Dissertação - Thálita Cardoso

Centro Brasileiro de Estudos em Ecologia de EstradasUniversidade Federal de Lavras - Lavras - MG - Brasil

THÁLITA DE RESENDE CARDOSO

Dissertação | 2013

1

DOES SHAPE AND SCALE OF ANALYSIS AFFECT THE CONSISTENCY OF

THE FACTORS INFLUENCING SMALL VERTEBRATE ROAD-KILLS?

Thálita de Resende Cardoso1,*, Clara Grilo2 Alex Bager1

1Centro Brasileiro de Estudos em Ecologia de Estradas, Universidade Federal de

Lavras. Campus

Universitário.

37200-000, Lavras, Minas Gerais, Brasil. 2 Departamento de Biologia & CESAM, Universidade de Aveiro, 3810-193

Aveiro, Portugal *Endereço para correspondência: [email protected]

ABSTRACT KEYWORDS: Helicops infrataeniatus, Trachemys dorbgni, Didelphis albiventris,

Myocastor coypus, Conepatus coypus, road segments, buffer analysis, road mortality,

GLM, HP.

1. INTRODUCTION

Wildlife-vehicle collisions (WVC) are considered the most noticeable impact caused

by roads. It is frequently pointed as the main cause of vertebrate mortality by direct

human influence (FORMAN; ALEXANDER, 1998; FAHRIG; RYTWINSKI, 2009). In

the last decades, researchers have used available locations of WVC to model

distribution patterns along roads in order to implement measures to minimize the road

mortality rate (Gunson et al. 2011). These analyses have indicated that WVC along

roads are not randomly distributed but are spatially clustered and therefore, some factors

may promote the road-kills likelihood (e.g. Joyce and Mahoney, 2001).

It is generally agreed that ecological patterns result from processes occurring at

multiple spatial and temporal scales (Collinge, 2001). Integrate knowledge across scales

is needed to characterize the full context of roads and wildlife interactions: from

sites/individuals to landscape/population (e.g. DeCesare 2012). Mortality studies are

mainly focus on one unit of analysis and there is a lack of understanding on the role of

different spatial scales and shapes of units in determining the factors that explain the

likelihood of WVC. This fact prevents accurate comparisons, assessments and

2

extrapolations of the patterns to other regions, and therefore creates problems in the

location of measures to minimize WVC (DANKS AND PORTER, 2010).

In general, differences in units of analysis are related mostly with two features:

shape and size. Two types of shapes in the road-kill analysis are commonly used: a

buffer area around the road-kill spot (e.g. LANGEN et al., 2009; COLINO-RABANAL

et al., 2011) or the surveyed road is segmented in sections with pre-defined length

where the presence of road-kills are assigned to each segment (e.g. GOMES et al., 2009;

GRILO et al., 2011). Regarding the size of the units, the majority of studies used

1mile/km (e.g. LANGEN ET AL., 2009; CARVALHO E MIRA, 2011). Most of the

unit sizes are chosen arbitrarily, ranging from 50m to 500m for buffer radius and from

50m to 1000m for road segments length (citação).

The observation of some phenomena in different scales may deeply affect the

values of prediction variable measured for the region a (Colino-rabanal, 2011, Danks e

Porter, 2010, Ramp et al., 2005) and also its interpretation. The success of the analysis

is condicioned to a careful selection of the best size and approach. However this corcern

is not frequent. Researches have been using analysis and prediction scales with different

sizes and kinds of approaches, sometimes related to the same species (e.g. Cureton II e

Deaton, 2012; Shepard et al., 2008) or animal class (e.g. Caro et al., 2000; Malo et al.,

2004) and often with no foundation to the use of an specific size (e.g. ). This way we

have found results in different scales (e.g. Pinowski, 2005; Cáceres et al., 2010) and

dealing different spatial characteristics related to the same species (e.g. Finder et al.,

1999; Nielsen et al., 2003).

Additionally, no research compared the results obtained by the use of different

shapes and unit sizes in the road-kill analysis. Only a few studies compared the unit size

with the buffer approach to identify the features that promote road-kills (e.g. FARMER

ET AL. 2006; JANET ET AL., 2008, LANGEN ET AL., 2009, DANKS & PORTER

2010; COLINO-RABANAL ET AL., 2011) with mixed results: JANET ET AL (2008)

with deer Odocoileus spp. and COLINO-RABANAL ET AL. (2011) with Iberian wolf

Canis lupus signatus found better predictions with larger scales. On the other hand

DANKS & PORTER (2010) with mouse Alces alces e LANGEN ET AL. (2009) with

reptiles and amphibians found it in smaller sizes.

We argue that the shape and size of unit analysis may influence the results.

Thus, the evaluation of the results consistency over different approaches is needed to

better understand the effect of shape and size on the road-kill analysis.

3

The main goal of this study is to examine consistency of the effect of the shape

and the size of unit analysis in determining the mortality risk for small vertebrate

species with different life-history traits. More specifically, we want to address the

following issues: (1) Do variables related to the road-kill risk vary accordingly to the

size and spatial scale used? (2) Does the weight of those variables are similar in units

with different shape and size? With these insights we intend to contribute for a better

understanding of the role of using different methodologies to determine the factor road

mortality risks and therefore, contribute to a successful application of mitigation

measures.

