Large-seeded, animal-dispersed (LSAD) trees include some of the most valuable and threatened species in the tropics, but they are chronically underrepresented in regenerating forests. Toucans disperse many LSAD species, so attracting toucans to regenerating forests should help re-establish more diverse tree communities. We ask: (1) What constitutes suitable toucan habitat in premontane southern Costa Rica? (2) How much do small-scale restoration strategies influence toucan visitation compared to landscape-scale habitat suitability outside of restoration sites? (3) How well does toucan visitation predict the richness of LSAD tree species recruiting into regenerating forests? We combined habitat suitability models with long-term toucan observations and comprehensive tree recruitment surveys to assess these questions in a multi-site forest restoration experiment. Restoration treatments included tree plantations, natural regeneration, and applied nucleation. Habitat suitability obtained by modeling for three sympatric toucan species was predicted by elevation and the extent and age of landscape forest cover. Within suitable landscapes, toucans visited areas restored via tree planting ≥5 years sooner and ≥2× more often than plots restored via natural regeneration. Tree plantations in suitable toucan habitat at the landscape scale had LSAD tree recruitment communities that were 2-3× richer in species than plantations in poor toucan habitat, and 71% (15/21) of all recruiting LSAD species were found only in plantations where landscape habitat was suitable for the largest toucan, Ramphastos ambiguus. Results support a multi-spatial-scale model for predicting toucan-mediated dispersal of LSAD trees. Tree planting increases toucan visitation and LSAD tree recruitment, but only within landscapes that represent suitable toucan habitat. More broadly, habitat suitability modeling for key seed dispersers can help prioritize restoration actions within heterogenous landscapes.

1 - Experimental restoration design: We selected 11 forest restoration sites established in 2004-2006 on former agricultural lands that had been farmed for ≥18 years. Study sites spanned a deforestation gradient ranging from 13-82% tree cover within 1 km and represented a randomized block design. Each site contained three 0.25-ha plots, which were separated from one another by ≥5 m. Plots were randomly assigned one of three experimental treatments: (1) natural regeneration, (2) applied nucleation, or (3) plantation. Natural regeneration plots were left to recover without tree planting. Applied nucleation plots were planted with six patches of tree seedlings (86  seedlings plot-1). Plantations were planted with rows of tree seedlings throughout (313 seedlings plot-1). All plots were fenced to prevent cattle incursion and were hand-cleared with machete for 2.5 years. Study sites were separated from one another by ≥0.7 km. Tree seedlings planted in applied nucleation and plantation plots consisted of two widely-planted native species, Terminalia amazonia (Combretaceae) and Vochysia guatemalensis (Vochysiaceae), and two naturalized legumes, Inga edulis and Erythrina poeppigiana (both Fabaceae). Inga produces an indehiscent fruit with an edible pulp that is consumed by primates, Terminalia and Vochysia produce winged seeds adapted for wind dispersal, and Erythrina produces brown seeds with no obvious dispersal mechanism. None of the planted tree species produce fruits regularly eaten by toucans. 

2 - Focal avian LSAD dispersers: Among several groups of large-bodied (>100 g) fruit-eating birds in southern Costa Rica, we investigated the three toucan species (Piciformes, Ramphastidae) that occur in the region: Aulacorhynchus prasinus, Pteroglossus frantzii, and Ramphastos ambiguus ssp. swainsonii. Toucans are consistently important seed dispersers for a variety of neotropical trees; collectively, they can remove as much as 65% of a tree’s seed crop. Despite being dependent on forest for nesting and foraging, they have the potential to fly long distances across open country, making them effective long-distance dispersal agents. We excluded large-bodied Cracids (curassows, guans, and chachalacas), which are either not found (or very rarely found) in regenerating forests in the study area, or are not forest-dependent (e.g., the chachalaca Ortalis cinereiceps) and may be less likely to disperse forest tree seeds. We also excluded Psittacids (parrots and parakeets), which sometimes disperse viable propagules over long distances and were often observed flying overhead, but were rarely observed in restoration sites. Finally, we excluded some large-bodied omnivores (e.g., the crake Aramides cajaneus) that occasionally eat fruit as part of a diet consisting primarily of invertebrates. 

