Birds play key roles in forest dynamics, favoring the resilience of forests to cattle ranching. Generalist species may benefit from changes in the vegetation structure caused by cattle grazing while specialist species are usually negatively affected. Therefore, studying changes in taxonomic and functional bird diversity can provide valuable information in order to propose sustainable cattle management practices. Here, we study the influence of different cattle management practices (Cattle Exclusion: FCE; Continuous grazing: FCG; Rotational grazing: FRG; Seasonal grazing: FSG) on the understory vegetation structure, and the bird taxonomic and functional diversity in humid Chaco forests of Argentina. Forests under grazing had an open (FCG) or semi-open (FRG and FSG) understory, whereas FCE presented a closed understory with higher plant richness. Generalized linear models revealed a decrease in bird richness and abundance under most cattle management practices compared to FCE. However, FRG was most similar to FCE in terms of vegetation structure and showed no differences in bird richness. FCE presented the lowest values of functional diversity while FCG and FRG showed the highest values. A fourth-corner analysis showed that FCE was associated with insectivorous species which forage and nest in the shrub layer, reflecting a possible loss of ecosystem services in forests with cattle grazing.  We propose FRG as the most appropriate cattle management option and the least detrimental to bird assemblages. However, maintaining protected areas that exclude cattle ranching activities will be key to striking a balance between the conservation of most specialist species and beef production.

Experimental design and surveys of birds and vegetation

We conducted surveys on ranches and at the INTA Experimental Station located in the Departamento Presidencia Plaza, Chaco province, Argentina (Figure 1). Crops and livestock, as well as livestock raising in a traditional continuous grazing regime, are the main activities in an extensive agricultural landscape composed of forest fragments. Sampling was conducted in forest patches under different cattle management practices:  (1) a control forest with cattle exclusion  (FCE) which consisted of a forest with 81-year old permanent exclosures located in four large paddocks (in total 200 ha) at the Estación Experimental INTA (26° 56′ S, 59° 45′ W); (2) forest with continuous grazing (FCG): cattle are left in five large paddocks, foraging in grassland and forest areas (in total 220 ha), which is the most widespread cattle management practice especially among small ranchers (Borrás et al. 2017); (3)  forest with annual rotational grazing (FRG): it has 60 paddocks of two to three ha each,  where one to two days of intense grazing are carried out in each paddock followed by 45 to 60 days of rest for the vegetation to recover (in total 220 ha); and (4) forest with seasonal grazing (FSG), with cattle grazing in the forest only in winter, since it is the only season in which cattle have access to water, whereas the rest of the year cattle graze in grassland paddocks (Cattle are moved to the forest in winter to allow the grassland to recover; in total 100 ha). Twenty sampling sites were randomly selected for each cattle management practice. All cattle management practices were embedded in a similar matrix of forest alternating with grasslands and pastures thus any confounding effects related to matrix composition were minimized.

In order to study the effects of these cattle management practices on the vegetation structure, a visual estimation of the cover percentage within each management practice was carried out at stations with a fixed radius of 20 m (Matteucci & Colma, 1982). At each station, the cover of “caraguatá” (dominant species of the herbaceous stratum: Pseudananas macrodontes, Bromelia serra, and Aechmea distichantha), shrub cover and tree cover were estimated, grouping the estimates in 10% intervals (0-10%, 11-20%, 21-30%, etc.). We used the classification guide of Ledesma et al. (2017) to identify the dominant species within each stratum. The understory plant species richness was calculated for each cattle management practice from presence and absence data of these species.

Bird sampling points were separated from each other by at least 150 m in order to ensure independence of records and avoid double counting (Bibby et al., 2000). We used a handheld GPS (error ±5 m) to measure distances. Bird sampling was conducted during the breeding season, covering September to December, in 2017. The same observer (MC) sampled each point count and recorded all birds seen or heard within a 25 m radius of the survey point during a 10-min period (following Bibby et al., 2000) except for individuals flying over 20 meters in height (Codesido & Bilenca, 2000; Codesido & Bilenca, 2004). The number of counts of individuals of each species was used as a proxy for abundance estimates.  Point counts were conducted during the four hours following sunrise, registering all birds from the moment of reaching the point. Birds flying away from the point count when the observer arrived were considered present at the count point (Hutto et al., 1986). Bird taxonomy followed Remsen et al. (2022).

