Species occurrence records

All available records of C. salta were used, including the few references in the literature (just two records from Salta Province: 17 km N of La Caldera, -24.5030 S -65.3351 W, and 22 km N of La Caldera, -24.5011 S -65.3180 W: Huber 2000) together with our own records (Table 1). Most of the latter were obtained in an ongoing ecological study in different sites in the central portion of Salta Province using the Garden-Vacuum method (Bolger et al. 2000; Bell et al. 2002) to collect spiders on vegetation. Additional samples were obtained from other yungas sectors: Jujuy and Salta Provinces for the northern yungas, and Tucumán representing the southern part. Records were georeferenced either in situ, using a Map-60 Garmin-GPS, or with the use of different digital gazetteers available in the Internet (mainly Google Earth ©). This dataset was arranged to be used within a geographic information system (DIVA-GIS 5.4, Hijmans et al. 2005a). The complete dataset consisted of 21 point records (Table 1), but due to duplicate records from the same gridcell being removed by the software during the analysis, valid effective records were restricted to 15 points. The specimens examined were deposited in the following Argentinean institutions (abbreviations and curators in parentheses): Instituto para el Estudio de la Biodiversidad de Invertebrados, Universidad Nacional de Salta (IEBI, J.A. Corronca); Colección Nacional Aracnológica, Museo Argentino de Ciencias Naturales “Bernardino Rivadavia” (C. Scioscia and M. Ramírez); and Colección Aracnológica de la Cátedra de Diversidad Animal I, Facultad de Ciencias Exactas, Físicas y Naturales, Universidad Nacional de Córdoba (L.E. Acosta).

Table 2.

Summary of the bioclimatic profile of Chibchea salta:

Environmental data

Values of 19 bioclimatic parameters were extracted from the WorldClim database (Hijmans et al. 2005b) at a resolution of 30 arc-seconds (∼1 km²). These parameters (bioclimatic variables) fall into two broad categories — temperature or precipitation variables — and are listed in Table 2. Elevation data were available for reference purposes, but not used as a predictor by themselves.

Modeling methods

The geographical range of C. salta was modeled with two widely used, presence-only methods, Maxent and Bioclim. The Maxent algorithm (described in detail by Phillips et al. 2006) yields results ranging from 0 to 1, indicating relative suitability of a given grid cell (high values indicate a higher presence probability) (Graham and Hijmans 2006; Phillips et al. 2006). Since probability is continuous, a threshold needs to be set to separate suitable from not suitable gridcells; in this study, the “maximum training sensitivity plus specificity” threshold rule was applied (Liu et al. 2005). Other relevant settings of the software were used in their default values, including the convergence threshold (10^-5); maximum background points (10,000); maximum iterations (1,500); replicated run type (subsample), output format (logistic), and “auto features” activated. Resulting predictions were visualized and mapped by importing the ASCII files into DIVA-GIS 5.4 grid format (Hijmans et al. 2005a). Version 3.3.0 of the Maxent software was employed (Phillips et al. 2009). To estimate the relative contribution of each variable in the final model a jackknife analysis was applied as a built-in functionality of Maxent (Phillips et al. 2006). The Bioclim model was built using its implementation in DIVA-GIS 5.4 (Hijmans and Graham 2006; Ward 2007; Acosta 2008; Echarri et al. 2009). Bioclim is a frequency distribution based algorithm, which extracts values of each bioclimatic variable from all localities and arranges them in a cumulative frequency distribution. The set of values of all variables defines the bioclimatic profile of the species, delimiting the so called “envelope”, i.e. the climatic conditions that bound all occurrence localities (Guisan and Zimmermann 2000; Fischer et al. 2001; Walther et al. 2004; Ward 2007). In the potential distribution maps, gridcells are scored as suitable (if within the envelope; i.e. the presence of the species can be expected) or unsuitable (if outside the envelope) (Acosta 2008). The “most limiting factor” analysis, available in Bioclim, was applied to detect, for a given gridcell, which variable is most critical to the inclusion of that gridcell within the resulting envelope.

