Prey capture success of Sandwich Terns Sterna sandvicensis varies non-linearly with water transparency: Optimum water transparency for Sandwich Tern

Ibis (2010), 152, 815–825 Prey capture success of Sandwich Terns Sterna sandvicensis varies non-linearly with water transparency MARTIN J. BAPTIST* & MARDIK F. LEOPOLD IMARES Wageningen UR, PO Box 167, 1790 AD Den Burg, the Netherlands Human impacts on water transparency may affect plunge-diving seabirds. We studied prey capture success of Sandwich Terns Sterna sandvicensis as a function of six environmental variables during the breeding season. We observed diving terns in the south eastern North Sea and found a non-linear optimum curve for the relationship between Secchi transparency and prey capture probability. High capture probability was found at 1.5–2.0 m, with an optimum of 63% at 1.74 m. At a minimum transparency of 0.4 m and at a maximum transparency of 3.2 m, capture probabilities were about halved. Conversion of transparency to total suspended matter (TSM) concentration showed that the optimum concentration for foraging terns would be 5–10 mg ⁄ L under mean summer conditions for chlorophyll-a. However, the summer-averaged TSM concentrations in the nearshore Dutch coastal waters range between 10 and 30 mg ⁄ L, which implies that foraging Terns in the breeding season do not encounter optimal foraging conditions. The Full Plunge Dive, with which the largest diving depth can be achieved, was applied dominantly in clear water, while the Partial Plunge Dive and Contact Dip were applied more frequently in turbid water, thus showing that Sandwich Terns adjust their diving technique in response to water transparency. Keywords: behaviour, diving, feeding, physiology, seabirds. Prey capture in visual hunters is constrained by ambient light conditions. Birds that capture prey under water need to deal with varying visibilities in air and water, and in many cases also need to deal with the special features of the water’s surface. Birds such as penguins, gannets, cormorants, terns, kingfishers and dippers that are generally assumed to rely on vision to catch prey feed during the daytime and stop feeding below certain light thresholds (Douthwaite 1976, O’Halloran et al. 1990, Cannell & Cullen 1998, Wanless et al. 1999, Garthe et al. 2000). The foraging efficiency of prey detection in piscivorous birds therefore varies with light conditions in plunge divers, strikers and underwater pursuit divers (Ainley 1977, Grubb 1977, Eriksson 1985, Katzir et al. 1989, Wilson et al. 1993, Cannell & Cullen 1998, Wanless et al. 1999, *Corresponding author. Email: martin.baptist@wur.nl ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union Lovvorn et al. 2001). Other physical variables, such as wind or sun glare, may affect prey visibility, as do several characteristics of the hunted prey, e.g. coloration, including bioluminescence, swimming speed and depth, including diel vertical migration patterns (Sumner 1934, Swenson 1979, Piersma et al. 1988, Reichholf 1988, Katzir et al. 1989, Kooyman et al. 1992, Wilson et al. 1993, Wanless et al. 1999, Lovvorn et al. 2001). As many different factors might simultaneously impact prey detectability and catchability under water it is often difficult to study the impact of any single parameter that is considered important. We argue that the following probabilities are likely to determine prey capture: 1 The probability of the presence of prey at a certain location (dependent on the spatial distribution, diel vertical migration, behaviour of the fish, presence of subsurface predators such as cetaceans or predatory fish, and transparency). 816 M. J. Baptist and M. F. Leopold 2 The probability of observing prey under water (dependent on experience of the predator, swimming depth of the prey, water depth, wind speed, wave height and transparency). 3 The probability of escape of prey (dependent on the visibility of the predator, the swimming depth and reaction time of the prey, water depth, wave height and transparency). In the field, one cannot distinguish between these individual probabilities and the observed prey capture success, i.e. whether a tern catches prey is dependent on a cumulative probability. Sandwich Terns Sterna sandvicensis are visual hunters that use aerial plunge diving as their main foraging technique (Taylor 1983). Prey capture success is assumed to increase with increasing transparency, as birds plunging from the air must continually adjust their position, rate of descent and diving technique to match the location of visually located prey (Ainley 1977). Haney and Stone (1988) supposed that the number of plunge divers should increase along a transect with increasing transparency, but did not find this relationship. This is in line with Eriksson (1985) who argued that changes in transparency should have little effect on the ability of plunge divers to detect prey. Stienen and Brenninkmeijer (1997) observed foraging Sandwich Terns in the Wadden Sea and did not find a significant linear relationship between the number of foraging terns and the local transparency. Brenninkmeijer et al. (2002a) studied the relationship between foraging success (number of prey caught per hour) and transparency in the Western Scheldt estuary in the Netherlands. The foraging success in turbid water (Secchi transparency S < 0.5 m) was not significantly lower than in ‘clear’ water (S > 0.5 m). In contrast, Brenninkmeijer et al. (2002b) found an increasing capture rate with