Common opossum population density in an agroforestry system in Bolivia

The common opossum, Didelphis marsupialis thrives near human settlements. Understanding its ecology could help planning management decisions about this species, especially in anthropogenic landscapes. Yet, there are no density estimations for this species throughout its distribution range in Bolivia. We estimated the density of D. marsupialis in a rural agricultural community, where agroforestry plantations and fallows cover most of the land. We counted individuals in line transects and used DISTANCE software to calculate density. We covered a total of 70.21 km in 143 night counts, obtaining 38 records of D. marsupialis. We estimated a density of 0.30 individuals ha-1 (SE = 0.062; range: 0.20 0.45 individuals ha-1), with a mean encounter rate of 0.54 individuals km-1. Encounter rate varied between habitats, with mean values of 0.20 in secondary forests and 0.64 in agroforestry plantations. Our density estimate is near the lower range of previously reported values for the common opossum in other countries. We argue that our results may reflect the response to the availability of food resources and predation pressure in agroforestry plantations.


INTRODUCTION
The common opossum, Didelphis marsupialis Linnaeus 1798, is a medium-sized nocturnal mammal that occurs from Mexico to northeastern Argentina, as well as Trinidad and the Lesser Antilles, from sea level to about 2000 m (Rocha and Rumiz 2010). It is a common species across its distribution range, and thrives well near human settlements and croplands (Rocha and Rumiz 2010). It feeds mainly on insects, fruits, small vertebrates, and occasionally on pollen and nectar (Julien-Laferriere and Atramentowicz 1990;Cáceres 2002;Ceotto et al. 2009). The common opossum is used in some regions as game meat, but it can also be considered as a harmful species by farmers, because it preys on poultry (Bezerra-Barros and de Aguiar-Acevedo 2014; Astua de Moraes et al. 2016).
Population density is one of the most important ecological parameters to understand ecosystems and to apply management strategies (Ojasti and Dallmeier 2000;Pramod et al. 2012). Despite the common occurrence of D. marsupialis in human-influenced tropical habitats, there are no recently published density estimates of this species in Bolivia (Rocha VOL. 50(3) 2020: 246 -251 ACTA AMAZONICA and Rumiz 2010). Most estimates were carried out in Central America during the 1980s and 1990s, using trap grid methods (e.g. Atramentowicz 1986;Sunquist et al. 1987;Julien-Laferriere 1991). Trap capture (live-trapping) is the most used method for studying marsupials because they are supposedly less prone to under-estimating density and abundance, as compared to other methods such as transect counts. The latter method depends on the probability of detection (i.e. the capacity of detecting an animal), which decreases with the distance between the observer and the animals (Thomas et al. 2013;Fauteux et al. 2018). Line transect counts are mainly used for estimating the density of large animals (Thomas et al. 2013;Hoffmann et al. 2010). Yet, comparative studies suggest that line-transect counts may be as reliable as grids for evaluating Didelphis marsupialis populations (Emmons 1984), and have been used in other studies on mammals (e.g. Chiarello 2000).
Density estimations for D. marsupialis in agroecosystems are absent in the literature. Agroecosystems in the lowland tropics are highly diverse, varying from monocultures (simplified crops) to diversified agroforests. Agroforests are a common type of agriculture in the Alto Beni region of Bolivia (Benavides et al. 2017). It has been suggested that these crop systems offer suitable foraging and dispersing habitats for several mammalian species, likely due to their structural complexity, which resembles a forest more than conventional crops, and provide food resources throughout the year (Gallina et al. 1996;Nyhus and Tilson 2004;Perfecto and Vandermeer 2008;Altieri 2009;Cassano et al. 2012). If resources for the common opossum are abundant within agroforestry plantations, it is reasonable to expect high population densities. We report the first density estimate of the common opossum in an agroforest at the piedmont area of the Bolivian Amazonian region. As information on our study area is very scarce, we also provide encounter rates for other mammalian species for which a density estimate was not possible.

