Resource landscapes explain contrasting patterns of aggregation and site fidelity by red knots at two wintering sites

Background Space use strategies by foraging animals are often considered to be species-specific. However, similarity between conspecific strategies may also result from similar resource environments. Here, we revisit classic predictions of the relationships between the resource distribution and foragers’ space use by tracking free-living foragers of a single species in two contrasting resource landscapes. At two main non-breeding areas along the East-Atlantic flyway (Wadden Sea, The Netherlands and Banc d’Arguin, Mauritania), we mapped prey distributions and derived resource landscapes in terms of the predicted intake rate of red knots (Calidris canutus), migratory molluscivore shorebirds. We tracked the foraging paths of 13 and 38 individual red knots at intervals of 1 s over two and five weeks in the Wadden Sea and at Banc d’Arguin, respectively. Mediated by competition for resources, we expected aggregation to be strong and site fidelity weak in an environment with large resource patches. The opposite was expected for small resource patches, but only if local resource abundances were high. Results Compared with Banc d’Arguin, resource patches in the Wadden Sea were larger and the maximum local resource abundance was higher. However, because of constraints set by digestive capacity, the average potential intake rates by red knots were similar at the two study sites. Space-use patterns differed as predicted from these differences in resource landscapes. Whereas foraging red knots in the Wadden Sea roamed the mudflats in high aggregation without site fidelity (i.e. grouping nomads), at Banc d’Arguin they showed less aggregation but were strongly site-faithful (i.e. solitary residents). Conclusion The space use pattern of red knots in the two study areas showed diametrically opposite patterns. These differences could be explained from the distribution of resources in the two areas. Our findings imply that intraspecific similarities in space use patterns represent responses to similar resource environments rather than species-specificity. To predict how environmental change affects space use, we need to understand the degree to which space-use strategies result from developmental plasticity and behavioural flexibility. This requires not only tracking foragers throughout their development, but also tracking their environment in sufficient spatial and temporal detail. Electronic supplementary material The online version of this article (10.1186/s40462-018-0142-4) contains supplementary material, which is available to authorized users.


APPENDIX 1. Outline of the functional response model
The functional response model that we used in this study is generally referred to as the Toxin-Digestive Rate Model (TDRM) [1], and was developed for red knots that forage under non-ad libitum circumstances and need to search for their prey. The full model is outlined in [1]. Below, we provide a concise explanation of the model, and a detailed explanation of its current parameterization.
The model is based on the idea of a forager searching for prey in an environment with different prey types of limited availability. The goal of the model is to estimate for each prey type the optimal acceptance probability ( ) such that long-term energy intake rate ( ) is maximized, and to calculate this rate.
To do so, each prey type is assumed to have a specific energy content ( , estimated for molluscs as the ash-free dry flesh mass, AFDMflesh), and a specific ballast mass ( , estimated for molluscs as the dry shell mass, DMshell). It is further assumed that the forager has a maximum ballast mass intake rate, referred to as digestive capacity ( , [mg DMshell/s]). Digestive capacity varies among individual red knots and scales to the square of gizzard mass [2,3]. Gizzard sizes were measured non-invasively by ultrasonography [4,5] immediately after the catch by AD.
Gizzard mass was estimated from the observed sizes as described in [4] and were lower in the Wadden Sea (mean ± SD, 7.0 ± 2.0 g) than at Banc d'Arguin (8.5 ± 1.8 g). From these values, the digestive constraint was estimated using an experimentally derived calibration curve [6] as 2.5 mg DMshell/s in the Wadden Sea and 3.7 mg DMshell/s at Banc d'Arguin. For one of the prey species that is absent from the Wadden Sea but abundant in Banc d'Arguin, Loripes lucinalis, the maximum intake rate is not set by digestive rate, but by its toxicity due to a high sulfur content 2 [7]. This toxin constraint is denoted by and has been experimentally derived as 0.1 mg AFDMflesh/s [6,7].

Given the density ( , [number/m 2 ]), handling time (ℎ , [s]), AFDMflesh mass ( , [g]) and
DMshell mass ( , [g]) of each prey type, the model uses a graphical procedure to derive the optimal combination of acceptance probabilities ( ) for all available prey types, such that the AFDMflesh intake rate (mg AFDMflesh/s) is maximized, but without surpassing the digestive constraint c and the toxin constraint q. For details of the optimization procedure, we refer to [1].
is derived from Hollings disc equation for multiple prey types [8], and is calculated as: is the searching efficiency for each prey type , estimated at 6.4 cm 2 /s in the Wadden Sea [9] and, due to obstruction by seagrass roots, at 2.0 cm 2 /s at Banc d'Arguin [10]. Similarly, ballast mass intake rate, which must remain below the digestive constraint, was calculated as: The toxin constraint was defined as where , and are the prey-type specific values for Loripes lucinalis.
Handling time of each prey type (ℎ ) was assumed be a function of shell size, previously estimated for Cerastoderma edule as 3.3 x length [cm] 2 [9]. Senilia senilis in Banc d'Arguin is similarly shaped, and therefore the same estimates were used. The other relevant molluscs at DMshell were estimated as a function of shell length for each species separately [2,11]. All parameters used in the model and their descriptions are given in Table A1.

APPENDIX 2. Additional sampling at foraging locations at Banc d'Arguin
The estimated range of resource patches in the Banc d'Arguin, 50 m, was smaller than the inter-sampling distance of 250 m. Given the low autocorrelation intercept (Moran's I = 0.18), resource patches may have been smaller than the sampling accuracy, approximately 10 m.
Therefore, it is expected that many resource patches were actually missed by the sampling grid.
To verify this, we additionally performed an alternative sampling scheme, based on the idea that red knots are the champions when it comes to finding resource patches. Sampling locations were determined in the field. Two observers with telescopes searched for tagged red knots in the field, careful not to disturb foraging flocks of red knots. When a tagged red knot was observed, usually from a distance of 150-250 m, the knot and its precise location was carefully observed. After the red knot flew away, one of the observers guided the other observer to the exact foraging location, without losing sight of the location through the telescope. The location was stored in a GPS, and eight wooden picks were placed at foraging traces (holes left by a red knot bill, droppings, or footprints). A sample was taken at each of the picks within the following week, according to the same protocol as described in the main text, but on foot during low tide rather than by boat during high tide.
Given that red knots need an average energy intake of 0.2 mg AFDMflesh s -1 to maintain a stable body mass in Banc d'Arguin [12], only 7% of the locations in the sampling grid in the Banc d'Arguin harboured enough mollusc biomass (Fig. A2b). Contrarily, at 70% (34 out of 44) of the locations where tagged birds were observed foraging, at least one sample surpassed this threshold (Fig. A1). This way of sampling uncovered many more resource patches than the grid (compare Fig. A1 with Fig. 4b in main text).  Arguin (b and d). Each dot denotes one residence patch. In panels a and b, each color denotes a different individual. Panels b and d show the same residence patches, but now each color refers to a single low tide period. Note that the spatial scale slightly differs between the left and the right panels.