Ana səhifə

Bioivers I ty rese arch


Yüklə 1.2 Mb.
səhifə2/3
tarix24.06.2016
ölçüsü1.2 Mb.
1   2   3

Aestivating habitat

hypothesis Spatial distribution of

breeding habitats hypothesis

Ecosection type hypothesis


Breeding habitat hypothesis

Aestivating habitat hypothesis


Spatial distribution of breeding habitats hypothesis

Figure 3 (a) Mean value and standard error of the Akaike’s weight of each model across all species. (b) For each species, the Akaike’s weight of each main hypothesis is shown.

Table 4 Relative importance of parameters obtained for species distribution models accounting for the reliability of non-detection records.

Ecosection



Pelobates cultripes

Discoglossus galganoi

Hyla meridionalis

Pleurodeles waltl

Triturus pygmaeus

Lissotriton boscai

0.0

1.4

100

0.2

76.2



5.8

Pond size

99.5

6.4

0.0

48.8

9.1

0.0

Hydroperiod

99.4

5.9

0.0

43.1

12.1

0.0

Forest

0.0

57.5

0.0

1.9

0.0

48.3

Scrub

0.0

57.5

0.0

1.9

0.0

48.3

Dunes

0.0

48.6

0.0

4.0

0.0

48.3

Surrounding vegetation

0.0

2.4

0.0

0.7

10.8

45.5

Distance to marshes

0.0

1.8

0.0

1.9

10.5

38.0

Distance to nearest pond

0.0

1.8

0.0

1.9

10.5

38.0

Number of ponds

0.0

4.8

0.0

10.4

9.6

37.6

Number of large ponds

0.0

6.7

0.0

9.4

9.6

37.6

Parameters with highest relative abundance are shown in bold.



one explained the distribution of L. boscai (Fig. 3 and Supporting Information for details). The presence of both species was positively related to dunes and negatively with forest, but showed contrasting responses to the presence of scrub habitat in the surroundings (see Supporting Informa- tion: Tables S2 and S3). Remarkably, the presence of L. boscai was positively associated with the number of ponds persisting in dry years (i.e. number of large ponds) but not with the number of ponds persisting in wet years (i.e. number of ponds).

DISCUSSION


This study illustrates how to account for the reliability of absence data in SDMs, taking the case of amphibian species in Mediterranean temporary ponds as an example. Our results evidenced a wide variability in the reliability of non-detection

records for different amphibian species when the number of surveys or its timing (choice of sampling months) is not the same for all inventoried sites (i.e. ponds in this study). As a consequence, one should not assume that all non-detections are valid absences and, instead, question their individual reliability. Although uncertainty in the reliability of absences was previously recognized as a potential source of problems in SDMs (Lobo, 2008), our results accurately show that non- detection records should be considered with particular care when a species’ detectability is low and/or can change over time, as it is the case of L. boscai and D. galganoi (following Go´ mez-Rodrı´guez et al., 2010d), both endemic species of the Iberian Peninsula. In addition, this study also indicates that accounting for absence data reliability is of special relevance in the case of rare species. Thus, the less common species (i.e. L. boscai or D. galganoi, following D´ıaz-Paniagua et al., 2006) had less reliable absences than more common species, as those




(a) (b)

60

P. cultripes
40

20

P* < 0.8 Absence



P* ≥ 0.8

Presence


1.0
0.0
–1.0

P. cultripes



0

Ephemeral
Long-duration

–2.0
Ephemeral


Long-duration

Intermediate

Zacallon

Intermediate

Zacallon

HYDROPERIOD

HYDROPERIOD



H. meridionalis

60

40
P* < 0.8 Absence



P* ≥ 0.8

Presence


H. meri

dionali

s






2000


0

20




0

1 3 4


5 6 7 8
–2000

1 3 4 5 6 7 8

ECOSECTION

60

ECOSECTION



P. waltl
40

20

P* < 0.8 Absence



P* ≥ 0.8

Presence


1.5

0.5

–0.5


P. waltl


0

Ephemeral


Long-duration

–1.5
Ephemeral


Long-duration

Intermediate

Zacallon


Intermediate

Zacallon


HYDROPERIOD

HYDROPERIOD



T. pygmaeus

60

40



20
P* < 0.8 Absence

P* ≥ 0.8

Presence


6000

2000


–2000
T. pygmaeus


0
160
120
80
1 3 4 5 6 7 8

ECOSECTION


P* ≥ 0.8
L. boscai P* < 0.8 Absence

Presence


–6000

3.0
2.0


1.0
1 3 4 5 6 7 8

ECOSECTION
L. boscai

Figure 4 (a) Number of observed presences and absences in each category of habitat factors with high relative impor- tance. Data are shown both for all absences and for absences with high reliability

(P* ‡ 0.8). (b) Partial effects (± standard

error) of categorical factors with high

40
0

Low Intermediate High

SURROUNDING VEGETATION

0.0


–1.0

Low



High

Intermediate

relative importance.

