Four estuaries (Nahoon, Gonubie, Qora and Xhora), all of which drain into the Indian Ocean, were selected within the warm temperate region of the Eastern Cape, South Africa (Fig. 1). Estuarine selection was based on similar geomorphological and biological features: All four estuaries had permanently open mouths and similar drainage basin areas, river and estuary sizes, and available habitats. The Nahoon Estuary (32°59′S; 27°57′E) is situated furthest south of the four estuaries sampled, and falls on the edge of the southern limit of mangrove distribution in Africa (Quisthoudt et al. 2013). Three mangrove species have been introduced into the Nahoon Estuary from Durban Bay since 1969, including the white, black, and red mangroves (Saintilan et al. 2014, Hoppe-Speer et al. 2015). The Gonubie Estuary (32°56′S, 28°02′E) is situated 10 km to the east of Nahoon Estuary and can be found north of the city of East London. The Gonubie Estuary is relatively similar to the Nahoon Estuary in terms of watershed size and available habitat, but there are no mangroves in the Gonubie Estuary. The Nahoon and Gonubie estuaries are situated within urban development areas, and this has led to the anthropogenic alteration of both systems. Whitfield and Baliwe (2013) described both the health and condition of the Nahoon Estuary as fair, and the same parameters of the Gonubie Estuary as good. The Qora Estuary (32°27′S, 28°40′E) is situated approximately 80 km northeast of Gonubie Estuary, and the literature shows no historical evidence of mangroves in this estuary (Ward and Steinke 1982). The Xhora Estuary (32°10′S, 29°00′E) lies north of the Qora Estuary, roughly 290 km northeast of East London, forming the northern boundary of this study. A number of mangrove stands are present in the lower reaches of the Xhora Estuary (Ward and Steinke 1982). The Qora and Xhora estuaries are located along the rural Wild Coast (formerly Transkei), where infrastructure and development are lacking, so humans have little impact on these estuaries. However, due to poverty and lack of management in the region, mangrove and salt marsh habitat are often cut down for wood and grazed by livestock. Whitfield and Baliwe (2013) described the Qora and Xhora estuaries as being in excellent and good condition respectively, and the health of both estuaries as good.

Field sampling took place over a three-year period from 2015 to 2017, with data collected over the first-quarter moon phase in January of each year. Prior to sampling, five fixed sites were chosen remotely along the length of each estuary and marked using a GPS. Sites were spaced at one-kilometre intervals, with the first site being situated approximately 500 m from each estuary mouth. Physico-chemical measurements including temperature (°C), turbidity (NTU), salinity (PSU), conductivity (S m^–1), pH, dissolved oxygen (mg L^–1), and total dissolved solids were measured at each site using a YSI-6600 multimeter. Habitat type was also recorded at each site based on sediment and vegetation type. Six different habitat types were identified, including mud, mud and mangrove, mud and rock, mud and Nanozostera, sand, and sand and mangrove. Habitat types in the upper reaches of each estuary were characteristically muddy and rocky, while the lower 2 km were mostly sand. The middle reaches were mostly muddy, with mangroves present in the Nahoon and Xhora estuaries and absent in the Gonubie and Qora estuaries.

Small fishes were sampled at each site using a 50x2 m seine net with a 12-mm stretched mesh. The seine net was deployed from a boat and pulled ashore, covering an estimated area of 400 m². A consistent deployment of the seine net was maintained at each site, while a heavy sinker line allowed the net to be dragged through eelgrass (Nanozostera capensis) beds and over the pneumatophores of mangroves in order to obtain quantifiable results among sites and estuaries.

All fishes were identified in situ to the species level, measured (in millimetres) and quantified prior to being released back into the estuary. Individuals which could not be identified in the field were placed in sample jars containing a 10% formalin solution for further identification in the laboratory. Fishes were identified in the laboratory by doing lateral line scale counts, as well as teeth counts for Mugilidae species following van der Elst and Wallace (1976). Once identified, all fishes were categorized into estuarine usage guilds following Potter et al. (2015), with categorization determined following Whitfield (1994b).

Prior to statistical analyses, all factors were tested for normality and homogeneity of variance using the Shapiro-Wilk test and Levene test, respectively. Of the environmental variables, temperature and dissolved oxygen met the assumptions of normality and homogeneity of variance, and so did turbidity after square-root transformation. Therefore, parametric tests (one-way ANOVA and Tukey test) were used for these variables. The remaining environmental variables did not meet parametric assumptions, even after transformation, so non-parametric tests were used. The Kruskal-Wallis test was used to compare non-parametric environmental variables among years, estuaries, sites and habitats. The Mann-Whitney test was then used to further explore the data if the Kruskal-Wallis test returned a significant p-value (p<0.05). The Bonferroni Correction was applied for post hoc use of Mann-Whitney tests at a 0.025 level of significance.

