Located on the northern coast of Santa Catarina State (26°02′-26°28′S, 48°28′-48°50′W), Babitonga Bay (Fig. 1) is one of the main estuarine complexes of southern Brazil. It shelters the southernmost of the large mangrove formations of South America, covering nearly 6200 ha (Cremer et al. 2006, IBAMA 1998). Considered a homogeneous water body, the bay has a water surface of 134 km², a storage volume of about 780 million m³ and a drainage basin of 1400 km² (Denit/IME 2004). In the eastern sector it is connected to the Atlantic Ocean through a 1850-m-wide mouth.

The average depth of Babitonga Bay is around 6 m, with a maximum of 28 m in the channel leading to the Port of São Francisco do Sul. The SW-NE axis of the bay (Fig. 1), called Palmital Channel, differs from other areas in that it receives the largest hydrologic contributions and is characterized as a sediment exporter (Schettini and Carvalho 1999). In contrast, the NW-SE axis, known as Linguado Channel, was blocked in 1935 during the construction of a road leading to the island of São Francisco do Sul. This intervention has permanently changed its original hydrological circulation and has induced intense sedimentation within the channel since then.

In southern Brazil, the wave regime is controlled by the Subtropical Atlantic High pressure system and the passage of synoptic cold fronts. The most energetic waves are from the south and are generated by strong winds associated with the passage of cold fronts (Pianca et al. 2010). Prevailing winds are from the east and northeast quadrants. Tides are characterized by a system of micro-tides of mixed type with semidiurnal dominance. Unequal heights occur for consecutive high and low tides, and the maximum height is less than 2 m. In front of the Port of São Francisco do Sul, for instance, the average tidal range is 85 cm, with maximum values of 128 cm (Truccolo and Schettini 1999).

The climate is characterized as humid, with a well-defined winter and summer, and the absence of a dry season. Though the area lacks a truly dry season, rainfall varies considerably throughout the year, reaching higher values in spring-summer and lower ones in autumn-winter. The local climate is driven by the dominance of Equatorial Continental, Tropical Atlantic and occasionally Tropical Continental air masses during the summer. The Equatorial Continental air mass leads to high temperatures and humidity, with heavy rainfall events; the Tropical Atlantic air mass is also associated with the occurrence of strong convective rainfall, but of lower intensity than that associated with the Equatorial Continental mass. The summer weather only becomes drier, with little or no rainfall, when the Tropical Continental air mass prevails (FATMA 2002). In winter, the passage of the Atlantic Polar Front precedes the arrival of the Atlantic Polar Mass, which brings rainfall with thunderstorms followed by cold, dry air (FATMA 2002). At this point, tropical air masses are displaced to the centre and north of the country.

Sampling and in situ experiments were performed on different temporal scales during the spring of 2012 and the summer of 2013, always at neap tide to minimize possible interferences. Sampling/experiment campaigns were performed in two months of each selected season (October and November; February and March), two weeks of each selected month and, finally, two days of each selected week. This class of experimental design is known as "fully nested”. It is effective in assessing the variability of environmental data at several temporal scales (Morrisey et al. 1994, Bell et al. 1997) and can be properly analysed by hierarchical ANOVA (Underwood 1997). All four temporal scales (seasons, months, weeks and days), with two levels each, were considered as random factors, and the sampling was repeated at four adjacent sites (A-D) located approximately 300 m apart from each other, in the central part of the estuary, where depth ranged from 6 to 7 m. Measurements obtained from each sampling site were true replicates, representing a source of variation to estimate changes in time.

Values (n=72) of physicochemical parameters, nutrient concentrations and primary production and respiration rates were compared by ANOVA on different timescales (day, week, month and season), using the R software package version 2.13.1 (R Development Core Team 2012) and the GAD (Sandrini-Neto and Camargo 2010). The coefficient of variation was also estimated in order to determine the timescale in which the greatest variability occurred. Normality and homoscedasticity of data were tested by Shapiro-Wilk and Cochran analyses, respectively, and data were transformed by log(x) or square root when these assumptions were not met.

