Predicting geographic distributions of Phacellodomus species ( Aves : Furnariidae ) in South America based on ecological niche modeling

Phacellodomus Reichenbach, 1853, comprises nine species of Furnariids that occur in South America in open and generally dry areas. This study estimated the geographic distributions of Phacellodomus species in South America by ecological niche modeling. Applying maximum entropy method, models were produced for eight species based on six climatic variables and 949 occurrence records. Since highest climatic suitability for Phacellodomus species has been estimated in open and dry areas, the Amazon rainforest areas are not very suitable for these species. Annual precipitation and minimum temperature of the coldest month are the variables that most influence the models. Phacellodomus species occurred in 35 ecoregions of South America. Chaco and Uruguayan savannas were the ecoregions with the highest number of species. Despite the overall connection of Phacellodomus species with dry areas, species such as P. ruber, P. rufifrons, P. ferrugineigula and P. erythrophthalmus occurred in wet forests and wetland ecoregions.


Introduction
Distribution and abundance of bird species depend on climatic characteristics and variability within temporal and spatial dimensions (WATKINSON et al., 2004).The influence of climatic variables on the distribution of organisms may be investigated by ecological niche modeling (ENM).Several ENM methods have been used to estimate the potential distributions of species based on occurrence points and on the analysis of environmental variables (TSOAR et al., 2007).Although EMN has been widely used to predict the distribution of bird species (ANCIÃES; PETERSON, 2009;CORREA et al., 2010;ECHARRI et al., 2009;FERIA;PETERSON, 2002;GRAHAM et al., 2010;LEE et al., 2010;MARINI et al., 2009;MARINI et al., 2010a, b;SCHIDELKO, et al., 2011;STRUBBE;MATTHYSEN, 2009), no studies examining ecological niches of Phacellodomus spp.(Aves, Furnariidae, Synallaxinae) have yet been developed.Such information and knowledge may be a great help in planning bird conservation strategies and may also be applied in health sciences since Phacellodomus nests are the habitats for some triatomine species which are vectors of Chagas´ disease (DI IORIO; TURIENZO, 2009;GURGEL-GONÇALVES et al., 2012).
Phacellodomus species occur in South America in open and generally dry areas (VAURIE, 1980).Some species, such as P. ruber (Vieillot, 1817), are associated with more humid environments.Phacellodomus striaticeps (Orbigny & Lafresnaye, 1838) and P. maculipectus Cabanis, 1883, occur in mountain areas at 2500-4200 m altitude.Phacellodomus rufifrons (Wied, 1821) has a discontinuous distribution (RIDGELY;TUDOR, 1994).These variations in the geographic distribution of Phacellodomus may be explored by using spatial analysis methods that consider ecological preferences of each species to better understand the distribution patterns.Current study aims at predicting the geographic distributions of Phacellodomus species in South America.
Records of nine Phacellodomus species that could be referenced to geographic coordinates with a reasonable degree of confidence (i.e., with an uncertainty of < 5 km, to a precision of 0.01°) were compiled.Records were georeferenced based on http://www.fallingrain.com/world.An occurrence data sample size criterion of 20 unique latitudelongitude points per species were set as a minimum to permit robust ENM development, based on previous analyses (STOCKWELL; PETERSON, 2002).

Ecological niche modeling
ENM uses associations between environmental variables and known occurrences of species to identify environmental conditions where populations may be maintained (TSOAR et al., 2007).Environmental datasets consisted of 'bioclimatic' variables characterizing climates during the 1950-2000 period, drawn from the WorldClim data archive (HIJMANS et al., 2005).The environmental database used in current analyses covers all South America at a spatial resolution of 2.5' (5 km).To avoid confounding effects of calibrating models in an overly dimensional environmental space (PETERSON; NAKAZAWA, 2008), only a subset of the 19 'bioclimatic' variables in the climatic data file was chosen: annual mean temperature, maximum temperature of the warmest month, minimum temperature of the coldest month, annual precipitation, precipitation of the wettest month, and precipitation of the driest month.According to Schidelko et al. (2011), these parameters mainly affect thermophile passerines in tropical and subtropical environments and represent dimensions in which potential physiological limits may be manifested.
Occurrence data were separated into two sets, one for model calibration (75% of points) and the other for model evaluation (25% of points).ENMbased distribution maps were produced using the maximum entropy method (PHILLIPS et al., 2006) as implemented in Maxent v. 3.3.3selecting default parameters which are considered appropriate for most situations (PHILLIPS; DUDÍK, 2008).Maxent software for species habitat modeling has been shown to perform well, in particular when sample size is low (HERNANDEZ et al., 2006).To emphasize the fact that omission error takes considerable precedence over commission error in niche modeling applications, a modified version of the least training presence thresholding approach was used (PEARSON et al., 2007).Specifically, models were thresholded at the suitability level that included 90% of the occurrence records of each species used in model calibration (PETERSON et al., 2008).All maps were edited using ArcView version 3.3 (ESRI, 2002).

