|Environmental factors inﬂuencing the distribution
of the Lesser Rhea (Rhea pennata pennata) in southern Patagonia
Julieta Pedrana A,D, Javier Bustamante B, Alejandro Travaini A, Alejandro Rodríguez C, Sonia Zapata A, Juan Ignacio Zanón Martínez A and Diego Procopio A
ACentro de Investigaciones Puerto Deseado, Universidad Nacional de la Patagonia Austral, CONICET, Avenida Prefectura Naval s/n, 9050 Puerto Deseado, Santa Cruz, Argentina.
BDepartment of Wetland Ecology, & Remote Sensing and GIS Lab (LAST-EBD), Estación Biológica de Doñana,
CSIC, Américo Vespucio s/n, E-41092 Sevilla, Spain.
CDepartment of Conservation Biology, Estación Biológica de Doñana, CSIC, Américo Vespucio s/n, E-41092 Sevilla, Spain.
DCorresponding author. Present address: Recursos Naturales y Gestión Ambiental, Instituto Nacional de
Tecnología Agropecuaria (INTA), Estación Experimental Agropecuaria Balcarce, CC 276, CP 7620 Balcarce, Argentina. Email: firstname.lastname@example.org
Abstract. The Lesser Rhea (Rhea pennata pennata) has suffered a marked decline in numbers over recent decades, probably mainly as a result of livestock production and overhunting. Our aim was to investigate the factors that determine the distribution of Lesser Rheas in southern Patagonia and to generate a predictive regional distribution map. We surveyed
8000 km of roads and sighted 795 Lesser Rhea individuals or ﬂocks. We also estimated environmental predictors from remotely sensed data and analysed the occurrence of Lesser Rheas in relation to these predictors. The predictors we examined
were associated with four hypotheses explaining the distribution of Lesser Rheas: the persecution by ranchers, primary productivity, topography, and anthropogenic disturbance hypotheses. We built models for each hypothesis. Our results suggest that the distribution of Lesser Rheas is not negatively affected by persecution by ranchers, as the species is more
abundant in areas with high stocking levels of sheep, but is positively inﬂuenced by primary productivity and negatively by the proximity of human habitation. The resulting distribution map can be used as a management tool for government agencies and highlights the conservation priorities for managing this declining and emblematic species.
Additional keywords: Argentina, large-scale habitat models, ratite ecology, species distribution maps.
The Lesser Rhea (Rhea pennata) is a large, ﬂightless, cursorial ratite endemic to South America. It has three subspecies: R. p. garleppi from southern Peru through south-western Bolivia to north-eastern Argentina; R. p. tarapacensis in northern Chile; and R. p. pennata endemic to the shrub-steppes and semi-deserts of Patagonia (Folch 1992). The Lesser Rhea is currently categorised as near threatened globally (IUCN, Gland, Switzerland; http:// www.iucnredlist.org/apps/redlist/details/141087/0, accessed 19
October 2011). In Argentina all subspecies were hunted without restrictions until 1975 and as a result they have suffered marked and progressive decline and subspecies garleppi is in danger of extinction (Folch 1992). Furthermore, some populations of sub- species pennata are at risk of local and regional extirpation (Bellis et al. 2006) and the subspecies has been considered functionally extinct as prey for native carnivores in north-western Patagonia (Novaro et al. 2000).
A diverse range of factors are responsible for the decline and fragmentation of Lesser Rhea populations, most of them related to
human activities (Martella and Navarro, 2006). Suggested rea- sons for the decline of Lesser Rheas are: loss of suitable habitat owing to livestock production (Bellis et al. 2006; Martella and Navarro 2006; Barri et al. 2009a), legal and illegal hunting above sustainable levels (Novaro et al. 2000; Bellis et al. 2004; Barri et al. 2008), and the development of oil industry (Golluscio et al.
1998; Funes et al. 2000).
The distribution and abundance of the Lesser Rhea has mainly been studied using ecological (Navarro et al. 1999; Funes et al.
2000; Bellis et al. 2006; Barri et al. 2008) and social surveys
(Martella and Navarro 2006). Navarro et al. (1999) reported that the density of Lesser Rheas populations increases towards south- ern Patagonia. Therefore, it is thought that Santa Cruz, the southernmost province of Argentine Patagonia, could currently hold the largest wild populations. The potential distribution of the Lesser Rhea in Argentine Patagonia, as estimated from ﬁeld surveys, covers 670 000 km2 (Navarro et al. 1999). Never- theless, large areas within this region have suffered a progressive degradation as a result of maintaining sheep numbers above
sustainable stocking rates (Golluscio et al. 1998; Abraham et al.
Research on the factors affecting Lesser Rhea conservation in Patagonia has been over fairly small scales (Bellis et al. 2006; Barri et al. 2008, 2009a, 2009b). However, large-scale studies are necessary to detect patterns of distribution (Scott et al. 2002; Rodríguez et al. 2007) and to link them with regional processes involved in the decline of the Lesser Rhea. It has been shown that models of species distribution can produce maps that improve knowledge of the distribution of species (Bustamante and Seoane
2004; Gottschalk et al. 2007) and are helpful for predicting where a given species could occur (Suárez-Seoane et al. 2002; Seoane et al. 2003).