MATERIAL AND METHODS
The study was conducted at Remolinos (15°41' -15°42´S, 67°04' − 67°06'W), a rural agricultural community within the Palos Blancos municipality, La Paz department, Bolivia. The community encompasses about 2000 ha of mountainous terrain, at a mean elevation of 450 m. Undisturbed vegetation and secondary forest patches are representative of Amazonian and sub-Andean forests, with species such as Brosimum sp. (Moraceae), Hura crepitans (Euphorbiaceae), Otoba parvifolia (Myristicaceae), among others (Quintana and Vargas 1995). Trees are younger in the secondary forest, around 20 -30 years old (C. Benavides, pers. obs). However, agroforestry plantations and fallows cover most of the community's lands. Fallows are dominated by various families of shrubs and pioneer tree species, such as Cecropia spp., Ochroma pyramidale, and various species of Inga (PIAF-El CEIBO 2002). Crops are varied and mainly under successional agroforestry systems, combining rice, corn, watermelon, banana, squash, manioc, peanuts, papaya and different species of Citrus. Cacao (Theobroma cacao) is the most common commercial crop, and 40% of its production is organic (Jacobi et al. 2014). Trees such as Swietenia macrophylla, Amburana cearensis, and Cedrela odorata (Quintana and Vargas 1995) are being regrown within crop areas with trees up to around 20 years old. Several species of palms and fruit trees (e.g. Attalea phalerata, Averrhoa carambola) are kept and planted in home orchards for local consumption (Benavides et al. 2017).
Between March and November 2012, we used line transect counts to estimate the density of pacas (Cunniculus paca) for another project (Benavides et al. 2017), following the DISTANCE sampling approach (Laake et al. 1996;Ruette et al. 2003;Thomas et al. 2013). During the counts, we also spotted other mammalian species, including the common opossum, which was the most frequently observed animal. This species was easily detected and individuals were less wary than other animals (e.g. Dasypus novemcinctus).
We established transects based on the participation of farmers in the project. We set nine different transects, ranging from 0.25 to 1.5 km in length (Table 1). Six transects were placed in agroforestry systems, one traversed a small portion of a grassland (with patches of trees) and a secondary forest, and the last two were placed through secondary forest patches. The overall size of the sampling area was ~650 ha. Agroforestry plantations represented 70% of the sampling area and the other 30% was considered as secondary forest, including the transect that traversed the grassland (Figure 1.). The total transect lengths were 2.5 km in agroforestry plantations and 2.3 km in secondary forests. The survey effort in agroforestry plantations totaled 54 km (112 surveys), whereas the effort in secondary forests amounted to 15.4 km (31 surveys) ( Table  1). Transects were separated at least 250 m from each other; based on the home-range size of C. paca (i.e. ± 4 ha, Beck-King et al. 1999). Considering that the home range for common opossums varies between 0.2 and 16.3 ha (Sunquist et al. 1987;O'Connell 1989;Cáceres and Monteiro-Filho 2001;Sanches et al. 2012), the distance between transects allowed for the possibility of recording the same individual twice in the same transect. However, only four or five transects were surveyed each night, following a different sequence each time. Additionally, we used data from transects on our way back to the camp only when no observations were made on the first passage and, as pointed above, common opossums usually walked away slowly when detected.
In total, 70.21 km were covered in 143 night counts between sunset and 5:00 a.m., which includes the activity period of D. marsupialis (Sunquist et al. 1987). Surveys were done walking at a speed of <1 km h -1 , avoiding making noise, while searching for individuals on both sides of the trail, stopping from time to time to listen. When an animal was spotted, we identified the exact site where it was when first observed and then measured the perpendicular distance to the line transect using a measuring tape. All night counts were carried out by two of the authors (C. Benavides and A. Arce), after about two months of training. Density and total encounter rate were estimated using the DISTANCE software (v.6.2) (Laake et al. 1996). We allowed DISTANCE to choose the best model under the Akaike Information Criteria (AIC). Although we also provide encounter rates for other mammalian species, we do not discuss those in detail because of small sample sizes, but we consider those data of scientific value since they are the only data available for mammals within this region in Bolivia.
For means of comparison of our results, we searched the open access literature on D. marsupialis density on Google Scholar®. We considered information on density estimation methods, as well as the country and habitat where the study was carried out (Table 3).