2. METHODS

2.1 STUDY SITE

The study site is located in the Rio Grande do Sul, Brazil (Fig. 1). This region is

characterized by coastal plains with low and rectilinear relief dominated mainly by

wetlands associated with freshwater and saltwater lakes. This landscape also comprise

extensive perennial and seasonal wet areas, seaside dunes and rice fields (TAGLIANI,

2001). The climate is humid with rainfall evenly distributed throughout the year. The

temperature ranges between -3ºC and 18ºC most of the year, reaching 22ºC during the

hottest months.

A total of 137 km of two Brazilian Federal paved roads were surveyed (33km

from BR392 and 104km from BR 471). The segment started in Pelotas city (52º19’

38’’O, 31º48’18’’S) and the end is on Santa Marta’s Farm (52º41’27’’O, S 32º49’32’’

S), in Santa Vitória do Palmar town. BR 471 is connected an excerpt to Taim ecological

station (ESEC Taim). The ecological station is a Protected Area with full protection and

it is between Mirim Lake and Atlantic Ocean (Fig. 1).

2.2 TARGET SPECIES

We selected five species vulnerable to road traffic which represent different life-

history traits. We analyzed records of two reptiles - water snake (Helicops

infrataeniatus) and the water tiger turtle (Trachemys dorbigni); and three mammal

4

species - white-eared opossum (Didelphis albiventris), nutria (Myocastor coypus) and

skunk (Conepatus chinga).

Water snake is an aquatic species, spread in southern Brazil (Rio Grande do Sul

and Santa Catarina counties), Uruguay and Argentina (Buenos Aires, Santa Fe, Entre

Rios, Corrientes, Chaco, Formosa, Misiones) (AGUIAR & DI-BERNARDO, 2004,

KAWASHITA-RIBEIRO ET AL., 2013). It is a species particularly vulnerable to road

mortality (KUNZ & GHIZONI-JR, 2009; MAINARDI & HARTMANN, 2009;

BAGER & ROSA, 2011, AGUIAR & DI-BERNARDO, 2004; 2005).

Water tiger turtle is a fresh water species, which inhabit swamps, lakes, and

slow-moving rivers in Brazil, Uruguay and Argentina (BAGER ET AL., 2012). Because

it moves slowly this species is a common victim of road traffic (LEMA, 2002,

HENGEMÜHLE & CADEMARTORI, 2008), mostly females during reproduction

season (LEMA, 2002). For example, BAGER & FONTOURA (2013) presented a rate

of 0.23 individuals/100km/day in the Brazilian road BR471.

White-eared opossum is a generalist (Vieira 2006), solitary and omnivorous

species (CABRERA & YEPES 1960). Although this species occupy a broad ecological

niche, it is mostly found in open deciduous forests (CERQUEIRA, 1985) from

Colombia to central Argentina (EMMONS & FEER 1990). With the deforestation raise,

individuals have been approaching urban areas and acquiring synanthropic habits

(SANCHES ET AL., 2012). The high road-kill incidence must be related to the tolerant

and opportunistic habit of the species (HENGEMÜHLE & CADEMARTORI, 2008;

ROSA & MAUHS, 2004; CHEREM ET AL., 2007).

Nutria is an aquatic rodent from South America, which was exported to Europe,

Asia, Africa and North America for fur production (CARTER & LEONARD, 2002).

Their occurrence in river sides and humid zones resulted from escapes and possible

releases from nutria farms (LORI ET AL., 2013). BAGER & FONTOURA (2013)

estimate a rate of 8.25 individuals/100km/day in BR471, Rio Grande do Sul, Brazil.

Skunk is a carnivore widespread in northern Argentina, Uruguay, southern

Brazil, Paraguay, southern Bolivia, South Peru and Northern Chile (EISENBERG &

REDFORD 1999; CHEIDA ET AL., 2006). They inhabit mostly low vegetation areas,

avoiding dense forests (MICHALSKI et al. 2007; LYRA-JORGE et al. 2008). BAGER

& FONTOURA (2013) estimate a road-kill rate of 0.03 individuals/100km/day in

BR471, Rio Grande do Sul, Brazil.

2.3 DATA COLLECTION

5

2.3.1 ROAD-KILL DATA

Road-kill data were obtained from “Estrada Viva” project database. Locations

were recorded weekly from January to December 2005, by car, at low speed

(approximately 50 km/h), with at least two observers, avoiding weekends, holidays and

rainy days. Road surveys were performed between 7am and 3pm and road-kills were

recorded with a handheld GPS (maximum error 5 m).