The aim of this study was to investigate the factors that determine the distribution of Lesser Rheas in southern Argentine Patagonia, and to generate a predictive distribution map at the regional scale. We also tested four hypotheses regarding the factors that inﬂuence the distribution of Lesser Rheas:
* The persecution-by-ranchers hypothesis states that Lesser Rhea distribution in the Patagonian steppes reﬂects a direct conﬂict with ranchers and indirect effect of competition with sheep. Between 1970 and 2005, sheep husbandry has declined across this region owing to a combination of natural cata- strophes and low prices for wool and meat (González and Rial
2004). As a result, sheep ranching was abandoned in many regions and, under this hypothesis, reduced persecution and competition could have allowed populations of Lesser Rheas to increase. This hypothesis predicts that the probability of ﬁnding Lesser Rheas is greater in areas with low stocking levels of sheep.
* The primary productivity hypothesis states that the distribution of Lesser Rheas is determined by the availability of more mesic environments within a semi-arid landscape. This hypothesis predicts a higher probability of ﬁnding Lesser Rheas in more productive environments, and close to wetlands, as these habitats offer abundant and better quality forage.
* The topography hypothesis states that the ruggedness of the terrain affects the distribution of Lesser Rheas, with the pre- diction that there is a higher probability of ﬁnding Lesser Rheas in ﬂat open areas, where detection of predators and a subse- quent quick escape are facilitated.
* The anthropogenic disturbance hypothesis postulates that un- regulated hunting and frequent disturbance is greater around centres of human activity, and predicts a lower probability of the occurrence of Lesser Rheas closer to places of high human density.
Materials and methods
The province of Santa Cruz (46–53oS, 65–73oW) has an area of
245 865 km2 (González and Rial 2004). The topography consists of hills and plains with vegetation dominated by a mixed steppe of
grass and shrubs rarely >0.5 m in height. The Nothophagus forests
that occur on the Andean slopes of the province were excluded from the study area. The climate is dry and cold, with strong predominantly westerly winds, and a marked gradient in precip- itation – decreasing from west to east – and temperature –
decreasing from north-east to south-west (González and Rial
2004). Since its colonisation by Europeans, sheep ranching has been the only economic activity across the study area until the
1980s, when oil extraction has increased markedly. Average human population density is 0.8 inhabitants km–2, concentrated in 11 urban areas with more than 2000 inhabitants. In the countryside, human density is <2 inhabitants per 100 km2 (INDEC; http://www.indec.gov.ar/webcenso/provincias_2/ provincias.asp, accessed 19 October 2011).
Field surveys and selection of sampling units
Road surveys were conducted during two consecutive spring–
summers (November 2004–February 2005, and December
2005–January 2006). We ﬁrst established which road segments would be surveyed by performing a stratiﬁed random sampling. We divided the study area into 12 regions, based on a combination of two environmental variables: mean NDVI (Normalised Dif- ference Vegetation Index) and mean slope. We used mean NDVI because we hypothesised that primary productivity could be an important driver of Lesser Rhea distribution, and mean slope because terrain irregularity could affect the detection of birds during surveys (Travaini et al. 2007).
Using a vector coverage of roads, we randomly selected road segments that totalled 4500 km of transects during the ﬁrst year.
To ensure all strata were properly sampled, 1500 km were equally distributed among survey strata (125 km on each stratum) and
3000 km were distributed proportionally to the area of each
stratum. During the second year, we randomly selected 3500 km of road segments not surveyed in the previous year. The strati- ﬁcation guarantees an unbiased distribution of survey effort considering that 90% of public roads were surveyed. Approxi- mately 10% of these roads are paved and trafﬁc density is <5 vehicles per day.
Surveys were done by two observers from a vehicle driven at a maximum speed of 40 km h–1. When Lesser Rheas were sighted we measured the distance to the animal or to the centre of the ﬂock with a laser range ﬁnder (Leica LRF 1200 Rangemaster, Leica, Solms, Germany) and the angle of the animal relative to our bearing. Our bearing was determined relative to north from the inertial compass in a global positioning system (GPS) unit (Garmin GPS MAP 76CS, Garmin, Olathe, KS, USA). Sightings were collected in a personal digital assistant (PDA; Tungsten T3, Palm Inc., Sunnyvale, CA, USA) using the free Cybertracker software (http://www.cybertracker.org/, accessed 19 October
2011). The PDA was synchronised with the GPS unit, which was used to record the precise location of the census track and the sightings, as well as date, time and vehicle speed.
We selected ten potential environmental predictors that sum- marised the most relevant environmental gradients and landscape features needed to test our hypotheses (Table 1).
We derived vegetation productivity variables from the Veg- etation sensor of the Spot 4 satellite (http://www.spot-vegetation. com, accessed 19 October 2011), which monitors terrestrial
vegetation cover at 1-km spatial resolution. We used NDVI images to estimate primary productivity (mean NDVI and its coefﬁcient of variation), the month at which the NDVI reaches its
Table 1. Hypotheses about the factors inﬂuencing the distribution of Lesser Rheas and variables used as predictors in models testing the hypotheses
Persecution-by-ranchers hypothesis Sheep_density. Sheep stocking level estimated from a model of sheep distribution (Appendix S1 in Pedrana et al.