Model evaluation

For Bioclim, the original data set was split and a subset of 30% presence points was set apart as a test sample. Pseudo-absence points were generated from the background using DIVA-GIS. The model was then run using the remaining 70% of the original presence data (training sample), randomly resampled in 20 repetitions (Maes et al. 2005; Pearson 2007; Acosta 2008; Echarri et al. 2009). Models obtained in these repetitions where overlaid to get a first visual evaluation of their mutual consistence and with the model built using the full dataset. Subsequently, the accuracy of the model was evaluated by calculating the AUC (area under curve) in a receiver operating characteristic plot, and the maximum Kappa (max-k). AUC values vary from 0.5 (model not better than random) to 1.0 (perfect accuracy as indicative that the model can discriminate perfectly between presences and absences of records); the max-k values over 0.75 are deemed to be excellent (Louto et al. 2005; Graham and Hijmans 2006; Randin et al. 2006). In the case of Maxent models, the program routinely calculates the AUC for each run.

Additional localities sampled

To empirically test whether localities in which the species was proven to be absent are correctly classed by the models as unsuitable, additional samplings were carried out in sites that are close to the record localities but correspond to a different ecoregion, the Chaco thorn-forest. These extra samples were obtained in central Salta using the above mentioned Garden-Vacuum method (i.e. with identical sampling effort). They consisted of 11 sites (Figures 9–10): General Güemes (three sampling sites at -24.6562 S -65.0035 W; -24.654 S -64.9917 W; -24.653 S -64.9892 W), along Juramento River (three sites: 25.0894 S -65.0025 W; -25.1345 S -65.0093 W; -25.1676 S -64.9861 W), near Cabra Corral dam (three sites: -25.1219 S -65.0361 W; -25.1205 S -65.0574 W; -25.1209 S 65.0921 W), south of La Merced (-25.0501 S 65.4962 W), and Castellanos (-24.7194 S 65.4367 W). The Chaco ecoregion is characterized by xeric and semi-deciduous forests, with shrub and herbaceous strata as well; all localities except one (General Güemes) belong to the “Sierra chaco” sub-ecoregion, extended on the basal slopes of the mountains. Sierra chaco interdigitates among yungas patches, having thus an important role in the connectivity of the northern and southern yungas sectors (Brown et al. 2006). As stated, in all 11 Chaco localities C. salta was not recorded (Figures 9–10).

New records of Chibchea salta

Nineteen new localities for this species, along with collection information and geographical coordinates are listed in Table 1.

Bioclimatic profile

The envelope of C. salta, as obtained with Bioclim, contains (with the default percentile threshold of 0.025) 66.7% of the presence records (i.e. 10 of 15 points fall within all possible bidimensional variable combinations). Table 2 summarizes the bioclimatic profile of this species by indicating minimum, maximum, and medians for all 19 bioclimatic variables. Quebrada de San Lorenzo, site 1 (K in Table 1 and Figure 1) has the highest score of extreme bioclimatic values of the envelope (n = 14). Many of these extreme values indicate this locality as the coldest and driest site (bel, 5–6, 8–14, 16, 18), as well as with highest isothermality and precipitation seasonality (bc3, 15); all these features were consistent with the elevation of the locality, the highest in the whole dataset (1905 m). Another locality with many extreme bioclimatic values (13 variables) is site U (San Javier, Tucumán), in this case showing the highest precipitation (bc12–14, 16–19), and lowest values of isothermality and temperature range (bc2–3, 7; see also Figure 2). Site T (road to Cabra Corral, in Salta Province), representing the record with lowest elevation and closest to the Chaco plains east of the mountains, is the warmest place (bel, 5, 8, 10–11) and has highest temperature seasonality (bc4; Figure 2), principally due to the high values in December and January.

Potential range

Models obtained with Maxent and Bioclim overall share a similar pattern (Figures 3–4), and both, in turn, match fairly well the yungas ecoregion (Figure 1). The highest probability (Maxent) or suitability (Bioclim) is consistently situated around the central area of Salta, i.e. where most records originate (Figure 1). The area with highest climatic suitability (0.80–1) recognized by Maxent is larger, extending from central Salta (Lerma Valley, San Lorenzo) to mid-southern Jujuy (Figure 3). In contrast, the highest suitability in the Bioclim model is much more concentrated in the Lerma Valley up to San Lorenzo, but reaches southern Jujuy only weakly (Figure 4). In both models the predicted range extends southwards, bordering northern and eastern slopes near Metán sierras. Bioclim and Maxent also agree in detecting one relevant potential area on the East, with no records of the species yet, close to the El Rey National Park and along the mountain group on the Jujuy-Salta border (Figures 1, 3–4). This area comprises several contiguous Sierras, interestingly bearing a large easternmost yungas isolate; it includes a small isolated mountain as well (Sierra de la Lumbrera), also recovered in isolation in the models (Figures 1, 3–4). Although with different intensity, both models project the species range north- and southwards. The projection into northern Salta (near the Baritú National Park area) is more continuous though with low probability in Maxent (Figure 3), whereas it is just represented by scattered dots in Bioclim (Figure 4). The southwards extension into Tucuman Province is separated from the core area in both models, and predicts one larger zone on the Northeast (Sierra de Medina) and a narrow strip on east faced slopes of the Aconquija-Calchaquíes range (Figures 3–4); only the Maxent model gives high probability on the surroundings of the sole record in that province (site U, Sierra de San Javier; Figure 1), while Bioclim remarkably ranks the latter as a marginal site bearing, as stated above, many extreme values for bioclimatic variables.

