Professional Documents
Culture Documents
First Records On Habitat Use of Semi-Feral Cattle in Southern Forests: Topography Reveals More Than Vegetation
First Records On Habitat Use of Semi-Feral Cattle in Southern Forests: Topography Reveals More Than Vegetation
2
3
First records on habitat use of semi-feral cattle in southern forests: topography reveals
5
6
Nicols Seoane
7
8
Laboratorio Ecotono, Universidad Nacional del Comahue, Quintral 1250, (8400) Bariloche,
10
11
12
13
14
15
16
17
18
19
20
21
22
23
NICOLS SEOANE
24
ABSTRACT
25
26
While the presence of cattle in forests is quite common, how they use the habitat is often
27
overlooked. When this is examined, most studies focus on measurements of the vegetation
28
variables influencing habitat selection. This current study provides a suitable model to study
29
habitat use by livestock in forested areas. The model was applied to data from semi-feral cattle in
30
order to obtain the first description of their habitat use in southern forests. Furthermore, the
31
model accounted for individual variability, and hinted at population patterns of habitat use. The
32
positions of four individual animals with GPS collars were recorded during four months in a
33
Nothofagus (southern beech) forest in Patagonia (Argentina). By projecting these GPS location
34
data into a geographical information system (GIS), a resource selection probability function
35
(RSPF) that considers topographic and vegetation variables was built. The habitat use of semi-
36
feral cattle in southern beech forests showed a large interindividual variability, but also some
37
similar characteristics which enable a proper description of patterns of habitat use. It was found
38
that habitat selection by cattle was mainly affected by topographic variables such as altitude and
39
the combination of slope and aspect. In both cases the variables were selected below average
40
relative to availability, suggesting a preferred habitat range. Livestock also tended to avoid areas
41
of closed shrublands and showed a slight preference for pastures (meadows). This result suggests
42
that cattle are showing adaptations to the major features of these mountainous forests due to
43
ferality. Furthermore, these results are the basis for management applications such as predictive
44
45
46
NICOLS SEOANE
47
KEYWORDS
48
49
Nothofagus forests; beef cattle; semi feral cattle; resource selection probability functions (RSPF);
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
NICOLS SEOANE
70
INTRODUCTION
71
72
Cattle (Bos taurus) are an important species for forest dynamics globally, and the originally
73
domesticated animals have transgressed to become semi-feral to feral again in many areas
74
(Decker et al. 2015, Berteaux and Micol 1992, Hernandez et al. 1999, Baker 1990, Atkinson
75
2001, Lazo 1994). Livestock can affect the diversity and structure of plant communities
76
(Coughenour 1991, Cingolani et al. 2008) mainly through diet selection (Rook et al. 2004), but
77
also due to indirect actions such as changing the distribution of nutrients and compacting the soil
78
by trampling (Milchunas et al. 1998, Cingolani et al. 2008). Thus, it is possibly one of the main
79
ecosystem engineers for these environments, as has already been demonstrated in grasslands
80
(Jones et al. 1997). Although the impacts of cattle are often studied, their habitat use, which
81
82
83
Southern beech forests (i.e., Nothofagus forests) in Patagonia have been used to raise
84
cattle since the late 19th century. The traditional management involves extensive cattle ranching
85
in the forest. But occasionally and for different reasons, some animals escape the herd,
86
establishing feral populations (Atkinson 2001, Veblen et al. 1996). Although feral cattle can be
87
found in many areas of the Nothofagus forests in Patagonia (Novillo and Ojeda 2008) and New
88
Zealand (Howard 1966, Veblen and Stewart 1982), no studies have been performed on their
89
behavior or habitat selection in southern forests. Additionally, many levels of ferality already
90
exist depending on contact with human settlements, hunting pressure, isolation time, etc.
91
92
The proportion of time an animal spends in a location to meet their biological demands
defines its habitat use (Beyer et al. 2010). As resources are often heterogeneously distributed,
NICOLS SEOANE
93
individuals need to explore multiple environments to satisfy their needs (Law and Dickman
94
1998). Therefore, the selection of habitats is a key feature of behavior and population dynamics
95
(Morris 1987). This is especially relevant for large herbivores, because their distribution on the
96
landscape can decisively influence the productivity and biodiversity of fields and pastures
97
98
99
There are multiple relationships between livestock and their environment but one of the
major ones is the change in the heterogeneity of vegetation produced by foraging (Adler et al.