*For each species, partial effects are obtained from a global model including all factors with



high relative importance (> 0.376) using

SURROUNDING VEGETATION

command plot.gam (library ‘gam’, R project).




with large occupancy (i.e. T. pygmaeus and H. meridionalis, following D´ıaz-Paniagua et al., 2006) or persisting longer in the ponds because of their long larval period (i.e. Pelobates cultripes). It should be noted that, following Go´ mez-Rodrı´guez et al. (2010d), there are two major causes of the unreliability of non-detection records: (1) ‘methodological constraints’, attrib-

uted to a low efficiency of the sampling survey, and (2)

‘phenological constraints’, attributed to inadequate survey timing (i.e. the pond was surveyed before the species had reached the pond for breeding). Identifying the cause of data unreliability would be useful to minimize errors in future surveys. However, we would like to stress that, once data are




collected, the cause of data unreliability is irrelevant as the consequence is always the same: absences should be taken with caution.

Unreliable absences, like those found in this study, may cause severe errors in SDMs, as one faces the risk of considering sites unsuitable that are in fact occupied by the species (i.e. detection failure). To avoid the error caused by such methodological absences (sensu Lobo et al., 2010), we propose a novel approach to explicitly account for absence reliability in SDMs, using objective and quantifiable criteria. This improved framework allows incorporating information on species detectability at each particular survey site into traditional modelling techniques (GLMs). The rationale behind it is that the relevance of information contained in a non-detection record is conditional on the probability of having detected the species at that site if it was really present. If the probability is high, we can be certain that the species was not there; otherwise, we would have detected it. In that case, the absence is reliable. We propose this approach as one step further from the suggestions in Lobo et al. (2010), who recommend the use of expert opinion or conceptual designs to avoid the indiscriminate inclusion of zeros from badly surveyed localities in model building.

One of the main advantages of our approach is that it does not dismiss all non-detection records but, instead, weights its relevance according to estimated reliability. In this regard, our approach constitutes an alternative to using presence-only models (see Elith et al., 2006) when the data are susceptible to false absences (see Gibson et al., 2007; Rota et al., 2011 as some recent examples). In fact, as it partially uses information on non-detection records to provide more confidence on the most reliable ones, our methods minimize the problem of placing random pseudo-absences in favourable sites (Engler et al.,

2004; Lobo, 2008). Rota et al. (2011) reported that the performance of logistic regressions diminished for less detect- able species because non-detections were ambiguous, an issue that highlights the need for accounting for absence data reliability in SDMs for these species. It must be stressed that this approach is not intended to replace an adequate survey nor occupancy models (see MacKenzie et al., 2006) that simultaneously estimate species occupancy and detectability from intensive sampling. On the contrary, it is a low-cost alternative for improving the reliability of data from monitor- ing schemes in which multiple surveys of all sites were not feasible.

Here, we also evidence that SDMs accounting for data reliability outperformed traditional regression models for anurans, especially for highly inconspicuous species, such as D. galganoi, although it was not clearly advantageous in the case of urodeles. Overall, SDMs did not provide good results for urodeles in this study, suggesting that alternative hypoth- eses to the ones considered could be necessary to explain their distribution in the study area. In this sense, past events and/or historic processes may be important for these species given their strong site fidelity and limited dispersal ability (Smith & Green, 2005). We would like to highlight that, for amphibians

