Along with the increase in the number of road motor vehicles there is also an increase in the number of collisions of vehicles and wildlife (WVC) that for some reason come out onto the road. The consequences of such collisions are sometimes very severe. Apart from material damage on the vehicle, wild animals also get injured, and there is great danger for human lives as well. The observed trend of the increasing number of traffic accidents involving collisions between vehicles and wildlife (WVC) requires proposals of measures in order to prevent the appearance of game at risky road sections. In order to take certain measures of preventing vehicle collision with the game (WVC), it is necessary to identify the high-risk road sections regarding the possible appearance of game. Therefore, a model of identifying dangerous sections on public roads, and measures of efficient management of the road network have been studied, and this will contribute to reducing the occurrence of wildlife on the roads. In practice, the question of liability for the damage caused by wildlife in collision with the vehicles or the vehicles colliding with wildlife (WVC) is often raised. The law stipulates that the game are the assets of interest for the Republic of Croatia and has its protection. The game lives in the hunting grounds with roads passing through. The hunting grounds are managed by the hunting licensees or hunting right owners, who are either legal or natural persons (craftsmen). The legislation in the Republic of Croatia, related to this issue is based on four laws of the Republic of Croatia. The area of compensation for the damage and liability caused by the vehicle colliding with the game is regulated by the Hunting Act, the Roads Act, the Road Traffic Safety Act, and the Obligations Act. The judicial practice of passing judgements lacks uniformity and there are different court judgements both in the first and the second court instances. According to the verdicts, sometimes the driver is to blame and sometimes the hunting licensee, and sometimes the legal entity that manages the road on which the vehicle collided with the game. The purpose of this doctoral thesis is to produce a maximally reliable model of forecasting the collision of vehicle with the wildlife (game) based on relatively easily accessible data. The objective of research is to propose measures that would significantly increase the traffic safety and reduce the number of vehicle-game collisions (WVC). Research hypotheses have been set, i.e. by in-depth analysis of data on traffic accidents caused by vehicle colliding with the game, it is possible to determine the elements of road and environment that highlight the high-risk sections of public roads and it is possible to develop a model of identifying the dangerous sections of public roads regarding the occurrence of wild animals on them. Lika-Senj County, selected as the research area, has central geographical position, and therefore an important connecting significance within the Republic of Croatia. Lika-Senj County occupies 9,46 % of the Croatian territory. Most of the County belongs to the mountainous area and includes the mountains of Velebit, Plješivica and Velika and Mala Kapela. The area of the County includes also the karst fields separated by the mountain ridges: fields of Lika, Gacka, Krbava, Drežnica, Korenica, Lapacand Gračac. The County also includes the Adriatic coast as well as a part of the island of Pag, i.e. a part of the territorial sea (596,63 km2 or 1.9 % of the Croatian sea area) and 2,29 km2 of the island area or 0,07 % of the area of all the islands of the Republic of Croatia. The mainland area of the Lika-Senja County covers an area of 535 113 ha, and stretches from 0 to 1 738 metres above sea level. Considering the division of climate according to Köppen, several different climate types change in the entire Lika-Senj County (climate type Cfb – 85,6 % of the area; climate type Cfa – 6,7 % of the area and climate type Df – 7,7 % of the Lika-Senj County area). The dominant type of the game habitat in Lika-Senj County are forests, which make up 65 % i.e. together with shrubs, brake-grown areas, and heaths the closed habitats make up almost 70 % of the observed area. Carbonate rocks make up 74 % of the researched area, which makes this part of the research area porous in terms of precipitation retention. Consequently, and due to a large range of the altitudes of the road network (0 – 1 011 metres a.s.l.), climatic differences, differences in terrain configuration and habitat of almost all species of game in the Republic of Croatia, the research of the doctoral dissertation was conducted on the roads of the Lika- Senj County. A prerequisite for determining the dangerous road sections regarding the occurrence of wildlife are the collected relevant data about the vehicle collisions with wildlife. The data on traffic accidents of vehicle collisions with wildlife have been collected by the employees of the Ministry of the Interior of the Republic of Croatia through going to the scene of accident, and for the purposes of this thesis the data have been collected by the Lika-Senj Police Department, i.e. police stations Gospić, Otočac, Senj, Donji Lapac, Korenica, Karlobag and Novalja. The data about vehicle collisions with wildlife have been taken from the police records on vehicle collisions with wildlife in the time period from 2012. to 2016.. There are 63 established hunting grounds in this area, which are managed by slightly fewer hunting licensees (some hunting licensees lease two or more hunting grounds). In order to verify the accuracy of the obtained data the hunting licensees of the hunting grounds in the Lika-Senj County have been surveyed, and because of the sensitivity of the data that