Productivity hypothesis Mean_NDVI. Mean Normalised Difference Vegetation Index calculated using the VGT-S10 product, which is a
10-day maximum composite value from the VEGETATION sensor of the Spot satellite (http://www.spot-vegetation. com) from April 1999 to March 2005
Growth_period. Length of the vegetation growth period deﬁned as the mean number of 10-day periods with NDVI
values > 85
CV_NDVI. Coefﬁcient of variation of NDVI
Season_MAX. Month at which the NDVI reaches its annual maximum value
Distance_wet meadow. Distance (km) to the nearest pond-bog-wet meadow obtained as a vector coverage from
Mazzoni and Vázquez (2004)
Topography hypothesis Altitude. Mean altitude in metres above sea level in a 1-km pixel acquired from the Shuttle Radar Topography Mission
Slope. Mean slope in degrees in a 1-km pixel acquired from the SRTM
Anthropogenic disturbance hypothesis
Distance_urban. Distance (km) to the nearest urban area with 2000 inhabitants. Data obtained from the Instituto
Geográﬁco Nacional de la República Argentina (http://sig.gov.ar/, accessed 19 October 2011)
Distance_oil. Distance (km) to the nearest oil camp. Data obtained from the Instituto Geográﬁco
Nacional de la República Argentina (http://sig.gov.ar/)
annual maximum, and seasonality in vegetation growth using 7 consecutive years of data (April 1999–March 2005). We acquired topographic data (mean slope and altitude) from the Shuttle Radar Topography Mission (SRTM; http://www2.jpl.nasa.gov/ srtm,accessed 19 October 2011). Distances from each 1-km cell to the nearest city (i.e. urban settlement with an estimated population size >2000 inhabitants), to the nearest oil camp, and to the nearest wet meadow (taken from Mazzoni and Vázquez 2004) were calculated in a geographical information system (GIS, IDRISI Kilimanjaro, Clark Labs, Worcester, MS, USA). The probability of contact with sheep in a cell, as a proxy of sheep stocking density, was taken from a predictive map built with data recorded during ﬁeld surveys (appendix S1 in Pedrana et al. 2011).
Multicollinearity of environmental predictors can make inter- pretation of alternative models difﬁcult (Lennon 1999). We
considered two predictors to be collinear when the Spearman rank correlation coefﬁcient (Rs) was >0.7. Among strongly correlated predictors, we retained those with the clearest eco- logical meaning for the species (Austin 2007).
Presence–absence data and factors inﬂuencing detectability
Tracks recorded with the GPS deﬁned the route of our survey. We used the distance to Lesser Rheas that were sighted to estimate the area effectively covered. We used the software DISTANCE 5.0 (Thomas et al. 2010) to ﬁt a detection function to the distance data (Buckland et al. 2001). A 300-m buffer on both sides of the track was chosen to deﬁne the effective area surveyed as 75% of all sightings of Lesser Rheas were within this area.
Presence–absence modelling requires deﬁning units in which presence or absence is recorded. We used a 1-km grid deﬁned by
the spatial resolution of NDVI data. We overlaid the surveyed
tracks with 300-m buffers on this 1-km grid and selected all cells that partially or totally overlapped with buffers. Lesser Rhea sightings (n = 795) were overlaid on selected cells. Grid-cells with
1 Lesser Rhea sightings were considered presences and the remaining cells were considered absences.
The probability of detecting an individual or ﬂock in a 1-km cell was affected by the proportion of the cell that was effectively surveyed. We calculated the variable ‘Area_surveyed’ as the fraction of the cell surface included in the 300-m buffer on both sides of the survey transect and this variable was included as a ﬁxed term in the models to correct for its effect on detection probability (Travaini et al. 2007; Pedrana et al. 2010). Although the survey protocol was standardised there are unavoidable survey variables that affect detectability of fauna that are rarely considered in species distribution modelling. For example, we tried to survey at a constant speed of 40 km h–1, but speed recorded by the GPS indicated that, within a cell, mean speed varied with road condition, weather and number of contacts with fauna. For this reason we analysed if vehicle speed (Speed), time of day (Time_day), or calendar date (Date) had any inﬂuence on detect- ability of individuals. Time of day could affect our results through the inﬂuence of light levels on detectability, and Date could increase detectability of Lesser Rheas towards the end of the reproductive season when birds increase their tendency to congregate.
We ﬁtted generalised additive models (GAM; Hastie and Tib- shirani 1990) using as a response variable the presence or absence of Lesser Rheas in a 1-km cell using binomial errors and a logit link. As the number of cells with presence (n = 482) was low compared to the number of cells with observed absence (n = 13 230), we used a re-sampling scheme to obtain a balanced sample (Liu et al. 2005), randomly choosing 482 out of the 13 230 cells with absence. We reserved a random sample of 20% of cells with presence and absence for model cross-validation and used the remaining 80% for model ﬁtting. This procedure was repeated
100 times. Predictors for the models were selected from the initial set by a backward-forward stepwise procedure (step.gam routine
in S-PLUS 2000; MathSoft 1999), starting from a full model that
included all potential predictors relevant to a particular hypoth-
esis. Predictors were initially included in the models as smoothing splines with 3 d.f. The Akaike’s Information Criterion (AIC) was used to retain a term (Sakamoto et al. 1986). From the 100 models built with the re-sampling procedure for each hypothesis, we selected those that ranked as the best model 10 times. Then we repeated this re-sampling procedure with each of the selected models, in which the predictors were ﬁxed, but progressive simpliﬁcation of the degrees of freedom of the splines was allowed. Again we retained the models that were selected 10 times. Finally, we used a single matrix with the complete dataset where original prevalence was maintained (Jiménez-Valverde and Lobo 2006) to compare alternative models within each hypothesis that were as good as the best model in terms of AIC (Burnham and Anderson 2002). We considered as competing models those for which the differences between AIC and the AIC (Di) of the best candidate model was 44.