Limiting factors and relative importance of variables

Separate models were built in Bioclim with either temperature (be 1–11) or precipitation variables (be 12–19), to investigate their relative contribution to the final model (Figures 5–6). Results show that precipitation variables are clearly restrictive on the West because of the decrease of rainfall (following the increase of elevation), and partly on the South — mainly in the sub-xeric Lerma Valley in Salta Province. On the contrary, these variables are remarkably permissive towards the East into the Chaco ecoregion (part of Santiago del Estero, eastern Salta, even entering the Bolivian territory: Figure 5), showing that precipitation in this area would be enough for the species. Models obtained with temperature variables alone (be1–11) result in a much narrower area more similar to the final model, but still with a remarkable permisiveness into xeric mountain valleys (mainly the “Valles Calchaquíes” region, between Salta and Tucumán Provinces; Figures 1, 5) where no montane forest exists. Temperature variables are critical to the species range mostly on the southeastern and northern part of the predicted range (presumably related to a latitudinal gradient), and towards the East, in this case preventing geographical expansion into the warm (though otherwise humid enough) Chaco plains (Figures 5–6). For example, bc6 (minimum temperature of the coldest month) is the most limiting factor on the margins of the San Francisco Valley with Chaco vegetation (Jujuy; Figure 1), and a small move into the valley represents an increase of >3° C for this variable which makes it to fall outside the species envelope; bc6 is also restrictive in southern Salta and northern Tucumán where the increase can be of 5° C when moving apart from the core area into the Chaco. From another viewpoint, the jackknife analysis performed in the Maxent run indicated that three variables linked to temperature (bc5: maximum T° of warmest month, bc6: minimum T° of coldest month, and be 10: mean T° of warmest quarter) and one of precipitation (bc15: precipitation seasonality) are the most relevant when the range is considered as a whole, i.e. they show the highest gain when analyzed individually; bc6 is the variable that decreases the gain the most when omitted (Figure 8). These results emphasize the major importance of temperature variables in the final models, constraining the climatic niche within a quite narrow thermic tolerance (cf. Table 2).

Model performance and comparisons

Both Maxent and Bioclim performed well and their resulting modeled ranges are consistent to each other (Figures 3–4). The range predicted with Maxent is 35% larger than with Bioclim, showing that in the latter, suitable gridcells tend to concentrate more around record points. The overlay of the resulting maps of 20 runs in Bioclim using training data (70% of stochastic original records) are highly consistent with the predicted range using all points. Bioclim values of AUC were of high accuracy (0.80–0.97; mean = 0.93, 20 replicates using training data), while max-k; proved excellent performance in average (0.61–0.99; mean = 0.86, same number of replicates). In Maxent, AUC values resulting from the training data were excellent (0.998– 0.999; mean = 0.998, 20 repetitions). These AUC values should be taken with caution, however, since they might be over-rated due to the low number of records. Aside from the statistical meaning, the modeling proved to be in reasonable agreement with the expected range, especially considering the distribution of yungas formations and the type of environment inhabited by this species (mountain forests and rainforests, G.D.R. pers. obs.). Although this study did not focus primarily on absence data, samplings available for 11 Chaco sites yielded no specimen of C. salta outside the predicted range (Figures 9–10). All Chaco sites placed east of the mountains (i.e. in the Chaco plains proper; C–D on Figures 9–10) matched negative areas of the models that reflects a correct prediction in this sector. This is especially remarkable for the row of sites along Juramento River (D) that seems to follow a narrow unsuitable corridor between suitable gridcells. These observations strongly suggest the models ability to make correct predictions on negative areas. However, this ability did not stand the same for Chaco sites placed on the west, inside the Lerma Valley: with Maxent both sites (Castellanos and south of La Merced: A–B on Figure 9) fall not only within the predicted positive area, but with high probability; in the Bioclim model, the former site was classed among presence gridcells too, though the latter (actually placed on the very limits of the suitable area) did not.