2.3.2 LAND USE DATA

We built a land use map based on an earlier database and maps from TAGLIANI

(2001). Tagliani’s map was created from Landsat 7 satellite images (2000) using

SPRING software. The size of the imaged scene is 185km x 185km, with spatial

resolution of 30x30m, 15x15m (panchromatic band), 60x60 (thermal band). The

representative land use classes analyzed in the study site were: Predominant Rice Field,

Sandbank vegetation, Seaside fields, Wetlands and Non-native vegetation. Predominant

rice field are areas predominantly covered by irrigates rice culture; Sandbank vegetation

are community’s placed high on dunes and slopes, on dry soil; Seaside fields are flood

fields of low grass; Wetlands are floodplains with fertile clay soils and Non-native

vegetation are forestation of eucalyptus and pine (Eucalyptus sp. and Pinus sp.).

2.3.3 SAMPLE UNITS

For this analysis we used two shapes (buffer and segment) and three unit sizes

related with daily movement, standard scale and dispersal ability (Table 1). We used a

buffer area around the road-kill location and divided the road surveyed in segments with

pre-defined length. Daily movement was estimated using information of the average

home-range size for each species accordingly to BISSONETTE & ADAIR (2008).

According to this study daily movement is the square root of the home-range size. The

standard scale was the measure used in the majority of studies, which was 1000m

length. Dispersal ability is seven times the length of the daily movement

(BISSONETTE & ADAIR 2008). For water snake, no information was available in

terms of home-range size. Thus we used the data of a research on a species of the same

family - liophis (Liophis poecilogyrus) (Hartz et al., 2001) which is present in a site

6

close to our study site. For Nutria the home range values presented in the literature vary

widely, thus we used the average area estimated resulting in 375000m².

Thus, we used the daily movement and dispersal ability of each species as well

the standard scale to define length of road segments and the diameter of buffers (Table

1). We estimate the number of buffers and segments with road-kills and select a similar

number of random buffers and segments without road-kills, using "Hawth's analysis

tools" of ArcGis 9.3 (Environmental Systems Research Institute, Inc., Redlands, CA).

For both approaches we extracted the area of each land use class. 2.4 DATA ANALYSIS

The analysis of consistency of results were performed through Generalized

Linear models (GLM) and hierarchical partitioning (HP). With GLM analysis we

obtained the best model that explain the relationship between presence/absence of road-

kills and land use classes. The proportion of variance explained independently and

jointly by each land use was determined using HP (OLEA, MATEO-TOMAS &

FRUTOS, 2010; MAC NALLY 2000; 2002). We used HP procedures due to the

tendency of linear models being seriously affected by multicolinearity among several

explanatory variables (GRAHAN, 2003). Therefore, GLM and HP were considered

complementary methods since GLM allow the selection of best models and HP identify

the contribution (weight) of each isolated (I’s) and joint (J’s) land use class to analyze

the effect of scales and shape of units to explain the road-kills likelihood.

Firstly, we tested the multicollinearity among land use variables, confronting

them via pair-plots (ZUUR, 2010). From pair of variables with correlation value higher

than 0.7, we just kept the land use variable with higher biological meaning in the GLM

analysis. We then run GLM using all possible combinations of the selected land use

variables. We considered the best model the one with the lowest Akaike Information

Criteria (AIC) (BURNHAM & ANDERSON, 2002).

We used log-likelihood as a goodness-of-fit measure in HP (MAC NALLY,

2002). To assess the statistical significance of variables, 1000 randomizations of the

data matrix were generated to compute I’s distributions for each predictor. Results were

expressed with α=0.05 (Z-score>1.96) (MAC NALLY, 2002). All statistical analyses

were performed using the R statistical software, version 2.9.1 (R DEVELOPMENT

CORE TEAM 2006, VIENNA, AUSTRIA). For GLM analyses we used MuMIn

7

package (Barton, 2012) and Hierarchical partitioning was performed using the

“hiert.part package” Version 1.0 (Walsh and Mac Nally, 2004) .

3. RESULTS

With the road survey, we recorded 186 water snakes, 67 water tiger turtles, 121

white eared opossums, 68 nutrias and 54 skunks. Thus, we defined buffers around the

road-kills and select the same number of random buffers without road-kills. The number

of road segments for the daily movement, standard and dispersal scales were

respectively: 164, 67 and 98 for water snake; 108, 72 and 64 for water tiger turtle; 174,

110 and 114 for white-eared opossum; 68, 64 and 30 for nutria and 64, 72 and 10 for

skunk. Half of these numbers represents the number of segments for each species with

road-kills.

The effect of size and spatial scale on the identification of the factors influencing

the small vertebrate road-kills

Based on the GLM analysis, small variation was found to explain road mortality

likelihood over two approaches and three scales (Table 2). Regarding the shape of units,

we reported that the buffer approach presented more consistent results than the segment

approach, since the number and type of variables in the models did not differ much over

the three scale of analysis (daily movement, standard and dispersal). However, the

segment approach presented better results regarding the correct classification of

observations and the area under curve. The best models were performed under the

standard and dispersal scale: the dispersal scale performed the most accurate model for

water tiger turtle, white-eared opossum and skunk road-kills whereas standard scale

obtained the most accurate model for water snake, nutria and also white eared opossum

road-kills. In terms of AIC weight, all best models did not differ much from the all

combination of variables run, ranging from 0.13 to 0.33, except for nutria at buffer/daily

movement scale (AICweight = 0.97).