100
2001). In the case of forests, despite having high heterogeneity and frequent cattle presence,
101
there is no systematic review of the resulting ecological processes due to the introduction of this
102
103
activity patterns of semi-wild cattle in general, which makes their management even more
104
difficult. Some studies that have been conducted in different forests account for diet selection
105
along seasons (Marquardt et al. 2010) or in relation to anthropogenic changes, usually due to
106
forest management practices (Roath and Krueger 1982, Kaufmann et al. 2013). In these studies,
107
vegetation type and topography were the most common variables to explain habitat selection by
108
cattle in forests.
109
A variety of technological and statistical tools can be applied to study habitat use. Among
110
the first, radiotelemetry and GPS devices allow for accurate and frequent data collection on
111
animal location, which facilitates the development of mechanistic models of habitat use and
112
movement (Beyer et al. 2010). Therefore, their use is widespread (Johnson et al. 2004, Ungar et
113
al. 2005, Cagnacci et al. 2010, Peinetti et al. 2011). In addition to these data collection methods,
114
habitat use can be modeled by resource selection functions (RSF) and resource selection
115
probability functions (RSPF) that estimate the probability that a given environment has been
NICOLS SEOANE
116
used by an animal (Boyce and McDonald 1999, Beyer et al. 2010). Thus, the RSFs provide a
117
framework for developing an ecological theory of habitat use and allow linking landscape
118
ecology and population biology (Boyce and McDonald 1999, Boyce 2006). One way to build a
119
RSPF with GPS data is to model the intensity of use in sampling units, relating the relative
120
frequency of records per unit to the environmental features of such sampling units (Nielson and
121
Sawyer 2013).
122
The objective of this study was to evaluate the habitat use of semi-feral cattle in southern
123
beech forests as no current data or a common methodology exists for this purpose. Furthermore,
124
the use of vegetation variables is usual in studies on habitat use by cattle, but the role of
125
126
Specifically, this study address the following questions: (i) What is the habitat use of semi-feral
127
cattle in Nothofagus forests?; and (ii) Are the vegetation variables of the landscape more
128
important than topography for habitat use by cattle?. These research questions were addressed by
129
using a utilization distribution approach and the construction of a resource selection probability
130
131
132
133
METHODS
134
135
- Study Area
136
The study was carried out in the Llodconto Valley, Northwest Patagonia (Rio Negro, Argentina).
137
The valley is within the Nahuel Huap National Park, between 4121' and 4127' south latitude,
138
and 7131' and 7141' west longitude, at an elevation ranging from 920 to 2000 m.a.s.l.. It is a
NICOLS SEOANE
139
mountainous area with cold-temperate to moist-cold climate with prevailing westerly winds. The
140
summers are dry and winters rainy, with a mean monthly temperature ranging from 2.4C in July
141
to 12.9C in January. The area is part of the Andean-Patagonian forest and includes a wide
142
variety of habitats such as forests, shrublands, meadows, burned forest, riparian and high
143
mountain areas. The dominant species is Nothofagus antarctica which is one of the main
144
components of both forests and shrublands. Other species such as the Nothofagus pumilio tree,
145
the Schinus patagonicus shrub, and the Chusquea coleou bamboo are also abundant. An
146
uncertain number of free-ranging cattle inhabit the valley and are traditionally and minimally
147
148
149
150
In order to capture the animals, an enclosure located at the center of the valley was used to herd
151
the animals into. Four female adults caught in 2012 and one more caught in 2013 were fitted
152
with custom-made GPS collars (GPS Module ZX4120 and Amicus GPS Shiled Antenna,
153
Crownhill Associates Ltd.). These GPS collars store data on board and recapture was necessary
154
to recover the data. Each collar was set to record location every ten minutes and these records
155
were later screened by hour in order to standardize the records and eliminate the unacquired data.
156
All capture and handling methods used in this study met the standards of animal welfare
157
practices recommended for scientific research (Rollin and Kessel 1998, Gannon and Sikes 2007).
158
All research was approved by the proper authorities of the protected area.
159
Topographic mapping and spatial statistical analysis were conducted using Quantum
160
GIS software (QGIS) version 1.8.0 using a digital elevation model (ASTER GDEM) as the
161
data source. Animal locations were overlayed on the digital elevation model and a minimum
NICOLS SEOANE
162
convex polygon (MCP) with a 200 meter buffer was used around the complete set of GPS
163
positions to define our study area. One thousand sampling units of 100 meters radii each (i.e.,
164
sampling unit area= 0.03km2) were allocated randomly inside this area.
165
Habitat types were assigned for each sampling unit by means of a visual classification on
166
Google Earth ( Google Inc.), identifying the following vegetation types: forests, shrubland,
167
burned areas, meadows, open areas (vegetation cover <25%) and high Andean vegetation. All
168
169
Corresponding values for elevation, slope, aspect, hill-shade and ruggedness were extracted from
170
GIS for each sampling unit. All these were used as predictor variables in the habitat use model.
