We evaluated whether seed mass, handling time, handling efficiency and profitability account for (a) preferences in controlled experiments, and (b) field-diet composition of four bird species of the Monte desert, Argentina. The question of whether birds maximise their energy intake rates while feeding on seeds is assessed. We used feeding experiments with six native seed species of 0.07 – 0.75 mg (i.e. the seed-size range consumed in nature), which account for 0.59 – 0.84 of the field diet of the four birds. We measured seed-handling times, and used published information on bird preferences and diets, and on seed chemistry, for further calculations. Bird preferences were always positively related to seed mass, and also to seed profitability in the two intermediate-sized birds. Diet composition correlated positively with seed mass and negatively with seed profitability in three species, but some birds also showed a flexible behaviour eating the most attractive seeds according to their availability. This behaviour is not genuinely opportunistic because it only focuses on a restricted fraction of the total seed species present in the field. Contrary to expectations of species coexistence due to resource partitioning, small and large birds showed similar feeding efficiencies when eating the smaller and the larger seeds. The positive association between seed mass and profitability in several studies suggests that most birds can maximise their energy reward, on average and in the long term, by preferring the larger seeds. A combination of potential feeding optimisation with certain flexibility in the field may characterise the feeding ecology of desert seed-eating birds.

Seed species

Birds and seeds tested came from the Ñacuñán Biosphere Reserve (34°03’S, 67°54’W), Mendoza province, Argentina, which is in the central Monte desert and has been effectively excluded from domestic grazing since 1972. The climate is dry and temperate, with hot summers and cold winters. On average, >75% of the annual rainfall occurs during the growing season (October–March; 273 ± 95(SE) mm, n = 47 yr). The main habitat type in the reserve is open woodland with dispersed Prosopis flexuosa and Geoffroea decorticans trees. The shrub stratum is dominated by Larrea divaricata, and the herbaceous stratum is mainly composed of perennial grasses and annual forbs (see Fig. S1, supplementary material).

Seeds offered in the laboratory trials came from six herbaceous plant species that are common in the soil seed bank of the reserve (Pol et al. 2014). Four of them are from perennial C₄ native grasses (Sporobolus cryptandrus, Pappophorum spp., Digitaria californica and Setaria leucopila) and the other two are from annual native forbs (Chenopodium papulosum and Parthenium hysterophorus) (Table 1). The seed-size range (0.07– 0.75 mg) offered was identical to the natural range of herbaceous seeds consumed by the birds in the reserve (Marone et al. 2008). The seeds used account for a high percentage of the granivorous fraction of the bird diets: 84% (P. ornata), 68 -77% (Z. capensis), 59-69% (S. multicolor), and 76-79% (D. diuca) (Marone et al. 2008, 2017). The grass seeds are protected by glumes and glumellas that surround the round (S. leucopila), oval (S. cryptandrus) or elongated (Pappophorum spp., D. californica) caryopses. The kernel of P. hysterophorus is surrounded by an achene or cypsela with two membranous glumellas, whereas the kernel of C. papulosum is suborbicular and covered by a thin, fragile membranous pericarp. Although the birds used in our trials always dehusk the seeds when feeding, eating the whole grain, most seeds of S. cryptandrus and C. papulosum lose their seed coats during primary dispersal, arriving at the soil dehusked (L. Marone, pers. obs.) and so, we offered dehusked seeds of S. cryptandrus and C. papulosum in our experiments. References to large and small seeds in the text correspond to seeds with high and low mass, respectively (Table 1) (see Fig. S2, supplementary material).

Seed-handling time experiments

We carried out handling-time experiments on P. ornata, Z. capensis, S. multicolor, and D. diuca, which differ in several body-size measurements (Table 2). Thirty-five individuals (8-10 for each species) were mist-netted in the Ñacuñán Reserve and kept in individual cages (30 × 20 × 20 cm) with a natural photoperiod for one week before the trials were carried out. At the lab, we provided all birds with commercial seeds (Setaria italica or Phalaris canariensis) and vitamin-enriched water ad libitum. The experiments for all the combinations of bird per seed species were made during the following 3-4 weeks to prevent captive individuals from becoming used to the laboratory diet (Cueto et al. 2001). After the experiments, we released all the birds in the same area where we caught them.