Analyzing at which scale unit was the best for each shape approach in terms of

model accuracy, we also found few differences. However, for buffer approach, the

standard scale fitted better for all species except for water tiger turtle, whereas for

segment approach the best scale of analysis was: daily movement for water snake;

8

standard for nutria, dispersal for skunk, both standard and dispersal for white-eared

opossum and all scales showed similar results for water tiger turtle.

In the assessment of consistency among GLMs, we observed that several models

showed the same variables. The majority of models with the same variables were

between standard and dispersal scales for the following species: water tiger turtle (two

models), white-eared opossum (one model), nutria (one model), and skunk (one model).

We reported similar models between daily-movement and standard scales for all species

(four models) except skunk. Equally, for daily-movement and dispersal we found

similarities for water tiger turtle (one model) and white-eared opossum (two models).

At least one variable was present throughout all the best models for each species,

except for nutria where the seaside field variable occurred in four of six models. The

variables that enter in every model were: seaside fields for water snake; predominant

rice field for water tiger turtle; seaside fields, sandbank vegetation and non-native

vegetation for white eared opossum; and predominant rice field for skunk.

The effect of size and spatial scale on the weight of variables to explain road-kills

When analyzing the independent and the joint effect in terms of variance

explained of variables for road-kill likelihood, we reported that the effect of

independent variables was consistent but the joint effects varied considerably. In fact,

standard scale and segment approach were the most consistent units of analysis. For

water snake, we found that variables have similar importance, except the joint effect of

the sandbank vegetation and non-native vegetation at segment approach, which

explained most of the variance in the standard and dispersal scale, respectively. No

variation on the effect of variables was found among the variance explained for water

tiger turtle among the shape and size of units. For white-eared opossum, we found joint

differences between buffer and segment approach regarding the daily movement scale,

namely for predominant rice filed and sandbank vegetation. For nutria there were no

huge differences on the independent and joint effects, except for the joint effect of

sandbank vegetation and wetlands in dispersal scale at buffer and segment approaches.

For skunk, the effects of variables did not changed along the scale and shape of

analysis, except the wetlands and sandbank vegetation joint effect at segment for the

daily movement scale and, the independent effect of non-native vegetation at buffer

approach in the standard scale.

9

When comparing the presence of variables in the best GLM with the percentage

of the variance explained by the joint effect of variables in HP, we found some

differences. In fact, we have reported many variables with high effect in the HP analysis

which were absent in the GLM. For example, variables with high effects in the water

snake HP analysis (predominant rice field and sandbank vegetation) were absent in the

GLM except at the buffer /dispersal and buffer/standard approaches, respectively. For

water tiger turtle, seaside field was an important variable in the HP analysis, which is

only present in the segment/daily movement approach at GLM. Wetlands have a high

effect on the HP analysis for white-eared opossum, but it was not selected in the best

GLM. Similar observation occurred when run the GLM for skunk. Sandbank vegetation

was an important variable in the HP analysis but only occurred in the buffer/dispersal

approach for nutria.

Summarizing, we found more consistency between GLM and HP analysis with

the buffer approach at standard and dispersal scales, since they presented more variables

with high variance explained in the HP and were also present in GLM. The best

examples for this fact were water tiger turtle and white-eared-opossum models (Figure

2).

4. DISCUSSION

The definition of unit analysis is a critical issue to understand the ecological

processes (McMahon & Diez, 2007). Inferences from an observation may be biased, if

the scale and shape of analysis does not detect the spatial species requirements (Cadotte

& Fukami 2005). There is little knowledge on the effect of scale and shape on the road-

kill analysis. Such information is valuable for accomplishing mitigation objectives as

the definition and placement of the most appropriate measure to prevent WVCs.

In this study we wanted to clarify this issue by using different shapes and scale

of analysis based on the spatial requirements of five small vertebrate species with

different life-history traits.

Overall, our results show that road-kills are better explained at broader scales

(standard and dispersal scales) and segments approaches. However, our findings show

that the definition of scale and shape of units is complex and may be species-specific.

The similarity of results among GLMs indicates that scale and shape seems to

have little influence on the identification of variables explaining the road-kill

10

occurrence of small vertebrates. However, Hierarchical Partitioning analyses show that

the importance of variables to explain road-kills occurrence seems to be affected by the

scales and shape of units. Moreover, in several cases, the most important variables in

the HP analysis were not present in the best GLM models. O uso combinado de GLM e

HP pode fornecer conclusões firmes sobre as relações espécie-ambiente (Frutos et al.,

2007). The incongruences between GLM models and HP analyses may have occurred

due to an undetected degree of multicollinearity, despite the removal of variables with a

correlation above 0.7. In fact, HP analyses are used as a tool against multicollinearity

(MacNally 2000; 2002; MacNally & Walsh, 2004), which prevent the exclusion of

ecologically important variables whereas variables that statistically better explain the

data are kept in the models (Graham, 2003). Therefore, we used HP to complement the

analysis of GLM determinando a contribuição independente que cada variável tem

sobre a variável resposta e separando da contribuição conjunta, resultante da correlação

com outras variáveis (Frutos et al., 2007). For instance, we detected through GLM that

the proximity with predominant rice field is the variable responsible for a high

likelihood of skunk road-kills. However, wetlands emerged as the variable with highest

effect on road-kills occurrence in the HP. This factor may have implication on the

definition and location of roadkill mitigation measures.

