Species and study site
T. chinensis is a dioecious and wind-pollinated species that is distributed in evergreen broadleaf forests. Every year, female plants bear axillary cones which, in autumn, develop into fleshy arils (commonly, although incorrectly, referred to as “fruits”) that contain a single seed. An average tree bears more than 4000 of these “fruits” (Li et al. 2015, 2019).
This study was conducted in a yew ecological garden (elevation 895–1218 m above sea level [a.s.l.], slope gradient 27°), located in the southern experimental area of the Meihua Mountain National Nature Reserve (25°15′–25°35′N, 116°45′–116°57′E) in the west part of Fujian Province, China. This site contains the largest natural population of T. chinensis in China (approximately 490 adults, distributed in the evergreen broadleaf forest), including 200 trees that are > 500 years old. A national forest garden of 15 ha was established by the government in 2003 to protect these endangered trees. Due to long-term of human use, the vegetation around the forest garden is highly fragmented. The remnant evergreen broadleaf forest patch is interlaced with bamboo patches and mixed bamboo and broadleaf patches to form a fragmented forest. The dominant tree species of the remnant evergreen broad-leaved forest is T. chinensis (Additional file 1: Fig. S1).
Microhabitat selection by Hypsipetes leucocephalus
To study post-foraging microhabitat selection of H. leucocephalus, field observations were made after the birds departed mature T. chinensis plants. We observed the post-foraging perching position of H. leucocephalus using a field scope (Leica 70, Germany) at distances of 50–100 m from the opposite mountain slopes. When the position of birds was recorded, we collected regurgitated seeds in the canopy of bird preferred microhabitats (regurgitated seeds refer to cleaned seeds without any aril/coat that had been totally digested by birds). Because we tried to show the relationship between bird habitat use and plant recruitment, we chose the site with regurgitated seeds as bird-preferred microhabitat and set 1 m × 1 m quadrats. To test whether bird selected habitat, we also set the quadrats in other available areas, and the position were confirmed by random number table. Totally, 30 used and 30 available quadrats were set to collect information on microhabitat factors in 2011. To exclude year-to-year variation in bird habitat selection, we recorded perching frequency at these 60 sites in the other study years (2012, 2018‒2019).
In both bird-use and available quadrats, we measured three qualitative factors: aspect (shade slope; sunny slope), vegetation type (bamboo forest; Chinese Yew forest; farmland; mixed bamboo and broadleaf forest) and heterogeneous tree species (other tree species, except T. chinensis trees). Moreover, we also measured 15 quantitative factors: elevation, slope, distance to water, distance to light gap, distance to roads, distance to nearest T. chinensis tree, distance to nearest heterogeneous tree, distance to nearest T. chinensis mature tree, distance to fallen dead tree, herb cover, herb density, shrub cover, shrub density, tree cover, and leaf litter cover.
For analyzing microhabitat selection by birds, we first compared three qualitative factors by Chi square test. The other 15 quantitative factors evaluated between bird used, and available quadrats were first analyzed by a t-test. All quantitative variables were evaluated with Principal Component Analysis (PCA) based on their correlation matrix with a varimax rotation to screen out the key factors in microhabitat selection of H. leucocephalus. PCA is a multivariate technique that produces a simplified, reduced expression of the original data with complex relationships, and has been widely applied in studies of wildlife habitats (Fowler et al. 1998). We also used logistic regression to explore the role of microhabitat factors for bird habitat selection.
Effects of microhabitat selection by birds on the early recruitment of Taxus chinensis
Independent of the seed dispersal study, seedling emergence was assessed experimentally in 2017 and 2018 beneath the 30 sites used by H. leucocephalus. At each site, 200 seeds were sown at a depth of 1 cm to avoid predation by rodents. The germinated seedlings were monitored weekly from spring to fall of the following year. Plant early recruitment was computed as the fraction of germinated seedlings that survived to the end of the first fall.
To study the effects of bird microhabitat selection on the early recruitment of plants, we used the t-test to compare the seedling emergence rate in the fragmented forest with natural conditions (seedling emergence rate: 10.86%, with 1000 seeds sowing under the canopy of 10 microhabitats in the natural habitat; Gao 2006). Random Forest model is an ensemble machine-learning method for classification and regression that operates by constructing a multitude of decision trees. It is appropriate for illustrating the nonlinear effect of variables, can handle complex interactions among variables and is not affected by multicollinearity. Random Forest can assess the effects of all explanatory variables simultaneously and automatically ranks the importance of variables in descending order. Then, we used the Random Forest (RF) algorithm to evaluate the quadrat habitat factors selected by birds in relation to the number of germinated seedlings (R package Random Forest) (Breiman 2001).