In order to verify the relationship between the environmental variables and phytoplankton taxa, a canonical analysis of principal coordinates (CAP) was conducted. Environmental variables were standardized and the values of phytoplankton abundance were transformed by log(x+1). The CAP analyses were performed from a matrix of Bray-Curtis similarity in the statistical package PRIMER and PERMANOVA 6.1.13+ version 1.0.3.

Temperature, salinity and pH were recorded in the field using a standard Hg thermometer, a refractometer (ATAGO S1000) and a pH-meter (DENVER UP-25). The latter was calibrated before each sampling using standard buffers (Sinth). Water transparency was measured with a Secchi disc. In addition, 1-L subsurface water samples were collected with a van Dorn bottle, stored in vials and kept refrigerated in the dark. In the laboratory, approximately 5 hours after sample collection, samples (250 ml) were filtered through Whatman GF/C filters (φ=47 mm). The filtered liquid was used to determine dissolved inorganic nutrients (ammonium, nitrate, nitrite, phosphate and silicate) following the colorimetric method described by Grasshoff et al. (1983) in a spectrophotometer (SHIMADZU UV/V-1601), and the suspended particulate matter (seston) was determined by the gravimetric technique (Strickland and Parsons 1972) from the material retained on the filters. To do this, the filters were maintained at 65°C for approximately 2 h, cooled for 2 h, and weighed on an analytical balance (Mettler H51AR). Additionally, ~120-ml sample aliquots were immediately fixed after collection, preventing the formation of bubbles and the exposure to light and temperature variation until the determination of the dissolved oxygen (DO) concentration in the laboratory, approximately 5 hours after sample collection, following the titration method described in Grasshoff et al. (1983). During the field campaigns, hourly values of irradiance, rainfall, intensity and direction of the winds were obtained from an automatic meteorological station run by the National Institute of Meteorology (INMET) located in Itapoá, Santa Catarina State, approximately 10 km from the bay, and subsequently reported as daily averages. Wind data were not recorded in early October and irradiance measurements could not be computed on 26 October due to a technical failure in the anemometer and the pyranometer of the meteorological station, respectively. The pyranometer (Kipp & Zonen^®, model CM6B) used has a spectral range of 310 to 2800 nm and is fully compliant with ISO 9060 specifications for a First Class pyranometer.

For the in situ incubation experiments with natural phytoplankton populations, at each of the four sampling sites seawater was collected from the entire water column using a transparent hose (3 cm in diameter) containing a closing valve at one end and a weight at the other. The hose was gently deployed into the water, maintaining the valve kept on board open until the opposite tip of the stretched hose had touched the bottom. At that time, the valve was closed and the hose containing a full sample of the entire water column was brought to the boat deck, where the valve was opened and the sample was transferred to a van Dorn bottle. The sampled material was then distributed to BOD (biochemical oxygen demand) bottles with calibrated volumes for the incubation experiments. The samples were transferred to the bottles gently in order to avoid the formation of bubbles. The BOD bottles were then placed in acrylic tubes coated with layers of semi-transparent plastic film of varying thickness, representing decreasing irradiance levels, and incubated on two independent floating frames for a period of 3 h, always from 10:00 am to 1:00 pm. The two sets of tubes simulating decreasing irradiance levels yielded slightly different attenuation ratios: (i) 0, 5.7, 16.4, 20.2, 38, 42.9, 67.9 and 100% (dark tube); and (ii) 0, 7.0, 20.2, 23.5, 40.4, 46.5, 68.1 and 100%. Zero and 100% attenuation were attained by tubes that were either uncovered or wrapped with several layers of black plastic film, respectively. The remaining attenuation levels were previously calculated in the laboratory by comparing the light intensity measured with a radiometer (Li-Cor LI-1400) inside and outside each tube. At each sampling/experimental site, one sample was immediately fixed to give the initial DO concentration at time zero. Thus, the rates of gross PP were simultaneously estimated at 7 levels of luminosity, plus the rate of respiration of the community (dark flasks), and represented as the average value obtained from the duplicate series of BOD bottles. For the conversion of PP and respiration rates from DO to carbon units, a photosynthetic coefficient of 1.2 and a respiratory quotient of 1.0 were used, according to the literature (Asmus 1982, Biddanda et al. 1994, Almeida et al. 2005)