The same procedure was used to build a general model starting with all relevant variables retained in the best models for each hypothesis.
The area under the curve (AUC) of the receiver operating characteristic (ROC) plot was computed for each of the 100 models with each set of validation data to estimate its predictive power through cross-validation (Murtaugh 1996). The AUC ranges from 0 (model discrimination is not better than random) to 1 (perfect discriminatory ability; Pearce and Ferrier 2000). Predictive models are considered usable if AUC 0.7 (Harrell
2001). The difference between the mean predictive ability of the model selected for each hypothesis and the general model was
tested with a Wilcoxon–Mann–Whitney test (Crawley 2002).
Distribution of Lesser Rhea in Santa Cruz
We used the best model to build a predictive map of the current distribution of Lesser Rheas in Santa Cruz Province. To produce this map we used the option in IDRISI Kilimanjaro (Eastman
2003) to export predictors as a data matrix to S-PLUS, applied the predict.gam procedure (MathSoft 1999) to make predictions
based on the new data matrix, and then exported the predicted
probability values from S-PLUS back to IDRISI. The estimated probability of Lesser Rhea occurrence was simpliﬁed into three probability classes to ease interpretation of the distribution.
On 8000 km of road transect we made 795 sightings of Lesser
Rhea individuals or ﬂocks, comprising a total of 3462 individuals.
We found a high correlation between the predictors Growth_ period and Mean_NDVI (Rs = 0.87), Growth_period and Sea- son_MAX (Rs = 0.89), and Mean_NDVI and Season_MAX (Rs =
0.93). We chose Mean_NDVI as the best ecological represen-
tative of these three predictors.
As expected, the probability of sighting a Lesser Rhea showed a signiﬁcantly non-linear decline with the proportion of the cell that was included in the 300-m buffer (model 1 in Table 2, Fig. 1). Among the survey-speciﬁc variables, we found that the proba- bility of detecting a Lesser Rhea was also affected by time of day, survey date and vehicle speed (model 1 in Table 2).
Contrary to the prediction of the persecution-by-ranchers hypothesis, the most parsimonious model showed that the probability of occurrence of Lesser Rheas varied little with low
Table 2. Competing GAM models obtained by stepwise selection for each hypothesis of the factors inﬂuencing occurrence of Lesser Rheas in the semi- arid steppes of Santa Cruz Province, Southern Patagonia
For each model Akaike’s Information Criterion (AIC) and the difference of AIC between the current model and the best model (Di) are given. Subscripts refer to the degrees of freedom of the smoothing spline and no subscripts refer to linear terms (d.f. = 1)
Model code Models AIC Di
1 Area_surveyed3 + Date3 + Speed3 + Time_day3 3882.05 0
2 Area_surveyed3 + Date3 + Speed3 + Time_day2 3883.07 1.02
3 Area_surveyed3 + Sheep_density3 3887.06 0
4 Area_surveyed3 + Sheep_density2 3890.64 3.58
Area_surveyed3 + Mean_NDVI3 + Distance_wet meadow
Area_surveyed3 + Mean_NDVI3 + Distance_wet meadow3
Area_surveyed3 + Mean_NDVI3
Area_surveyed3 + Slope + Altitude3
Area_surveyed3 + Altitude3
Anthropogenic disturbance hypothesis
Area_surveyed3 + Distance_urban + Distance_oil3
Area_surveyed3 + Distance_urban3 + Distance_oil3
Area_surveyed3 + Mean_NDVI3 + Distance_urban + Distance_wet meadow3 + Distance_oil3 +Altitude
Area_surveyed3 + Mean_NDVI3 + Distance_urban3 + Distance_wet meadow3 +Distance_oil3
General models combined with survey-speciﬁc variables
Area_surveyed3 + Mean_NDVI3 + Distance_urban + Distance_wet meadow3 + Distance_oil3 + Speed3 + Date3 + Time_day3
Area_surveyed2 + Mean_NDVI3 + Distance_urban + Distance_wet meadow + Speed3 + Date3 + Time_day3
20 60 100 0 70 140
Area_surveyed Mean_NDVI Distance_wet meadow
0 500 1000
2 (d )
0.4 0.6 0.8 1.0 0 75 150 0 62 125
Area_surveyed Distance_urban Distance_oil
Fig. 1. Partial effects of predictors included in the most parsimonious models for each alternative hypothesis of the factors that inﬂuence the occurrence of Lesser Rheas: (a) persecution-by-ranchers hypothesis (model 3 in Table 2); (b) productivity hypothesis (model 5 in Table 2); (c) topography hypothesis (model 8 in Table 2); and (d) anthropogenic disturbance hypothesis (model 10 in Table 2). Dashed lines represent 95% conﬁdence intervals for the mean prediction.
to moderate values of sheep abundance (probability of sheep presence <0.5, assuming that this probability is positively correlated with abundance; Pedrana et al. 2011) but increased with moderate to high values of sheep abundance (model 3 in Table 2, Fig. 1a probability of sheep presence 0.5).