171
An additional variable called "NWness" was created to take into account hillside exposure. The
172
rationale for this is based on strong west winds and large thermal amplitudes due to sunlight
173
exposure in north facing sites that shape the rigorous weather in these zones. NWness was
174
calculated as the cosine of the hillside aspect minus 310 degrees. Thus, this variable accounts for
175
the exposure of the slope with a value ranging from 1 (maximum exposure) to -1 (minimum
176
exposure).
177
Highly correlated variables (r > 0.6) were not used together in the same model. Squared
178
elevation and slope were included in the models for the purpose of evaluating preferred ranges of
179
use by animals. In order to account for a strong regional climatic pattern, the interaction between
180
slope and NWness was also considered since steep and northwest exposed slopes are generally
181
182
183
- Data analysis
184
There are several ways to model habitat use. The most widely used include the GLMs and related
NICOLS SEOANE
185
models (Guisan and Zimmermann 2000), but multivariate approaches (Clark et al. 1993, Burke
186
et al. 2013), GIS-based habitat suitability models (Osborne et al. 2001, Store and Jokimki
187
2003), individual based models (Railsback and Harvey 2002), artificial neural networks
188
approaches (zesmi and zesmi 1999), and maximum entropy modeling (Baldwin 2009) also
189
exist. Here I applied a utilization distribution approach described by Nielson and Sawyer (2013)
190
to develop the RSPF because it accounts for intensity of use among habitats and is unbiased in
191
192
193
relative frequency of positions within each sampling unit as an empirical estimator of the use
194
distribution (Millspaugh et al. 2006). This relative frequency of positions was used as a response
195
variable in a generalized linear model (GLM) with negative binomial distribution that accounts
196
for over-dispersed count data. In order to relate the observed counts in a sampling unit with the
197
total number of positions recorded for each animal, an offset term was included, thus allowing for
198
199
200
201
yi ~ NegativeBinomial i ,
202
203
204
where:
205
yi = number of observations at the i-th sampling unit; = relative frequency in each sampling
206
unit by animal; = overdispersion parameter; total = number of recorded positions in the whole
207
NICOLS SEOANE
208
Several models exploring different combinations of predictor variables were fitted for
209
each animal and compared using the Akaike information criterion (AIC). The overdispersion
210
parameter of the negative binomial distribution was estimated simultaneously with the models
211
using the glm.nb function in R (Venables and Ripley 2002, R Core Team 2013 version 3.0.2). A
212
set of ecologically plausible models was chosen to work with all animals based on a stepwise
213
regression. The package glmulti (Calcagno and Mazancourt 2010) was used as a tool for quickly
214
identifying the best models for each animal. The deviance (Guisan et al. 2000) was used as a
215
216
217
experimental unit, which reflects the individual nature of resource selection (Marzluff et al.
218
2004, Millspaugh et al. 2006). A single set of covariates was selected considering the comparison
219
among models and the biological relevance of the variables. The estimated coefficients, variance
220
and confidence intervals were calculated for each parameter of this general proposed model.
221
222
1
k = kj
n
223
where kj is the estimated coefficient k for each individual j (j=1,2,3,4) and n is the total number
224
of animals (n=4).
225
The variance of the estimated coefficients for the population model was computed as:
226
227
228
229
Var ( k )=
1
2
( ij ij )
n 1 j= 1
NICOLS SEOANE
230
RESULTS
231
232
A total of 2575 position records (an 87% fix rate for the GPS collars) was obtained from 4 of the
233
animals. One of the collared animals could not be relocated during the study and therefore was
234
not recaptured.
235
Based on the fitted models for each animal, the general pattern observed that topographic
236
variables were always more important than vegetation. Given the full additive model (not shown)
237
as an example, the only meaningful variables for all animals were elevation, aspect and the
238
interaction between slope and NWness. In all cases, topographic variables had negative
239
coefficients because the animals preferred elevations below the available mean in the landscape
240
and avoided hillsides with northwestern exposure and steep slopes. The vegetation variables
241
showed different tendencies for each individual and in most cases were not significant (p > 0.05).
242
The burned areas and shrubs had a tendency of being avoided by cows 1 and 2 but were
243
preferred by cow 3, which also preferred wet meadows. Forest habitat had low coefficients for all
244
the animals with values near zero, which showed no tendency to be selected, either in favor or
245
against.
