The study area was located in the lower reaches of the Swartkops Estuary (33°52′S; 25°38′E), on the warm temperate coastline of South Africa (Fig. 1). The permanently open estuary is approximately 16 km in length with a tidal range of <1.5 metres (Beckley 1983). The area experiences bimodal rainfall with peaks in autumn and spring (Whitfield 1998). The lower reaches of the estuary are characterized by extensive mudflats, salt marshes and large stands of submerged aquatic vegetation dominated by halophytic eelgrass Zostera capensis and ricegrass Spartina maritima (Talbot and Bate 1987, Colloty et al. 2000). The Tipper’s Creek arm of the lower estuary (Fig. 1) supports the highest percentage (59%) of this vegetation (Talbot and Bate 1987).

Juvenile fishes were sampled in the lower reaches of the Swartkops Estuary in consecutive months, twice per season for a period of one year, spanning a total of eight sampling trips between March 2013 and January 2014. Fishes were collected in four habitat types, and each habitat type was sampled in two separate locations in the lower estuary. Each sampling trip was conducted on an incoming tide during the day. Fishes were sampled using a 30×1.5 m seine net (mesh size=10 mm) at eight sites accessed by boat in the lower 3 km of the estuary. Two replicate sites were selected to represent four predetermined habitat types: sand flats (S), mud flats (M), vegetated areas (V) and shallow estuarine creeks (C) (Fig. 1). The sand and mud habitat types were selected based on a consistent substrate of either sand or mud, respectively, with no vegetation cover. Estuarine creek habitats were characterized by shallow, unvegetated drainage channels that were permanently supratidal and surrounded by an adjacent Sarcocornia spp. saltmarsh with a muddy substrate. The creek habitats differed from the mud habitats as they were not situated in the main channel but maintained tidal connectivity with the main channel. Vegetated habitats were selected in areas where there was consistent and dense coverage of submerged Zostera capensis and/or Spartina maritima ranging in cover from 90% to 100%. It was impossible to separate Spartina maritima and Zostera capensis vegetation as they co-occurred along the margins. Sampling of vegetated areas involved dragging a seine net directly through the vegetated areas. Sites were selected to ensure no overlap in habitat types and habitats were replicated to ensure that species occurrence patterns were persistent with habitat type, if applicable, and were treated as separate samples for analyses. Water sampling depth for fishes and physico-chemical variables at each site did not exceed 1.2 m on all sampling occasions. On each sampling occasion, temperature (°C), total dissolved solids (TDS), salinity, pH, turbidity (NTU) and conductivity (µS/cm) were measured at each site 20 cm below the water surface using a YSI multi-parameter meter (660 V).

A single seine net haul was conducted at each site on each occasion. The seine net was manually operated by two to four people in a semi-circle over the selected habitat area and pulled perpendicular to the shoreline. Each fish was identified to the lowest possible taxon (Van der Elst and Wallace 1976, Smith and Heemstra 1995) and was measured to the nearest 1.0 mm total length (TL) in the field. All catches were expressed as catch per unit effort (CPUE) referring to the number of fishes per seine haul. All fish were released except for individuals that could not be positively identified. These were fixed in 10% formalin for later laboratory identification and measurement. All positively identified fishes were categorized into estuary association categories based on guild according to Potter et al. (2015). The categories applicable to this study include, in the marine category, the marine straggler group, the marine estuarine-opportunist group and the marine estuarine-dependent group. Within the estuarine category, the relevant groups include the solely estuarine group, the estuarine and marine group and the estuarine migrant group. Finally, catadromous species and freshwater estuarine-opportunist species also feature (Potter et al. 2015).

Descriptive and inferential statistics for both the physico-chemical and biological data were generated using the Statistica software package version 11, 2012. Data were assessed for normality using a visual normal probability plot and the Shapiro-Wilk test and homogeneity of variance was tested using Levene’s test following square root and log transformations. All biological and physico-chemical data did not conform to parametric assumptions and therefore only non-parametric tests were used. Multivariate and diversity analyses were done using the Primer statistical package version 6.1 (2001) (Clarke and Gorley 2001).

