Abstract
The spatial distribution of summer flounder (Paralichthys dentatus) in relative abundance in survey samples differs over time with changes in environmental factors, such as bottom depth, bottom salinity, bottom temperature and sea surface temperature (SST). In 1991–2014 NOAA Fisheries have collected data on fish abundance and environmental covariates through their Fall and Spring bottom trawl surveys. We use a conditional autoregressive (CAR) model with these environmental covariates under the R-INLA framework and fit the observations over an irregular survey strata lattice. Results indicate that distributions of summer flounder stock seasonally correlate well with regional-climate-driven changes in bottom depth, bottom temperature and sea surface temperature. Estimating spatial autocorrelation and a second-order random walk in time both as fixed and random effects improved model performance. However, our study shows that such models can often inadvertently be over parameterized when including higher order interaction terms between spatial and temporal random effects. This can lead to inflated variances in the estimates and predictions as well as lengthening model convergence times. Therefore, when constructing models of this type, care should be taken in identifying the level of model complexity as well as the structural and statistical assumptions being made.