246
By means of a step-by-step selection procedure over all the models, five plausible
247
ecological models were selected for each animal in order to use them as a RSPF for semi-wild
248
cattle (Table 1). These models were among the best ten for each animal. It can be seen that all
249
models have as variables elevation, some aspect indicator, its interaction with slope, and in some
250
cases the vegetation types with greater selection: shrubs and meadows. The coefficients of these
251
252
The most parsimonious models for each animal (Table 2) were significantly different
NICOLS SEOANE
253
from the null model (p<0.01) and explained 46% of the total deviance for cow 1, 37% for cow 2,
254
53% for cow 3 and 49% for cow 4. All the animals selected lower elevations than available in
255
the landscape. Cow 1 also avoided steep slopes and hillsides with northwest exposure. Cow 2
256
also avoided slopes with northwest exposure, and the square of the slope shows that it had a
257
range of preferred slopes below the available mean at the site. In addition, it had a tendency to
258
avoid shrubs and wet meadows. Cow 3, on the contrary, selected wet meadows, shrub sites and
259
steep slopes, avoiding northwest hillsides with steep slopes. Cow 4 used slopes with northwest
260
exposure. The confidence intervals of the mentioned variables for each individual do not include
261
262
Models with the lowest AIC were different for each individual (Table 1). However, model
263
"D" (i.e. including elevation, NWness:slope, shrubland and meadow) is relatively well-ranked
264
for all animals and includes a set of variables that summarize the habitat use patterns observed.
265
Therefore, model "D" was chosen to build the final predictive RSPF with all the individuals. This
266
general proposed model (i.e. model D, Table 3) shows a common selection for an elevation
267
below the average available (Figure 1) and against sites with northwest exposure. These were the
268
only variables that allow a strong inference, since their confidence intervals excludes zero.
269
Nevertheless, some interesting trends of this population model are noteworthy. Shrubland, for
270
example, reduces the probability of use for a given site and the opposite occurs with meadows
271
(Table 3). A high variability in resource selection can be observed among individuals. Some
272
variables (as the interaction slope:NWness) were selected with the same tendency while others
273
274
275
NICOLS SEOANE
276
DISCUSSION
277
278
Using GPS logs and an approach that takes into account the intensity of use, a RSPF that
279
accounts for topography and vegetation was built for semi-wild cattle in the Andean-Patagonian
280
forest. This study shows that habitat use by livestock in these forests has a large inter-individual
281
variation but also common features that would allow a description of a pattern of use.
282
Topographic variables affected the probability of habitat use by cattle to a greater extent than
283
vegetation types. In the general proposed model, for instance, a change in a single standard
284
deviation unit of elevation (~ 300 meters) resulted in a threefold increased in the probability of
285
use (Table 3) and the squared slope has a similar role in some of the alternative models (Table 1).
286
This suggests that terrain features have a leading role in defining the use of space by livestock,
287
perhaps as an adaptation due to the long history of use and because topography is important in
288
289
Some common features such as the preferred elevation range and the tendency to prefer
290
meadows suggests that it would be possible to describe a population-level home-range for a more
291
comprehensive description of habitat use. In this regard, the squared elevation would play an
292
important role in defining a preferred elevation range, although there seems to be a high
293
variability among individuals. This variability could be modeled by a RSPF with hierarchical
294
formulation. Cattle tend to slightly avoid shrubland (Table 3). This may be due to a lower
295
proportion of trails in this type of thick vegetation, in contrast to other more open areas or those
296
that offer better protection from snow, such as forest. Moreover, these paths are unusable in the
297
winter. In this scenario, the availability and network of trails would play an important role in
298
defining the population-level home-range of cattle, and therefore, in determining habitat use. At
NICOLS SEOANE
299
the same time, use by livestock has created the paths over time, so there is a reciprocal process
300
301
The response variable used in this study proved to be appropriate and useful as it
302
describes the habitat use of cattle in probabilistic terms. However, the experimental design has
303
limitations that must be made explicit. This includes the low sample size of animals. However,
304
the different trajectories and spaces used by the sampled individuals suggest that the study was
305
able to capture considerable variability in resource use by cattle (Figures 1 and 2) and even
306
studies with one GPS-collared animal have accomplished some general features in the behavior
307
of a herd (Moritz et al. 2010). Another limitation is the temporal span of the data collection. As
308
the data collection period occurred between the months of April and July, it can be assumed that
309
this covers a typically cold season (corresponding to the austral autumn and winter). Therefore,
310
all models and conclusions of this work should be most reflective of that season.
311
A disadvantage of using logistic regressions such as RSPFs is that the model coefficients
312
decrease as the availability of a resource increases, and may even change the sign (Beyer et al.