One explanation of this dissimilarity is the home-range size and the generalist

character of the species. Home range is the area used by an individual to forage,

reproduce and perform all of its daily activities (BURT 1943, citação). This area varies

according to animal size and feeding habits (citação). Analyzing in the following

sequence: water snake, water tiger turtle, white-eared opossum, nutria e skunk, we

observed that while the home ranges increased, also increased the differences among the

variables in the models (especially regarding HP results). Moreover, the species which

presented models completely equal on scales and approaches were the two species with

similar and intermediary home ranges (17424m² for water tiger turtle, 23300m² for

white-eared opossum). Species with larger home ranges and dispersal capacities are

generally less specialists regarding habitat, being able to explore distinct places

according to resource supply (KREBS & DAVIES, 1996, citação). The result

differences can be explained by a larger diversity of explored habitats by the species

with larger home ranges (nutria e skunk). We observed that large home-ranges are

realted with high variations among the most important variables which suggest that

species with higher home ranges, the differences among scales and approaches may be

more substantial. We also haven’t found any relation between the scale of analyses and

11

dispersal ability of the species. Standard scale seems to be a reasonable solution as it

worked well for species with both low and high home-range values. However, several

studies with larger body sized species are in line with our findings (JANET ET AL.,

2008, COLINO-RABANAL ET AL. 2011). For example, among three scale of analysis,

COLINO-RABANAL et al. (2011) found better results at broader scale (buffer of 1000

and 5000 meter radius) for Iberian wolf. Equally, JANET ET AL. (2008) using road-kill

data of North American deer identified, among the sizes of 100, 200, 400 and 800m

radius, found better results with the highest radius.

In fact, our results are based on data of small vertebrate species with territories

ranging from 0.0003 to 1.65km². Despite this similarity with our results, both researches

were done with large mammals which have high dispersal ability (according to

Mattisson et al. (2013), from 259 to 1.676 km² home range for the Wolf, and according

to Walter et al. (2013) from 8 to 16.2Km² for the deer). There were also other

researches which obtained complementary results according to the analyses they

intended to do. DANKS & PORTER (2010, used data from the moose species Alce

alces, among scales with 0.25, 0.5, 1, 2.5 e 5 Km radius. They identified that the scale

with 2.5km radius better predict the landscape variable composition, furthermore, the

scale with 5.0km better predict the landscape variable configuration. In the other hand,

FARMER et al. (2006), using data from Odocoileus hemionus sitkensis, among the

scales (50, 500 and 1000m radius), the habitat factors in scales with 500 and 1000m

scales had a higher effect on adult and young female roadkill, whilew the characterístics

of nearby habitats had a higher effect on adult male roadkill. No research had as a

result the best explanation in fine scale. Then we can conclude that the results are better

as larger the area of the analysis unit is, regardless the dispersion capacity of the

species. This factor shows the need of the species to use the landscape to explore

different places according to the resource supply to satisfy different needs. (KREBS &

DAVIES, 1996).

Most studies do not present any scientific support to explain why some image

resolution was chosen to generate the land-use map used to analyze data on a species

(citação). In fact, we can raise the hypothesis that some important fine-scale differences

may have not been detected since we used a land-use map generated from an image

with 30m resolution (Tagliani, 2001). As smaller the animal’s dispersal capacity is, the

larger should be the scale of the land use variables so the differences may be perceived

(LANGEN ET AL. 2009). When analyzing the relationship between herpetological

12

fauna and land use LANGEN ET AL. (2009) using a high resolution map of 1m,

obtained better results with smaller scales (100m) than for 500 and 1000m.

There is also another factor related to study area that may explain this disparity

in results. Our area is inserted in a region of pioneer vegetal formations with seasonal

floods and strong culture on irrigated rice. The rice culture in the area represents 79% of

the country’s production (EMBRAPA, 2004; CORDEIRO E HASENACK, 2009). There

are a few differences in the soil use of the region, having a strong relation with

agriculture. Evidences of this agricultural use are irrigation/drainage channels and

“taipas” resultants from the irrigated rice culture in wetlands and seaside fields, as well

as in terraces on dry fields. (CORDEIRO E HASENACK, 2009). The studied species often

showed use of these environments since they are well adapted to seasonal water

conditions (citação). Seaside fields and predominant rice fields are the more frequent

variables on the models. Both variables have a strong water factor, and it is a habitat

characteristic limiting to the development and nesting of most of the studied species,

which corroborates with the similarity found in the results.

Management Implications

In general we found out that broader scales and segments shown to be the best

analysis units for small vertebrates. The best analysis scales are particular of each

species and the ideal would be preliminary multi-scale studies. Complementarily, we

sugest new researchs using land-use maps with better resolutions, analysing species

with different home ranges and in a region with a very heterogeneous landscape.

