Combining sources of high-resolution fishery-dependent data from the northeast United States to develop a catch rate time series

Abstract

Fishery-dependent data, despite being a valuable resource, is often underutilized in addressing scientific inquiries comprehensively. Similar to citizen science data, it possesses significant potential to enhance our comprehension of changing species abundances and distributions. In this study, we present an illustrative example of harnessing available fishery-dependent data from the northeast United States to generate a valuable scientific output. Our approach involves combining data sets from two high-resolution fishery-dependent research and monitoring programs: (1) the Northeast Fisheries Science Center’s Study Fleet Program and (2) the Northeast Fisheries Observer Programs. By focusing on data collected from vessels employing bottom trawl gear, we construct a novel data set and establish a nominal catch per unit effort index specific to black sea bass (Centropristis striata), a commonly targeted species. To augment this data set, we incorporate additional variables pertaining to ecological and economic factors that could influence catch rates. The expanded data set is then used to develop a catch rate standardization using a generalized additive model. This study not only exemplifies how these unique, high-resolution data sets can be effectively leveraged for scientific purposes but also provides a detailed account of the methodology employed to compile these extensive data sets. We suggest that similar techniques could be applied to other species within the region or to analogous data sets from different regions. Advancing methodologies for utilizing fishery-dependent data in scientific research is a cost effective means for improving our understanding of species abundance and ecosystem dynamics. Moreover, it presents an opportunity to demonstrate the value of incorporating the knowledge and experiences of fishers and other stakeholders quantitatively into the scientific process. By tapping into the wealth of information provided by fishery-dependent data, we can make significant strides in expanding our scientific knowledge and informing sustainable management practices.

logbook, catch rates, fishery dependent data, northeast United States, observer programs, reference fleet

Introduction

There is growing recognition of the direct and indirect benefits of pursuing science-industry research collaborations (SIRCs; sensu Steins et al. 2020). Engaging the fishing community in collecting scientific data and using those data to construct novel scientific products is a form of SIRC. This type of SIRC can be undertaken with limited additional effort by fishers because it can fit in the natural rhythm of fishing activity and therefore has the potential to be maintained over a longer period of time (Johnson and Van Densen 2007). These endeavors parallel citizen science efforts where data sets collected outside traditional research surveys are used for a variety of scientific purposes (Hyder et al. 2015, Kaiser et al. 2019, Dalby et al. 2021, Rufener et al. 2021, Karp et al. 2023). In the same way that citizen science can help to answer research questions when our ability to observe a phenomenon is hampered by its scale, SIRCs can help advance understanding of marine ecosystems that are difficult to observe by filling data gaps and producing estimates of abundance independently or in combination with other sampling programs (Fletcher et al. 2019, Brick et al. 2022).

Catch rate index development, where catch and effort information are utilized to understand trends in the abundance of a stock, is a unique application of fishery-dependent data and an example of a potential SIRC. Specifically, data collected for fishery monitoring or research can be put to use to better understand the ecology of commercially important marine organisms. Direct scientific benefits of collaboratively developed catch per unit effort (CPUE) standardizations include indices that may contribute to stock assessment models or other quantitative aspects of the science process (Stephenson et al. 2016, Mercer et al. 2023). Indirect benefits include increased confidence in stock assessment inputs and processes, the opportunity to better communicate the details of specific stock assessments (e.g. model assumptions or data limitations), and potentially increased trust between scientists, managers, fishers, and other stakeholders (Steins et al. 2020). These benefits must be weighed against concerns about potential sources of bias that are inherent to data sets that represent directed fishing efforts (Maunder and Punt 2004, Hoyle et al. 2024). Sources of bias for these types of data sets are well documented and range from imprecise or inaccurate reporting to the impacts of management measures on targeting or avoidance (Conn et al. 2017, Pennino et al. 2019, Alglave et al. 2022). Assessment scientists are especially cautious with CPUE-based indices because non-linear relationships between abundance and a CPUE index can lead to hyperstability and inaccurate patterns of abundance (Hutchings and Myers 1994, Rose and Kulka 1999). While the balance between the benefits and challenges of interpreting and applying CPUEs is stock-specific, their development and honest evaluation provide valuable insights about fishery dynamics and help further discussion about potential differences in perception between fishers and scientists.