Before each experiment, each bird was maintained without food for 2-5 hours. At the beginning of every trial, we moved one individual to an observational acrylic cage (40 × 40 × 40 cm) in darkness and, after one minute, the observer turned the light on and left the bird to feed for 10 min. In every trial, there were 50 seeds of one plant species on the ground of the cage. We assigned the order in which seed species were offered randomly, and we did not test the same individual with other seed species during the following 24 hs. Bird feeding activity was filmed using a video camera with a chronometer (±0.1 s), at a velocity of 30 photograms per second. Images were digitalised and assessed using a photogram-by-photogram inspection. Handling time for seeds that disperse with husks was the interval from when the bird picked it up from the ground with its bill until it was peeled and swallowed (Benkman and Pulliam 1988). For seeds that disperse without any structures attached, handling time was the interval from when the seed was picked up until the bird started the head movement to search for another seed. Data were not collected when seeds were not eaten (Benkman and Pulliam 1988). Given that several seeds were usually eaten by the same individual during a trial, handling times were averaged for every individual and seed species. We used these averages to calculate bird species-specific mean handling time (n = number of individuals of a given bird species assessed).

The calculation of seed handling time by the four seed-eating birds allowed us to estimate (a) seed handling efficiency (mg s^-1), which is the amount of mass of each seed species incorporated by a given bird species per unit of time, and (b) seed profitability (kJ s^-1), which is the energy that a bird species gains per unit of time when eating a certain food item. To estimate the profitability values, we used the calculations of the energy per unit of mass (kJ g^-1) provided by seeds of every plant species, reported in Ríos et al. (2012). Seed masses used for these estimations were measured on dehusked seeds (Table 1).

Bird-seed preferences and diet

Information on preferences for the six-seed species by the four-bird species comes from the published levels of seed consumption in a choice experiment (see Fig. 1 in Cueto et al. 2006). In those experiments, we offered an equal number of seeds (20) of every plant species simultaneously, which were dispersed homogeneously on the experimental arena, avoiding the biases of uncontrolled studies (Díaz 1990; Cueto et al. 2001). Seed preferences were established according to the mean proportion of seeds of each species consumed by each bird species. Field-bird diets in the Ñacuñán Reserve have been already published (see Table 2 in Marone et al. 2008), and are expressed as the proportion of the total seed mass eaten by each bird species that corresponds to each of the six plant-species evaluated.

Data analysis

Plausible causes of bird seed preferences and of the granivorous composition of the bird diets were explored by evaluating their relationship with four independent or “explanatory” variables (i.e. seed mass, seed-handling time, seed-handling efficiency, seed profitability) by using Generalized Linear Models (GLMs) with binomial distribution due to preferences and diets expressed as proportions (Crawley 2013). Each bird species was modelled in a separate GLM. Independent variables were standardized to make the estimates of the regression coefficients comparable, directly weighing the effect of each variable on the response. Variable selection was carried out by stepwise regression. The values of some of the independent variables were correlated according to the Spearman ordinal test: seed mass with seed-handling time in P. ornata (r = 0.94, p < 0.05), and seed mass with seed-handling efficiency in S. multicolor (r = 0.89, p < 0.05), so we only retained one of the variables in the final models of both preference and diet to avoid problems with multicollinearity. In such cases, the variable with the highest standardized regression coefficient was retained (Montgomery et al. 2021). All statistical calculations were conducted in R with the ‘glm’ function - ‘stats’ base package, and ‘stepAIC’ function - ‘MASS’ package (Venables and Ripley 2002).

Different indicators of feeding efficiency may be used to assess whether birds fit model 1 or 2 for seed consumption: the time taken to deal with a food item (Grant et al. 1976), the number of seeds consumed per unit of time (Grant 1986), or the mass of seeds consumed per unit of time (Schluter 1982; Pulliam 1985). We assessed the way feeding efficiency of different-sized birds varies with small and large seeds using three indicators: handling time, handling efficiency and seed profitability, which made it possible to test the robustness of different indicators when the four bird species ate the smaller (S. cryptandrus, C. papulosum) or the larger seeds (Pappophorum spp., S. leucopila). As we were interested in contrasting feeding efficiency with the largest and smallest seeds, the two intermediate-sized seeds (D. californica, P. hysterophorus) were discarded for this analysis. The averages of each indicator for different bird species were compared with one-way ANOVA. Raw data were log-transformed on some occasions to accomplish ANOVA assumptions. Simple linear correlations reported in the text are always Spearman ordinal correlations.