Data analysis
The quality of the models generated was evaluated using the Receiver Operating Characteristic (ROC) curve, which relates two characteristics of model performance, namely, sensitivity and specificity (PHILLIPS et al., 2006).Sensitivity is defined as the proportion of true presences in relation to the total number of presences predicted by the model.It is also a measure of the absence of omission error; good quality models should show greater sensitivity.Specificity is the proportion of true absences in relation to the total number of absences predicted by the model; 1 -specificity is a measure of the degree to which predicted areas exceed observed occurrence.The area under this curve (AUC) provides a measure of model performance, ranging between zero and one: AUCs close to 1 indicate high performance, while readings around 0.5 indicate poor performance (ELITH et al., 2006).
The predictive power of the models was also compared with a random null hypothesis.The developed model was checked to see whether test points fell into areas predicted to be present more often than expected at random, given the overall proportion of pixels showing predicted presence vs. predicted absence for that species.In addition to the model significance (departure from random predictions), model accuracy was assessed by examining the proportion of test points falling into regions of predicted presence (ANDERSON et al., 2002).
Using the Jackknife test and adopting procedures described by Pearson et al. (2007), variables that most influenced the distribution of the species of Phacellodomus were identified.Additionally, an analysis was made by reporting the presence of Phacellodomus species within the terrestrial ecoregions of Latin America (WWF, 2012).An intersection between the distribution range of Phacellodomus species and the terrestrial ecoregions shapefile was designed with ArcView v 3.3.
The models indicated higher climatic suitability for the occurrence of Phacellodomus species in opendry areas when compared to rain forest areas.Jackknife tests showed that annual precipitation and minimum temperature of the coldest month were the variables that most contributed to the models.
Although Phacellodomus species occurred in 35 ecoregions and seven macrohabitats of South America, the Chaco and Uruguayan savannas were the ecoregions with the highest number of species (Table 1).In fact, grasslands and savannas were the main macrohabitats of Phacellodomus species, even though at least 14 (40%) of the 35 ecoregions where Phacellodomus species occurred were wet forests and wetlands.
Ecological niche models indicated distant and discontinuous areas when compared to the species´ actual distribution as, for instance, P. erythrophthalmus, P. ferrugineigula, and P. striaticollis (along the Andes in Bolivia and Peru); P. maculipectus and P. sibilatrix (in the Caatinga of northeastern Brazil).These areas should have been predicted by the models because they had the climatic characteristics of the areas where the species actually occurred.The above overprediction error in ENM approaches derives from potentially habitable regions correctly predicted as presence, but probably outside the species´ dispersal area and, consequently, not inhabited.The examination of congruence or discordance between predicted and actual distributions evaluates the roles of ecological and historical factors in determining the species' eographic distribution (ANDERSON et al., 2002).
Figure 1.Geographic distributions of Phacellodomus species in South America predicted by ecological niche modeling (Maxent).The areas in gray tones show the distribution predicted according to climate suitability: light gray (low), dark gray (high).The white areas represent the absence predicted by the models and the squares represent the occurrence records used in modeling.In the case of P. rufifrons, different subspecies are indicated by different symbols: P. r. specularis (square), P. r. sincipitalis (circle), P. r. rufifrons (triangle), P. r. inoratus and P. r. castilloi (dotted circle), P. rufifrons peruvianus (dotted square).In the case of P. dorsalis, only occurrence points were presented since it was not possible to predict the distribution due to the sample size criterion of 20 unique latitude-longitude points.Among the species studied, P. dorsalis occurred in a single ecoregion (Peruvian yungas), a mountainous area of Peru over 2,500 