impact the value of the hunting grounds, no newer relevant data could be obtained. In the research period there were 548 accidents involving vehicles colliding with wildlife, and the largest number of accidents of vehicle and wildlife collisions occurred on state roads, as many as 441, mostly collisions with roe deer (Capreolus capreolus) and wild boar (Sus scrofa). Having in mind the frequency of traffic accidents and the possibility of fatal outcomes of the traffic accidents, the work started on developing a model for recognizing dangerous sections on state roads, for the cases of collisions with roe deer, wild boar and large game in total. Special attention was paid to the temporal and spatial patterns of vehicle collisions with wildlife that served as the basis for developing the model of recognizing the dangerous sections regarding the appearance of wildlife. In order to determine the relevant minimum section length for the calculation of the collision probability estimate, state roads were divided into sections of 200, 500, 1 000, 2 000 and 12 000 m. Circles (cells) of radii of 100, 250, 500, 1 000 and 6 000 m were drawn around the centres of the sections and for each circle (cell) independent variables, that is, predictors were determined. In developing temporal patterns regarding the time of the occurrence of vehicle – wildlife collision, the 24-hour day was divided regarding the time of dawn and dusk into day, night and dusk. The time of sunrise and sunset was calculated by means of the algorithm provided by the Zagreb Observatory website for every day of the research period. When calculating the lunar phases, the international standard of the US Maritime Oceanographic Portal (lunar cycle of eight lunar months) was used. Analysing spatial patterns is far more complex than the temporal patterns and required the use of a large number of independent variables, i.e. predictors (predictors of the road, landscape, relief, number of wildlife, and number of collisions in the cell). The observed characteristics of the road included: average annual daily traffic (AADT), average summer daily traffic (ASDT) and curves parameter. The habitat data that were used included: share of water, shores, bare grounds, heaths and brake-grown areas, thickets, forests, grasslands, builtup land, neglected agricultural land, arable land. The index of topographic position (TPI or TOPEX) was used for the relief as a predictor of topographic characteristics of road. Regarding the index of topographic position and slope, the terrain has been classified into six categories: valleys, less steep terrains, medium steep terrains, extremely steep terrains, upper parts of the slopes and ridges. The data on the number of wildlife are relatively unreliable as predictor of population density, and therefore the data on game shooting in individual hunting grounds were used; however, they have been reduced to the unit of the hunting area. In order to obtain a maximally precise model of identifying dangerous sections on state roads, spatial patterns of vehicle collisions with large game were used. The spatial data were prepared in the software package Arc GIS 9.2., and processed in the software package Statistica 13.4.014 TIBCO Software Inc., 2018. The doctoral thesis proposes two types of patterns of recognizing dangerous sections regarding appearance of wildlife. These are: collision probability estimate model and vehiclegame collision number estimate. For the selection of the most reliable model of recognizing dangerous road sections regarding appearance of wildlife the software tools Akaike Information Criterion (AIC) was used. The selection of a reliable model followed if ΔAIC < 2 units. Akaike weight (wi) was also determined, and it represents the probability that the model is the best, that is, the most reliable compared to other models. Logistic regression was used to calculate the prediction of the probability of a vehicle colliding with wildlife. AIC analysis provided collision estimate models for every cell radius separately, and the logistic regression provided reliability of the results in estimation percentages. For the estimate of the collision probability with roe deer the smallest road section would be 2 000 m (cell radius 1 000 m), and the prediction accuracy is 68,20 %. The used independent variables (predictors) are the number of roe deer, share of the neglected agricultural land and the shares of bare land and the sea. The number of collisions with roe deer increases with the population density of roe deerfor radiiof 500 and 1 000 m, roe deer and wild boarfor radius of 250 m and wild boarfor radii of 6 000 m; the share of neglected agricultural land for sections of 500, 1 000 and 2 000 m; share of heaths and brake-grown areas on road sections of 500 and 1 000 m; distances from possible watering places (fresh water) on road sections of 500 and 1 000 m; AADT on sections of 500 and 12 000 m and with lower TPI (TOPEX) value, i.e. in valleys and less steep terrain, but this applies only to road sections of 12 000 m. The number of collisions with roe deer gets reduced with the increase of curves on a section (sections of 1 000 m) or road share (sections of 2 000 m); share of bare land (on all sections, except sections of 12 000 m); share of the sea (on all sections, except sections of 12 000 m) and built-up land on sections of 500 m. In the wild boar collision estimation model, AIC analysis provided quite a lot of collision estimation models for each cell radius separately, and the logistic regression gave reliable results only for road sections of 12 000 m. The collision probability on the section can be estimated in 73,33 % of cases. In the wild boar collision estimation model the game population density proved to be the key independent