However, dealing with each species at a time, we saw more contrasting results

with distinct variables on the models. Those variations become more important when

we deal with the application of those results. The application of mitigation measures

based on researches with arbitrary use of scales is fragile. We highlight with our

research the importance of a criterion for the use of scale and shapes of analysis units on

roadkill studies. Our research opens new grounds when we discuss different scales and

shapes for the analysis units, considering species with different life-history traits. For

this reason it was supposed to have complex results and a large raise of hypothesis and

discussions.

5. ACKNOWLEDGEMENTS

13

6. REFERENCES

ANDREWS, K.M.; GIBBONS, J.W.; JOCHIMSEN, D.M. 2008. Ecological effects of roads on amphibians and reptiles: a literature review. Herpetological Conservation 3:121-143.

Barton, K., 2012. Package ‘MuMIn’. Model selection and model averaging based on information criteria. R package version 1.7.11.

Beaudry, F., P. G. de Maynadier, and M. L. Hunter, Jr., 2008. Identifying road mortality threat at multiple spatial scales for semi-aquatic turtles. BIOLOGICAL CONSERVATION 141:2550-2563.

BENÍTEZ-LÓPEZ, A.; ALKEMADE, R.; VERWEIJ, P.A. 2010. The impacts of roads and other infrastructure on mammal and bird populations: a meta-analysis. Biological Conservation 143 (6): 1307-1316.

CÁCERES, N.C. 2002. Food habitats and seed dispersal by the White-Eared Opossum Didelphis albiventris in southern Brazil. Stud. Neotrop. Fauna & Environm. 37:97-104.

CARO, T. M. ; SHARGEL, J. A.; STONER, C. J. 2000. Frequency of Medium-sized Mammal Road Kills in an Agricultural Landscape in California. The American Midland Naturalist, 144(2):362-369.

CLEVENGER, A. P.; CHRUSZCZ, B.; GUNSON, K. E. 2003. Spatial patterns and factors influencing small vertebrate fauna road-kill aggregations. Biological Conservation 109: 15-26.

COELHO, I.P.; KINDEL, A.; COELHO, A.V.P. 2008. Roadkills of vertebrate species on two highways through the Atlantic Forest Biosphere Reserve, southern Brazil. European Journal of Wildlife Research, 54(4): 689-699.

Colino-Rabanal, V. J.;Lizana, M.; Peris, S. J. 2011. Factors influencing wolf Canis lupus roadkills. Eur J Wildl Res (2011) 57:399–409

Collinge SK (2001) Spatial ecology and biological conservation - Introduction. Biol Conserv 100:1–2

Cureton II, J. C.; Deaton, R. 2012. Hot moments and hot spots: Identifying factors explaining temporal and spatial variation in turtle road mortality. The Journal of Wildlife Management; DOI: 10.1002/jwmg.320

Danks, Z.D. and W.F. Porter. 2010. Temporal, spatial, and landscape habitat characteristics of moose-vehicle collisions in western Maine. Journal of Wildlife Management 74:1229-1241.

DEWOODY, A.; NOGLE, J. M.; HOOVER, M.; DUNNING, B. 2011. "Monitoring and Predicting Traffic Induced Vertebrate Mortality Near Wetlands". Joint Transportation Research Program. Paper 1114.

FAHRIG, L.; RYTWINSKI, T. 2009. Effects of roads on animal abundance: an empirical review and synthesis. Ecology and Society 14:1–21.

FINDER, R.A., ROSEBERRY, J.L. & WOOLF, A. 1999. Site and landscape conditions at white-tailed deer/vehicle collision locations in Illinois. Land. Urban Plan. 44:77-85.

14

Glista, D.J., DeVault, T.L., DeWoody, J.A., 2007. Vertebrate road mortality predominantly impacts amphibians. Herpetological Conservation and Biology, (3) 77-87.

GRILO, C.; CRAMER, P. C;. BISSONETTE, J. A. 2011. Mitigation measures to reduce impacts on biodiversity. pp. 73-114 in: Frank Columbus (ed.) Highways: Construction, Management, and Maintenance. Nova Science Publishers, Inc.

GRILO, C.; BISSONETTE, J.A.; SANTOS-REIS, M. 2009. Spatial-temporal patterns in Mediterranean carnivore road casualties: Consequences for migration. Biological Conservation 142: 301-313.

GUNSON, K.E.; MOUNTRAKIS, G.; QUACKENBUSH, L. J. 2010. Spatial wildlife-vehicle collision models: A review of current work and its application to transportation mitigation projects. Journal of Environmental Management.

JAARSMA, C.F.; LANGEVELDE, F.V.; BAVECO, J.M.; EUPEN, M.V.; ARISZ, J. 2007. Model for rural transportation planning considering simulating mobility and traffic kills in the badger Meles meles. Ecological Informatics 2: 73–82.