Selection of functional traits

We selected 11 functional traits (see Table S1) related to the life history of the species and based on previous studies of functional diversity or responses to natural habitat replacement (Cofre et al., 2007; Corbelli et al., 2015; Feeley et al., 2007; Luck et al., 2013; Petchey & Gaston, 2006). The selected traits are relevant to understand how bird species may respond to environmental changes (Luck et al., 2013). We considered the following four groups of traits: 1) Feeding traits: body weight, which is associated with metabolic rate, foraging behavior, longevity and territory size; foraging substrate, meaning where the species carry out their feeding activities and provides information on the sensitivity of species to the loss of different types of habitats that affect food availability; diet, which  indicates the place of the species in the food chain and the ecosystem services it provides; 2) Breeding traits: migratory status, which can influence large-scale nutrient cycling and the provision of services over large regions; nesting habitat, which provides information about the sensitivity of species to the loss of different types of habitats that affect the availability of nesting sites; clutch size, species with low reproductive rates tend to be less resilient to habitat changes than those species with higher rates; 3) Habitat traits: number of habitats used, generalist species tend to be more resilient to habitat changes than specialist species as they can use different types of habitats; main habitat, primary habitat where the species can be found, including anthropic habitats; 4) Vulnerability traits: abundance, population size is strongly related to the risk of extinction of the species; distribution, the species with a restricted distribution are at greater risk of extinction than those with wider distributions;  sensitivity to human disturbance, meaning how species respond to habitat modification, replacement and contamination by anthropogenic processes. All traits were categorical variables, and each category was binary (present or absent for each species). Traits with multiple states were subdivided, and each state was treated as a single binary trait. For example, if a species feeds primarily on fruits and seeds, we assigned 1 to both "insectivore" and "frugivore", and 0 to all other feeding categories. All trait categories were mutually exclusive except "diet," "feeding substrate," and "nesting habitat." Table S1 lists the references from which trait information was taken for each species.

Statistical analysis 

All analyses were performed using the R statistical software (R Core Team, 2022). The influence of the cattle management practices on understory vegetation species richness (response variable) was studied by testing a generalized linear model (GLM) with a one-factor Poisson distribution (management type factor with four levels). The "glmmTMB" function was used to generate the models and the "DHARMa" package (Hartig, 2021) was used to test the assumptions of these models.  To study the ordination of the sites based on the set of forest and understory species recorded and the coverage of the different strata we carried out a redundancy analysis (RDA). RDA is a direct ordination method that arranges the sites in relation to one or more continuous variables (axes), which are linear combinations of the original variables, to optimize the visualization of the data (Leps and Smilauer 2003). This analysis provides a plot showing how strongly and in which direction the vegetation variables are associated with each axis, and the location of the sites on the axes indicates how they are characterized in terms of these variables (Leps and Smilauer 2003).

To study differences in bird taxonomic diversity among cattle management practices, we fit General Linear Models with Poisson distribution to species richness and abundance per site as response variables, and cattle management practice as the explanatory variable. The "glmmTMB" function was used to generate the models and the "DHARMa" package (Hartig, 2021) was used to test the assumptions of these models. A posteriori comparisons were performed using Tukey's test with the "glht" function in the multcomp package (Hothorn et al., 2008). An indicator species analysis (De Cáceres & Legendre, 2009) was used to identify the characteristic bird species for each cattle management practice. This analysis estimates an indicator value for each species based on its frequency and relative abundance. This value ranges from 0 (no association) to 100 (perfect association, i.e., the species was observed at all sites for a determined cattle management practice and absent from all sites of the different management conditions). We used the multipatt function of the "indicspecies" package and only those species with indicator values greater than 35 and significant (P ≤ 0.01) were taken into account (Codesido et al., 2013). This function also allows the evaluation of indicator species for combinations of cattle management practices as described in De Cáceres (2013).

To estimate functional diversity, we used the multidimensional functional dispersion index (FDis) developed by Laliberté & Legendre (2010). FDis is the mean distance of individual species to the centroid of all species in a multidimensional trait space, this index takes into account both species relative abundances and their functional traits. FDis is not affected by species richness, it is not strongly influenced by outliers, it can include any number and type of traits, and it can be calculated with any measure of distance or dissimilarity (Anderson et al., 2006). Higher values for this index correspond to higher functional trait diversity (Anderson et al., 2006). To estimate FDis for each site, we first calculated the distance matrix between species from the original species-trait matrix, using the Jaccard index (vegdist function in the R package "VEGAN") since it is suitable for categorical data and omits double zeros (Legendre & Legendre, 1998). The calculation of the FDis index for each site was performed with the "FD" package using the distance matrix between species and the matrix of sites by species (Laliberté & Legendre, 2010). To study the differences in functional diversity between the management practices, a linear model was generated with the FDis value per site as the response variable and cattle management practice as the explanatory variable. After testing the assumptions of homoscedasticity and normality, an analysis of variance (Anova) and a posteriori comparisons were performed with the "emmeans" function.

Finally, we analyzed associations between functional traits and management practices using a fourth-corner analysis (Brown et al., 2014), which tested for relationships between species traits and cattle management by comparing three matrices: sites-by-species, species-by-traits and sites-by-traits matrices. The significance is tested by a permutation procedure using model 6 (Brown et al., 2014). We performed separate analyses for each of the four trait groups. The sign of the correlation coefficients obtained with this analysis indicates whether the association between a site and a trait is positive or negative, and their value indicates the strength of the association.

Differences were considered significant at P<0.05 with the exception of the indicator species analysis (P<0.01).