variable (predictor). The number of wild boar collisions will be larger if the population density of roe deer, wild boar or large game in total is higher; higher share of neglected agricultural land and higher ASDT. The number of collisions will be smaller if the share of built-up land is higher; higher share of forests; higher share of roads; higher share of the sea (exceptions are cells of radius of 6 000 m) and higher AADT. For the models estimating the probability of collision with large game in total, all the large game killed during the research period were included as well as grey wolf (legally grey wolf is not game, but is included in the model development due to similar consequences of the collision). For models of estimating the collision probability with large game in total the wildlife population density as independent variable comes in all cell radii. Respecting the results of the logistic regression, it may be said that the shortest section on which the probability of collision with large game in total may be reliably determined is a section of 2 000 m. On this section it is possible to predict the collision of vehicle on wildlife with a certainty of 70,11 %, based on the relative density of game, share of bare land, share of the sea, and the proximity to the nearest watering place. The number of collisions with large game in total will be greater if the population density of roe deer or large game is higher; higher share of neglected agricultural land; higher share of heaths and brake-grown areas; greater distance to the watering place; higher AADT. The number of collisions on large game in total will be smaller if the road features more curves or more intersections; higher share of bare land; higher share of built-up land; higher share of the sea and more indented relief (higher TPI value). Although the collision or non-collision location model can be estimated with an accuracy of 70,11 %, there are also certain model errors. Three outcomes of the model operation can be expected, and these are: the model has estimated correctly that on a certain section there will not occur or there will come to a collision; there were no collisions on the section during the research period, but the model predicted that a collision might occur there (collision overestimation error) and on the section collisions were recorded during the research period, but the model predicts that no collision can occur there (collision underestimation error) Each model for itself is relatively deficient. The first type of model (collision probability estimation) estimates quite accurately the collision occurrence probability. If another model were used for the collision probability estimation (estimation of the number of collisions) then the combination gives quite high accuracy in estimating the danger of vehicle collision with wildlife. The proposed models will enable the legal entities that manage public roads to get a view of dangerous sections and to act through measures of maintaining and equipping the roads for traffic safety. The road sections at risk of the appearance of wildlife have been ranked by multicriteria analysis by applying the Analytic Hierarchy Process (AHP method), using the software tools Expert Choice. Before using the software tools the objective was set, and it is to rank the dangerous road sections on state roads of the Lika-Senj County regarding the appearance of wildlife. The criteria that have been used are the number of collisions, section length, number of collisions per 100 km and number of collisions per 100 km annually. Appropriate matrices have been developed comparing the criteria with each other in relation to the set objective and based on these the values from the Saaty evaluation scale were added. The program requires also setting of variants, and they are several dangerous sections. Field research on critical road sections examined the impact of certain maintenance measures on the appearance of wildlife on the roads. The tested maintenance measures include installation of protective wire fences along the road; thorough cleaning of the protective road belt; mounting of optical and sound sensors on signposts; installation of adequate traffic signs – game on the road; planting of plant species that repel game, and also the impact of the location of game feeding grounds on the wildlife population density along the road was investigated. A new public road maintenance procedure has been proposed that would reduce wildlife crossing the public roads by increasing the patrol hours on high-risk sections; installation of protective fences to prevent the appearance of wild animals on the roads; cleaning the full profile of road protection zone in order to reduce the occurrence of wildlife on the roads; planting of game-repellent plant species along the road; installation of optical and sound sensors on signposts that drive off the game, and amendments to the Ordinance on traffic signs, signalization and road equipment. By proving the set objective and hypotheses, this doctoral dissertation opens up the possibilities of further research in preventing the wildlife to appear on the roads by developing a model with and without large predators; classification of killed game by gender and estimated age of the individual; unification of temporal and spatial patterns; separation of spatial patterns of collision according to the season, time of day and phases of the moon. Furthermore, in further research it is of extreme importance to separately study the killing on the roads of fallow deer, as a separate species. According to the conducted research in this doctoral thesis, this is the game which suffers major casualties in vehicle collisions, although they are small in number and living in a confined area.