The study area covers Howe Sound and eastern Georgia Strait in southern British Columbia, Canada. From mid-July 2008 to September 2009, we conducted SCUBA-based timed counts along permanently marked line transects; one diver counted lingcod and rockfish and a second diver counted demersal crustaceans. Transects were spread over nine reefs at depths of 15–27 m below chart datum. Four reefs were within 1–2 km from each other and five were 5–15 km from other sampled reefs. Five reefs had two distinct sections differing in structural complexity where separate transects were established (i.e. 14 transects total). Individual transects were treated as independent spatial units and were sampled 12–22 times during the study (see the electronic supplementary material, table S1). Structural habitat complexity, which affects rockfish positively (Marliave & Challenger 2009
), was indexed with rugosity measurements (see the electronic supplementary material, Supplemental Methods).
Analyses aggregated quillback and copper rockfish and focused on subadults (total length 8–19 cm) because their size-specific vulnerability to lingcod is greater than for larger rockfish (Beaudreau & Essington 2007
). Also, demersal crustaceans are more important diet items for subadult than for adult rockfish (Murie 1995
), and our methods tallied demersal crustaceans accurately. Demersal shrimps—namely Pandalus danae
and smaller-bodied genera unidentifiable from stomach contents—are major prey items year round (Murie 1995
). At our sites, small-bodied genera of shrimp consisted primarily of Eualus
, which we aggregated into one group named EUHEL. Although rockfish use the inside of rocky crevices year round, during winter this habitat preference increases and activity levels decrease (Carlson & Barr 1977
; see the electronic supplementary material, figure S1).
We partitioned data into four periods. Period 1 covered mid-July to early December 2008 (summer and autumn). Period 2 covered late December 2008 to early April 2009 (winter). Period 3 covered mid April to late June 2009 (spring). Period 4 replicated summer (early July to early September 2009).
From the natural history of the system and the frameworks of trophic cascade and apparent competition, we made the following predictions. First, the abundance of subadult rockfish is negatively affected by the abundance of lingcod but positively affected by the abundances of Pandalus and EUHEL (i.e. rockfish prefer sites with more food resources) and by structural habitat complexity. Because rockfish prey on shrimp, however, subadult rockfish negatively affect the abundances of Pandalus and EUHEL (i.e. rockfish-shrimp relationships are non-recursive or bidirectional). If the above direct relationships between trophic levels were to be found, we expected rockfish to mediate an indirect positive effect of lingcod on shrimps, indicative of a trophic cascade, and an indirect negative relationship between the abundances of Pandalus and EUHEL, indicative of apparent competition. Finally, we expected data to support these predictions year-round, except for winter, when rockfish use of crevices increases and activity levels decrease.
We tested predictions using counts per minute (CPM) for each taxon (transformed as log10
(CPM + 1)) and rugosity measures to build a structural equation model (s.e.m. or path analysis) with the RAMONA procedure of Systat
12 (Browne 2007
). The model included direct recursive paths (assuming unidirectional causation) from lingcod to rockfish and rugosity to each of rockfish and both shrimp groups, and non-recursive direct paths between rockfish and each shrimp group. A Pandalus
-to-EUHEL direct path was included because the former may prey on the latter (Jensen 1995
). Stomach content data suggest that lingcod consume rockfish never, rarely and commonly when total lengths are less than 40 cm, 40–49 cm, and more than 50 cm, respectively (Beaudreau & Essington 2007
). The s.e.m., therefore, excluded lingcod with total lengths less than 50 cm.
RAMONA cannot include nominal variables (e.g. for transect location) and, consequently, rugosity values controlled for location effects. This approach is justifiable because rugosity values, which were distinct for most locations, are a biologically meaningful representation of location effects (see the electronic supplementary material, figure S2). Relative abundances of each taxon varied temporally within most transect locations (see the electronic supplementary material, figure S3). Consequently, results reflect the combined effects of time and space. (Partitioning these effects would have required greater sample sizes than ours.)
Indirect path coefficients were calculated by multiplying direct path coefficients that involved mediating variables. Variances and 90% CI for indirect path coefficients were calculated, respectively, with eqns (4) and (6) of Mackinnon et al. (2007)
We first applied the s.e.m. to non-winter periods aggregated. The robustness of results was confirmed by applying the same s.e.m. to each period separately.