Within the northeast United States, a number of long-running fishery-independent data sources exist to assess the distribution and abundance of commercially important species (Grosslein 1969, Politis et al. 2014). The availability of these synoptic surveys and concerns about the reliability of preferentially sampled data sets have led to minimal use of CPUE indices in stock assessments (Richardson et al. 2014, Cadrin et al. 2020), as well as the limited use of fishery catch rates in other scientific studies such as species distribution models (McHenry et al. 2019, Lowman et al. 2021, Braun et al. 2023). This underutilization of fishery catch rate data overlooks a potential source of valuable information that could provide a complementary view of the marine ecosystem and potentially fill in years of data when fishery-independent information is not available (Link et al. 2021, Saba et al. 2023). Exploring catch rate indices may reveal patterns in species distribution and contribute an alternative perspective of abundance that better aligns with the reality that fishers are experiencing on the water. Fishery-dependent data can provide an important comparison to indices of abundance derived from scientific surveys or intercept surveys of recreational catch. Working through these data sets enhances the discussion of trends in abundance, economic realities of a fishery, and longer-term changes in the ecological or social systems (Johnson and van Densen 2007, Steins et al. 2020).

At the same time that quantitative explorations of CPUE indices in the northeast United States have failed to develop, the quantity and quality of fishery-dependent data have increased greatly. In addition to multiple layers of quality control that have been built into the synoptic vessel trip reporting (VTR) process, other data systems now integrate mandatory dealer reporting and VTRs to produce comprehensive records of catch and effort (Palmer 2017, Hocking et al. 2024). The data fields collected through VTRs are focused on meeting management needs and thus provide information at a relatively coarse level, often at a scale with fishing efforts binned into management grids larger than 50 km squared.

On the finer scale of an individual fishing effort, which is typically a few kilometers in length, two other fishery-dependent data sources have emerged as sources of information from which to construct CPUE indices and advance understanding of fishery and ecosystem dynamics: (1) the Northeast Fisheries Science Center’s Study Fleet Program and (2) the Northeast Fisheries Observer Programs. The Study Fleet Program run by the Northeast Fisheries Science Center’s Cooperative Research Branch is a SIRC that was initiated in 2006 to engage fishers in collecting detailed catch and effort data from individual gear tows for scientific research (Palmer et al. 2007, Jones et al. 2022). While the Study Fleet was originally focused on the groundfish fishery in the Gulf of Maine, the diversity of focal fisheries and geographic range of the Study Fleet have greatly increased over time (Jones et al. 2022). The number of trips sampled by the Observer Program has also increased through time in an effort to reduce the uncertainty in regional estimates of discarded catch. This has included the development of an at-sea monitoring program for the groundfish catch share program that complements traditional observer deployments, as well as other industry-funded monitoring programs for the scallop fishery (Brooke 2012).

The availability of these time series of high-resolution catch and effort information has the potential to greatly improve CPUE estimates through precise location information and to allow catch rates to be linked to a number of other ecosystem and socioeconomic variables that are potentially useful for CPUE standardization models. Specifically, using high-resolution catch and effort data allows us to bring in covariates from bathymetric and oceanographic data sets that are available and ecologically significant at fine scales. Thus, observer and Study Fleet data provide an advantage over more synoptic data sets (such as dealer reports and VTRs), which do not collect catch and effort at the tow level and therefore cannot be linked to bathymetric or oceanographic data (but see Hyun et al. 2014, Hansell et al. 2020). Because of the extensive fishery-independent sampling within the region, this and similar efforts could provide a valuable avenue for testing CPUE methods. This ability to compare results to independent sampling makes CPUE products from this region potentially insightful to a number of other regions and systems.

Here, we provide an example of how to use multiple sources of fishery-dependent data to construct standardized CPUE indices for application in a stock assessment (Fig. 1). First, we combine catch, effort, and a set of ecosystem and socioeconomic covariates likely to be important to standardizing catch and effort information. Next, we align the coding, convert units, and ensure that sampling schemes are producing similar estimates of effort and catch. We then refine the full data set down to a relevant set of records by focusing on a single focal species and gear type and fit CPUE standardization models. Finally, we qualitatively compare CPUE indices to those developed using fishery-independent surveys. Through this process, we hope to provide an example of how fishery-dependent data can be synthesized and used to address scientific questions. This work also serves as a practical example of the utility of these data sets when combined in a thoughtful way. This type of data synthesis and analysis is a key component of SIRCs, in that it takes the information collected from industry and applies it to scientific questions that are essential to the management of commercially important stocks.