increasing transparency in the coastal waters of Guinea-Bissau, where Sandwich Terns spend the winter. The capture success (in percentage of successful dives) was positively related to transparencies and the relationship was linear. We argue that the relationship between prey capture success and water transparency might not be a simple linear relationship because other factors will also play a role. Prey fish may see a predator coming and will try to escape if given enough time (Katzir & Camhi 1993). Safina and Burger (1988) observed that prey fish routinely avoid the surface, and that they tend to swim deeper at increasing water transparency. This would increase their ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union chances of escape from diving predators. We thus expect a non-linear optimum relationship between capture success and transparency. Other studies pointed out that besides transparency, wind speed and wave height (Stienen et al. 2000) and diving technique (Taylor 1983) play a role in determining capture success. Taylor (1983) found a decreasing capture rate at increasing wind speed in the range 3–16 m ⁄ s, and suggested that the increasing wave height diminishes the capture success; he observed shallower dives at higher wind speeds. Dunn (1972a, 1973), on the other hand, found an increasing capture rate with an increasing wind speed in the range of 0.5–7.0 m ⁄ s. Capture rate was higher at moderate wave conditions (wind chop) compared with a calm sea (surface smooth or slightly rippled). The latter might seem counterintuitive, but could be explained by the visibility of the predator, which is strengthened by the more vigorous hovering action at low wind speed, and by the reflective properties of a smooth sea surface (Dunn 1973). Prey fish might react to this by moving down in the water column, preventing easy capture. Likewise, Stienen et al. (2000) and Stienen (2006) report an increasing chick provisioning rate in the Wadden Sea at increasing wind speeds up to 8 m ⁄ s, but subsequently a decrease at higher wind speeds, declining rapidly at speeds over 14 m ⁄ s. Thus, we also expect non-linear optimum relationships between capture success and wind speed and between capture success and wave height. Human impacts on natural resources and habitat quality continue to increase, in concert with efforts to protect vulnerable species, for instance in Natura 2000 networks. This development generates demands for studies on effects of specific activities that might affect the survival of protected species, for instance by affecting their possibilities to find food. Coastal engineering, for example, is likely to change water clarity and this is likely to affect the feeding efficiency of local seabirds. Such effects on feeding efficiency particularly impact birds in the breeding season, when they are working at near-maximum capacity to rear their young (Drent & Daan 1980). Being central place foragers in the breeding season, such birds cannot easily avoid feeding in near-colony waters and even though they probably breed in protected colonies on land, they still have to face the consequences of human engineering in the sea nearby. There is thus a need to evaluate the consequences of changing Optimum water transparency for Sandwich Tern water clarity near protected seabird colonies and consider these in the light of specific protection given to these birds. The objective of this study is to determine the prey capture success and foraging technique of Sandwich Terns for a range of environmental variables. METHODS Study area Sandwich Terns were observed off the southern coast of the Dutch Wadden island Texel (c. 5259¢N, 442¢E; Fig. 1), during the breeding season. A small breeding colony was located in De Petten, a nature reserve situated some 2.5 km inland, in which a layer of shell fragments has been supplied to facilitate breeding of terns and gulls. In May 2007, 337 Sandwich Tern nests were counted (IMARES unpubl. data). The first chicks were observed on 4 June 2007. This was the onset for regular foraging flights of the adults to feed their chicks. Sandwich Terns have a clutch size of either one or two, and an average daily provisioning rate of 10 prey fish per chick (Stienen et al. 2000). 817 Adults also have to fend for themselves, so during the chick phase from June to mid-July, several hundreds of terns made thousands of plunge dives daily to find prey in the study area. Field methods A fast and manoeuvrable charter fishing vessel with a length of 12 m and a maximum speed of 25 knots was used, fully equipped and certified to work offshore. Manoeuvrability and speed were important to locate and pursue diving terns to observe prey capture success at close range. Measurements were carried out on 11 days between 8 June and 19 July 2007. Each measuring day started at 08:00 h and lasted until approximately 18:00 h. During these hours the supply rate of food to a breeding colony of Sandwich Terns is considered to be practically constant (Stienen 2006, fig. 2.8c), so we would not expect an effect of time of day. The applied method was to navigate in the study area searching for foraging Sandwich Terns without using a preset course or transect. Foraging terns were identified by their flight altitude, speed and downward looking Secchi disc transparency in metres Depth in metres >0 –5 – – 4 –20 – – 15 –1 – 0 –6 – – 5 –25 – – 20 –2 – – 1 –7 – – 6 –30 – – 25 –3 – – 2 –10 – – 7 –40 – – 30 –4 – – 3 –15 – – 10 < – 40 0.0 – 0.5 0.5 – 1.0 1.0 – 1.5 1.5 – 2.0 2.0 – 2.5 2.5 – 3.0 > 3.0 De Petten 2.5 km 5 km 10 km 7.5 km Figure 1. Secchi disc transparency of all protocolled diving locations. De Petten is the location of the breeding colony and concentric rings give distance from colony. Bathymetry data (depth in metres) courtesy of Dutch Rijkswaterstaat. ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union 818 M. J. Baptist and M. F. Leopold posture. It was attempted to have one or more foraging tern(s) within 100 m distance. Sandwich Terns often forage in pairs, but solitary individuals or small, rather loosely packed, groups occur as well (Gochfeld & Burger 1982). A measurement started with the first observed plunge dive (or attempt) of a randomly selected tern near the vessel. Visual observations were made with binoculars. The type of dive was recorded as: Full Plunge Dive (FPD), Partial Plunge Dive (PPD) or Contact Dip (CD), as defined in Taylor (1983). We distinguished a fourth type, the Broken-off Dive (BD), which is a dive that is abandoned during the descent. In the FPD, the Tern descends to the water vertically and becomes completely submerged. This technique results in relatively large diving depths. In the PPD, the Tern checks its descent and levels out partially so that entry is made in the water at a shallower angle and only the front half of the body becomes submerged. In the CD, the Tern descends to the water at an even shallower angle and only its bill comes into contact with the water, snatching the prey from the surface. The capture success of the dive was recorded as caught (prey visible in the Tern’s bill) or missed. While one observer watched the dive, the other observer and the skipper noted the location where the Tern hit the water. The research vessel then navigated to this diving position and here a small buoy with attached iron chain of 0.5 m length was thrown overboard. This buoy moved with the upper 0.5 m of the water column, influenced by the tidal current and the wind, marking the moving water mass into which the Tern dived. For each diving location, six environmental variables were recorded: Secchi depth (see below), time of dive, significant wave height, distance to shore, wind speed and water depth. Time of dive was converted to decimal day for further analysis. Significant wave height was estimated by the observers for each dive and ranged between 0.1 and 1.2 m. Distance to shore was calculated from the GPS coordinates of each dive and calculated to the nearest beach line, including the beach of the large sandy islet De Razende Bol offshore. Data on the 10-min averaged wind speed (m ⁄ s) were gathered from a weather station of the Royal Dutch Meteorological Institute, no. 229 Texelhors, located on the southern tip of Texel, and ranged between 1.0 and 13.8 m ⁄ s. Water depth was recorded from the ship’s echo sounder corrected for the draught. ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union Secchi disc transparency measurements The transparency of the water column was determined with a standard 30-cm-diameter, white perspex Secchi disc. The disc was attached to a rope with 0.5-m markings. It was lowered on the shadow side of the vessel until it was no longer visible. It then was gently raised until the disc was just visible again. The transparency S was recorded with 0.1 m accuracy and values were later corrected for rope shrinkage. Measurement errors may occur due to waves, bubble screens generated by the propeller of the vessel or foam at the surface. In the case of bubbles or foam, the observer waited until the view was cleared. In the case of waves, the reading of the rope was complicated by the oscillating surface level. On the other hand, when the Secchi disc was held at the appropriate depth, the alternating visibility with rising and falling surface level also helped to determine Secchi depth. Statistical analysis We used an information theoretical approach in which we defined candidate models based on biological hypotheses (Johnson & Omland 2004). As outlined in the introduction, we expected that transparency, and possibly also wind speed and wave height, would affect prey capture success. We further hypothesized that non-linear optimum relationships may result. We therefore applied a special form of logistic regression, the Gaussian logit curve (Ter Braak & Looman 1986). Such a model can be used to correlate a binary response variable (0 or 1) with one or more explanatory variables. In this study the binary response is ‘prey caught’ or ‘prey missed’, and the explanatory variable is an environmental variable at the diving location. In this form the logit-transformation on probability is given by a quadratic function: log PðxÞ ¼ b0 þ b1 x þ b2 x2 1  PðxÞ in which P(x) is the probability of success as function of the explanatory variable (x) and b0, b1 and b2 are regression coefficients. This form of logistic regression may produce a bell-shaped curve, depending on the regression parameters. This is a realistic form for this study, given the hypothesized optimum relationship for capture success with transparency, wind speed or wave height. Optimum water transparency for Sandwich Tern The above equation can be rewritten as: PðxÞ ¼ RESULTS expðb0 þ b1 x þ b2 x2 Þ 1 þ expðb0 þ b1 x þ b2 x2 Þ Correlations between environmental variables and can be extended with regression coefficients b3, b4, etc., for explanatory variables y, z, etc., in the case of multivariate models. When the regression coefficients are known, the