Spatial and seasonal variability in temperature (°C), TDS, salinity, pH, turbidity (NTU) and conductivity (µS/cm) were assessed using a Kruskal-Wallis test by rank. A Kruskal-Wallis test was also used to test for differences in fish abundance among habitats and seasons. Species included in these analyses (dominant species) were identified as those that showed a total catch of more than 40 individuals throughout the study period in order to allow for a large enough sample size. Species showing significant habitat partitioning were explored for relationships with environmental variables also showing a significant difference between habitats using multiple linear regression. Shannon-Wiener diversity indices (H’) were compared among habitats and seasons using a non-parametric Kruskal-Wallis test.

For community analysis, abundance data were transformed by log(x+1) and grouped by both habitat type and season for the multivariate analyses. This transformation was used to reduce the contribution of highly abundant species. Abundance data were grouped into a similarity matrix using the Bray-Curtis similarity measure. A 3-D non-metric multidimensional scaling (nMDS) plot of abundance in each habitat type was generated. Species that were not frequently caught during the study period (caught on less than three occasions) and sites that showed zero catch during the study period were excluded from the nMDS so that clear community trends could be observed. A two-way crossed ANOSIM analysis was conducted to assess habitat and seasonal similarities in community composition and a two-way crossed SIMPER analysis by Bray-Curtis similarity was used to assess individual species contributions to the similarity attributes among these groups.

Of the six physico-chemical parameters measured during the study, TDS (H=14.89; n=64; P<0.01), salinity (H=14.91; n=64; P<0.01) and conductivity (H=14.40; n=64; P<0.01) showed a significant difference between the four habitat types (df=3). Temperature, turbidity and pH showed no significant variation between habitats, but temperature and pH varied significantly between seasons (temperature H=42.64, n=64, P<0.01; pH H=49.38, n=64, P<0.01; df=3). As expected, the highest water temperatures occurred in summer (mean=25.32°C) and the lowest in winter (mean=14.34°C), with spring (mean=19.05°C) and autumn (mean=20.67°C) showing moderate temperatures. The ranges of environmental variables recorded in each habitat are shown in Table 1. During the summer sampling period, the highest temperatures were recorded in the shallow creek sites, but vegetated sites showed the highest temperatures in winter and spring. The lowest temperatures were generally recorded at the sand sites (Table 1). The range of temperatures recorded was fairly consistent among the different habitat types. TDS was variable among the different habitat types but consistent throughout the seasons and also showed a large range of values within specific habitat types, exceeding 20 TDS units in some cases, particularly in the creek habitats (Table 1). The highest salinity and conductivity values recorded were in the vegetated habitats, which maintained stable salinities throughout the seasons. The most variable salinities were recorded in the creek and sand habitats (Table 1). pH was fairly stable throughout the different habitat types (range not exceeding 1.0 unit), but changes >1 unit were observed between seasons, with the lowest average pH in the system recorded in spring and summer and the highest pH in the colder seasons (Table 1). Turbidity was highly variable throughout the habitat types and seasons but this variability was not significant.

A two-way crossed ANOSIM also revealed a significant difference between seasonal groups (Global R=0.21; significance level (SL)=0.10%, averaged across habitat). Pairwise tests showed that the seasons that differed significantly included autumn and spring (R statistic=0.26; SL=1.60%), autumn and summer (R statistic=0.28; SL=0.70%), winter and summer (R statistic=0.36; SL=0.10%) and spring and summer (R statistic=0.31; SL=0.20%) combinations. Comparisons between autumn and winter and winter and spring showed no significant differences. SIMPER analysis showed that the summer season differed significantly from the other seasons due to high catches of the marine estuarine-dependent Sparidae Rhabdosargus holubi (26.66%) and marine estuarine opportunists Diplodus capensis (21.87%) and Rhabdosargus globiceps (15.46%) (Table 3). The winter and spring seasons were characterized by higher catches of estuarine resident Gobiidae species compared with other seasons (Table 3).

A two-way crossed ANOSIM revealed that there was a significant difference between habitat groups (Global R=0.27; significance=0.10%, averaged across season). Pairwise tests comparing habitats showed that the habitats that differed the most were the creek and sand habitats (R statistic=0.39; SL=0.10%), the sand and vegetated habitats (R statistic=0.53; SL=0.10%) and the mud and vegetated habitats (R statistic=0.27; SL=0.40%). The comparison of creek and mud habitats as well as mud and sand habitats showed no significant differences. Abundance data in a 3-D nMDS analysis showed a distinct grouping of sites by habitat based on similarities in species abundance (Fig. 4). Although habitat groupings based on dissimilarity are not distinct, there is subtle grouping of sand, creek and vegetated habitats with mud habitats tending to fall in between these groups.