in DNP, traditional SDMs seemingly failed to identify relevant hypotheses as the one that received the largest support for most species was the ‘ecosection’ hypothesis. If this were indeed the case, it would imply a lack of active habitat selection by the species. On the contrary, using the enhanced modelling framework, we found evidence of active habitat selection although we did not find environmental correlates that were valid for all the amphibian species considered. This confirms previous studies that reported species-specific responses to habitat factors (i.e. Beja & Alcazar, 2003; Van Buskirk, 2005). It also stresses the importance of multifactor approaches for modelling amphibian distributions, since the characteristics of both the pond and the landscape (terrestrial and surrounding aquatic habitat) proved important for the different species. The ecosection hypothesis obtained high support from H. merid- ionalis and T. pygmaeus, as these species occurred in almost every pond in the ‘wet stabilized sands at higher elevation’. The breeding habitat hypothesis obtained high support for P. waltl and P. cultripes. Our results were in accordance with previous studies reporting that these species require ponds with long hydroperiod for breeding success (D´ıaz-Paniagua et al., 2005), usually of large size, but avoid the artificial permanent water bodies, especially P. cultripes. Terrestrial characteristics were the most important predictors of D. galganoi and L. boscai occurrence. However, as the shape of the responses was contrary to expected in some cases, the relevance of this hypothesis may not be solely related to terrestrial habitat usage by adults (i.e. movements between ponds or terrestrial residence). Instead, it could also be indirectly associated with the hydrological dynamics of DNP (i.e. large density and variability of temporary ponds in the area nearby the dunes, see Go´ mez-Rodr´ıguez et al., 2010a). The occurrence of L. bos- cai, a species reported to breed in ponds of intermediate hydroperiod in the area (D´ıaz-Paniagua et al., 2005), would be also related to the formation of ponds of intermediate/long hydroperiod in the surroundings but not to the formation of ephemeral ones that only flood in wet years.

It should be noted that, for most species, the best models had low explanatory power although still providing reasonable discrimination ability, appropriate for many uses (following Pearce & Ferrier, 2000). Low explanatory power is expected in binomial models where probability values ranging from 0 to 1 (predictive values) are compared to a binary variable (pres- ence–absence). An alternative explanation could be the potential lack of compliance of the study system with the equilibrium assumption of SDMs: suitable habitats should be occupied and unsuitable habitats should be empty. In fact, in our study area, absence of a species from suitable habitats may also be due to environmental stochasticity, a characteristic of Mediterranean ecosystems, or because the large availability of breeding sites in wet years (Go´ mez-Rodrı´guez et al., 2008,

2010b) might result in a system where more habitat patches are available than necessary given the number of individuals (i.e. empty ponds as a result of an absence of colonization rather than non-suitability, see Pulliam, 2000). This scenario would result in many contingent absences (sensu Lobo et al., 2010)






being included in the model as unsuitable sites when they are not really so. On the other hand, unsuitable habitats may also be occupied when a species does not always choose the best available site (Pulliam, 1996). As typical r-strategists, amphib- ian species in the study area may be reproducing in as many different ponds as possible to increase their reproductive success. Notwithstanding, we should not disregard that model performance might have been increased by including addi- tional predictors, such as biotic interactions attributed to the presence of competitors or predators (e.g. Wells, 2007) or by modelling larval density rather than simple species occurrence (see Van Buskirk, 2005).

To conclude, we proposed and successfully illustrated a novel implementation for incorporating the reliability of non- detection records in SDMs of amphibian species using Mediterranean temporary ponds as a model system. We showed that estimating the reliability of absences, an exercise that had been previously seen as a na¨ıve goal in SDMs (Lobo et al., 2010), may be feasible and affordable. A critical point is that, because it is based on a double-sampling scheme, financial costs are reduced by intensively sampling only a limited set of sites. It should be noted, however, that accounting for absence reliability minimizes just one of the potential sources of uncertainty that may affect the perfor- mance of SDMs in these stochastically and dynamically changing environments. SDMs are static in nature (Guisan & Zimmermann, 2000), and hence prediction errors are inevi- table if there is temporal variability in the habitat relationships (Fielding, 2002), as happens in our system of temporary ponds (Go´ mez-Rodrı´guez et al., 2009). As a consequence, future studies should also try incorporating the uncertainty caused by temporal variability in both habitats and populations into SDMs.

ACK N OWLEDG EMENTS
We thank Margarita Florencio Dı´az, Alexandre Portheault and Carlos Marfil Daza for assistance with fieldwork. The Spanish Ministry of Science and Innovation and the EU-FEDER programme (projects REN 2002-03759/GLO, CGL2006-04458/ BOS and CGL2006-02247/BOS, and Fellowship grant AP-2001-3475 to C.G.-R.) and Junta de Andalucı´a (Excellence Research Project 932 to C.D.-P. and Excellence Short-Stay to C.G.-R.) financed this work. A.G. received support from the European Commission ECOCHANGE project.

1   2   3


Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©atelim.com 2016
rəhbərliyinə müraciət