optimum for a univariate model is given by: optimum ¼ 819 b1 : 2b2 Univariate as well as multivariate logistic models were constructed. We used model selection based on Akaike’s information criterion (AIC). AIC is a criterion for choosing between competing models (Akaike 1974) that takes into account both the statistical goodness-of-fit and the number of parameters that have to be estimated to achieve this particular degree of fit, by imposing a penalty for increasing the number of parameters. The preferred model is indicated by the lowest AIC, i.e. the one with the fewest parameters that still provides an adequate fit to the data. Although the AIC is widely used to infer statistical correlation, it does not infer causation (James & McCulloch 1990). Model selection based on the AIC alone may fail to predict biologically plausible patterns (Johnson & Omland 2004), particularly if AIC values for competing models are very similar. Gaussian as well as ordinary logistic regression were applied to analyse the use of FPD in relation to environmental variables. Additionally, the relative use of diving technique FPD, PPD, CD and BD was calculated as a function of transparency. Statistical analysis was carried out with R (R Development Core Team 2009). Correlations are considered statistically significant if P < 0.05. Measurement locations of Secchi disc transparency for each observed dive (n = 189) are plotted in Figure 1. In general, shallow waters were expected to be closer to shore and have a lower transparency than deeper waters. Transparency also depends on weather conditions. Increasing wind speed builds up wave action over time, bringing sediments into suspension. However, our data show contrasting relationships (Table 1). We did not record deeper water further away from shore because of the peculiar bathymetry of our study area. Moreover, our data suggested that deeper waters were more turbid, which is against the general expectation. Our data further showed that increasing wind speed led to higher waves and that wave height increased in deeper waters. There was no significant correlation between wave height and transparency, whereas there was between wind speed and transparency. This implied that in our data, increasing wind speed generated more turbid water without a mechanistic link to wave height. A probable cause may be that in shallow waters, transparency may drop fast in response to increasing wind speed at relatively low wave height. However, this does not comply with our observation that transparency increased at shallower water depth. Our data further showed that wind speed and wave height both increased during the day, leading to decreasing transparency. This cannot be a fixed, daily mechanism and only occurred in our measurement period. Finally, we found that wind speed increased significantly closer to shore. Obviously, we have a number of artefacts in our dataset. The contrasting relationships have to do with the complex morphological setting of our study area Table 1. Correlation matrix for six environmental variables. Linear correlation values for r are given in the lower triangle of the matrix, and the probabilities P of the correlation coefficients are given in the upper triangle. Transparency Transparency Distance Depth Wave Wind Time Distance 0.384 0.064 )0.187 )0.114 )0.216 )0.151 0.077 0.129 )0.166 0.055 Depth Wave Wind Time 0.010 0.291 0.118 0.077 0.032 0.003 0.022 0.073 < 0.001 0.038 0.450 0.770 0.030 < 0.001 0.156 0.131 0.021 0.492 0.158 0.337 ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union 820 M. J. Baptist and M. F. Leopold and varying weather conditions over the measurement period of 11 days. Table 2. Results of univariate and multivariate Gaussian logistic regression on candidate models of prey capture success for six environmental variables. Capture success Variable(s) Six environmental variables (transparency, wave height, wind speed, water depth, distance to shore and time of day) were tested for their influence on the capture success of Sandwich Terns. The complete multivariate Gaussian logistic model relating capture success to all six environmental variables scored an AIC of 265.39. Candidate models were defined for each univariate variable as well as for multivariate combinations of transparency with additional variables. The model with the lowest AIC is a univariate model for transparency, having an AIC of 260.15. Next best models, with AIC scores less than two points away, are univariate models for distance to shore and for wind speed, as well as multivariate combinations of transparency with wind speed, with distance to shore and with water depth (Table 2). Adding more than one additional variable to transparency did not improve AIC. Transparency seemed to be the prime factor determining capture success, although AIC values for this model and for alternative candidate models are close to each other. Considering the significant correlations for decreasing transparency at increasing wind speed, decreasing transparency at increasing water depth and increasing wind speed with decreasing distance to shore, transparency is a covariate of wind speed, water depth and distance to shore, which shows in the AIC scores. Model selection on the basis of the AIC alone cannot be supported, as the AIC scores are very similar. Therefore, we proceeded with investigating univariate logistic model results. Univariate logistic regression applied on the six environmental variables gave statistically significant