313
2010). This could explain some of the inter-individual variation observed in the case of meadow
314
use (Figure 2), with animals 1 and 2 both negatively selecting this vegetation, and moving into
315
an area where the abundance of meadows is higher in relation to the areas where individuals 3
316
and 4 moved (unreported movement data). A similar case is that of the forest variable, which was
317
not significant in any model (Table 1) and yet it is a very abundant vegetation type and widely
318
used by cattle. To overcome this limitation it would be necessary to develop models that also
319
account for animal movement, as habitat selection and movement are not independent processes.
320
Methods for estimating movement patterns and habitat selection simultaneously have recently
321
NICOLS SEOANE
322
Comparing the results of this study with others conducted on cattle in forests we find
323
interesting relationships in the applied variables and scales. Studies on seasonal diet selection
324
(Marquardt et al. 2010) in Bolivian forests for example, show that different functional groups of
325
plants are preferred in different seasons. It would be possible to infer species selected by
326
livestock throughout the year in the Andean-patagonian forest if direct observations of consumed
327
plants was added to this study. Roath and Krueger (1982) also used topographic and vegetation
328
variables to study the behavior of cattle in a forested mountainous environment, finding that the
329
type of vegetation and water availability determined the degree of use by livestock. These
330
apparently opposite results are consistent if we consider the differences between the study areas.
331
While Roath and Krueger (1982) worked in a dry, arid vegetation and elevational range of 300
332
meters, Llodconto Valley is an area with high water availability, abundant streams, and altitudes
333
ranging from 800 to 1900 masl. Kaufmann et al. (2013) found that in silvopastoral systems cattle
334
prefer places with forest cover, avoiding places with high slope and low biomass, particularly
335
logged sites. A comparison with this study would make an analogy between logged and burned
336
areas, both without forest cover. Nevertheless, with the available data we cannot make inferences
337
about the type of selection on burned areas so far (Tables 1- 3). It may be noted, however, that
338
other studies in Patagonia (Blackhall et al. 2008, De Paz and Raffaele 2013) found a significant
339
influence of livestock on vegetation in burned environments, making it clear that cattle forage in
340
these areas. Interestingly, Johnson et al. (2004) conducted a study on resource selection by
341
caribou at two spatial scales, finding that topographic variables such as elevation and slope are
342
more important at a landscape scale, while vegetation variables were more important at the patch
343
scale. Something similar could be occurring in this present study, which is closer to the
344
landscape scale.
NICOLS SEOANE
345
As mentioned, cattle can increase landscape heterogeneity, especially when the spatial
346
pattern of vegetation is already heterogeneous (Adler et al. 2001) as in the case of the currently
347
studied forests. This has implications for forests dynamics because although there are studies on
348
cattle inhabiting forest (Bartolom et al. 2011, Kauffmann et al. 2013, etc.), there still remains a
349
lack of knowledge regarding how forest dynamics are affected by the distribution of wild cattle
350
worldwide. Recently, a study conducted in Patagonia developed a model that postulates distance
351
to meadows as an estimator of forest use by livestock (Quinteros et al. 2012). While meadows
352
are of crucial importance to the Andean-Patagonian forests and Quinteros et al. (2012) make a
353
practical contribution in relation to forest management, their model has a spatial limitation, being
354
355
scale and the model fails to consider other relevant variables such as topography, which proved
356
to have greater importance in resource selection by cattle (Tables 2-3, Figure 2). Nonetheless, the
357
approaches are consistent, because as the distance to meadows increases, the intensity of use by
358
livestock decreases (Quinteros et al. 2012), although altitude is also higher, which is predicted by
359
360
There remains a need for tools that enable animal management in agreement with
361
362
livestock farming has been identified as an activity with potential to be carried out in the forest.
363
Yet, there have been only some pioneer experiences of silvopastoral handling (Fertig and Guitart
364
2006), and current livestock handling is still scarce (Ormaechea et al. 2009). The main
365
limitations for management worldwide are a lack of understanding of animal foraging decisions
366
and the spatial scale of diet selection (Rook et al. 2004). Although effective management
367
depends heavily upon knowledge about the interaction of the animal with its environment, it is
NICOLS SEOANE
368
common to have poor information on the activity patterns of semi-wild cattle (Moyo et al. 2012).
369
This current study provides a direct contribution to this knowledge-gap, and direction for future
370
research in these systems. As an example, a practical application of this work is the production of
371
maps showing intensity of land use by livestock. Similar maps have been developed for other
372
regions using equivalent methodologies (Nielsen et al. 2003, Sawyer et al. 2009), and therefore
373
offer a management and research tool within protected areas or productive forests.