Figure 1.

Graphic depicting the steps in data processing and CPUE analysis. (1) Raw data sets are compiled to create a tidy data set with information about the species total weight per tow. Additional relevant information available in only a single data set is also included (e.g. empirical bottom temperatures from the Study Fleet data set and empirical estimates of fished net width from the Observer program). (2) Program codes and data elements are consolidated to a single standard. Metrics of associations or other criteria are used to subset these data to a relevant set of records. (3) Elements needed to calculate a catch rate are estimated from each data set. For example, with black sea bass, a swept area is estimated from gear size, tow duration, speed of vessels while towing, and the total catch of black sea bass per tow. (4) Covariates of interest are incorporated into the data set. These include ecologically important variables like bottom water temperature from modeled sources, as well as sociologically relevant metrics like the price of fuel and the price harvesters received for black sea bass. (5) With a fully developed data set, the information is visualized, and models are fit to the data. In this example, this involved splitting the data between black sea bass regions and fitting CPUE standardization models. (6) Once a product is developed, it goes through a process of comment and feedback, eventually reaching a versioned state of completion.

Methods

Example stock

For this work, we focus on the northern stock of black sea bass (Centropristis striata) as an example of how to develop and apply catch rate standardizations for a stock assessment. The spatial and temporal dynamics of black sea bass, including potential environmental drivers, are complex (Miller et al. 2016). Thus, seasonal fishery-independent surveys are insufficient for understanding the population dynamics of this stock. Fortunately, there are abundant high-resolution fishery-dependent data for this species, making it an ideal candidate for developing a standardized catch rate. A commercial CPUE has not been included in recent stock assessments for the northern stock of black sea bass, but the black sea bass fleet is actively engaged in data collection to support CPUE standardizations to better understand the dynamics of the stock and fishery (NEFSC 2017).

The northern stock of black sea bass occupies the region from Cape Hatteras, North Carolina, to the southern portion of the Gulf of Maine, with most of the catch coming from the Mid-Atlantic Bight (Shepherd and Terceiro 1994). This stock is currently further split into northern and southern regions for the stock assessment (NEFSC 2017). The southern region of the northern stock of black sea bass extends from Cape Hatteras, North Carolina, to roughly the Hudson Canyon, and the northern region then continues from there to the Gulf of Maine (Fig. 2). Historically, the abundance of black sea bass has been limited on Georges Bank and in the Gulf of Maine; however, warming ocean temperatures have expanded the northern edge of their range (Bell et al. 2015). Fish from both regions follow a seasonal migratory pattern, where in the spring fish are typically near the shelf edge, then as shelf waters warm in the late spring and early summer fish migrate inshore (Moser and Shepherd 2008, Fabrizio et al. 2014). This cycle coincides with a number of life history transitions, and differences in ocean conditions are hypothesized to affect survival at specific stages (Miller et al. 2016). Black sea bass that are of sufficient size to be caught in the commercial fishery are traditionally associated with complex benthic features (Cullen and Guida 2021). Conversations with fishers, however, suggest that the recent increase in black sea bass biomass has resulted in expanding distribution and availability of the species across bottom types.

Figure 2.

Spatial breakdown of northern and southern regions used in the black sea bass stock assessment along existing statistical area boundaries. The catch in the NEFSC’s bottom trawl survey across the years included (2003–2023) is shown as purple points. The commercial catch is depicted as a grid of total landings across all years included in the study. To maintain confidentiality, cells with fewer than three commercial catch records were removed.

