Zero-altered modeling of an aquatic parasite host with application to invasive species risk assessments

Abstract

A fundamental consideration in aquatic invasive species risk management is the distribution of invasive taxa relative to the risk assessment area. However, sampling the distribution of aquatic invasive species (AIS) is costly and time consuming, especially when they are rare or clustered. While random sampling removes bias in most estimates, it could lead to many zeroes in the response variable if the target species is not evenly distributed. We surveyed the distribution of Tubifex tubifex , the obligate secondary host of Myxobolus cerebralis , an invasive parasite that causes whirling disease in salmonids. We used intensive grid and random sampling in a lake and two river systems but collected many samples with zero T. tubifex . Zero-altered models identified a single source of zero T. tubifex counts in each dataset and improved model validation compared to Poisson or negative binomial models. While T. tubifex counts were not predictable using the covariates we measured, the binomial processes identified zero odds of a T. tubifex occurrence at lake sites with water depths > 2 m and stream sites with channel slopes > 3%. These covariates could be used to stratify the landscape for future sampling, which would save survey time and resources and likely reduce uncertainty in parameter estimates. Model-based sample stratification could address some of the challenges with AIS surveys when species are at low abundance or are clustered.