374
In summary, the present study demonstrated that the habitat use by semi-feral cattle in
375
southern beech forests has a large inter-individual variability but also similar characteristics
376
which allow the description of a pattern of use. Specifically, the results indicate that vegetation
377
variables are no more important than topography features. Conversely, habitat selection by cattle
378
is mainly affected by topographic variables such as altitude and the combination of slope and
379
aspect. These results suggest an adaptation by free-ranging cattle to the major features of the
380
381
382
383
384
ACKNOWLEDGEMENTS
385
386
Funding for this research was provided by a PhD grant from the CONICET (Argentina). The
387
author thanks to Juan Manuel Morales for helpful comments on the manuscript.
388
389
390
NICOLS SEOANE
391
LITERATURE CITED
392
393
394
395
396
397
398
Adler, P., D. Raff, and W. Lauenroth. 2001. The effect of grazing on the spatial heterogeneity of
vegetation. Oecologia 128(4):465479. doi:10.1007/s004420100737
Atkinson, I. E. 2001. Introduced mammals and models for restoration. Biological Conservation
99(1):8196. doi:10.1016/S0006-32070000189-0
Bailey, D. W., and F. D. Provenza. 2008. Mechanisms determining large-herbivore distribution.
In Resource Ecology pp. 728. Springer.
399
400
Baldwin, R. A. 2009. Use of maximum entropy modeling in wildlife research. Entropy 11(4):
401
402
854866. doi:10.3390/e11040854
Bartolom, J., J. Plaixats, J. Piedrafita, M. Fina, E. Adrobau, A. Aixs, L. Polo. 2011.
403
404
405
Berteaux, D., and T. Micol. 1992. Population studies and reproduction of the feral cattle Bos
406
407
408
409
410
411
Blackhall, M., E. Raffaele, and T. T. Veblen. 2008. Cattle affect early post-fire regeneration in a
NICOLS SEOANE
412
413
414
415
416
Boyce, M. S. 2006. Scale for resource selection functions. Diversity and Distributions
12(3):269276. doi:10.1111/j.1366-9516.2006.00243.x
Boyce, M. S., and L. L. McDonald. 1999. Relating populations to habitats using resource
417
418
53479901593-1
419
420
Multivariate models of adult Pacific salmon returns. PloS One 8(1): e54134.
421
doi:10.1371/journal.pone.0054134
422
Cagnacci, F., L. Boitani, R. A. Powell, and M. S. Boyce. 2010. Animal ecology meets GPS-based
423
424
425
doi:10.1098/rstb.2010.0107
426
427
Calcagno, V., and C. De Mazancourt. 2010. glmulti: An R Package for Easy Automated Model
Selection with Generalized Linear Models. Journal of Statistical Software 34(12):129.
428
429
430
18(3):253271.
431
Clark, J. D., J. E. Dunn, and K. G. Smith. 1993. A Multivariate Model of Female Black Bear
432
Habitat Use for a Geographic Information System. The Journal of Wildlife Management
433
57(3):519526.
NICOLS SEOANE
434
De Paz, M., and E. Raffaele. 2013. Cattle change plant reproductive phenology, promoting
435
436
437
438
MacNeil. 2015. Origins of cattle on Chirikof Island, Alaska. bioRxiv p. 014415. Cold
439
440
441
442
443
444
445
446
Fertig, M., and E. Guitart. 2006. Cuaderno de Campo para Vacunos p. 112. Esquel: INTA EEA
Esquel.
Gannon, W. L., and R. S. Sikes. 2007. Guidelines of the American Society of Mammalogists for
the Use of Wild Mammals in Research. Journal of Mammalogy 88(3):809823.
Guisan, A. and N. E. Zimmermann. 2000. Predictive habitat distribution models in ecology.
Ecological Modelling 135(2):147186. doi:10.1016/S0304-38000000354-9.
Hernandez, L., H. Barral, G. Halffter, and S. S. Colon. 1999. A note on the behavior of feral
447
cattle in the Chihuahuan Desert of Mexico. Applied Animal Behaviour Science 63(4):259
448
267. doi:10.1016/S0168-15919900015-5
449
450
451
Howard, W. E. 1966. Ecological changes in New Zealand due to introduced mammals. Towards
a New Relationship of Man and Nature in Temperate Lands, 219-240, UICN.