Commercial data

Here, we focus on records from commercial fishing vessels using trawl gear to capture black sea bass and other similarly distributed species within this northeast US region. Bottom otter trawl gear is commonly used throughout the region by fishers targeting a wide variety of species. Because of this, the region’s major fishery sampling programs tend to sample the vessels using trawl gear to a greater extent than other gears such as hook and line or pot and trap (Wigley et al. 2007, Wigley et al. 2021). In addition to being well sampled by high-resolution fishery sampling programs, the effort metrics for this gear are well defined, and sampling intensity can be easily quantified. Specifically, an estimate of the sampling area (often referred to as swept area) can be constructed using the speed a vessel travels, the time that the vessel is actively fishing, and the effective width of the gear. Finally, catch from vessels using trawl gear makes up ∼45% of the total black sea bass landings (Fig. 3, NEFSC 2017), and many of the captains and vessels using this gear have been fishing in a consistent manner for an extended period of time (Jones et al. 2022).

Figure 3.

(a) Commercial black sea bass landings taken by trawl gear or other gears (pot, trap, gillnet) through time. (b) The proportion of the total commercial landings from trawl gear that are reported at higher resolution by either the Study Fleet or the Observer programs in the northern and southern regions for black sea bass. Note the y-axis in panel (a) represents total commercial landings across all gears, and in (b) represents only total commercial landings from trawl gear.

We leverage two similar but distinct high-resolution fishery-dependent data sets that both record a vessel’s catch and effort at a single fishing effort (gear retrieval): the Northeast Fisheries Science Center’s (NEFSC) Observer Programs (Brooke 2012) and the NEFSC’s Study Fleet Program (Palmer et al. 2007, Bell et al. 2017, Jones et al. 2022). Observer Program data are collected by at-sea biologists deployed on a subset of trips from selected federally managed fisheries (Wigley et al. 2007). The Study Fleet Program is a long-standing SIRC that trains collaborating captains to independently collect catch, effort, and environmental data via an electronic logbook system. Captains are compensated to collect data at a higher resolution than is mandated by regulations and to record all components of the catch (both kept and discarded catch). The data models (how data are collected and stored) and specific data elements for both the observer and Study Fleet programs are quite similar, making it possible to combine information from them. Moreover, comparisons of co-sampled trips suggest that catch weight estimates from these programs are often quite similar (Bell et al. 2017). Together these data represent the highest resolution information about commercial fisheries catches for the northeast United States. They provide information on a subset of the total trips but are thought to provide a representative sample of fishing effort (Wigley et al. 2021, Jones et al. 2022). To maximize the time series for comparison, we used Observer data from 2003 to 2023 and Study Fleet data from 2007 to 2023. Catch data were subset according to the methods provided in the Supplementary Material.

Catch rates

The swept-area biomass-per-tow (CPUE_i [kg/km²]) was calculated as

$$\begin{eqnarray} \mathrm{ CPU{{E}}_i} = \frac{{{{k}_i} + {{d}_i}}}{{{{t}_i}{{s}_i}{{f}_i}c}}, \end{eqnarray}$$

where k_i and d_i are the kept and discarded weights of black sea bass on tow i, t_i is the active fishing duration (from when the gear is locked in place to the beginning of the haul back), s_i is the tow speed, f_i is the footrope length, and c is a conversion factor to estimate fished wing spread from footrope length. The width at which a trawl net is fished (wing spread) can be estimated precisely with modern net mensuration equipment; however, this information is rarely collected by even the more detailed fishery sampling programs (no information is collected by the Study Fleet program, and very few records exist from the Observer program). Available data were used to generate estimates of wing spread (Fig. S1), and these numbers were similar to other estimates of footrope to fished width conversion factors that exist in the wider literature (Gómez and Jiménez 1994). For the vast majority of tows, the speed of the vessel was reported in the data set; however, for a subset of records, the mean speed derived from both data sets was used instead.

Catch rate standardization

To explore inter-annual changes in abundance, statistical standardizations attempt to account for differences in commercial fishing behavior between years (Maunder and Punt 2004). Developing appropriate standardizations are challenging tasks that require a number of choices (Forrestal et al. 2019), and recent publications highlight the choices that most commonly lead to successful analyses (Hoyle et al. 2024). The impact of a standardization on the annual trends in catch rates is often investigated by comparing a nominal catch rate to a standardized catch rate, and we calculate both nominal and standardized catch rates from the high-resolution commercial trawl catch data. An annual CPUE generally consists of the average catch in a fishery in a given year, and this mean can be calculated as a stratified (e.g. area weighted) or simple mean for a fishery. Here, we weighted catch values by the size of the statistical area in which they were collected, following standard practice for commercial discard estimates and fishery-independent survey indices within the region (Bell et al. 2017). The annual catch rate (CPUE_y) for year y was calculated as