RESULTS
Overall we obtained 38 sightings of D. marsupialis and an encounter rate of 0.54 (95% CI 0.20 − 0.45) individuals per km. Animals were spotted at a mean distance of 7.3 m from the transect. Relative density (encounter rate) of D. marsupialis was 3.6 times higher in agroforestry plantations (0.64 individuals km -1 ) than in secondary forests (0.20 individuals km -1 ) (see Table 1). The mean distance of observation was 7.28 m in agroforestry plantations and 0.33 m in secondary forest. DISTANCE chose the Half-normal key detection function model, with k(y) = exp (-y 2 /2σ 2 ), where "y" is distance and "σ" is the model parameter, and estimated a global density of 0.30 individuals ha -1 [Standard Error (SE) = 0.062), 95% CI = 0.20 − 0.45 individuals ha -1 ]. The encounter rates for opossums and other four medium-sized mammals ranged from 0.05 for Dasypus novemcinctus to 0.41 individuals km -1 for Cuniculus paca (Table 2).   Table 3 for details), but below the mean value from all the previous studies (0.9 individuals ha -1 ). Our density estimate is similar to those found in a mature tropical forest disturbed by natural and selective logging in Peru by Emmons (1984). Agroforests can be similar in vegetation richness and structure to undisturbed forest, and several mammal species have been observed to use this human-influenced habitat as foraging sites and migration routes (e.g. Cassano et al. 2012). For a generalist species like the common opossum, we expected higher densities in a habitat that may offer rich and stable resources due to the high diversity of fruiting species with year-round production (Harvey et al. 2006;Harvey and González Villalobos 2007;Pardini et al. 2009;Cassano et al. 2012;Caudill et al. 2015).
It has been suggested that density estimates based on line transects may be biased when the species studied are easily frightened and difficult to be detected (e.g. when vegetation cover is dense) (Chiarello 2000), affecting the detection function of line-transect counts (Thomas et al. 2013;Fauteaux et al. 2018). In general, trap captures (live-trapping) have been preferred over line transects to study marsupials such as Didelphis (e.g. Sunquist et al. 1987;Malcolm 1990; Julien-Laferrier 1991; Table 3). However, Emmons (1984) showed that night counts and live trapping result in similar abundance estimates for D. marsupialis. Also, the common opossum is a conspicuous species, and was rarely frightened by human presence at our study site, suggesting they may be somewhat used to human presence and are not usually hunted locally. It is notable that our encounter rate was much higher in agroforestry plantations than in secondary forests. Visibility of animals was lower in secondary forest, and our survey effort was mainly concentrated in agroforestry plantations, owing to the higher access difficulty to the secondary forest patches. Both circumstances likely decreased the probability of observing common opossums in secondary forest, and may have led to the underestimation of density estimates in these habitats. Nonetheless, the apparent selection and higher frequency of use of agroforests over secondary forests by common opossums in our study area deserves further investigation. We propose that opossums may be more abundant in agroforestry plantations as a function of higher resource availability (food) and suitable vegetation cover as compared to forest patches.
The relatively low overall density of D. marsupialis estimated for our study site may be also affected by the presence of domestic dogs (Canis familiaris) and other less common predators, such as ocelots, Leopardus pardalis (Moreno et al. 2006;Foster et al. 2010;Cassano et al. 2012;Cassano et al. 2014). Local residents in Remolinos commonly employ domestic dogs for hunting, but only occasionally kill common opossums for medicinal purposes (C. Benavides, pers. obs). However, when dogs are left alone in crops, they hunt to eat, and we confirmed that their prey include the common opossum by observing hunting events. Common opossums are also killed by farmers when found near their homes, because they occasionally prey on poultry. However, they are not considered as a great problem by farmers (C. Benavides, pers. obs.). Future research should also be focused on the effect of predation on the common opossum populations.

CONCLUSIONS
We estimated the density of the common opossum, Didelphis marsupialis using line transects and distance sampling in agroforestry plantations in the Amazonian piedmont in Bolivia as 0.30 individuals ha -1 . The estimate lies below the average of the values already reported in literature. The density of D. marsupialis may be responding to a combination of resource availability (food and vegetation cover) and predation pressure at our study site.

ACKNOWLEDGMENTS
This study was funded by Ecotop S.R.L. We thank Fortunato Velásquez for all his help and support. We thank all the people at Remolinos for receiving us kindly. Comments from two anonymous reviewers helped to improve an earlier version of the manuscript. Finally, we thank Jorge Salazar-Bravo, Ezgi Ogutcen, Preston Mcdonald, two anonymous reviewers and the editor for improving previous versions of this manuscript.