Johnson, C. J., D. R. Seip, and M. S. Boyce. 2004. A quantitative approach to conservation
452
planning: Using resource selection functions to identify important habitats for mountain
453
454
Jones, C. G., J. H. Lawton, and M. Shachak. 1997. Positive and negative effects of organisms as
NICOLS SEOANE
455
456
Kaufmann, J., E. W. Bork, P. V. Blenis, and M. J. Alexander. 2013. Cattle habitat selection and
457
458
459
doi:10.1016/j.applanim.2013.03.014
460
Law, B. S., and C. R. Dickman. 1998. The use of habitat mosaics by terrestrial vertebrate fauna:
461
462
333. doi:10.1023/A:1008877611726
463
Lazo, A. 1994. Social segregation and the maintenance of social stability in a feral cattle
464
465
466
Seasonal variation in activity and habitat use of free-ranging cattle in southern Bolivian
467
468
469
probabilistic measure of space use: Forest fragments and Stellers Jays. Ecology 85(5):
470
14111427. doi:10.1890/03-0114
471
Milchunas, D. G., W. K. Lauenroth, and I. C. Burke. 1998. Livestock grazing: animal and plant
472
473
474
475
NICOLS SEOANE
476
477
478
Moorcroft, P. R., and A. Barnett. 2008. Mechanistic home range models and resource selection
analysis: a reconciliation and unification. Ecology 89(4):11121119. doi:10.1890/06-1985.1
Moritz, M., E. Soma, P. Scholte, N. Xiao, L. Taylor, T. Juran, and S. Kari. 2010. An Integrated
479
Approach to Modeling Grazing Pressure in Pastoral Systems: The Case of the Logone
480
481
482
483
484
communal areas of the Eastern Cape Province, South Africa. African Journal of
485
486
487
488
489
490
491
492
493
management on landholdings in south Patagonia with ire Nothofagus antarctica forest. In:
494
495
496
Osborne, P. E., J. C. Alonso, and R. G. Bryant. 2001. Modelling landscape-scale habitat use using
GIS and remote sensing: a case study with great bustards. Journal of Applied Ecology
NICOLS SEOANE
497
498
38(2):458471. doi:10.1046/j.1365-2664.2001.00604.x
zesmi, S. L., and U. zesmi. 1999. An artificial neural network approach to spatial habitat
499
500
doi:10.1016/S0304-38009800149-5
501
502
503
504
00021.1
505
Quinteros, C. P., P. M. Bernal, M. E. Gobbi, and J. O. Bava. 2012. Distance to flood meadows as
506
507
508
R Core Team. 2013. R: A Language and Environment for Statistical Computing. Vienna, Austria.
509
510
511
512
513
514
515
516
517
NICOLS SEOANE
518
519
520
521
522
523
524
525
526
Store, R., and J. Jokimki. 2003. A GIS-based multi-scale approach to habitat suitability
modeling. Ecological Modelling 169(1):1-15. doi:10.1016/S0304-38000300203-5
Ungar, E. D., Z. Henkin, M. Gutman, A. Dolev, A. Genizi, and D. Ganskopp. 2005. Inference of
527
Animal Activity From GPS Collar Data on Free-Ranging Cattle. Rangeland Ecology and
528
Management 58(3):256266.
529
Veblen, T. T., C. Donoso, T. Kitzberger, and A. J. Rebertus. 1996. Ecology of southern Chilean
530
and Argentinean Nothofagus forests. The Ecology and Biogeography of Nothofagus Forests.
531
pp. 293353.
532
533
534
535
536
537
538
Veblen, T. T., and G. H. Stewart. 1982. The effects of introduced wild animals on New Zealand
forests. Annals of the Association of American Geographers 72(3):372397.
Venables, W. N., and B. D. Ripley. 2002. Modern Applied Statistics with S. New York: Springer.
Retrieved from http://www.stats.ox.ac.uk/pub/MASS4
NICOLS SEOANE
539
Table 1. Alternative models for free-range cattle habitat use in the study area; the table shows
540
Akaike information criterion value (AIC), number of predictor variables (K ) and ranking of
541
542
Models
Animal 1
Animal 2
Animal 3
Animal 4
1 619.9 2.4
7 188.9 1.4
2 622.7 5.2
4 189.5 2.0
3 617.5 0.0
6 192.4 4.9
4 619.8 2.3
6 196.3 8.8
5 620.8 3.3
Model A + shrubland +
B meadow
Elevation +
C NWness*slope
Model C + shrubland +
D meadow
Elevation + slope +
Nwness + forest +
E shrubland + meadow
543
544
545
546
547
NICOLS SEOANE
548
549
Table 2. Coefficients and 95% confidence intervals of the most parsimonious RSF models for
550
selection of topographic and vegetation resources during autumn by free-range cattle in the study
551
area.
