Regression and point pattern models of moose distribution in relation to habitat distribution and human influence in Ida-Viru county, Estonia

Abstract

Attempts are presented to model the location pattern of moose for Ida-Viru county, Estonia, using kriging interpolation, the GLM regression model, and point pattern generation. The characteristics of the neighbourhood, in addition to the properties of the local habitat, were taken into account in choosing the regression model. The neighbours' density distribution was applied as the dynamic control of point process. The density of moose covaried positively with the amount of habitat in the neighbourhood (Wald statistic = 860), distance from human settlements (W = 153), distance from blasting activity in open quarries (W = 144), distance from underground mines where explosions are performed (W = 74), and negatively with distance to roads (W = 19). The total estimated number of moose in Ida-Viru county, according to interpolated density, is 730 (mean density in habitats 3.20 moose per 10 km 2) and according to the regression model, 823 individuals (mean density in habitats 3.61 moose per 10 km 2). In earlier neighbourhood studies the vicinity of a location is usually treated as one homogenous zone, or the influence at different distances is integrated by a formal weighting function. We suggest calculation of the weights of the influence of distance zones from observed data by comparing the neighbourhood characteristics of two groups of sites, e.g. sites where a species occurred and sites where the species was not found. The difference in the mean coverage of habitat in the vicinity of sites with moose and without moose was found to be the largest at indicative neighbourhood distances 1 … 10 km from a site. The highest significance of the effect of the amount of habitat in the surroundings and the extent of indicative neighbourhood indicate the preference of extensive (hundreds of sq. kilometres) habitat areas by moose to a landscape of habitat patches. We generated various point patterns of a possible location of moose presuming three types of influence as the point process parameters: habitat suitability as estimated by the regression model, mean relative bias of the density of neighbours from expected density, and a random effect. Comparison of the generated patterns with census data indicates that inclusion of the effect of neighbours' density enables to reduce the random component of point process.