correlations only for transparency (Table 3). Considering the significance of the univariate model for transparency, we rated highly the candidate model with transparency for explaining capture success. The highest capture probability was found at transparencies between 1.5 and 2.0 m, with an optimum value at 1.74 m of 63% (Fig. 2). The mean capture probability was 54%. The 95% confidence intervals show that the decrease in probability in turbid water follows more reliably from the data than does the decrease in clear water. This difference is explained by the smaller number of observations in clear water (Fig. 2). Transparency Transparency + wind speed Distance to shore Transparency + distance to shore Transparency + water depth Wind speed Transparency + wave height Transparency + time of day Water depth Wave height Time of day Full model ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union AIC 260.15 260.69 261.03 261.33 262.03 262.07 262.35 263.12 263.86 263.94 265.08 265.39 Table 3. Results of Gaussian logistic regression on prey capture success for six environmental variables in order of lowest AIC. Value Transparency b0 )1.48 b1 2.33 )0.67 b2 Distance to shore 0.72 b0 b1 )1.02 0.21 b2 Wind speed b0 0.53 b1 0.02 )0.01 b2 Water depth b0 0.20 b1 0.04 )0.004 b2 Wave height b0 0.74 )2.13 b1 1.42 b2 Time of day b0 )0.24 0.15 b1 1.34 b2 sd Pr (> |z|) 0.72 0.99 0.31 0.041 0.019 0.031 0.30 0.65 0.26 0.018 0.12 0.42 0.66 0.25 0.02 0.42 0.93 0.59 0.49 0.13 0.006 0.69 0.77 0.50 0.41 1.74 1.57 0.07 0.22 0.37 4.69 18.43 17.64 0.96 0.99 0.94 Relationship between transparency and type of dive We applied both an ordinary as well as a Gaussian multivariate logistic model for a binary response ‘Full Plunge Dive (FPD)’ or ‘other dive’. We defined candidate models for univariate variables as well as for multivariate combinations of transpar- Optimum water transparency for Sandwich Tern 821 Table 5. Multivariate model for Full Plunge Dive selected by lowest AIC. Regression coeffcient Value sd Pr (> |z|) b0 b1 (transparency) b2 (distance) b3 (water depth) )0.466 1.464 )0.851 )0.060 0.534 0.334 0.305 0.039 0.383 < 0.001 0.005 0.127 Table 6. Results of ordinary logistic regression on use of Full Plunge Dive for six environmental variables in order of lowest AIC. Value Figure 2. Relationship between transparency and prey capture probability. The solid line presents the results of the logistic regression, the dotted lines give the 95% confidence intervals, the histograms on the top and bottom give, respectively, the number of caught prey and missed prey and the dots give the prey capture probability for each histogram class. Table 4. Results of univariate and multivariate ordinary logistic regression on candidate models of Full Plunge Dive for six environmental variables. Variable(s) Transparency + distance to shore water depth Transparency + distance to shore water depth + time of day Transparency + distance to shore water depth + wave height Transparency + distance to shore water depth + wind speed Full model Transparency Distance to shore Water depth Wave height Wind speed Time of day AIC + 223.19 + 225.00 + 225.18 + 225.18 228.94 230.47 248.69 249.26 254.86 255.34 255.83 ency with additional variables. The model with the lowest AIC (223.19) was an ordinary logistic model for a combination of transparency, distance to shore and water depth (Table 4). Adding more variables did not improve the AIC. Transparency is an important variable without which the AIC increases substantially. Table 5 presents the values for the regression coefficients of the selected multivariate model, their standard deviations and their P-values. Subsequently, we applied ordinary univariate logistic models for all six environmental variables Transparency b0 )1.46 1.53 b1 Distance to shore b0 0.94 )0.72 b1 Water depth 1.04 b0 b1 )0.096 Wave height b0 0.73 )0.64 b1 Wind speed b0 0.76 b1 )0.05 Time of day 0.92 b0 b1 )0.88 sd Pr (> |z|) 0.44 0.34 < 0.001 < 0.001 0.24 0.28 < 0.001 0.009 0.27 0.04 < 0.001 0.011 0.28 0.57 0.010 0.258 0.36 0.06 0.037 0.372 0.83 1.57 0.267 0.576 (Table 6). Both regression coefficients for the model with transparency were highly significant. Distance to shore and water depth show significant P-values for both coefficients as well. We analysed the trend in the relative use of four diving types. We subdivided the data in seven equal-sized (n = 27 observations) classes and calculated the frequency (%) for each diving type. We found that FPD is dominant in clear water conditions and its incidence decreases significantly at lower transparency classes, favouring the relative use of PPD. In the case of highly turbid water (S < 0.75 m), the CD and BD were used relatively often, thus showing that Sandwich Terns adjust their diving technique in response to water transparency (Fig. 3). DISCUSSION We found a statistically significant optimum curve for capture probability as a function of transparency. ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union 822 M. J. Baptist and M. F. Leopold Figure 3. Relative use of diving technique. The 189 observations were put in ascending order and subdivided into seven equal classes of 27 observations each. For each observation class the percentage of FPD, PPD, CD and BD was calculated as a function of transparency. The line presents the ordinary logistic regression for the frequency of use of Full Plunge Dive as a function of transparency. Such a relationship was to be expected because of the known behaviour of both predator and prey, but has not been previously described. Brenninkmeijer et al. (2002b) found that capture success (percentage successful dives) was significantly higher at greater transparencies according to a linear relationship. They did not find indications for an optimum transparency, but this may be attributed to small sample size in clear waters. We observed Sandwich Terns in waters with a wide range of transparencies, but mainly between 0.75 and 2.0 m (Fig. 2), which is in accordance with Stienen (2006). Mean capture success in our study was 54%. This is similar to the value of 46% found by Taylor (1983) off the Scottish east coast in the breeding season. However, Stienen and Brenninkmeijer (1994) found a mean capture success of only 12% in Guinea-Bissau and Dunn (1972b) found 13% and 17% for juvenile and adult Sandwich Terns, respectively, wintering in Sierra Leone. The lower success in the wintering areas is hard to explain. It may be the result of an abundant food supply inducing terns to dive more often, and take greater risks (Stienen et al. 1993), or of high water transparency in the tropics. Alternatively, terns may need to work harder in the breeding season than in winter when life seems to be easier, in energetic terms (Brenninkmeijer et al. 2002b). ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union Based on the available literature we know that wind speed and wave height can have a direct effect on capture success, affecting the flight of the terns as well as the visibility through the water’s surface. We did not find a significant effect of wind speed as a univariate variable in the range of 1.0–13.8 m ⁄ s nor of wave height in the range of 0.1–1.2 m. We studied a wide enough range for parameters to compare with the studies by Dunn (1972a, 1973), Taylor (1983) and Stienen (2006). Time of day may influence the capture probability, if fish have a diel vertical migration pattern (cf. Piersma et al. 1988). In our study it was assumed that observations between 08:00 h and 18:00 h do not show an effect of time of day (cf. Stienen 2006, fig. 2.8c), as these times are sufficiently distant from sunrise and sunset, respectively. As we have not measured the presence of fish, we cannot be sure of the absence of this effect. However, logistic regression applied to time of day as a univariate variable did not show an effect within the time span covered. We did not find a significant effect of water depth between 1.5 and 23.3 m on the capture success. We also did not find a significant effect of distance to shore as a univariate logistic variable in the range of 0.032–3.268 km. In the Netherlands, 99% of the prey of Sandwich Terns consists of Herring Clupea harengus, Sprat Sprattus sprattus and sandeels Ammodytidae (Stienen et al. 2000). The decrease in capture probability at increasing water transparency can be explained by the decreasing availability of prey. Stienen and Brenninkmeijer (1997), later summarized in Stienen (2006), found that the swimming depth of the prey of Sandwich Terns in the Netherlands was lower in clear water than in turbid water. A similar relationship was found for Smelt Osmerus eperlanus in Lake IJssel, the Netherlands (Piersma et al. 1988, Mous 2000). This might be an anti-predator reaction: fish avoid swimming near the surface to avoid being caught by plunge-diving birds. In addition, fish swimming in clear water might be better at escaping an approaching tern, further reducing the capture probability. We found that the FPD, which allowed the largest diving depth to be achieved, was applied dominantly in clear water, while the PPD and CD were applied more frequently in turbid water (S < 0.75 m). Taylor (1983) found that the diving technique of terns changed in relation to wind speed. In our study, wind speed was not a factor of importance either in the multivariate or Optimum water transparency for Sandwich Tern 823 in the univariate regression for the use of FPD. Our observations show that Terns adjust their diving technique in response to the transparency of the water. Practical conservation implications As the Sandwich Tern is a protected species in Natura 2000 areas, human activities that might affect turbidity on feeding grounds around breeding colonies need an appropriate assessment. Activities such as sediment disposal, port dredging, sand mining and sand nourishment may lead to increased levels of total suspended matter and thus increase turbidity (e.g. Kuo et al. 1985). To assess the relationship between prey capture success and suspended matter concentration, values for transparency were converted. First, the measurements were converted to the extinction coefficient, KD (m)1), of Lambert Beer’s law. An empirical relationship for the Southern North Sea was determined by Visser (1970): 1:25 KD ¼ S in which S is the Secchi disc transparency (m). Secondly, an empirical relationship between extinction from total suspended matter concentration ((TSM) in mg ⁄ L) and chlorophyll-a concentration ((CHL) in lg ⁄ L) in Dutch nearshore waters was used (Suijlen & Duin 2001): TSM ¼ 20ðKD  0:03½CHL  0:04Þ Algal concentration, expressed as [CHL], is introduced to this equation because of its effect on the extinction of light. This concentration is highly variable due to algal blooms. A long-term summer average of 10 lg ⁄ L was determined for Dutch coastal waters (Suijlen & Duin 2001). The optimum capture probability was found at total suspended matter concentrations of 5–10 mg ⁄ L in combination