The most parsimonious model of Lesser Rhea occurrence among the models being evaluated under the productivity hy- pothesis included Mean_NDVI and Distance_wet meadow (model 5 in Table 2). The probability of Lesser Rhea occurrence increased non-linearly with the mean NDVI and decreased linearly with the distance to the nearest wet meadow (Fig. 1b).
The best model among the models being evaluated under the topography hypothesis included slope and altitude (model 8 in
Table 2). Lesser Rhea occurrence was negatively related to mean slope and positively related to altitude (Fig. 1c).
Two variables were retained in the best model under the
anthropogenic disturbance hypothesis: Distance_urban and Dis- tance_oil (model 10 in Table 2), indicating that Lesser Rhea occurrence strongly increased with distance to the nearest city and distance to an oil camp (Fig. 1d).
When all predictors were considered in a model of Lesser Rhea presence, the general model included: Mean_NDVI, Distan- ce_urban, Distance_wet meadow and Distance_oil, and three survey-speciﬁc variables (time of day, census date, and vehicle speed; model 14 in Table 2, Fig. 2). The ﬁt of model 14, which included survey-speciﬁc variables, was superior to the ﬁt of model
12 that did not include them (Table 2). All models of Lesser Rhea presence were improved with the inclusion of survey-speciﬁc variables. These variables, however, did not alter the identity, sign, or the relative strength of the predictor effects (see Appen- dix) except the effect of altitude on Lesser Rhea occurrence was no longer signiﬁcant (model 8 in Table 2). The probability of
detecting Lesser Rheas ﬂuctuated seasonally, decreased with car speed, and reached a peak in the hours around midday (Fig. 2).
The predictive models for Lesser Rhea distribution ﬁtted the data well, with a mean validation AUC (±s.e.) better than a null model for every set of predictors: productivity model (0.72 ± 0.03), topography model (0.70 ± 0.02), anthropogenic disturbance model (0.73 ± 0.03), and general model (0.86 ± 0.02). The strong ﬁt suggests that the models were robust and could be considered useful for predicting the distribution of the species (Harrell 2001). Among the general models, model 14 (Table 2) had the highest predictive ability, signiﬁcantly higher than the predictive ability of ﬁnal models representing a single hypothesis, either the productivity (Z = 9.41, P < 0.001), topography (Z = 11.94, P < 0.001) or anthropogenic disturbance hypotheses (Z = 7.07, P < 0.001).
Predictive mapping of the distribution of Lesser Rheas
The predictions of model 14 (Table 2) were translated to a GIS
assuming that whole 1-km cells were effectively surveyed at
30 km h–1, at the most favourable date in the middle of spring–
summer (i.e. 20 January) and at the most favourable time (i.e.
1200 hours) (Fig. 3). This map shows that, although Lesser Rheas have been sighted almost everywhere in Santa Cruz Province, the species is not uniformly distributed (Fig. 3). The probability of occurrence of Lesser Rheas increased from north to south in Santa Cruz Province, although there are small areas of high probability of occurrence, close to the Andean slopes in the west and near wetlands scattered across the centre of the region.
1.0 20 80 140 0 70 140
Mean_NDVI Distance_wet meadow
80 130 180
0 30 60
Fig. 2. Partial effects of predictors included in the best general model of Lesser Rhea presence corrected for survey-speciﬁc variables (model 14 in Table 2). Dashed lines represent 95% conﬁdence intervals for the mean effect.
80 °W 60 °W 40 °W
0 50 100 200
Fig. 3. The distribution of the Lesser Rhea in Santa Cruz Province, Argentina. Values represent the probability of recording a Lesser Rhea in a 1-km cell as predicted by Model 14 (Table 2). Probabilities are categorised in three classes (low, <0.33; medium, 0.33–0.66; high, >0.66). White represents areas of the region omitted from predictions: sea, lakes, forested areas or beyond the model’s environmental space.
The continuing decline of Lesser Rhea may cause local or regional extirpations unless conservation measures are undertaken (Bellis et al. 2006).
Our results highlight the main factors inﬂuencing the current distribution of Lesser Rheas and allow us to draw conclusions
about the causes of this decline. In agreement with the produc- tivity hypothesis, the occurrence of Lesser Rheas was positively associated with mean primary productivity and distance to the
nearest wet meadow, in a regional context dominated by dry steppe habitat. Preference for wetlands has been observed in other populations of Lesser Rheas, presumably because wetlands
provide the best quality forage for adults and their chicks (Bellis
et al. 2006; Barri et al. 2008, 2009a).
The occurrence of Lesser Rheas strongly increased with distance from the nearest city or oil camp. In accordance with the anthropogenic disturbance hypothesis, areas with low prob- ability of Lesser Rhea occurrence were especially common in northern Santa Cruz, where oil exploitation currently concen- trates (González and Rial 2004). Oil extraction is preceded by the development of a large number of roads in otherwise inaccessible areas. This activity not only affects wildlife by increasing trafﬁc casualties and habitat degradation, but also by increasing access for poachers (Novaro et al. 2000; Bellis et al. 2004; Martella and
Navarro 2006). Funes et al. (2000) suggested that one of the major causes of the decline of the Lesser Rhea in north-western Pata- gonia was the opening of new roads associated with the expansion of the oil industry. The fact that Lesser Rheas stayed away from cities indicate that these birds might actively avoid areas where they are intensively hunted or disturbed by people (Funes et al.