$$\begin{eqnarray} \mathrm{ CPU{{E}}_{\mathit{ y,a}}} = \frac{{\mathop \sum \nolimits_{i = 1}^{{{n}_{\mathit{ y,a}}}} \mathrm{ CPU{{{\rm{E}}}}_{\mathit{ y,a,i}}}}}{{{{n}_{y,a}}}}, \end{eqnarray}$$

$$\begin{eqnarray} \mathrm{ CPU{{E}}_{\mathit{ y}\ }} = \frac{{\mathop \sum \nolimits_{a = 1}^A \mathrm{ CPU{{E}}_{\mathit{ y,a}}}\times \ {{m}_\mathit{ a}}}}{{\mathop \sum \nolimits_{a = 1}^A {{m}_a}}}, \end{eqnarray}$$

where a is the geographical statistical area (indexed from 1 to A), n_y,a is the number of tows that occurred in area a in year y, and m_a is the size [km²] of area a.

CPUE standardization, which generally attempts to account for the effect of confounding variables on catch rates, can be completed with a variety of statistical methods, and a primary focus of standardization studies is to compare the fits of different model types (e.g. models fit with complements of covariates). Generalized linear models (GLMs) have historically been the most commonly used (Maunder and Punt 2004). These models allow for fitting with a range of parameters and many distinct error distributions to accommodate different distributions of catches. Generalized Additive Models (GAMs, Wood 2006) allow for similar link functions but also for nonlinear responses that are common in social and biological systems. GAMs are an effective tool when the functional relationship among predictors and response variables is not known. Additionally, incorporating nonlinearity in a GAM is helpful in that it potentially avoids including many interactions that would be needed in a traditional GLM, allowing users to better visualize the relationships between the univariate terms of the GAM and the dependent variable. This is especially important for CPUE standardization, where researchers are interested in evaluating multiple models and assessing the impact of specific variables on catch. For this field, where more sophisticated models often improve model fit to a limited extent, model interpretability may be more important when choosing among different modeling tools.

Recently developed software such as the mgcv package (Wood 2011) allows for GAM fitting in a fairly straightforward way with commonly available statistical languages like R (R Core Team 2022). This software facilitates the application of GAMs by automating the selection of important parameters (e.g. the number of knots to use and the penalty to be applied to achieve the appropriate degree of smoothness). This software also features the ability to fit models using a diversity of error distributions that are commonly tested in CPUE standardizations.

Forward model selection or other stepwise selection methods are commonly used in fisheries and CPUE standardization (e.g. Forrestal et al. 2019); however, this type of model fitting has been criticized recently because the order of variable addition can impact the perceived importance of the variable, and it can be difficult to rigorously define selection criteria for the variables (Tredennick et al. 2021). Here we took the approach of a priori defining a list of important variables based on structured conversations with black sea bass fishers and previous literature (e.g. Cullen and Guida 2021), and fit a full model with those variables present. This method is sometimes referred to as expert opinion for variable selection. The details of covariate data curation can be found in the Supplementary Material.

Catch rates were modeled independently for each region (north and south) using GAMs employing a Tweedie error distribution and a log link function with fixed categorical effects for the year the vessel fished (year), data source (source), target species for the trip (target), port at which the vessel landed (port), and vessel identifier (permit), as well as smoothed functions for location (lat, lon), depth (depth), month (month), weekly price of diesel (diesel), weekly price of black sea bass (price), a location matched terrain ruggedness index (tri), a bottom temperature—either empirical or modeled (temp), and the gross tonnage of the vessel (tons):