Catch data were separated into marine and estuarine usage guilds, which were analysed independently to avoid any confounding effects of different estuary use by fishes that are resident versus immigrant. Fish communities were explored using diversity indices (Margalef species richness and Shannon-Wiener diversity) in PRIMER v.6 (Clarke and Gorley 2006). Catch data did not meet the assumptions of parametric tests, so non-parametric tests were used. The Kruskal-Wallis and Mann-Whitney tests were used to compare species richness, abundance and diversity among years, estuaries, sites and habitats.

Generalized additive models were used to explore trends of species richness, abundance and diversity in relation to physico-chemical variables, as well as the influence that mangrove presence or absence and habitat type had on the same catch parameters. Catch data of fishes from each guild and of dominant species were added to the generalized additive models using a negative binomial distribution with log link. Rare species were removed from the model. The physico-chemical variables for each GAM were determined using forward stepwise variable selection. When habitat type was added to the model, dispersion was tested with ‘dispersiontest’ by Cameron and Trivedi (1990). A Poisson distribution with log link was used when there was under-dispersion and a Quasi-poisson distribution with log link when there was over-dispersion. The Akaike information criterion was used to select the model with the best fit. Statistical analyses were conducted using R and RStudio (R Core Team 2017) with the packages mgcv, VEGAN, and ggplot2 (Wickham 2009, Wood et al. 2016, Oksanen et al. 2017).

Within estuaries, horizontal physico-chemical variables exhibited a relatively uniform gradient (Fig. 2). Temperature in the Gonubie Estuary was the only variable with significant variability among sites, with upper sites (Sites 1 and 2) being significantly warmer than lower sites nearest the mouth (Sites 4 and 5) (p<0.05).

Among the estuaries, the variables temperature, salinity, pH and dissolved oxygen varied significantly, though all four estuaries shared a similar climate regime due to their geographical location. Mean temperatures were significantly warmer in the more northern (Qora and Xhora) estuaries than in the Gonubie Estuary in the south of the study area (p<0.01). Mean salinity varied significantly between the Nahoon and Qora estuaries, as well as between the Gonubie and Qora estuaries (p<0.025), with the Nahoon and Gonubie estuaries exhibiting a more uniform mean salinity gradient throughout the sites sampled. The Xhora Estuary had a significantly higher mean pH than the remaining three estuaries (p<0.025). Mean dissolved oxygen (mg L^–1) was only significantly higher in the Qora Estuary than in the the Nahoon Estuary (p<0.025). Physico-chemical variations in these parameters were also evident inter-annually among estuaries (p<0.025).

Over the three-year study period, a total of 11625 fishes were caught among all four estuaries sampled, including fishes from both the marine and estuarine guilds. Catches consisted of 52 taxa represented by 26 families, with only 17 taxa making up 97% of the total catch (Table 1). The most speciose families were Gobiidae, Mugilidae and Sparidae, which were comprised of eight, seven and six species respectively. The marine guild made up 67% of the overall catch, representing 7763 individuals from 40 taxa. The estuarine guild had a lower overall abundance and species richness, comprised of 3862 individuals from 12 taxa. Rhabdosargus holubi was the most abundant species, with a total of 5136 individuals captured, accounting for 44% of the overall catch. Gilchristella aestuaria was the second most abundant species, with 2145 individuals accounting for 18% of the catch.

Within the estuarine guild, significant differences in size were recorded among estuaries for G. aestuaria and Glossogobius callidus. The mean size of G. aestuaria caught in the Gonubie Estuary was significantly smaller (54.4 mm mean length) than that of those caught in the Nahoon, Qora and Xhora estuaries (p<0.01), which had mean lengths of 62.0, 60.9 and 62.6 mm, respectively. The mean size of G. callidus individuals caught in the Qora Estuary was significantly larger (71.7 mm mean length) than that of individuals caught in both the Gonubie and Xhora estuaries (p<0.01), which had mean lengths of 65.3 and 64.8 mm, respectively.

In the marine guild, significant differences in length were recorded for Pomadasys commersonnii, the mugilids Chelon dumerili, C. richardsonii and C. tricuspidens, and R. holubi. The largest mean size of P. commersonnii was recorded in the Qora Estuary (121.9 mm) and the smallest in the Xhora Estuary (83.1 mm). Within the Mugilidae, similar size ranges were recorded in all the estuaries sampled. The largest mugilid species was C. dumerili, with the greatest mean length being recorded in the Gonubie Estuary (197.0 mm). The smallest mean length of a mugilid species (C. richardsonii) was recorded in the Qora Estuary (77.1 mm). The mean size of the dominant sparid R. holubi was greatest in the Nahoon Estuary (86.2 mm), followed by the Gonubie Estuary (83.3 mm), with the Xhora Estuary having the smallest mean size (62.9 mm).