2000; Novaro et al. 2000; Bellis et al. 2004). Navarro et al. (1999)
also reported that the density of Lesser Rhea populations was negatively correlated with human density.
Our results suggest that the probability of Lesser Rhea oc- currence is greater at higher elevations and in ﬂat open areas, as predicted by the topography hypothesis. But, this association is not as strong as the one with primary productivity, and was not included in the ﬁnal model. Because Rheas have a ‘watch and run’ anti-predator strategy (Bruning 1974), open ﬂat areas favour vigilance and quick escape. However, given that the mesic, productive pasturelands that this species seems to prefer typically occur in ﬂat terrain, and that the effect of topographical predictors became weaker in the general models, the good ﬁt of the topo- graphical models could be confounded with the distribution of high-quality forage, at least in part.
Contrary to our expectations, the persecution-by-ranchers models show that sheep ranching did not have a negative effect
on the distribution of Lesser Rheas. The distribution map indi- cates that areas with high probability of Lesser Rhea occurrence are concentrated in the southern sector of Santa Cruz Province,
which is an area with above average rainfall where productive pastures abound (González and Rial 2004). These productive pastures are mainly devoted to extensive sheep ranching with high stocking rates. Traditionally, some species such as Guanaco
(Lama guanicoe) and Upland Goose (Chloephaga picta), have been considered pests by ranchers on the basis of assumed competition with sheep (Baldi et al. 2004; Blanco and De la
Balze 2006) and, as a consequence, they were actively persecuted. Predictive habitat models for the Guanaco indicate that its distribution is restricted to areas of low productivity with low
sheep stocking levels (Travaini et al. 2007; Pedrana et al. 2010). In contrast, the Lesser Rhea might not have suffered the same persecution, because there is little competition for food resources
between Lesser Rheas and other herbivore species, such as Upland Geese, Guanaco and sheep (Bellis et al. 2004). In addition, it seems that agricultural activities and sheep ranching did not inﬂuence Lesser Rhea occurrence (Bellis et al. 2004).
Moreover, Barri et al. (2009b) found out that sheep stocking at moderate levels (0.25–1 sheep ha–1) did not affect species repro- ductive success (Barri et al. 2008, 2009b).
Our study also shows that even when care is taken to stan- dardise the survey protocol there are unavoidable survey factors that inﬂuence the results. Here we show that these factors can be
controlled statistically and researchers should check that their conclusions are robust and do not change when correction factors are included in the models.
Species distribution models of the Lesser Rhea in Patagonia suggest that: (1) primary productivity is the main driver of the distribution of the species in the arid steppes of Santa Cruz; (2) mesic habitats like wet meadows are selected habitats, and allow
Lesser Rheas to occupy otherwise unproductive steppes; (3) urban areas and oil camps may have a negative effect on the distribution of Lesser Rheas or, at least, that Lesser Rheas appear
to be more disturbed near these areas; and (4) current levels of competition with sheep and of direct persecution by ranchers have no noticeable effect on the distribution of Lesser Rheas.
Finally, we believe that our statistical distribution model generates a map of the distribution of Lesser Rheas that can be a useful tool for governmental agencies to establish conservation management priorities for the species, and for identifying regions where local-scale ecological studies of this species should be conducted.
This work was funded by the Banco Bilbao Vizcaya Argentaria (BBVA) Foundation through a grant under the Conservation Biology Programme. Additional support was provided by the Universidad Nacional de la Patagonia Austral, Consejo Nacional de Investigaciones Cientíﬁcas y Técnicas (CON- ICET) and Comisión Nacional de Actividades Espaciales (CONAE). We thank R. Martínez Peck, E. Daher, M. Yaya and M. Brossman for ﬁeld assistance; Miriam Vásquez for providing the supervised classiﬁcation of wetlands habitats; and J. Navarro, P. E. Osborne and C. A. Hagen for their constructive criticism of an earlier draft of the manuscript. We appreciate the improvements to the writing made by Jeffrey Lusk through the editorial assistance program of the Association of Field Ornithologists.
