Spatio-temporal models reveal subtle changes to demersal communities following the Exxon Valdez oil spill

Abstract

Toxic pollutants such as crude oil have direct negative effects for a wide array of marine life. While mortality from acute exposure to oil is obvious, sub-lethal consequences of exposure to petroleum derivatives for growth and reproduction are less evident and sub-lethal effects in fish populations are obscured by natural environmental variation, fishing, and measurement error. We use fisheries independent surveys in the Gulf of Alaska to examine the consequences of the 1989 Exxon Valdez oil spill (EVOS) for demersal fish. We delineate areas across a range of exposure to EVOS and use spatio-temporal models to quantify the abundance of 53 species-groups over 31 years. We compare multiple community metrics for demersal fish in EVOS and Control areas. We find that areas more exposed to EVOS have more negative trends in total groundfish biomass than non-EVOS areas, and that this change is driven primarily by reductions in the abundance of the apex predator guild. We show no signature of increased variability or increased levels of synchrony within EVOS areas. Our analysis supports mild consequences of EVOS for groundfish communities, but suggests that long time-series and assessments of changes at the community level may reveal sub-lethal effects in marine communities.

Spatial community structure of groundfish is conserved across the Gulf of Alaska

“…We used smoothed estimates of AFSC survey data generated by a geostatistical model developed by Shelton and colleagues (Shelton et al 2014, 2018, Ono et al 2016 for several reasons, as follows. These geostatistical models accounted for imprecision in the data (ex pressly because the raw survey data is variable), environmental covariates (e.g.…”

“…Because most species were absent from a large number of observed trawls, we modeled the presence and distribution of species density separately, using a delta-GLMM approach with 2 submodels (Pennington 1983, Maunder & Punt 2004, Shelton et al 2014. The first sub-model ('occurrence model') described the probability of occurrence of individual species in a given year and at each given location; the second sub-model ('positive model') described the CPUE (kg hectare −1 ) of species in a given year and at each location conditional on the occurrence of at least 1 individual (please see specific equations in Shelton et al 2017Shelton et al , 2018. Both submodels included spatial random effects that follow a first-order autoregressive process and account for variation not explicitly included in the model (e.g.…”

“…We projected model estimates to the centroid of 2 × 2 km grid cells covering the entire GOA from 1000 Markov Chain Monte Carlo (MCMC) samples generated from the joint (both sub-models) approximate posterior density for each species (Shelton et al 2018). We calculated unconditional expectations for CPUE (kg hectare −1 ) of each species in each year and at each location by multiplying each MCMC sample from the occurrence and positive model (Shelton et al 2018). We properly accounted for the uncertainty in these estimates by sampling from the joint posterior distribution.…”

“…We properly accounted for the uncertainty in these estimates by sampling from the joint posterior distribution. These predicted CPUEs for each 2 × 2 km grid cell were combined to generate mean estimates (and standard deviations) of CPUE for each species in each year within discrete study areas (see Section 2.2) (Shelton et al 2014, 2018, Ward et al 2015. Further information on model structures, formulations, and projections can be found in Shelton et al (2018) and in the code from that publication (Shelton et al 2017).…”

“…While other datasets such as vessel monitoring system data with geographically referenced locations might provide higher resolution, these data do not exist for the entire time period of this study and so were not used. Instead, using the CFEC catch data, we calculated the total groundfish mass sold (sum) from trawl and longline fisheries to get a catch value (metric tons km −2 ) representing fisheries removals for each broad region and study year (1985−2014) (Shelton et al 2018). For some analyses, we then took the average across years or sites within a catch area for each year.…”