**3.3 Habitat use – movement distances - movement patterns – habitat suitability**

More specifically data analyses were used in order to test whether the highway construction phase affected:

(a) the dispersal ability, (b) preferences on habitat use and (c) distributional patterns of the species.

To estimate **potential changes in habitat use range** we estimated home range polygons (95% Kernel core area ). Additionally group home range estimates were based on home range size. We also calculated min distance of polygons from road using t-test and ANOVAs. Data were organized and grouped according to the sex of the individuals and the seasons. Adequate data to perform statistical analyses were found for males in spring and summer and for females in summer.

The analysis was repeated for males and females and for data collected at different seasons

To estimate **potential changes in movement distances** we analyzed day and night movement distances but also home ranges of each individual (estimated by using Kernel based methods) to examine whether the distance from the highway is an important indicator of the quality or quantity of brown bears activity levels. Mean and maximum moving distances according to the time of activity and distance from the highway were examined as well as variations in mean direction with respect to the distance from highway (angular analysis of point patterns). We used ANOVAs and the analysis was repeated for males and females and for data collected on daytime or at night.

Telemetry as a Tool to Study Spatial Behaviour and Patterns of

most frequently used/visited habitat units by bears.

"factors" and the rest as "covariates" (continuous numeric variables).

Deviance, Pearson Chi-Square statistics (Quinn and Keough, 2002).

main approaches:

pixel.

pixel of the study area.

Brown Bears as Affected by the Newly Constructed Egnatia Highway – N. Pindos - Greece 317

• (%) of contribution of 2nd and 3rd rank vegetation type over the total number of recorded vegetation types within 5 and 10 pixels radius from the central pixel. Bear telemetry data sets were incorporated on these pixel maps. For each map pixel with a bear radiolocation we extracted all relevant information related to topographic and vegetation characteristics, distances from the highway and the values of all neighborhood statistics variables as described above. It is important to note that in a random pixel of the study area there may be more than one radiolocation indicating bear presence. This may be attributed to selection and repetitive use (by one or more bears) of a given pixel due to its specific attributes and characteristics. Therefore it is interesting to investigate the effect of pixels attributes upon the probability of their use by bears (preference/avoidance) but also the frequency of their use. For this we have developed prediction models focusing on various characteristics related to spatial behavior and presence of bears, according to two

• To what extent the selected variables allow prediction of abundance of bear presence in given areas. This allows to identify which variables contribute most in the selection of

• prediction models emphasized only on the presence or absence of bears in each pixel without taking into account the frequency of use (stationary of transitional) of each

For the first approach we used General Linear Models (GLM) which allow the development of linear relations between the dependent variable and a group of categorized or qualitative factors but also with continuous variables (covariates) through specific operational connection functions (Quinn and Keough, 2002). These models allow a non-normal distribution of the dependent variable. We used Pearson's correlation coefficient to investigate the correlation degree between all variables. In total 15 variables were kept in our analysis. Their utility to predict abundance of bear presence over the study area was examined. Of them, two (vegetation types and aspect) were introduced in the model as

The possibility of implementation of those models and their prediction efficiency in bear presence abundance and habitat pixels use according to the explanatory variables, was evaluated following different statistical tests such as: likelihood-ratio chi-square test,

For the second approach we developed a Logistic Regression model (LR) which is only applied in the case of binary data (presence/absence). We used a logic function to interrelate the key variable (presence and bear activity) with the group of descriptive variables. We performed a group of diagnostic tests such as Hosmer-Lemeshow goodness-of-fit statistic, improvement of chi-square test in order to examine the suitability and efficiency of the model as a predictor tool for bear presence or absence in a given

Additionally we developed Classification Trees (CT) by using bear presence and absence as the dependent variable and we examined probable classification rules for the explanatory

• Shannon (vegetation) diversity index within a 10 pixels radius from the central pixel. • Shannon (vegetation) diversity index within a 5 pixels radius from the central pixel. • Number of different vegetation types within a 5 pixels radius from the central pixel. • Number of different vegetation types within a 10 pixels radius from the central pixel. • (%) of contribution of the dominant vegetation type over the total number of recorded

vegetation types within 5 and 10 pixels radius from the central pixel.

To estimate **potential changes in movement patterns** angular analysis of point patterns (changes in mean direction with respect to the distance from road) was performed according to the following protocol:


To **assess habitat suitability** in relation to bear presence and prediction of use of a certain point (or area) of the HR, we used a series of digital sources to derive potential predictor variables (land use, topographical, vegetation). In addition, 17 variables were calculated by using neighbourhood statistics techniques. The significance of distance from highway and of the former predictor variables upon species distribution and habitat use were assessed by using Generalized Linear Models (GLM), Logistic Regression (LR) and Regression and Classification Trees (CART). Relative abundance - (Generalised Linear Models)(Naves et. al., 2003, Wiegandet al., 2004, Nielsen et al., 2006).