Abraham, E., Macagno, P., and Tomasini, D. (2005). Experiencia argentina vinculada a la obtención y evaluación de indicadores de desertiﬁcación. In
‘Desertiﬁcación: Indicadores y Puntos de Referencia en América Latina y el Caribe’. (Eds E. Abraham, D. Tomasini and P. Macagno.) pp. 81–85. (Secretaria de Ambiente y Desarrollo Sustentable: Mendoza, Argentina.) Austin, M. (2007). Species distribution models and ecological theory: a critical assessment and some possible new approaches. Ecological Model-
ling 200, 1–19. doi:10.1016/j.ecolmodel.2006.07.005
Baldi, R., Pelliza-Sbriller, A., Elston, D., and Albon, S. (2004). High potential for competition between guanacos and sheep in Patagonia. Journal of Wildlife Management 68, 924–938. doi:10.2193/0022-541X(2004)068 [0924:HPFCBG]2.0.CO;2
Barri, F. R., Martella, M. B., and Navarro, J. L. (2008). Effects of hunting, egg harvest and livestock grazing intensities on density and reproductive success of Lesser Rhea Rhea pennata pennata in Patagonia: implications for conservation. Oryx 42, 607–610. doi:10.1017/S0030605307000798
Barri, F. R., Martella, M. B., and Navarro, J. L. (2009a). Nest-site habitat selection by Lesser Rheas (Rhea pennata pennata) in northwestern Patagonia, Argentina. Journal für Ornithologie 150, 511–514. doi:10.1007/s10336-009-0374-6
Barri, F. R., Martella, M. B., and Navarro, J. L. (2009b). Reproductive success of wild Lesser Rheas (Pterocnemia - Rhea - pennata pennata) in north- western Patagonia, Argentina. Journal für Ornithologie 150, 127–132. doi:10.1007/s10336-008-0327-5
Bellis, L. M., Martella, M. B., Navarro, J. L., and Vignolo, P. E. (2004). Home range of Greater and Lesser Rhea in Argentina: relevance to conservation. Biodiversity and Conservation 13, 2589–2598. doi:10.1007/s10531-004-
Bellis, L. M., Navarro, J. L., Vignolo, P., and Martella, M. B. (2006). Habitat preferences of Lesser Rhea in Argentine Patagonia. Biodiversity and Conservation 15, 3065–3075. doi:10.1007/s10531-005-5398-5
Blanco, D. E., and De la Balze, V. M. (2006). Harvest of migratory geese
Chloephaga spp. in Argentina: an overview of the present situation. In
‘Waterbirds Around the World’. (Eds G. C. Boere, C. A. Galbraith and
D. A. Stroud.) pp. 870–873. (The Stationery Ofﬁce: Edinburgh, UK.) Bruning, D. F. (1974). Social structure and reproductive behaviour in the
Greater Rhea. Living Bird 13, 251–294.
Buckland, S. T., Anderson, D. R., Burnham, K. P., Laake, J. L., Borchers, D. L., and Thomas, L. (2001). ‘Introduction to Distance Sampling.’ (Oxford University Press: Oxford, UK.)
Burnham, K. P., and Anderson, D. R. (2002). ‘Model Selection and Multi-
Model Inference: A Practical Information-Theoretic Approach.’ (Spring- er: New York.)
Bustamante, J., and Seoane, J. (2004). Predicting the distribution of four
species of raptors (Aves:Accipitridae) in southern Spain: statistical models work better than existing maps. Journal of Biogeography 31,
Crawley, M. J. (2002). ‘Statistical Computing.’ (Wiley: New York.) Eastman, J. (2003). ‘IDRISI Kilimanjaro: Guide to GIS and Image
Processing.’ (Clark Laboratories, Clark University: Worcester, MA.) Folch, A. (1992). Family Rheidae (Rheas). In ‘Handbook of the Birds of the
World. Vol 1: Ostrich to Ducks’. (Eds J. Del Hoyo, A. Elliott and
J. Sargatal.) pp. 83–84. (Lynx Edicions: Barcelona.)
Funes, M. C., Rosauer, M. M., Aldao, G. S., Monsalvo, O. B., and Novaro, A. J. (2000). ‘Manejo y Conservación del Choique en la Patagonia: Análisis de los relevamientos poblacionales.’ (C.E.A.N, Dirección Gen- eral de Supervisión Técnico Administrativa, Subsecretaria de Producción y Recursos Naturales, Secretaría de Estado de Producción y Turismo: Río Negro, Argentina.)
Golluscio, R. A., Deregibus, V. A., and Paruelo, J. M. (1998). Sustainability and range management in the Patagonia steppes. Ecología Austral 8,
González, L., and Rial, P. (2004). ‘Guía geográﬁca interactiva de Santa Cruz.’ (Ediciones Instituto Nacional de Tecnología Agropecuaria (INTA) and Universidad Nacional de la Patagonia Austral: Santa Cruz, Argentina.)
Gottschalk, T. K., Ekschmitt, K., I_sfendiyaroglu, S., Gem, E., and Wolters, V.
(2007). Assessing the potential distribution of the Caucasian Black Grouse Tetrao mlokosiewiczi in Turkey through spatial modelling. Journal für Ornithologie 148, 427–434. doi:10.1007/s10336-007-0155-z
Harrell, F. E. (2001). ‘Regression Modelling Strategies.’ (Springer: New York.)
Hastie, T., and Tibshirani, R. J. (1990). ‘Generalized Additive Models.’
(Chapman and Hall: London.)
Jiménez-Valverde, A. J., and Lobo, J. M. (2006). The ghost of unbalanced species distribution data in geographical model predictions. Diversity & Distributions 12, 521–524. doi:10.1111/j.1366-9516.2006.00267.x
Lennon, J. J. (1999). Resource selection functions: taking space seriously?
Trends in Ecology & Evolution 14, 399–400. doi:10.1016/S0169-5347 (99)01699-7
Liu, C., Berry, P. M., Dawson, T. P., and Pearson, R. G. (2005). Selecting
thresholds of occurrence in the prediction of species distributions. Eco- graphy 28, 385–393. doi:10.1111/j.0906-7590.2005.03957.x
Martella, M. J., and Navarro, J. L. (2006). Proyecto Ñandú: Manejo de Rhea americana y R. pennata en la Argentina. In ‘Manejo de fauna silvestre en la Argentina y programa de usos sustentables’. (Eds M. L. Bolkovic and D. Ramadori.) pp. 39–50. (Dirección de Fauna Silvestre, Secretaría de Medio Ambiente y Desarrollo Sustentable: Buenos Aires.)
