, the study of Tolimieri and Levin (2006), which explored patterns of groundfish assemblage structure in relation to longitude, latitude, depth, interannual variation and bottom temperature along the west coast of the USA, identified five assemblages (a deepwater group, a southern shallow water group, a shallow water mid-latitude group, a shallow northerly group and a shallow-mid-depth and mid-latitude southern group). The deepwater group identified by Tolimieri and Levin (2006) and the deep habitats identified in this study (Offshore Habitat and the Highly Variable Upwelling Habitat) had in common some species forming deepwater assemblages (Pacific grenadiers, giant grenadiers and longspine thornyheads). This was expected since Tolimieri and Levin (2006) showed that the deepwater assemblages along the west coast of the USA had broad latitudinal ranges. In contrast, the species forming their shallow northerly assemblage, darkblotched rockfish, arrowtooth flounder and Pacific ocean perch (the only shallow water assemblage that could be compared with this study), were not characteristic of the nearshore habitats identified in this study. The exception was arrowtooth flounder, which was an indicator species of the Highly Variable Habitat. The fact that both studies did not share the same spatial coverage and that Tolimieri and Levin (2006) log transformed and standardized (by dividing by mean and subtracting the standard deviations) the species abundances and we only log transformed the species abundances could explain the differences in the species characterizing nearshore habitats. By not standardizing the abundance data we did not minimize the effect of species with large biomasses and thus we used them as indications of productive areas.
Although groundfish may be adapted to a wide range of environmental variability and may use different habitats, they may still prefer some regions over others. What are the main factors creating preferable habitats or essential fish habitats for groundfish? Oceanographic processes are likely only one part of the explanation. Many studies have shown how seafloor habitat characteristics can explain groundfish distributions (Stein et al., 1992; Yoklavich et al., 2000; Tissot et al., 2007). In addition, exposure to predators (Bax, 1998; Emmett et al., 2006), prey abundance, degree of reproductive potential and likelihood of finding conspecifics in a particular habitat (Rice, 2005) may all influence the abundance of groundfish. Other oceanographic variables that were not available in this study, such as oxygen concentrations, can influence groundfish habitats. This exploratory analysis has attempted to broaden groundfish habitat science, which mainly focuses on groundfish associations with seafloor characteristics, by incorporating water column information to get a more complete understanding of what factors make up habitats for groundfish. Future studies should use the combined information of seafloor habitat characteristics and water column characteristics.
Although we assumed that groundfish distributions and abundances were stable within the cold-regime period as an attempt (1) to identify broad spatial patterns in groundfish distributions associated with cold periods and (2) to improve the knowledge of groundfish ecology and their habitats, we acknowledge that interannual variations in oceanographic conditions can also affect groundfish distributions (Smith et al., 1991). Improved monitoring programs are needed annually to monitor fish on the bottom and those in the water column and to collect ocean data throughout the water column at the same time and place as the fish surveys. Until then we must rely on the climatologies as proxies to characterize ocean processes.
Future studies should incorporate combined seafloor and water column habitat characteristics and take into account that oceanographic conditions are potentially more variable over time than substrate type in order to improve understanding of groundfish ecology and their habitats. This analysis could be expanded to examine groundfish distributions and abundance during summertime warm-regime periods and compare those results with this study to identify indicator species for each regime. We expect some species with wide tolerance for environmental changes to be able to withstand regime variability; in contrast, other species might not be able to sustain this variability. It will be very useful for fisheries management to know which species are more vulnerable to regime variations to manage these species differently from those that are more adaptable to environmental changes. Overall, by understanding the complex relationships between regime variations and groundfish distributions and abundance, researchers and managers can better anticipate the effects of regime shifts on the ecosystem and the effects of the ecosystem on fisheries, to support ecosystem-based management initiatives.
We thank the many people who assisted with the gathering of the data presented here: the Northwest Fisheries Science Center Fishery Resource Analysis and Monitoring Division and technical support group for providing the fisheries data and Andrew Thomas (University of Maine) and Bill Peterson (NOAA Fisheries) for providing some of the oceanographic data. We are also very grateful to many colleagues who helped us with data processing and analyses: Steve Pierce, Renato Castelao, Otto Gygax, Robert O’Malley, Doug Reese and Chris Romsos. We have pleasure in thanking Nick Tolimieri for his helpful comments and statistical advice on using the CAP program, Bruce McCune for statistical advice on the NMDS analyses and useful clarifications with the PC-ORD program, and Patrick Ressler for his suggestions, comments and insights on the manuscript. This work was supported under grant NA17FE1527 from NOAA Northwest Fisheries Science Center, Fishery Resource Analysis and Monitoring Division through OSU’s Cooperative Institute for Marine Resources Studies.