Can a habitat selection model predict the distribution of moose Alces alces over multiple years?

Abstract

The Fennoscandian population of moose Alces alces has been growing exponentially for decades. It is the centre of a conflict between stakeholders in the logging and hunting industries, who respectively prefer a low and high number of individuals to maximise economic gain. Population management is therefore of concern to financial stakeholders as well as wildlife management bodies. So far, management has been focussed on increasing the moose population, to benefit the hunting industry. Predicting distributions of populations is an important tool for management. It is commonly accepted that species distribution is closely linked to habitat selection. Even so, few studies have investigated whether habitat selection models can predict distribution. This study investigates whether the winter distribution of a heavily managed ungulate species is predictable using models based on habitat factors. Focussing on three management sites in Norway of approximately 40km2 each, a measure of time spent by moose in a patch (100m2) was generated using the number of moose pellets in 960 patches. Using GLMs validated by AIC values, and habitat data from 2012 and 2015 – as well as data of moose distribution in 2012 to 2015, and finally 2017, – the key factors selected for by moose were identified. These factors showed a high explanatory power over moose distribution. The parameters of the model provided accurate descriptions of distribution for three years before accuracy began to fall. Despite this, the predictions of the model for all years showed a low accuracy when compared to observed distribution. The accuracy was not improved by using newer habitat data. The cause of this is likely to be that moose show spatial autocorrelation in their distribution and that selection may not be strong enough to determine distribution. As these results show, predicting the future population of a large ungulate yields varying results. It is important that wildlife managers account for this when creating management strategies. These results also show that future studies attempting to model habitat selection must test their predictions against real world data before attempting to use them to create management strategies.