SIMPER analysis of habitat types showed that sand habitats were characterized by sand-associated species such as Psammogobius knysnaensis (37.41%) and Heteromycteris capensis (31.34%), whereas these species did not contribute significantly to catches in the other habitats (Table 3). The mud habitats showed high catches of the estuarine resident Gobiidae as well as Rhabdosargus holubi (23.71%) (Table 3). However, the highest catches of marine Sparidae were in the vegetated habitats separating this habitat type from the others in terms of species catch composition (Table 3). The creek habitats differed from other habitats in the high abundance of the estuarine species Gilchristella aestuaria (18.90%), in addition to the moderate abundance of some marine Sparidae (Table 3). Other predatory piscivorous fishes were found only in the creek habitats in low numbers.

Seasonal use of temperate estuaries where abundance of juveniles increases in spring and peaks in summer, with numbers declining in autumn, after which abundance and diversity declines markedly as the water temperature cools, is a worldwide phenomenon coupled to recruitment from the ocean and productivity maxima within the estuary, typically linked to rainfall patterns and therefore food availability (Whitfield 1999, Ramos et al. 2006, Strydom 2015). This phenomenon was particularly evident for juveniles of marine spawned Mugilidae, Soleidae and Sparidae utilizing the estuary as a nursery. Schooling adults of resident pelagic Atherinidae and Clupeidae species within the estuary, however, compounded catches in the winter months in the Swartkops. Atherina breviceps breeds in spring and summer in the estuary and juveniles can be found in high abundances in autumn and winter (Neira et al. 1988, Whitfield 1998). Using the guild approach as described by Potter el al. (2015) to categorize fish prior to analysis allows better elucidation of use patterns without the compounding effects of resident species on seasonal catches of marine recruits.

There was clear evidence of partitioning among juvenile fishes within the four habitat types common to the lower reaches of the Swartkops Estuary. Eight species showed significantly higher abundance in particular types of habitat and included species of both estuarine and marine origin. The highest catches throughout the study period occurred at shallow creek sites, as is supported by habitat use observations in an Atlantic estuary (Sogard and Able 1991) but has not yet been described locally. The shoaling estuarine resident Gilchristella aestuaria often occurred in high numbers in creeks. It has been suggested that species also select habitats based on the degree of protection afforded by either the habitat directly (vegetation) or the behaviour of the species to avoid predation (Weinstein and Brooks 1983, Orth et al. 1984). In the case of resident G. aestuaria, this species is able to limit predation by shoaling in shallow creeks (Paterson and Whitfield 2000). However, creeks were also important areas for other species, evidenced by the high species diversity and the larger size range within species in creek habitats, particularly in the resident Gobiidae and the marine estuarine-dependent Sparidae and Mugilidae. The mixed community within creeks is contrary to findings by Paterson and Whitfield (2000), who found saltmarsh creeks dominated by taxa mainly of marine origin as well as to findings by Sogard and Able (1991) and Weinstein and Brooks (1983), who both found that although density of fish in creeks was high, catches were dominated by only a few species and that these habitats showed a lower diversity of species when compared with vegetated sites.

Zostera capensis and Spartina maritima rich vegetated habitats also supported high abundances of juveniles, although the abundance was species-specific. All estuary guilds (Potter et al. 2015) were recorded in vegetated areas, and these habitats appear to be very important for juvenile marine species, especially the marine estuarine-dependent sparid Rhabdosargus holubi and the marine estuarine opportunist Diplodus capensis, a finding also supported by Paterson and Whitfield (2000) in the nearby Kariega Estuary. Aside from the solely estuarine resident species, these Sparidae contributed the most to catches in these areas containing submerged vegetation, which was also reported by Beckley (1983). However, vegetated habitats are seemingly less important for other marine estuarine opportunist species such as the sparid Rhabdosargus globiceps, which tends to occur more often in creek habitats.

