Integrating trawl and longline surveys across British Columbia improves groundfish distribution predictions


credit: Northeast Seamounts Expedition Partners

Supplementary Material


Thompson P. L., S. C. Anderson, J. Nephin, C.K. Robb, B. Proudfoot, A.E. Park, D.R. Haggarty, and E. Rubidge. 2022. Integrating trawl and longline surveys across British Columbia improves groundfish distribution predictions. Canadian Journal of Fisheries and Aquatic Sciences.

This Shiny app is an interactive version of the paper. It allows users to visualize the estimated response curves for each environmental gradient, spatially compare the different models, visualize the probability of occurrence and uncertainty in the final selected model, and visualize the coastwide species richness and groundfish assemblages for the 65 groundfish species included in the analysis. Users can select different groundfish species and environmental covariates to compare.

Abstract

Predictions of the distribution of groundfish species are needed to support ongoing marine spatial planning initiatives in Canadian Pacific waters. Data to inform species distribution models are available from several fisheries-independent surveys. However, no single survey covers the entire region and different gear types are required to survey the range of relevant habitat. Here we demonstrate a method for integrating presence absence data across surveys and gear types that allows us to predict the coastwide distributions of 65 groundfish species in British Columbia. Our model leverages data from multiple surveys to estimate how species respond to environmental gradients while accounting for differences in survey catchability. We find that this method has two main benefits: 1) it increases the accuracy of predictions in data-limited surveys and regions while having negligible impacts on accuracy when data are already sufficient, 2) it reduces uncertainty, resulting in tighter confidence intervals on predicted occurrences. These benefits are particularly relevant in areas of our coast where our understanding of habitat suitability is limited due to a lack of spatially comprehensive long-term groundfish surveys.

Data and Code Availability

Data from the trawl and HBLL surveys were obtained from the gfdata package. Data from the IPHC surveys were obtained from the gfiphc package.

Data and code are openly available:
DFO Groundfish Hard Bottom Longline Survey data
DFO Groundfish Synoptic Bottom Trawl Survey data
International Pacific Halibut Commission Independent Setline Survey data
Deep Substrate Model of the Pacific Canadian Shelf
The Salish Sea MEOPAR Model
Code used for analysis
Code used for Shiny app

Selected Model for Individual Species Distribution and Uncertainty



Probability of occurrence: spatial species occurrence predictions using the final selected model for each species at a 1 km resolution (in areas between 10 - 1400 meters depth and mean summer salinity > 28 PSU). Predictions were made for 2012-2015 and then averaged across years in each grid cell. Predictions should be interpreted as the probability of catching a species in a given location if sampling using the survey identified as most appropriate in the model selection process.
Occurrence hotspot probability: estimated areas where each species is most likely to be found. To identify occurrence hotspots for the individual species, we calculated the proportion of simulated values in each grid cell that exceeded a high occurrence threshold. This threshold was the lower of: 1) 0.8, or 2) the 80th percentile of the predicted probability of occurrence values across all grid cells that had a probability of occurrence greater than 0.05. Excluding grid cells with a probability of occurrence lower than 0.05 ensured that habitats that were unsuitable were not included. Using an upper bound of 0.8 for the hotspot threshold ensured that thresholds were not unreasonably high for species with high occurrence probability where it is predicted to be found (e.g. > 0.999 for Longspine Thornyhead).
Map projection is NAD83 / BC Albers and the coastline polygon was produced by the Canadian Hydrographic Service, DFO.

Interactive Map for Groundfish Species Richness


Map of the predicted number of species in each location (i.e., species richness). Predicted species richness is derived from summing the predicted species occurrences within each grid cell across all the selected final models.
This value should be interpreted as the number of species that would be expected to be caught if that location were surveyed using the most appropriate survey method for each species. This assumes, for example, that if one species has a probability of occurrence of 0.6 and another species has a probability of 0.8 in a given location, then a survey would catch an average of 1.4 species in that location if sampled repeatedly.
Predictions are made at a 1 km resolution. Map projection is NAD83 / BC Albers and the coastline polygon was produced by the Canadian Hydrographic Service, DFO.

Groundfish Species Assemblage at Selected Point


Enter coordinates in WGS84 (decimal degrees). See Species Richness plot mouse over for lat/long values that lie within the prediction extent.

Panal a shows the location of the selected point (red) and the extent of the prediction area (blue). Panal b shows the estimated occurrence probability of groundfish species for the 3 km grid cell that contains the selected latitude and longitude. Points indicate the mean predicted occurrence, thin bars represent the 95% confidence interval, thick bars represent the 50% confidence interval.

Estimated Environmental Response Curves





Estimates are with all other predictors set at their mean values. The models trained on just one survey are shown in yellow, the integrated gear models (trained on the HBLL and IPHC together, or the trawl and trawl (SoG) together) are shown in red, the fully integrated models that are trained on all four surveys are shown in blue. Each row shows the survey (fixed effect) used to make the predictions, labelled on the right hand side. The lines show the mean estimated value and the bands show the 95% confidence interval, which include intercept and main effect uncertainty but exclude random field uncertainty. The ticks at the top of the panels show the environmental conditions for all of the individual survey sets. Note that only depth is included in the single model for the SoG trawl.

Spatial Comparison of Different Models



Panel a shows the locations of the sets for each of the four surveys and sets that contain the chosen species are shown in red, and sets without the chosen species are shown in black.
Panel b shows the predicted probability of occurrence of the chosen species across the full coast at a 3 km scale. The three types of models - single, integrated gear, and fully integrated - are shown in different columns. Each row shows the survey (fixed effect) used to make the predictions, labelled on the right hand side. Note not all species will have all model and/or survey types.
Map projection is NAD83 / BC Albers and the coastline polygon was produced by the Canadian Hydrographic Service, DFO.