NG, J.W.; NIELSEN, C.; CLAIR, C.C.S.T. Landscape and traffi c factors infl uencing deer–vehicle collisions in an urban enviroment Human–Wildlife Confl icts 2(1):34–47

JOYCE, T.L.; MAHONEY, S.P. 2001. Spatial and temporal distributions of moose-vehicle collisions in Newfoundland. Wildlife Society Bulletin 29, 281e291.

LANGEN, T.A.; OGDEN, K.M.; SCHWARTING, L.L. 2009. Predicting hot spots of herpetofauna road mortality along highway networks. Journal of Wildlife Management 73: 104-114.

Mac Nally, R., 2000. Regression and model-building in conservation biology, biogeography and ecology: the distinction between and reconciliation of ‘predictive’ and‘explanatory’ models. Biod. Conserv. 9, 655–671.

Mac Nally, R., 2002. Multiple regression and inference in ecology and conservation biology: further comments on identifying important predictor variables. Biod. Conserv. 11, 1397–1401.

MALO, J. E.; SUÁREZ, F.; DÍEZ, A. 2004. Can we mitigate animal–vehicle accidents using predictive models?. J Appl Ecol 41:701–710.

NIEDZIALKOWSKA, M.; JEDRZEJEWSKIA, W.; MYSŁAJEKB, R. W.; NOWAKB, S.; JE˛DRZEJEWSKAA, B.; SCHMIDT, K. 2006. Environmental correlates of Eurasian lynx occurrence in Poland – Large scale census and GIS mapping. 33: 63 69.

Nielsen CK, Anderson RG, Grund MD (2003) Landscape influences on deer-vehicle accident areas in an urban environment. J Wildl Manage 67:46–51

PINOWSKI, J. 2005. Roadkills of vertebrates in Venezuela. Revta Bras. Zool. 22(1):191-196.

RAMP, D.; CALDWELL, J.; EDWARDS, K.A.; WARTON, D.; CROFT, D.B. 2005. Modelling of wildlife fatality hotspots along the snowy mountain highway in New South Wales, Australia. Biol Conserv 126:474–490.

RAMP, D.; WILSON, V.K.; CROFT, D.B. 2006. Assessing the impacts of roads in periurban reserves: road-based fatalities and road usage by wildlife in the Royal National Park, New SouthWales, Australia. Biological Conservation 129: 348-359.

15

ROGER, E.; RAMP, D. 2009. Incorporating habitat use in models of fauna fatalities on roads. Divers Distrib 15:222–231.

ROGER, E.; LAFFAN, S. W.; RAMP, D. 2011. Road impacts a tipping point for wildlife populations in threatened landscapes. Popul Ecol 53:215–227.

Shepard, D.B., Dreslik, M.J., Benjamin, C.J., Phillips, C.A., 2008. Reptile road mortality around an oasis in the Illinois corn desert with emphasis on the endangered eastern massasauga. Copeia 2, 350-359.

SNOW, N. P.; ANDELT, W. F.; GOULD, N. P. 2011. Characteristics of road-kill locations of San Clemente Island foxes. Wildlife Society Walsh, C., Mac Nally, R., 2004. “The hier.part Package” Hierarchical Partitioning. Documentation for R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, http://www.rproject.org.

Bulletin 35(1): 32-39.

Walsh, C., Mac Nally, R., 2004.“The hier.part Package” Hierarchical Partitioning. Documentation for R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, http://www.rproject.org.

Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A., Smith, G.M., 2009. Mixed Effects Models and Extensions in Ecology with R. Springer, New York.

16

TABELAS E FIGURAS

Figure 1- Study area.

17

Table 1. Average home range size (m²) for each species obtained through the literature and the estimated daily movements, standard, and dispersion ability scales accordingly to BISSONETTE & ADAIR (2008).

Common name Home range Reference Daily movement Standard DispersalWater Snake 3000 Hartz et al., 2001 62 1000 383

Water tiger turtle 17424 Bager et al. (2012) 150 1000 924White-eared opossum 23300 Sanches et al. (2012) 172 1000 1068

Nutria 375000 Nolfo-Clements (2009) 692 1000 4286Skunk 1650000 Kasper et al. (2011) 1450 1000 8991

18

Table 3 – Best GLM models for buffer and segment approaches and for the three scales of analysis to estimate the road-kill likelihood of species Water Snake,Water tiger turtle, White-eared opossum, Nutria and Skunk.

Species Approach Scale Variable Coefficients S.E. P -value AICweight Correct Classification

AUC

Water snakeBuffer

Daily movement (Intercept) -1.611216 0.250351 0.31 0.70 0.75Seaside Fields 0.000898 0.000108 0.00Wetlands 0.000575 0.000124 0.00

Standard (Intercept) -2.111000 0.291000 0.28 0.78 0.86Seaside Fields 0.000005 0.000000 0.00Sandbank vegetation -0.000006 0.000002 0.01Non-native vegetation 0.000002 0.000002 0.11

Dispersal (Intercept) -3.031000 0.797900 0.16 0.77 0.80Predominant Rice Field 0.000011 0.000008 0.00Wetlands 0.000018 0.000008 0.00Seaside Fields 0.000037 0.000007 0.00Non-native vegetation 0.000024 0.000010 0.01