Fish community composition and habitat use

Species diversity showed a decreasing trend from the mouth towards the upper reaches within each estuary, but there were no significant differences among estuaries or among habitats (Fig. 3). The Gonubie Estuary had the greatest species diversity (H’=1.4) and the Qora Estuary the lowest (H’=1). The Gonubie Estuary also had the highest mean species richness (ten species), closely followed by the Nahoon Estuary (nine species). The Qora and Xhora estuaries both had a mean species richness of seven species.

Nahoon Estuary had the greatest habitat complexity at the sites sampled, with a total of five different habitat types. The Qora and Xhora estuaries each had four recorded habitat types, while the Gonubie Estuary showed the lowest habitat complexity with only three habitat types. The sand and mangrove habitat in the Nahoon Estuary had the greatest mean species diversity (H’=1.8), followed by the sand and mud habitats in the Gonubie and Qora estuaries, respectively (H’=1.6). Mud habitat had the most consistently high mean species diversity across all four estuaries sampled.

There was no significant difference in the catch per unit effort (CPUE) of fishes within both the estuarine and marine guilds among habitat types (Fig. 4). Marine species, especially the marine estuarine-dependents, exhibited a high degree of habitat plasticity and dominated the catch throughout all habitat types. Fishes classified within the estuarine and marine category, as well as solely the estuarine category, also displayed habitat plasticity throughout the habitats sampled. Marine stragglers and marine estuarine opportunists were more abundant in the lower reaches of the estuaries sampled, where the sand habitat was predominant, than further upstream, where mud and rocks were predominant.

The habitats contributing the highest percentage of the total CPUE (combining fishes from all guilds) included mud and Nanozostera (26.5%), mud and rock (25%) and mud (21.1%). Mangrove habitats made little contribution to the total CPUE, with the mud and mangrove and sand and mangrove habitats contributing only 7.6% and 4.6%, respectively, to the overall CPUE combining fishes across all guilds.

When the CPUE of dominant species from the estuarine guild was compared between habitat type, it was found that mud and rock habitat contributed the highest percentage of the overall CPUE of dominant species (13%), followed by mud (8%) and mud and Nanozostera (6%) (Fig. 5A). Dominant fishes from the estuarine guild exhibited habitat plasticity, barring A. dussumieri, which can be explained by a single, large capture event at one site in the Nahoon Estuary.

Within the marine guild, mud and Nanozostera habitat contributed the highest CPUE of dominant species (19% overall), followed by the mud and sand habitats, which each contributed 13%, closely followed by the mud and rock habitat, which contributed 12% (Fig. 5B). Dominant marine species also displayed a high degree of habitat plasticity, although the majority of marine species were recorded at habitats nearest to the mouth of the four estuaries sampled. These habitats included sand, and habitats with mangroves in the two respective mangrove estuaries. Fewer marine species were found further upstream, which was characterized by habitats of mud or the combination of mud and rock.

Relationship between fish abundance and environmental variables

Generalized additive models were used to explore the influence of physico-chemical parameters and habitat type on species distributions. The presence of mangroves was included as a factor in all generalized additive models. The response variables analysed were species richness of all taxa and within guilds, as well as the abundance of dominant species. Species diversity was excluded from the models due to the high similarity among estuaries and habitat types.

The abundances of fishes within the marine guild were best described by a model using conductivity (Table 2 and Fig. 6). Peaks in species abundance were observed at conductivities between 41 and 48 S m^–1 for species in the marine guild. The model that best fitted the abundance of fishes from the estuarine guild included temperature, salinity, dissolved oxygen and habitat type. Peaks in abundance for the estuarine guild were observed between temperatures of 22°C and 24°C, with a smaller peak occurring at 26.5°C. Within the same model, there was a peak in abundance between conductivity readings of 29 and 31 S m^–1, with a smaller peak between 10 and 15 S m^–1. Although dissolved oxygen was not significant in the model, abundance was shown to increase with increasing dissolved oxygen concentrations.

For the dominant species within the estuarine guild, the abundance of Atherina breviceps was best explained by models incorporating temperature, dissolved oxygen and habitat. For G. aestuaria, a model with temperature, salinity, dissolved oxygen and habitat best explained abundance, while a model including turbidity and total dissolved solids best explained C. gilchristi abundance. Within the marine guild, temperature and salinity were significant variables influencing the abundances of P. commersonnii, C. dumerili and Pseudomyxus capensis. Temperature also influenced abundances of C. tricuspidens and R. holubi, while salinity had an impact on P. olivaceus and D. capensis. Habitat type also affected the abundances of both P. commersonnii and S. turbynei, the latter of which was also influenced by turbidity and dissolved oxygen. C. richardsonii was the only species for which conductivity had the greatest influence on abundance.