For each bear habitat pixel we calculated the following parameters using neighbourhood statistics:


$$\boxed{H^\circ = \sum\_{i=1}^{S} p\_i \cdot \ln p\_i}$$

Where: *Η* the Shannon's index value


To estimate **potential changes in movement patterns** angular analysis of point patterns (changes in mean direction with respect to the distance from road) was performed according


For each bear habitat pixel we calculated the following parameters using neighbourhood

• Average altitude within a 5 pixels radius from the central pixel. This variable allows to characterize the altitude of the central pixel based not only on its proper value but also on the values of the neighboring pixels as for the pixel selection by a bear depends also

• Altitude coefficient of variance within a radius of 5 pixels from the central pixel. This

• Altitudes range within a 5 pixel radius from the central pixel. This variable examines the max and min altitude differences in an area and functions as an indicator of

• Coefficient of variance of the average slope value within a 5 pixels radius from the

• Coefficient of variance of the average slope value within a 25 pixels radius from the

• Vegetation types variability index within a 5 pixels radius from the central pixel. Vegetation types variability was calculated after Shannon's (H) index as follows:

> = = ⋅ *S*

ln


*i <sup>i</sup> <sup>i</sup> H p p* 1

'

on its accessibility which is related to the altitudinal variation (ruggedness).

variable allows the quantification of altitude variance in a wider area.

• Average altitude within a 20 pixels radius from the central pixel.

• Average slope value within a 5 pixels radius from the central pixel.

• Slope range values within a 5 pixels range from the central pixel. • Slope range values within a 15 pixels range from the central pixel. • Slope range values within a 25 pixels range from the central pixel.

overall vegetation characteristics within in the same pixel.

selection or avoidance of movement in the given area.


to the following protocol:

statistics:

central pixel.

central pixel.

Where: *Η* the Shannon's index value



2003, Wiegandet al., 2004, Nielsen et al., 2006).

highway (0-2000m, 2000-4000m, 4000-6000m & >6000m)


Bear telemetry data sets were incorporated on these pixel maps. For each map pixel with a bear radiolocation we extracted all relevant information related to topographic and vegetation characteristics, distances from the highway and the values of all neighborhood statistics variables as described above. It is important to note that in a random pixel of the study area there may be more than one radiolocation indicating bear presence. This may be attributed to selection and repetitive use (by one or more bears) of a given pixel due to its specific attributes and characteristics. Therefore it is interesting to investigate the effect of pixels attributes upon the probability of their use by bears (preference/avoidance) but also the frequency of their use. For this we have developed prediction models focusing on various characteristics related to spatial behavior and presence of bears, according to two main approaches:


For the first approach we used General Linear Models (GLM) which allow the development of linear relations between the dependent variable and a group of categorized or qualitative factors but also with continuous variables (covariates) through specific operational connection functions (Quinn and Keough, 2002). These models allow a non-normal distribution of the dependent variable. We used Pearson's correlation coefficient to investigate the correlation degree between all variables. In total 15 variables were kept in our analysis. Their utility to predict abundance of bear presence over the study area was examined. Of them, two (vegetation types and aspect) were introduced in the model as "factors" and the rest as "covariates" (continuous numeric variables).

The possibility of implementation of those models and their prediction efficiency in bear presence abundance and habitat pixels use according to the explanatory variables, was evaluated following different statistical tests such as: likelihood-ratio chi-square test, Deviance, Pearson Chi-Square statistics (Quinn and Keough, 2002).

For the second approach we developed a Logistic Regression model (LR) which is only applied in the case of binary data (presence/absence). We used a logic function to interrelate the key variable (presence and bear activity) with the group of descriptive variables. We performed a group of diagnostic tests such as Hosmer-Lemeshow goodness-of-fit statistic, improvement of chi-square test in order to examine the suitability and efficiency of the model as a predictor tool for bear presence or absence in a given pixel of the study area.

Additionally we developed Classification Trees (CT) by using bear presence and absence as the dependent variable and we examined probable classification rules for the explanatory

Telemetry as a Tool to Study Spatial Behaviour and Patterns of

also table 1 and 2).

Brown Bears as Affected by the Newly Constructed Egnatia Highway – N. Pindos - Greece 319

simulation for exact P<0.05). Males home range sizes were significantly larger than females with all estimate methods (Kruskal-Wallis test: MCP100: χ2=11, P=0.012, MCP95: χ2=10.76, P=0.013, FK95: χ2=9.6, P=0.022 and CA50: χ2=9.33, P=0.025).(Giannakopoulos et al.2011). (see

In addition we found that the bears (2 males and 2 females) who kept collars more than one year seemed to maintain the same territories. Moreover the spatial patterns and distribution of home ranges between males and females were delineated in most of the cases by natural barriers and landmarks such as rivers, big streams, county roads and in some cases

Map 4. Home ranges of 22 bears of the sample versus highways network in the study area

**Sex Age N Gps Lo MCP100 MCP95 FKM CA50**  Males Adult 5 22083 271±26,1 200±14,5 130±15,1 30±4,2 Females Adult 3 15502 118±48,8 72±28,2 39±13 7±3,2

Data from the above table (1) refer only to the bears of the sample (males and females) that have kept their collar for an entire year cycle. A more analytical presentation of data on

Table 1. Annual home range sizes of GPS collared bears (2007-2009) estimated with (MCP100, MCP95, FKM and CA50) in Northeastern Pindos mountains Greece (n=8)

seasonal home range sizes on the overall sample are presented in table (2).

according with the topographic complexity (Giannakopoulos et al. subm.).

(descriptive variables). Here again we performed a group of diagnostic tests in order to examine the efficiency of the produced rules from the aforementioned analysis. This type of analysis is based on artificial intelligence methods (machine learning techniques principle) (Vayssieres κ.α. 2000, De'ath & Fabricius 2000, Thuiller κ.α. 2003, Mazaris κ.α. 2006).
