The department of Caldas is located in the central west Andes of Colombia (4°4'19”N; 75°57'26”W). It is composed of mountainous areas belonging to the Central and Western Andes Mountain Ranges and the inter-Andean valleys of the Cauca and Magdalena Rivers. The department covers a surface area of 7457 km², with an altitudinal range between 140 to 5350 m of altitude. It presents an annual precipitation average of 2800 mm, with two rainy seasons (March - June and September - December) and two less rainy periods (January - February and July - August), and a temperature that varies from 13 °C - 27 °C (Jaramillo-Robledo et al., 2011). In Caldas, agriculture spans 69.26% of the territory, with seasonal and permanent crops, while 21.88% is covered by native and/or planted forests. Coffee (Coffea arabica L.) is the main productive system in the department, followed in order of economic importance by plantain (Musa sp.), fruit trees (citric), and cocoa (Theobroma cacao L.) (Ministerio de Agricultura y Desarrollo Rural, 2006). In order to encompass this environmental diversity, we conducted bird surveys in 11 municipalities of the department: Pácora, Victoria, Samaná, Pensilvania, Manizales, Palestina, Chinchiná, Villamaría, Marmato, Riosucio, and San José, throughout an altitudinal range of 551 to 2679 m.

For sampling habitats selection, we used images from Google Earth version 7.1. Three types of habitats were selected from the images: type I agroecosystems (ATI, n=8), type II (ATII, n=5), and secondary forests (SF, n=8) (Figure 1). Eight habitats were defined as type I agroecosystems, namely mono-cropping with bare soil (ATI₁: cocoa; ATI₂: avocado; ATI₃: lemon; ATI₄: pasture; ATI_5: papaya; ATI₆: citric; ATI_7-8: coffee; Table 1). Five habitats were assigned as type II agroecosystems, including mixed-cropping, grazing pastures with dispersed trees, and forest plantations with developed understory (ATII₁: corn, cassava, plantain, and forage species; ATII₂: coffee and plantain; ATII₃: grazing pastures with weeds and dispersed tree cover; ATII₄: urapán (Fraxinus chinensis) plantation; and ATII5: patula pine (Pinus patula) plantation; Table 1). Finally, eight secondary forests were also selected (SF₁-SF₈; Table 1). All of the selected agroecosystems were located at a distance no greater than 245 m from a secondary forest. The distance between agroecosystems varied from 10 to 120 km, except for ATI₅ and ATI₆ that were located in the same sampling site. A detailed description of each habitat is given in Table 1. All agroecosystems included a secondary forest (control) within the same site.

Figure 1

Table 1

The avifauna present in each habitat type (type I and II agroecosystems and secondary forests) was assessed by mist net sampling, which has been previously used in agroecosystem and secondary forest studies (Blake and Loiselle, 2001; Barlow et al., 2007; Castaño-Villa et al., 2014). Between November to December of 2015 and January to April of 2016, we established 12 capture points within each site, and at each point, a mist net (12 × 2.5 m × 36 mm) was extended for 10 hours. The nets were randomly installed at each site and operated between 600 h and 1730 h. Each site was visited for four to five days until completing 120-net hours. Within each habitat type, mist nets were extended in an area of 0.1 ha. The total sampling effort for ATI was 960-net hours, 600-net hours for ATII, and 960-net hours for SF. The nets were not operated under rainfall, wind, or intense cold or heat. The birds captured were individualized through a small cut on the first tail rectrix in order to prevent counting more than once, and then released in the capture site. Birds were identified using the identification guide of Birds of Northern South America (Restall et al., 2007). Taxonomic classification of bird species was done according to Remsen et al. (2017).

We compared species richness, abundance and similarity between the two agroecosystem types and their corresponding secondary forests, in order to describe the contribution of agroecosystems to avifauna conservation. Species were categorized based on the criterion of sensitivity to disturbance (high, moderate, low), according to Stotz et al. (1996). Species with a high sensitivity to disturbance are known to be most affected by anthropogenic habitat modification (Castaño-Villa et al., 2014), so they are considered vulnerable and relevant to conservation at a local level. To determine if bird species richness statistically differed between the three habitats studied, we graphed lower and upper confidence intervals at 84% of the estimated species richness (S _est), using Mao Tao SD values, according to MacGregor-Fors and Payton (2013) and subsequent applications (Hanula et al., 2015; Fontúrbel et al., 2016). Mao Tao SD values were obtained from rarefaction analysis using EstimateS version 9.1.0 (Colwell, 2013). When the confidence intervals did not overlap, these were considered statistically significantly different with an alpha of 0.05. We considered the capture points within each site to be replicas (12 points per site). Habitats were compared by the same number of replicates (60), which was the minimum number of replicates obtained for ATII. Additionally, observed species richness was compared between type I and II agroecosystems, according to sensitivity to disturbance, through Fisher's Exact Test. The number of individuals captured within each habitat (i.e. abundance of all species captured, standardized for the sampling effort), shown as medians, was compared through of a Generalized Linear Model (Poisson distribution). Abundance data referred to the number of captures, not to real abundance, which is unknown. The differences in bird assemblage similarity (in terms of species abundance) between habitats were assessed by a One-Way Analysis of Similarity (ANOSIM), which constitutes a non-parametric permutation test (Clarke, 1993). For this analysis, a Bray-Curtis distance was used, which has been previously used in other studies to compare bird assemblages (Barlow et al., 2007; Castaño-Villa et al., 2014). The ANOSIM test was carried out with PAST 2.15 (Hammer et al., 2001). Additionally, we used a Non-metric Multidimensional Scaling analysis (NMDS) to visualize differences in the composition among habitat types. GLM and NMDS analyses were carried out on R version 3.4.3 (R Development Core Team, 2017).