$$\begin{eqnarray} && {{U}_i}\sim\ \textit{Tweedi}{\mathit{ {e}}_p}\left( {{{u}_i},\varphi } \right)\ ;\ \mathit{ lo{{g}}_\mathit{ e}}\left( {{{u}_i}} \right)\ \sim\ 0\ + \ \mathit{ yea{{r}}_\mathit{ i}}\\ &&\quad + \ \textit{sourc}{{\mathit{ e}}_i} + \ \textit{targe}{{\mathit{ t}}_i} + \ \mathit{ por{{t}}_\mathit{ i}} + \ \textit{permi}{{\mathit{ t}}_i}\\ &&\quad + te\left( \mathit{ {la{{t}}_\mathit{ i}},lo{{n}_\mathit{ i}}} \right)\ + \ s\left( {\textit{dept}{{\mathit{ h}}_i}} \right) + s\left( {\textit{mont}{{\mathit{ h}}_i}} \right)\\ &&\quad + \ s\left( {\textit{diese}{{\mathit{ l}}_i}} \right)\ + \ s\left( {\textit{pric}{{\mathit{ e}}_i}} \right)\ + \ s\left( {\mathit{ tr{{i}}_i}} \right)\ + \ s\left( {\mathit{ tem{{p}}_\mathit{ i}}} \right)\\ &&\quad + \ s(\mathit{ ton{{s}}_\mathit{ i}}). \end{eqnarray}$$

A Tweedie error distribution allows for the modeling of a skewed continuous variable with a point mass at zero, which lends itself well to the modeling of catch rate data (Shono 2008, How et al. 2022). It has a power parameter (p) and a dispersion parameter (φ), which make it a flexible modeling framework. The intercept was suppressed for the model so that the year variable represents the mean estimate for each time slice. To help guide models toward larger trends in the data and avoid overfitting, models were fit with restricted maximum likelihood, a more limited set of knots (k was set to 5 for all smooths except the tensor for latitude/longitude), and a package option was used to further penalize the likelihood criterion to shrink linear effects out of the model when it is justified (see Marra and Wood 2011). The basis function for the smooth of month in each model was fit with a cyclic cubic spline (bs = “cc”), while all other smooth terms were fit with thin plate splines (bs = “tp”).

To generate an annual index of abundance, the mean values of the focal term of the year were predicted using the ggeffects package (Lüdecke 2018). The ggeffects package uses mean values for continuous variables and averages across categorical variables. For the continuous variables representing the spatial tensor of latitude and longitude, a single mean value was also compared to a grid of spatial points as well as all locations where data was collected. This altered the magnitude of the annual value of the abundance estimates and had subtle effects on the confidence limits associated with these values, but did not impact the annual trends in abundance. To further explore the sensitivity of the mean annual abundance estimates, we iteratively fit the models for each region and calculated a Mohn’s Rho value for each index using the icesAdvice package (Magnusson et al. 2022). This average of the relative bias for each region was calculated using five peels and compared these values to those from simulation work to help interpret these findings (Hurtado-Ferro et al. 2015).

Index comparison and model validation

The catch rate indices developed for each region were compared to two other major indices of abundance used in the black sea bass stock assessment: (1) a model based fishery-independent trawl survey index and (2) a CPUE derived from recreational catches (NEFSC 2017). The fishery-independent bottom trawl survey index was developed using data from the NEFSC Bottom Trawl Survey, a primary source of relative abundance information for many stock assessments in the northeast United States (Politis et al. 2014), as well as a number of regional state trawl surveys. The recreational CPUE is developed using similar criteria (i.e. a set of positive records expanded using species associations similar to the Jaccard index) to those used here for recreational angler catches of black sea bass (NEFSC 2017). The trawl survey and recreational CPUE indices were both included in the most recent black sea bass stock assessment (NEFSC 2024).To compare the commercial CPUE indices developed here with the trawl survey and recreational CPUE indices, we first separated the data into comparable footprints and then ran a correlation analysis to quantify the relationship between the indices. For both the trawl survey and recreational CPUE indices, we used the years that were available from the most recent black sea bass stock assessment (2003–2023).

To evaluate model predictions more quantitatively, we performed a series of analyses using subsets of the existing data for each region. First, the data were reduced by selecting 50 random tows from each statistical area in each region for each year. These datasets were then replicated 20 times and randomly split into training and test subsets, with 70% of the data in each training set and 30% in the test set. A series of 13 models were then fit to each training dataset, with each successive model including an additional term from the full model. In total, 520 models of varying complexity were fit across both regions and then evaluated for their performance using total deviance explained. Finally, a metric of the relative error of the catch rate index was generated for each replicate model in each region by comparing predicted catch rate values to those observed in the test datasets. For this metric of index relative error, we calculated the root of the squared difference between the predicted and observed index values.

