Bridging the gap between commercial fisheries and survey data to model the spatiotemporal dynamics of marine species

Abstract

Monitoring and assessment of natural resources often require inputs from multiple data sources. In fisheries science, for example, the inference of a species’ abundance distribution relies on two main data sources, namely commercial fisheries and scientific survey data. Despite efforts to combine these data into an integrated statistical model, their coupling is frequently hampered due to differences in their sampling designs, which imposes distinct bias sources in the estimator of the abundance distribution. We developed a flexible species distribution model (SDM) that can integrate both data sources while filtering out their relative bias contributions. We applied the model on three different age groups of the western Baltic cod stock. For each age group, we tested the model on (1) survey data and (2) integrated data (survey + commercial) as a means to compare their differences and investigate how the inclusion of commercial fisheries data improved the spatiotemporal abundance estimator and parameter estimates. Moreover, we proposed a novel validation approach to evaluate whether the inclusion of commercial fisheries data in the integrated model is not in direct contradiction with the survey data. Following our approach, the results indicated that the use of commercial fisheries data is suitable for the integrated model. Across all age groups, our results demonstrated how commercial fisheries supplied additional information on cod’s spatiotemporal abundance dynamics, highlighting sometimes abundance hot spots that were not detected by the survey model alone. Additionally, the integrated model provided a reduction of up to 20% and 10% in the uncertainty (SE) of the predicted abundance fields and fixed‐effect parameters, respectively. The proposed model represents thus a valuable benchmark for evaluating spatiotemporal dynamics of fish, and strengthens the science‐based advice for marine policymakers.

Projecting species distributions using fishery‐dependent data

“…species abundance) (Conn et al, 2017;Diggle et al, 2010;Pennino et al, 2019). Such preferential sampling violates a statistical assumption that sampling locations have been chosen independently of the value expected at a given location and can result in biased predictions of abundance and distribution (Alglave et al, 2022;Conn et al, 2017;Diggle et al, 2010;Pennino et al, 2019;Rufener et al, 2021).…”

The shadow model: how and why small choices in spatially explicit species distribution models affect predictions

“…Further development of methods, such as Bayesian sampling approaches ( e.g ., Margossian et al, 2020 ), that improve computation are critical advancements, especially for working with large datasets (as is common in ecology). Data assimilation: Exploring methods for combining multiple data sources ( e.g ., Grüss & Thorson, 2019 ; Webster et al, 2020 ; Rufener et al, 2021 ) in the estimation process is another frontier that will greatly further SDM development. The need for such methods arises commonly because there are limited resources for data collection, sampling methods and intensity change over time (resulting in spatially and temporally imbalanced datasets), and researchers are often interested in phenomena at broader ecological scales than can be evaluated using a single data source.…”

“…Evaluating the fit of predictive process SDMs and validating assumptions about structure involves considerations that are shared among many statistical models, but also includes others that are specific to—or particularly important for—spatial or spatiotemporal modeling. Examples of diagnostics for SDMs include the analysis of temporal or spatial autocorrelation in residuals ( Cressie & Wikle, 2015 ; Ward et al, 2018 ), randomized quantile residuals ( Dunn & Smyth, 1996 ; Hartig, 2021 ), one-step ahead residuals ( Thygesen et al, 2017 ; Breivik et al, 2021 ), residuals from MCMC draws from the posterior ( e.g ., Rufener et al, 2021 ), and examining evidence of non-stationarity. The choice of additional metrics of fit often follows from previous decisions about model structure.…”

sdmTMB: An R Package for Fast, Flexible, and User-Friendly Generalized Linear Mixed Effects Models with Spatial and Spatiotemporal Random Fields

“…For state-space models, these residuals have known statistical issues (Thygesen et al 2017) but are quick to calculate. Simulation-based residuals (Hartig 2021) (dharma_residuals()), one-step-ahead residuals (Thygesen et al 2017), or residuals from a single MCMC draw (e.g., Rufener et al 2021) are also possible, but are slower. (5) The simulate.sdmTMB() method can simulate from fitted models and the sdmTMB_simulate() function can simulate entirely new data to which models can be fit to ensure identifiability, evaluate bias and precision in parameter estimation, or evaluate the consequences of model misspecification.…”

“…Simulation-based residuals (Hartig 2021) (), one-step-ahead residuals (Thygesen et al . 2017), or residuals from a single MCMC draw (e.g., Rufener et al . 2021) are also possible, but are slower.…”