MathSoft (1999). ‘S-Plus 2000. User’s Guide.’ (Mathsoft Data Analysis
Products Division: Seattle, WA.)
Mazzoni, E., and Vázquez, M. (2004). ‘Ecosistemas de mallines y paisajes de la Patagonia austral (Provincia de Santa Cruz).’ (Ediciones Instituto Nacional de Tecnología Agropecuaria (INTA): Santa Cruz, Argentina.) Murtaugh, P. A. (1996). The statistical evaluation of ecological indicators.
Ecological Applications 6, 132–139. doi:10.2307/2269559
Navarro, J. L., Cardón, R., Manero, A., and Clarke, R. (1999). Estimación de la abundancia poblacional del choique en la vida silvestre. Report to Direc- ción de Fauna y Flora Silvestres, Secretaría de Recursos Naturales y Desarrollo Sustentable, Buenos Aires, Argentina.
Novaro, A. J., Funes, M. C., and Walker, R. S. (2000). Ecological extinction of native prey of a carnivore assemblage in Argentine Patagonia. Biological Conservation 92, 25–33. doi:10.1016/S0006-3207(99)00065-8
Pearce, J., and Ferrier, S. (2000). Evaluating the predictive performance of
habitat models developed using logistic regression. Ecological Modelling
133, 225–245. doi:10.1016/S0304-3800(00)00322-7
Pedrana, J., Bustamante, J., Travaini, A., and Rodríguez, A. (2010). Factors inﬂuencing guanaco distribution in southern Argentine Patagonia and implications for its sustainable use. Biodiversity and Conservation 19,
Pedrana, J., Bustamante, J., Rodríguez, A., and Travaini, A. (2011). Primary productivity and anthropogenic disturbance as determinants of Upland Goose Chloephaga picta distribution in southern Patagonia. Ibis 153,
Rodríguez, J. P., Brotons, L., Bustamante, J., and Seoane, J. (2007). The application of predictive modelling of species distribution to biodiversity conservation. Diversity & Distributions 13, 243–251. doi:10.1111/ j.1472-4642.2007.00356.x
Sakamoto, Y., Ishiguro, M., and Kitagawa, G. (1986). ‘Akaike Information
Criterion Statistics.’ (KTK Scientiﬁc Publishers: Tokyo.)
Scott, J. M., Heglund, P. J., and Morrison, M. I. (2002). ‘Predicting Species Occurrences: Issues of Accuracy and Scale.’ (Island Press: Washington, DC.)
Seoane, J., Viñuela, J., Díaz-Delgado, R., and Bustamante, J. (2003). The effects of land use and climate on Red Kite distribution in the Iberian Peninsula. Biological Conservation 111, 401–414. doi:10.1016/S0006-
Suárez-Seoane, S., Osborne, P., and Alonso, J. C. (2002). Large-scale habitat selection by agricultural steppe birds in Spain: identifying species-habitat responses using generalized additive models. Journal of Applied Ecology
39, 755–771. doi:10.1046/j.1365-2664.2002.00751.x
Thomas, L., Buckland, S. T., Rexstad, E. A., Laake, J. L., Strindberg, S., Hedley, S. L., Bishop, J. R. B., Marques, T. A., and Burnham, K. P. (2010). Distance software: design and analysis of distance sampling surveys for estimating population size. Journal of Applied Ecology 47, 5–14. doi:10.1111/j.1365-2664.2009.01737.x
Travaini, A., Bustamante, J., Rodríguez, A., Zapata, S., Procopio, D., Pedrana,
J., and Martínez Peck, R. (2007). An integrated framework to map animal distributions in large and remote regions. Diversity & Distributions 13,
Appendix. Competing GAM models obtained by stepwise selection for each hypothesis of the factors that inﬂuence the occurrence of Lesser Rheas in the semi-arid steppes of Santa Cruz Province, including corrections for survey-speciﬁc variables
For each model Akaike’s Information Criterion (AIC) and the difference of AIC between the current model and the best model (Di)
are given. Subscripts refer to the degrees of freedom of the smoothing spline and no subscripts refer to linear terms (d.f. = 1)
Models AIC Di
Area_surveyed3 + Sheep_density3+ Date3 + Speed3 + Time_day3 1033.90 0.00
Area_surveyed3 + Mean_NDVI3 + Distance_wet meadow3 + Date3 + Speed3 + Time_day2 984.09 0.00
Area_surveyed3 + Mean_NDVI3 + Distance_wet meadow3 + Date3 + Speed3 + Time_day2 985.50 1.41
Area_surveyed3 + Mean_NDVI3 + Date3 + Speed3 + Time_day3 986.25 2.15
Area_surveyed3 + Slope + Date3 + Speed3 1028.49 0.00
Area_surveyed3 + Altitude3 + Date3 + Speed3 1031.33 2.84
Area_surveyed3 + Altitude + Date3 + Speed3 1032.36 3.88
Anthropogenic disturbance hypothesis
Area_surveyed3 + Distance_urban + Distance_oil3 + Time_day2 + Date3 + Speed3 1012.76 0.00
Area_surveyed3 + Distance_urban3 + Distance_oil3 + Time_day2 + Date3 + Speed3 1014.19 1.43