Merrill Rudd¹ Jim Thorson^2
1Scaleability LLC
²NOAA
April 2nd, 2019 9:00 (PST): FSH 203

Spatiotemporal models of populations in stream networks

Abstract: Spatial and spatiotemporal models of stream networks are vital in the management and conservation of freshwater and anadromous fishes, often threated by terrestrial development along connected waters upstream. An ideal approach to modeling stream networks would 1) allow for spatial correlations over time or autocorrelations as a function of distance, 2) distinguish between process and observation error, and 3) overcome computational challenges to analyze large networks while estimating covariance structures. The latter two issues for estimating spatiotemporal densities are addressed using the Vector Autoregressive Spatio-Temporal (VAST) model. VAST has many features that make it an accessible and reproducible tool for single and multivariate spatiotemporal models with covariates. Previous applications have only used Gaussian Markov random fields to approximate spatial and spatiotemporal variation. We added a stream network spatial model that uses the Ornstein-Uhlenbeck process, a stochastic process that implies a child node is correlated with its parent node as a function of distance, where variation includes an exponential rate of decay in correlation between child and parent nodes and an asymptotic variance for two sites that are far apart. This presentation will provide an overview of updates to the VAST R package considering stream networks as a spatial model and preliminary results of its application for longfin eels in the Waitaki River catchment in New Zealand. The Waitaki River catchment has been heavily affected by dams, the largest of which was installed in 1948. Longfin eels are a cultural keystone species for Maori. In the face of increasing terrestrial development, impacts of dams, and continued harvest, spatiotemporal modeling including multiple types of information is vital for estimating longfin eel densities. This work will contribute to ongoing studies of New Zealand longfin eel populations and later other freshwater species. This modeling approach is also being applied to model juvenile and adult densities of coho salmon in Oregon. This study could benefit from the perspectives of scientists with stream network modeling experience or spatiotemporal modeling experience in other contexts.