m.According to Vaurie (1980) and Ridgely and Tudor (1994), ecological data on P. dorsalis are scarce.This species is endemic to Peru, apparently restricted to the Marañon Valley, from southern Cajamarca to northern La Liberdad, but probably also in the neighboring Amazonas Department.More information on the occurrence of this species is needed Phacellodomus erythophthalmus occurred in Uruguayan savannas, Serra do Mar coastal forest, Araucaria forests, and Bahia interior forests.According to Simon et al. (2008), P. erythophthalmus has not been observed in Bahia since its description in 1821: our models clearly indicate that the interior forests of Bahia do not seem to have suitable climatic conditions for its occurrence.The potential distribution of P. ferrugineigula was very similar to that of P. erythophthalmus.Although P. ferrugineigula was long considered a subspecies of P. erythophthalmus, Simon et al. (2008) presented morphological evidence that it should be acknowledged a valid species.Apparently, the two species occur in sympatry in the Brazilian states of São Paulo, Rio de Janeiro, and Minas Gerais.Our results not only confirm overlapping potential, but even suggest that the area of co-occurrence could be even larger, including the southern Brazilian states of Rio Grande do Sul, Santa Catarina and Paraná.
The distribution of P. maculipectus lies in southwestern Bolivia and northwestern Argentina, in forests dominated by Alnus, Podocarpus and Miconia, at altitudes between 1,800 and 3,100 m in Bolivia and between 1,900 and 2,900 m in Argentina (BARNETT et al., 1998).
Phacellodomus ruber is widely distributed across 11 ecoregions.The species is common in the flooded areas of the Brazilian cerrado, associated with Mauritia flexuosa palm trees, where it builds its nest (GURGEL-GONÇALVES; CUBA, 2011;GURGEL-GONÇALVES et al., 2012).
Phacellodomus rufifrons, considered a single, polytypic species, is the most widespread species of Phacellodomus in South America, covering at least 19 ecoregions, such as La Costa xeric shrublands, Cerrado and Serra do Mar coastal rainforest.Although quantitative data are lacking, the subspecies may be fully detectable by plumage characteristics (RIDGELY;TUDOR, 1994;VAURIE 1980).Records of each subspecies and associated distributional estimates indicated little overlap between their distributions.Future studies applying ecological niche differentiation and phylogeography could provide evidence for the separation of some of these subspecies.Phacellodomus striaticeps occurred in seven ecoregions, the only species of Phacellodomus with records on the Central Andean wet puna and Central Andean dry puna.Phacellodomus striaticollis and P. sibilatrix occurred in temperate and tropical savannahs and grasslands with a similar distribution.Our results indicate that the area of co-occurrence of these species may be larger than that described by Vaurie (1980).It may even include northeastern Argentina, southern Brazil and Uruguay.As predicted in our study, P. sibilatrix was only recently recorded in Brazil (BELLAGAMBA; DE OLIVEIRA, 2012).
The database spatial resolution in current study could limit the accuracy of niche models.For example, some species occur in habitat patches and only a few meters would separate the limits of their habitat, as the swamp-occurring P. ferrugineigula.Coarse raster resolutions fail to capture sharp environmental gradients that occur at terrestrial habitats, while ecological niche models built from low-resolution data sets tend to overestimate species ranges (SEO et al., 2009).However, the effects of these scale-related uncertainties will differ according to the scale of analysis (WIENS et al., 2009).Some uncertainties may be averaged out if the grain size of predicted distributions is large, as in current study, where the geographic distributions of Phacellodomus species were estimated on a continental scale.

Conclusion
Current study updated the geographic distributions of Phacellodomus species in South America.Despite the overall connection of Phacellodomus species with dry areas, the species P. ruber, P. rufifrons, P. ferrugineigula and P. erythrophthalmus occurred in humid forests and wetland ecoregions.The resulting ecological niche models showed high specificity and sensitivity, which indicated a high predictive ability and summarized the ecological conditions for the occurrence of Phacellodomus species in South America.