SegmentDaily movement (Intercept) -0.737145 0.237514 0.21 0.67 0.68

Seaside Fields 0.000493 0.000114 0.00Standard (Intercept) -0.638300 0.368600 0.28 0.63 0.64

Seaside Fields 0.000002 0.000001 0.02Dispersal (Intercept) -0.710400 0.327000 0.33 0.58 0.66

Seaside Fields 0.000010 0.000004 0.02Non-native vegetation 0.000050 0.000032 0.05

Water tiger turtle Buffer

Daily movement (Intercept) 0.574300 0.237900 0.20 0.66 0.70Predominant Rice Field -0.000136 0.000040 0.00Wetlands -0.000043 0.000029 0.19Non-native vegetation -0.000204 0.000104 0.01

Standard (Intercept) 0.737900 0.270700 0.20 0.64 0.71Predominant Rice Field -0.000003 0.000001 0.00Wetlands -0.000002 0.000001 0.13Non-native vegetation -0.000007 0.000004 0.03

Dispersal (Intercept) 0.734900 0.269100 0.21 0.65 0.71Predominant Rice Field -0.000004 0.000001 0.00Wetlands -0.000002 0.000001 0.15Non-native vegetation -0.000008 0.000004 0.02


Predominant Rice Field -0.000107 0.000043 0.00Seaside Fields 0.000065 0.000029 0.02

Standard (Intercept) 0.489000 0.282900 0.13 0.69 0.69Predominant Rice Field -0.000003 0.000001 0.00

Dispersal (Intercept) 0.587400 0.309400 0.23 0.70 0.72Predominant Rice Field -0.000004 0.000001 0.00

White-eared opossum Buffer

Daily movement (Intercept) -0.734700 0.218000 0.23 0.63 0.66Seaside Fields 0.000048 0.000013 0.00Sandbank vegetation 0.000114 0.000043 0.00Non-native vegetation 0.000073 0.000042 0.07

Standard (Intercept) -0.845900 0.232100 0.17 0.64 0.68Seaside Fields 0.000002 0.000000 0.00Sandbank vegetation 0.000006 0.000002 0.01

Dispersal (Intercept) -1.011000 0.257700 0.17 0.63 0.69Seaside Fields 0.000002 0.000000 0.00Sandbank vegetation 0.000006 0.000002 0.00Non-native vegetation 0.000002 0.000002 0.14


Seaside Fields 0.000057 0.000016 0.00Sandbank vegetation 0.000111 0.000048 0.01Non-native vegetation 0.000054 0.000037 0.14

Standard (Intercept) -0.822500 0.329200 0.20 0.68 0.68Seaside Fields 0.000002 0.000001 0.02Sandbank vegetation 0.000005 0.000003 0.14Non-native vegetation 0.000004 0.000003 0.07

Dispersal (Intercept) -0.826000 0.327100 0.20 0.59 0.68Seaside Fields 0.000001 0.000001 0.03Sandbank vegetation 0.000009 0.000005 0.01Non-native vegetation 0.000004 0.000003 0.12

19

NutriaBuffer

Daily movement (Intercept) -0.862400 0.278000 0.33 0.67 0.71Seaside Fields 0.000005 0.000001 0.00

Standard (Intercept) -0.934800 0.286600 0.28 0.68 0.71Seaside Fields 0.000003 0.000001 0.00

Dispersal (Intercept) -0.773300 0.357200 0.22 0.65 0.71Seaside Fields 0.000000 0.000000 0.00Sandbank vegetation -0.000001 0.000000 0.14


Seaside Fields 0.000002 0.000002 0.23Standard (Intercept) -0.226400 0.289700 0.10 0.61 0.69

Wetlands 0.000002 0.000002 0.11Dispersal (Intercept) -0.457400 0.497700 0.16 0.67 0.62

Wetlands 0.000000 0.000000 0.15Skunk

BufferDaily movement (Intercept) -0.340900 0.262400 0.20 0.59 0.61

Predominant Rice Field 0.000001 0.000000 0.05Standard (Intercept) -0.524500 0.288200 0.20 0.60 0.62

Predominant Rice Field 0.000001 0.000001 0.08Non-native vegetation 0.000007 0.000004 0.05

Dispersal (Intercept) -0.137700 0.380400 0.22 0.65 0.62Predominant Rice Field 0.000000 0.000000 0.01Non-native vegetation 0.000000 0.000000 0.12


Predominant Rice Field 0.000001 0.000000 0.01Standard (Intercept) -0.826800 0.375000 0.26 0.64 0.72

Predominant Rice Field 0.000002 0.000001 0.01Sandbank vegetation 0.000010 0.000007 0.05

Dispersal (Intercept) 1.557000 1.383000 0.29 0.90 0.96Predominant Rice Field 0.000001 0.000002 0.13Non-native vegetation -0.000002 0.000001 0.01

20

Figure 2 –Percentage of variance in the occurrence of road-kills for each species explained independently (I) and jointly (J) by the five land use variables.

Documents

Dissertação - Thálita Cardoso