Dalarna University's logo and link to the university's website

In environmental impact assessments, it is important to be able to estimate influence of anthropogenic activities on animal populations. To quantify the influence, it is common to estimate how far, in distance, from a given disturbance source there is an influence on the animals’ habitat selection through estimating a zone of influence (ZOI). Usually, ZOI is estimated for one disturbance source at a time. In this work, we demonstrate how threshold regression modelling can be used for estimating ZOI from several possible sources of disturbances, simultaneously. Based on the theoretical properties of different estimation methods for the estimation of threshold regression we select a set of estimation methods and compare their merits through a simulation study and a real data example. The simulation results revealed that Adaptive Lasso, and Hierarchical likelihood (HL) methods, are two reasonable methods for dealing with the problem. HL performed better than Adaptive Lasso in that it had much higher success rate in identifying correct threshold with small sample size whereas Adaptive Lasso requires large sample to assure good performance. While Adaptive lasso needed to be aided with suitable weights, which are not easy to find, HL method did not require any prior weights. These two methods were applied to estimate the ZOI around 40 wind turbines and surrounding public roads on reindeer habitat selection in winter, by using GPS positioning data from 42 reindeer in north of Sweden in December to March (2012-2015). The results showed that both the disturbance sources have a negative effect on reindeer habitat selection in winter. The HL approach showed that the negative ZOI from the nearest wind turbine was 1.8 km (approx.), however the trend of higher selection of areas further away from the wind turbines was evident up to 4 km (approx.) from the active wind turbines.