Mud habitats also showed a higher diversity of juvenile fish species. In contrast to creek and vegetated habitats, sand habitats appeared to support the lowest abundance (excluding the large catch of shoaling marine and estuarine Atherina breviceps on a single occasion) and diversity of species. Sand habitats are the most uniform in terms of complexity and typically only support specialist sand species, namely the sand goby Psammogobius knysnaensis and the Soleidae. Based on these characteristics, vegetated and mud habitats tend to fall in between these extremes. This was evident in the study by Hanekom and Baird (1984) in the Kromme Estuary, in which they compared Zostera beds with adjacent non-Zostera areas of muddy substrate. They found no significant differences in fish community. These habitats often share species composition and it could be that fish move between the two for feeding and refuge. Diet studies on species utilizing both habitat types will provide much insight into microhabitat use and habitat partitioning.

Habitat partitioning with ontogeny showed that species use of habitats varied depending on their estuary association category. The solely estuarine and estuarine migrants, such as the Gobiidae, Atherinidae (namely A. breviceps) and Clupeidae (namely G. aestuaria), were frequently sampled in all habitats in both juvenile and adult stages, suggesting less segregation based on size. Mugilidae, many of which are dependent on estuaries as juveniles, were also commonly caught in all habitats in sizes ranging from recently transformed juveniles to older juvenile stages, with some adults also recorded in catches.

In contrast, the marine estuarine-opportunist and dependent Sparidae species occurred mostly as small juveniles: no adults were caught (>190 mm TL) across habitats and larger specimens were mostly associated with vegetated and creek habitats. The average size of D. capensis is less than the average size of R. holubi and R. globiceps. This suggests that the marine estuarine-opportunist D. capensis may utilize estuarine nurseries for a shorter period of time during their juvenile phase when compared with the other two estuarine-dependent sparid species co-occurring in the nursery area due to their reliance on estuaries. This probably also contributes to the success of sharing habitats as diet and gape size will vary. Diplodus capensis is not wholly dependent on estuaries, often moving into tide pools and gullies in the ocean at larger sizes (Strydom et al. 2014), whereas R. holubi has a fully dependent estuarine phase in its early life history. Sparidae typically also show ontogenetic shifts in habitat that are critical to survival as these species recruit into estuaries as postflexion larvae (Strydom 2015). Smaller-sized individuals were recorded in the open water areas at sand and mud sites. Strydom (2003) proposed that the high numbers of recruiting postflexion Sparidae in open water areas as opposed to Zostera beds was a predation avoidance mechanism as transparent larvae are less visible in open water than against the darker background of the Zostera beds, which are already rich with juvenile predators. Upon completion of settlement, the Sparidae move into vegetated and creek habitats, where they occur mostly as larger juveniles, as was observed in this study. Open shallow water over sand and muddy substrata therefore becomes a transient habitat during ontogeny for marine species recruiting as larvae into estuaries (Strydom 2003), after which they filter into vegetated or creek areas as juveniles, thereby utilizing multiple habitats throughout their early development following an ocean-estuarine habitat continuum (Able 2005, Baillie et al. 2015, Strydom et al. 2015).

Surprisingly, temperature was not found to be an important driver of habitat use. This is likely because all habitats were shallow and located in the lower reaches of the estuary, so they were subject to similar tidal and solar radiation effects and therefore had similar temperatures. Despite there being no significant difference in temperature among habitat types, the highest temperatures were recorded in creek and vegetated habitats, which also yielded the highest catches. As the Sparidae contributed substantially to catches in these two habitats, it is likely that this is a reflection of the temperature sensitivity in growth for these species (Strydom et al. 2014) and these species also select these habitats based on subtly higher temperatures. The value of generally warmer estuarine nursery areas compared with the neighbouring ocean must be emphasized in species with temperature-dependent growth. These conditions are generally more suitable for the faster growth of larvae and juveniles using shallow-water nurseries such as estuaries and shallow embayments, thereby enhancing chances of survival (Strydom et al 2014, Baillie et al. 2015). Temperature most probably plays a more significant seasonal role rather than being a contributor to habitat occurrence on a fine scale, as in this study, which focuses on the lower estuary within the nursery area. Higher turbidity in creek sites, subject to tidal drainage of the salt marsh, may also provide additional protective isolation for young fishes from visual predators, lending to the value of creeks along with vegetated areas for refuge, but this value is not statistically measureable on such fine scales. Since environmental variables such as salinity, conductivity and TDS, which were significantly different among habitats, also showed no clear trends in structuring fishes in these areas, it is highly likely that other factors such as behaviour, linked to predator avoidance and coupled with feeding, are significant drivers of use in these habitats. Behaviour and feeding need to be considered in future studies focussed on fine-scale habitat partitioning.