with a summer-averaged chlorophylla concentration of 10 lg ⁄ L. Figure 4 shows the TSM concentrations converted from the range of observed transparencies in this study. Note that the shape of the curve is different from Figure 2 and that lower values for TSM correspond to higher values for transparency. Figure 4 shows that capture success is very sensitive to changes in turbidity in the lower range of TSM concentrations Figure 4. Prey capture probability as a function of total suspended matter (TSM) concentration. TSM was calculated by an empirical equation for the range of observed Secchi disc transparencies, without chlorophyll-a (no Chl-a) and with a summer-averaged concentration of 10 lg ⁄ L chlorophyll-a (with Chl-a). (up to 15 mg ⁄ L), including changes in algal concentration, and not very sensitive in the higher range of TSM concentrations. When the background TSM concentration is low, a small change in turbidity might therefore have a large effect on capture success. The summer-averaged TSM concentration in the nearshore zone of the Dutch North Sea ranges between 10 and 30 mg ⁄ L (Suijlen & Duin 2001). This is higher than the optimum TSM concentration found. Foraging terns in the breeding season thus do not encounter optimal transparency: the water is too turbid. The coastal zone of the North Sea has clearer water than the adjacent Wadden Sea due to the natural shallowness of the Wadden Sea, high silt loads and current velocities. This may be the incentive for Sandwich Terns to fly large distances to catch fish, e.g. more than 13 km from the main breeding colony at Griend, Central Wadden Sea, out into the North Sea (Veen 1977). Theoretically, the effect of water transparency on prey capture success might not be a causal relationship but rather reflects the trade-off between foraging costs and water turbidity. On average, there is an increasing water transparency with distance offshore (Suijlen & Duin 2001), but our study was limited to the nearshore zone. Distance to shore did not give a statistically significant correlation with capture success (Table 3). Still, similar capture success at greater distance implies a higher ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union 824 M. J. Baptist and M. F. Leopold cost of foraging due to increasing costs for longer flights. Sandwich Terns forage in nearshore coastal waters with abundant food supply and in relatively close proximity to their breeding colony (Garthe 1997), despite the suboptimal transparency. This research was financed by the Ministry of Transport, Public Works and Water Management, National Institute of Public Works and Water Management (RWS), the Netherlands. The research was executed as part of a programme to evaluate the impacts of sand mining on the Dutch coastal ecosystem in general and on visual predators such as Sandwich Terns in particular. The authors acknowledge Marcel Rozemeijer, George Ellerbroek, Peter Meininger, Cor Berrevoets and Joy Seegers of RWS for their guidance of the study. The authors thank the Royal Netherlands Institute for Sea Research (NIOZ) for permission to use their Secchi disc and skipper Rutgert Oosterhuis of Het Sop for his cooperation. We thank our IMARES colleagues Jenny Cremer, Jannes Heusinkveld, Martin de Jong, Cor Smit and Hans Verdaat for their field observations of diving Terns and Erik Meesters and Geert Aarts for their help with the statistical analyses. Finally, we thank Eric W. M. Stienen and an anonymous reviewer for critically reviewing our manuscript. REFERENCES Ainley, D.G. 1977. Feeding methods in seabirds: a comparison of polar and tropical nesting communities in the eastern Pacific Ocean. In Llano, G.A. (ed.) Adaptations within Antarctic Ecosystems. Proc. 3rd Symp. Antarctic Biol. Sci. Comm. Antarctic Res: 669–685. Washington, DC: Smithsonian Institute. Akaike, H. 1974. A new look at the statistical model identification. IEEE Trans. Automat. Contr. 19: 716–723. Brenninkmeijer, A., Doeglas, G. & De Fouw, J. 2002a. Foerageergedrag van sterns in de westelijke Westerschelde in 2002. Report 346 [in Dutch]. Veenwouden: Altenburg & Wymenga. Brenninkmeijer, A., Stienen, E.W.M., Klaassen, M. & Kersten, M. 2002b. Feeding ecology of wintering terns in Guinea-Bissau. Ibis 144: 602–613. Cannell, B.L. & Cullen, J.M. 1998. The foraging behaviour of Little Penguins Eudyptula minor at different light levels. Ibis 140: 467–471. Douthwaite, R.J. 1976. Fishing techniques and foods of the Pied Kingfisher on Lake Victoria in Uganda. Ostrich 47: 153–160. Drent, R. & Daan, S. 1980. The prudent parent: energetic adjustments in avian breeding. Ardea 68: 225–252. Dunn, E.K. 1972a. Effect of age on the fishing ability of Sandwich Terns Sterna sandvicensis. Ibis 114: 360–366. Dunn, E.K. 1972b. Studies on terns, with particular reference to feeding ecology. PhD Thesis, Durham University. Dunn, E.K. 1973. Changes in fishing ability of terns associated with windspeed and sea surface conditions. Nature 244: 520–521. Eriksson, M.O.G. 1985. Prey detectability for fish-eating birds in relation to fish density and water transparency. Ornis Scand. 16: 1–7. ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union Garthe, S. 1997. Influence of hydrography, fishing activity, and colony location on summer seabird distribution in the southeastern North Sea. ICES J. Mar. Sci. 54: 566–577. Garthe, S., Benvenuti, S. & Montevecchi, W.A. 2000. Pursuit plunging by Northern Gannets (Sula bassana) feeding on Capelin (Mallotus villosus). Proc. R. Soc. Lond. B 267: 1717–1722. Gochfeld, M. & Burger, J. 1982. Feeding enhancement by social attraction in the Sandwich Tern. Behav. Ecol. Sociobiol. 