552
Variables
Intercept
Animal 1
Animal 2
Animal 3
Animal 4
-39.91 -67.66 -12.16 -2.73 -4.05 -1.61 -2.22 -3.44 -1.07 -1.59 -2.38 -0.83
-1.11 -1.43 -2.79 -0.33 1.45 0.68 2.38 0.73 0.20 1.30
Slope
NWness
-0.00 0.77 0.12 1.46 1.15 -0.01 2.54 2.01 1.29 2.73
Shrubland
Meadow
NW*Slope
553
554
555
556
557
558
559
1.04
NICOLS SEOANE
560
561
562
563
564
565
566
567
568
569
570
571
SE
Lower limit
Upper limit
Intercept
-9.368
1.136
-12.409
-6.327
Elevation
-3.016
0.864
-4.776
-1.255
Slope:Nwness
-4.772
1.319
-8.867
-0.677
Slope
-5.958
1.799
-13.576
1.660
NWness
-8.165
1.598
-14.180
-2.151
Shrubland
-0.060
0.795
-1.432
1.545
Meadow
1.521
1.361
-2.839
5.882
NICOLS SEOANE
572
573
574
575
576
Figure 2. General proposed model estimates (black) and individual model estimates (grey) with
577
90 % confidence intervals.
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
NICOLS SEOANE
595
596
597
Figure 1
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
NICOLS SEOANE
618
619
620
Figure 2
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
NICOLS SEOANE
641
642
643
############################################################
644
645
646
############################################################
647
648
yi ~ NegativeBinomial i ,
649
ln i = ln total + 0 + j .variablei
650
where:
651
y = number of observations per sampling unit; = relative frequency in each sampling unit by animal; =
652
overdispersion parameter; total = number of recorded positions in whole study area by animal; i = sampling unit
653
654
655
656
657
658
659
Model
NICOLS SEOANE
660
661
662
663
664
# Animal 1
Model A
Model B
Model C
Model D
Model E
Intercept
-72.577*
-75.991*
-11.633*
-11.422*
-10.516*
Elevation
-130.564*
-137.462*
-5.748*
-5.718*
-5.394*
Elevation2
-62.145*
-65.576*
Slope
-2.343
-2.657
-3.063*
-3.051*
-2.536*
Slope2
-0.101
-0.236
NWness
-0.060
-0.043
-2.852*
-2.621*
-0.010
Shrubland
-0.259
-0.284
-0.252
Meadow
-1.681
-1.127
-1.218
Forest
0.174
NWness*Slope
-2.279*
-2.117*
Theta
0.212
0.238
0.164
0.168
0.160
Theta SE
0.098
0.113
0.067
0.069
0.067
Model A
Model B
Model C
Model D
Model E
Intercept
-3.174*
-2.686*
-3.614*
-3.272*
-3.126*
Elevation
-1.577*
-1.585*
-1.967*
-1.937*
-1.572*
665
666
# Animal 2
NICOLS SEOANE
0.127
Elevation2
-0.078
Slope
-1.515*
-1.442*
Slope2
-0.231
-0.252
NWness
0.650*
0.783*
-1.334*
-1.227*
-1.167*
1.039*
1.312*
0.784
Shrubland
-0.627
-0.685
-0.275
Meadow
-0.203
-0.232
0.105
Forest
0.387
NWness*Slope
0.403
0.516
Theta
0.062
0.063
0.063
0.065
0.064
Theta SE
0.011
0.011
0.011
0.012
0.012
Model A
Model B
Model C
Model D
Model E
Intercept
-2.834*
-2.683*
-1.943*
-2.264*
-2.939*
Elevation
-1.512*
-1.636*
-1.713*
-1.698*
-1.711*
Elevation2
-1.288*
-1.597*
Slope
1.652*
1.969*
2.344*
3.010*
2.927*
Slope2
1.196*
1.043*
NWness
0.874*
1.444*
0.233
1.034*
0.994*
667
668
# Animal 3
Shrubland
-0.672
-0.712
-0.202
Meadow
2.929*
3.937*
4.166*
Forest
1.340
NICOLS SEOANE
NWness*Slope
-0.161
-0.873*
Theta
0.040
0.044
0.031
0.037
0.036
Theta SE
0.008
0.008
0.006
0.007
0.007
Model A
Model B
Model C
Model D
Model E
Intercept
0.073
-1.003*
-1.594*
-1.817*
0.230
Elevation
-3.173*
-3.131*
-3.545*
-3.420*
-2.639*
Elevation2
-1.273*
-0.782*
Slope
1.471*
1.591*
0.739*
0.783*
1.364*
Slope2
0.714*
0.830*
NWness
0.214
0.132
2.012*
1.740*
-0.177
669
670
# Animal 4
Shrubland
1.120*
0.634
-0.053
Meadow
0.212
0.794
0.764
Forest
-0.647
NWness*Slope
671
672
673
1.996*
1.872*
Theta
0.072
0.075
0.078
0.079
0.068
Theta SE
0.007
0.007
0.007
0.008
0.006