Based on the irradiance values (in W m^–2), the amount of photosynthetic active radiation (PAR, in E m^–2 d^–1) was estimated using a conversion factor of 0.1944, considering that the PAR spectral segment contribution to the global solar radiation ranges from 48.87 to 49.26% (mean=49%) in southern Brazil (Escobedo et al. 2011), and assuming a below-water-surface total quanta : total energy (Q:W) ratio of 2.5×10¹⁸ quanta s^–1 W^–1 (Morel and Smith 1974).

From the converted PAR values, we calculated the depths corresponding to the simulated irradiance levels on each experiment day, using the model Z=([ln E₀)-(ln E_z)]/K, where Z is the depth corresponding to a given level of irradiance; K is the attenuation coefficient of light, calculated from the water transparency (Secchi) and a conversion factor of 1.44 (Holmes 1970); E₀ is the amount of PAR at the surface; and E_z is the PAR at a given simulated light level. Next, the rates of productivity were calculated for each depth and extrapolated for the entire photic zone through trapezoidal integration. Finally, PP measured during the incubation period was extrapolated to the entire light period of the day using the daily curve of PAR, as estimated from the data of the meteorological station. In addition, in order to investigate the variation in light availability during the study period, irradiance values were daily-averaged and expressed as standard PAR unit (µE m^–2 s^–1).

Aliquots (200 mL) of the water sampled with a transparent hose along the entire water column, as described above, were placed in dark flasks containing Lugol’s solution to a final concentration of 2% and stored for latter phytoplankton analysis. Phytoplankton enumeration and identification were performed on an inverted microscope (Zeiss Axiovert) using Utermöhl counting chambers after sedimentation of 10- or 25-mL samples, depending on the cell density and seston concentration. The microphytoplankton (>20 μm) was identified to the lowest possible taxon according to specific references such as Tomas et al. (1997), Steidinger and Jangen (1997) and Hoppenrath et al. (2009). A minimum of 500 cells of this size fraction was counted per sample in the entire counting chamber at 200x magnification, although the most abundant species were counted separately in transects. Less abundant species, such as the mixotrophic, toxic dinoflagellates belonging to the Dinophysis acuminata complex and their prey, the ciliate Mesodinium rubrum, were counted in parallel after settling an additional 50-mL sample aliquot, as described above. Finally, the abundance of nanoplankton cells (<20 µm) was determined by analysing multiple transects at a higher magnification (400×) until ~300 cells were reached. In general, the counting accuracy remained at ≥90% confidence limit (Edler and Elbrachter 2010).

Over the spring of 2012 and the summer of 2013, October was the driest month, with a cumulative monthly rainfall of only 28.6 mm (Fig. 2A). Rainfall increased in November, December and January (177.2, 244.4 and 170.8 mm, respectively), reaching a maximum in February and March (405.4 and 377.8 mm). From late October to December, the predominant wind direction was SW, with daily-averaged intensity ranging from 0.5 to 2 m s^–1 (Fig. 2B). In February, the same pattern was observed, although there was a shift in wind direction to E in the second week. In March, unlike other months, winds were predominantly from the E quadrant, with a maximum daily-averaged intensity of 1.5 m s^–1 (Fig. 2B). The average PAR for the entire sampling period was 740.7 µE m^–2 s^–1, with the highest values being recorded in November. During the sampling dates, daily-averaged values ranged from only 83.4 µE m^–2 s^–1 on 5 March, a fairly cloudy day, to 1281.7 µE m^–2 s^–1 on 20 November (Fig. 2C).