10: 15–17. Grubb, T.C. 1977. Weather-dependent foraging in Ospreys. Auk 94: 146–149. Haney, J.C. & Stone, A.E. 1988. Seabird foraging tactics and water clarity: are plunge divers really in the clear? Mar. Ecol. Prog. Ser. 49: 1–9. James, F.C. & McCulloch, C.E. 1990. Multivariate analysis in ecology and systematics: panacea of pandora’s box? Annu. Rev. Ecol. Syst. 21: 129–166. Johnson, J.B. & Omland, K.S. 2004. Model selection in ecology and evolution. Trends Ecol. Evol. 19: 101–108. Katzir, G. & Camhi, J.M. 1993. Escape response of Black Mollies (Poecilia sphenops) to predatory dives of a Pied Kingfisher (Ceryle rudis). Copeia 1993: 549–553. Katzir, G., Lotem, A. & Intrator, N. 1989. Stationary underwater prey missed by Reef Herons, Egretta gularis: head position and light refraction at the moment of strike. J. Comp. Physiol. A. 165: 573–576. Kooyman, G.L., Cherel, Y., Le Maho, Y., Croxall, J.P., Thorson, P.H., Ridoux, V. & Kooyman, C.A. 1992. Diving behavior and energetics during foraging cycles in King Penguins. Ecol. Monogr. 62: 143–163. Kuo, A.Y., Welch, C.S. & Lukens, R.J. 1985. Dredge-induced turbidity plume model. J. Waterw Port Coastal Ocean Eng. 111: 476–494. Lovvorn, J.R., Baduini, C.L. & Hunt, G.L.J. 2001. Modeling underwater visual and filter feeding by planktivorous shearwaters in unusual sea conditions. Ecology 82: 2342– 2356. Mous, P.J. 2000. Interactions between fisheries and birds in IJsselmeer, The Netherlands. PhD Thesis, Wageningen University. O’Halloran, J., Gribbin, S.D., Tyler, S.J. & Ormerod, S.J. 1990. The ecology of Dippers Cinclus cinclus (L.) in relation to stream acidity in upland Wales: time-activity budgets and energy expenditure. Oecologia 85: 271–280. Piersma, T., Lindeboom, R. & Van Eerden, M.R. 1988. Foraging rhythm of Great Crested Grebes Podiceps cristatus adjusted to diel variations in the vertical distribution of their prey Osmerus eperlanus in a shallow eutrophic lake in The Netherlands. Oecologia 76: 481–486. R Development Core Team 2009. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. Reichholf, J.H. 1988. Die Wassertrübung als begrenzender Factor für das Vorkommen des Eisvogels (Alcedo athhis) am unteren Inn. Egretta 31: 98–105. Safina, C. & Burger, J. 1988. Ecological dynamics among prey fish, Bluefish and foraging Common Terns in an Atlantic coastal system. In Burger, J. (ed.) Seabirds and Other Marine Vertebrates: 95–173, New York: Columbia University Press. Optimum water transparency for Sandwich Tern Stienen, E.W.M. 2006. Living with gulls: trading off food and predation in the Sandwich Tern Sterna sandvicensis. PhD Thesis, Groningen University. Stienen, E.W.M. & Brenninkmeijer, A. 1994. Voedselecologie van de grote sterns (Sterna sandvicensis): onderzoek ter ondersteuning van een populatie-dynamisch model. Report IBN-120, IBN-DLO Texel (in Dutch). Stienen, E.W.M. & Brenninkmeijer, A. 1997. Fluctuaties in de lokale voedselbeschikbaarheid in relatie tot de populatiedynamiek van de Grote Stern Sterna sandvicensis: resultaten 1995–1996. Report BEON 97-1, BEON project IBN 95 H 24 (in Dutch). Stienen, E.W.M., Brenninkmeijer, A. & Geschiere, C.E. 1993. De relatie tussen de foerageermogelijkheden van de grote stern Sterna sandvicensis en het doorzicht van het water, alsmede de invloed van kleptoparasitisme door de kokmeeuw Larus ridibundus op de visaanvoer en het broedsucces van de grote stern: voorlopige resultaten. Report DG-512, IBN-DLO Texel (in Dutch). Stienen, E.W.M., Van Beers, P.W.M., Brenninkmeijer, A., Habraken, J.M.P.M., Raaijmakers, M.H.J.E. & Van Tienen, P.G.M. 2000. Reflections of a specialist: patterns in food provisioning and foraging conditions in Sandwich Terns Sterna sandvicensis. Ardea 88: 33–49. Suijlen, J.M. & Duin, R.N.M. 2001. Variability of near-surface total suspended matter concentrations in the Dutch coastal zone of the North Sea; Climatological study on the suspended matter concentrations in the North Sea. Report RIKZ ⁄ OS ⁄ 2001.150X, Rijkswaterstaat RIKZ. 825 Sumner, F.B. 1934. Does ‘protective coloration’ protect? – results of some experiments with fishes and birds Proc. Nat. Acad. Sci. USA 20: 559–564. Swenson, J.E. 1979. The relationship between prey species ecology and dive success in Ospreys. Auk 96: 408–413. Taylor, I.R. 1983. Effect of wind on the foraging behaviour of Common and Sandwich Terns. Ornis Scand. 14: 90–96. Ter Braak, C.J.F. & Looman, C.W.N. 1986. Weighted averaging, logistic regression and the Gaussian response model. Vegetatio 65: 3–11. Veen, J. 1977. Functional and causal aspects of nest distribution in colonies of the Sandwich Tern (Sterna s. sandvicensis). Behaviour. Suppl. 20: 1–193. Visser, M.P. 1970. The turbidity of the Southern North Sea. Ocean Dynam. 23: 97–117. Wanless, S., Finney, S.K., Harris, M.P. & McCafferty, D.J. 1999. Effect of the diel light cycle on the diving behaviour of two bottom feeding marine birds: the Blue-eyed Shag Phalacrocorax atriceps and the European Shag P. aristotelis. Mar. Ecol. Prog. Ser. 188: 219–224. Wilson, R.P., Puetz, K., Bost, C.A., Culik, B.M., Bannasch, R., Reins, T. & Adelung, D. 1993. Diel dive depth in penguins in relation to diel vertical migration of prey: whose dinner by candlelight? Mar. Ecol. Prog. Ser. 94: 101–104. Received 29 June 2009; revision accepted 20 July 2010. ª 2010 The Authors Journal compilation ª 2010 British Ornithologists’ Union