3 - Toucan surveys: In 2011-2012, we quantified the occurrence of three toucan species in 55 forest fragments in our study region in order to develop habitat suitability models. The forest fragments represented two uncorrelated gradients in fragment size and forest cover within a 1000-m radius, stratified across elevation. The area covered by a 1000-m radius approximates the upper range of home range sizes reported in previous studies. The fragments included the 49 fragments in Kormann et al. (2018), plus six additional fragments which we added to make the sample more balanced altitudinally. Fragments were selected according to a stratified-random design, based on aerial-based forest cover maps (2-m resolution). A priori, we assigned fragments to a size class (small: <5 ha, large: >30 ha), and from both categories we selected fragments that represented a forest cover gradient (5 – 80%), included two elevational bands (880 – 1100 m a.s.l. and >1100 – 1500 m a.s.l.), and were spatially spread across the study area. Due to this selection procedure, all correlations between altitude, forest cover, and fragment size were low (all Pearson’s r <0.5). We assessed toucan abundance using 50 m fixed-radius point counts and stopping rule-based walkabouts, all performed by the same experienced observer (Jeisson Figueroa Sandí, JFS). Forest fragments were visited once between May and June (2011, 2012) in the early morning in a randomized order. Walkabouts lasted an average of 80 minutes in small fragments, and 155 minutes in large fragments. We performed 12-minute point counts; three in each small fragment and six in each large fragment. We also assessed toucan visitation three times per year (Apr-May, Jul-Aug, and Nov-Dec) in each of 11 restoration sites from 2010-2018. This monitoring was used to test habitat suitability model predictions and in turn predict large-seeded tree recruitment. Within each site, we actively searched each of three experimental plots for 20 min, walking along pre-established trails, and recorded when toucans were seen or heard within the plot. When possible, we also recorded the tree or plant species in which the bird was observed and its foraging and reproductive behavior. During each sampling effort, we visited sites and plots in random order. We made observations from 0600 to 0900 hours during mild weather, including light fog or mist but excluding hard wind or rain. A single skilled observer (JAR) conducted all surveys in restoration sites. 

4 - Seedling surveys: In February to April 2017, RAZ and JAR comprehensively surveyed seven 50 × 50 m tree plantation restoration plots for LSAD tree species recruitment. Species were classified as LSAD based on dispersal syndrome (zoochorous) and seed width (>10 mm; N = 21 species). We selected a 10-mm threshold for “large” seeds because this is approximately the maximum diameter that smaller birds such as tanagers and thrushes are capable of swallowing. We selected all LSAD species rather than filtering for those with documented toucan dispersal because toucans characteristically have broad diets and the specific diets of toucans in our region have not been quantified. All naturally-recruiting individuals ≥10 cm in height were recorded in each plantation. We did not survey LSAD tree species recruitment in applied nucleation or natural regeneration treatment plots, largely due to the challenge of thoroughly detecting all recruits within the dense grass cover found in parts of these treatments. 

5 - GIS variables: We used four local and four landscape-level environmental variables to build habitat suitability models for the three focal toucan species. At the local scale, we used: (1) elevation and (2) forest age (secondary vs. old-growth) at the point of occurrence; (3) the fragment area; and (4) the area of the fragment covered by old-growth forest. At the landscape scale, we used the area covered by forest irrespective of forest age within a buffer of (5) radius = 500 m and (6) radius = 1000 m around the point of occurrence. Further, we used the area covered by old-growth forest only within a buffer of (7) radius = 500 m and (8) radius = 1000 m around the point of occurrence. Elevation was derived from a digital elevation model (EROS 2017). Orthophoto-based forest cover maps were derived from Hadley et al. (2014) (2-m resolution), and forest age was assigned based on regional forest age classification maps based on aerial photography (Zahawi et al. 2015). We considered forest to be old-growth if it had been forested continuously from 1947-2015. We filled in forest age coverage gaps for about 10% of the study area using the Costa Rican National Forest Inventory (REDD/CCAD-GIZ-SINAC 2015).

See Read Me file.