Generally, dissolved inorganic nutrients and other physicochemical parameters exhibited similar values among the sampling points for each collection day. Water temperature (daily-averaged values) increased from spring (20°C) to summer (27°C) (Fig. 3), as expected. The analysis of variance components stressed that most of the variability was on the scale of months within seasons (42.6%), followed by the seasons (31.1%, Table 1). The concentration of DO was higher in October and November (6.24 and 6.71 mg L^–1, on average) than in the remaining months. In the summer, DO levels were similar between months and significantly lower than those measured in spring (P=0.004, CV=45.1%, Table 1). On a few occasions, Secchi transparency was greater than 2 m, as occurred on 5 February (2.6 m), although no significant differences between seasons, months within seasons or weeks within months were found. The variability was only significant (P<0.001) on the shortest timescale, between days within weeks (CV=44.2%, Table 1).

Overall, there was a low variability in seston concentrations (monthly-averaged values of 25.2, 24.9 and 26.6 mg L^–1 in November, February and March, respectively), with the exception of October, when the average concentration (41.4 mg L^–1, Fig. 3) was significantly higher (P=0.03, Table 1). Though most of the variability was due to the differences between months of a same season (CV=43%), a significant variation also occurred on the scale of days (P=0.002, Table 1). Salinity and pH exhibited a slight tendency to increase from the innermost to the outermost region of the bay, although the difference was not statistically significant. Salinity values varied widely between months within seasons (CV=31%, Table 1), although there was little variation between seasons. The highest daily-averaged salinity value (32) was recorded on 19 and 21 October and the lowest (25) on 20 March. Finally, the highest monthly-averaged pH value (8.14) occurred in November, with similarly lower values (around 7.9) registered during all other months. For both the pH and salinity, the temporal variability was significant only on the shortest scale, among days, as indicated by the ANOVA analysis (Table 1).

Ammonium concentrations were relatively low in the first week of October (mean of 3.78 µmol L^–1), increasing considerably thereafter and remaining higher until the end of summer (Fig. 4). Nitrate and nitrite concentrations exhibited a similar pattern of variation over the sampling period (Fig. 4), with the lowest values recorded in October (mean of 1.33 and 0.58 µmol L^–1, respectively) and the highest ones in March (mean of 5.54 and 2.56 µmol L^–1, respectively). The variation pattern in total dissolved inorganic nitrogen (DIN) was similar to that of ammonium, the most abundant and representative nitrogen compound throughout the period of study. Overall, little variation was observed in the mean concentrations of nitrogen compounds either between spring and summer or between months within seasons. Marked differences were mainly noticed on the scale of weeks, representing more than 50% of the CV (Table 2) for nitrate, nitrite, ammonium and DIN. The variation was also high between days of a same week, but to a lesser extent (29.5%, 30.5%, 18.5% and 27% for nitrate, nitrite, ammonium and DIN). Phosphate varied significantly on the scale of days (P <0.001), but mostly between months within seasons (44.1% of the variability; Table 2). The highest concentration of phosphate in October, with average of 1 µmol L^–1 (Fig. 5) resulted in the lowest N:P ratio calculated over the entire study period (average of 8.7). In all other months, N:P ratios were generally >16, with the maximum value being recorded on 7 November (33.2). Silicate concentrations were significantly lower (P=0.005, CV=33.4%) in the spring (8.94 and 7.91 µmol L^–1) than in the summer (15.65 and 16.06 µmol L^–1). Large variations were also observed on the scale of weeks within months (CV=31.1%).

The daily-averaged values of primary production (PP) rate, integrated to the entire water column, varied mostly on the scales of months within seasons (P=0.043) and days within weeks (P=0.003, Table 3). These significant differences between the monthly averages were observed both in spring and in summer. The highest PP rates occurred in November and February, when the monthly-averaged values were 994.9 and 974.7 mg C m^–2 d^–1, respectively (Fig. 6). The greatest variation in PP rates within a single month was recorded in March, ranging from 237.9 mg C m^–2 d^–1 on 3 March to 1001.3 mg C m^–2 d^–1 on 20 March (Fig. 6). Respiration rates in the water column were higher in summer than in spring (P=0.003, Table 3), reaching the highest monthly-averaged value of 547.8 mg C m^–2 d^–1 in February, and the highest daily-averaged value of 825.6 mg C m^–2 d^–1 on 16 February (Fig. 6). In March, respiration rates decreased slightly, although they remained higher than in October (251.5 mg C m^–2 d^–1) and November (343.3 mg C m^–2 d^–1).

Large variability in the total abundance of phytoplankton cells was only observed on smaller timescales, between days for nanophytoplankton and between days and weeks for microphytoplankton. Monthly-averaged values of nanophytoplankton abundance were similarly high (~1.1×10⁶ cells L^–1) throughout the study, with the exception of March (0.8×10⁶ cells L^–1). Daily-averaged values ranged from 0.4 to 1.4×10⁶ cells L^–1. Microphytoplankton abundance was generally lower (<0.3×10⁶ cells L^–1) than nanophytoplankton abundance, with the exception of the second weeks of November and February, when expressive differences between the daily averages were observed and high cell abundances were recorded (0.7-1.8 and 0.7-1.7×10⁶ cells L^–1, respectively, Fig. 6).

In total, 53 different microphytoplankton taxa were identified in this study, including 41 diatoms (Bacillariophyceae), 10 dinoflagellates (Dinophyceae), one euglenophyte and one silicoflagellate. Some taxa were frequent and relatively abundant throughout the study period, such as the benthic diatoms Cylindrotheca closterium, Paralia sulcata, Psamodictyon panduriformis, Diploneis spp., Navicula spp., the planktonic Thalassionema frauendeldii, Thalassionema nitszchioides, Thalassiosira spp., and the dinoflagellates of the genera Gymnodinium, Gyrodinium, Prorocentrum and Protoperidinium.

Other taxa were more frequent and abundant during specific periods. For instance, T. frauenfeldii, the benthic diatoms Cyclotella spp. and Cymatodiscus planetophorus, and other larger (>30 µm) centric diatoms, as well as the dinoflagellate Oxyphysis oxytoxoides, were abundant in spring but rare or absent in summer. These taxa formed a well-defined spring phytoplankton assemblage associated with higher salinity, phosphate and seston concentrations and lower DIN values (Fig. 7). In addition, another diatom, Eucampia cornuta, was rarely detected in summer but was recorded at high cell densities during an intermittent bloom episode (0.2×10⁶ cells L^–1) in the second week of November (spring), when DO concentrations were higher (Fig. 7). Conversely, Skeletonema costatum was rarely observed in spring, but attained very high cell densities in summer (up to 1.3×10⁶ cells L^–1 in the first week of February). In addition to S. costatum, Corethron criophilum, Leptocylindrus minimus and other small-sized centric diatoms, as well as an unidentified euglenophyte species, formed a characteristic summer phytoplankton assemblage associated with lower DO values and higher temperature, rainfall and silicate concentration (Fig. 7).

Finally, toxic dinoflagellates belonging to the Dinophysis acuminata complex were frequently recorded at low cell densities (10 to 310 cells L^–1) and their abundance was not clearly related to any of the physicochemical parameters investigated (Fig. 7). Instead, the abundance of this mixotrophic organism was somehow dependent on the availability of its prey, the ciliate Mesodinium rubrum, as suggested by the temporal anti-clockwise orbits partially formed when the cell abundances of both taxa were compared (Ishikawa et al. 2014) in Figure 8. At least in summer months, an increase in D. acuminata abundance led to a decrease in M. rubrum cell density and vice-versa (Fig. 8). M. rubrum was present in all samples over the period of study, at cell densities ranging from 10 to 440 cells L^–1.