Results

Subsetting the data

Simple pooling of all bottom trawl data from the Observer and Study Fleet data sets provided a large volume of information for the time period we examined (2003–2023), with 327 168 individual tow records from fine scale sources (119 162 trips from Study Fleet and 208 006 trips from the Observer Program). Black sea bass catches ranged in size from zero to ∼7000 kg per tow.

In total, 89 641 records remained after subsetting the full data to include only those tows where either black sea bass or scup were caught (see Supplementary Materials). Removal of records with at least one missing or improbable value (across all covariates) reduced the total number of records to 84 488. This drop likely reflects the nature of the data set being used (i.e. this is data reported by fishers and fisheries observers splitting their time between collection and other fishing activities) and the large number of fields that need to be collected on each fishing effort to complete the data set. The final data set had 77 032 records from the northern region and 7 456 from the southern region, representing approximately 10% and 40%, respectively, of the total fishing activity in a given year (Fig. 3). Records primarily come from the offshore portions of the southern region and both inshore and offshore portions of the northern region, similar to the spatial distribution of the total trawl catches represented in Fig. 2. Records came from catches on 381 total vessels with some turnover in vessels over time (the mean number of vessels was 74 per year).

Methodological considerations

CPUE metrics can be estimated in a number of ways. Here, we estimated catch per unit of swept area using the best available information. Swept area estimates were not linearly related to black sea bass catch. Instead, catch increased to a maximum of around 150 kg and then decreased as swept area further increased. Because tow duration and speed are key elements in this calculation, we inspected the data to ensure data were similar among programs. The Observer mean tow duration (2.25 h ± 1.10) was slightly higher than the Study Fleet mean (1.95 h ± 0.95). Similarly, the Observer mean tow speed (3.04 knots/h ± 0.27) was marginally higher than the Study Fleet mean (2.88 knots/h ± 0.16). Gear size in terms of trawl net footrope length also played a large role in effort metrics and was similarly distributed among programs (observer mean = 103 ft ± 30; Study Fleet mean = 102 ft ± 29). For the subset of records with reported wing spread (net horizontal opening, n = 360), the relationship between footrope length and net width was strong (r² = 0.96) with an average ratio of horizontal opening to footrope length of 0.61.Together these pieces of information support the simple pooling of data sources as well as the assumptions that were used in developing estimates of fishing effort and catch rates.

Standardization fits and partial effects

Model fits, in addition to the inspection of residuals and other model diagnostics (e.g. quantile–quantile plots), suggested that the Tweedie error distribution was appropriate and that the models captured key elements of the system, with slightly higher deviance explained in the southern region (46.7% for the south and 34.1% for the north, Table S1). Patterns were present in the residuals and were worse for the northern region than the south (Fig. S2). Diagnostics also suggested some negative bias in residuals. Informal explorations where variables were added to the model in a stepwise fashion suggested that location (latitude and longitude), vessel permit, catch month, and bottom temperature were the most influential variables in each model. As expected, these variables and smooths were associated with the largest effective degrees of freedom (Table S2).

Smooth effects included in each model produced similar patterns across regions for several covariates (Fig. 4), and similarity among regions increased for most covariates when the areas with the most data were compared. For example, catch rates showed a cyclical relationship with month, where there was an increase in the winter, peak in the spring, followed by a decrease in the summer and fall. This cycle started earlier in the southern region as we might expect. The impact of the price of diesel fuel and the price of black sea bass were all relatively similar across regions as well. Generally, catch rates decreased with both the price of diesel and the price of sea bass in both regions. The negative relationship between catch and black sea bass price may be an effect of seasonally higher catches that potentially drive down the price. In both regions, catch rates were fairly flat across a range of vessels sizes, and apparent patterns may reflect the gear configuration and targeting behavior of different size vessels and expertise in harvesting black sea bass. Differences in vessel sizes within the sample may also play a role in shaping these relationships, as more data points came from smaller vessels in the north.

Figure 4.

GAM smooth effects from both regional CPUE standardizations. The smooths for longitude and latitude are shown first (a) and (b), and then the smooth for each variable included in the standardizations below (c). Plots were generated using the “smooth_estimates” and “draw” function in the package “gratia” (Simpson 2022). Plots of ecological and sociological covariates (c) include rug plots at the top of each panel to show the regions with the highest density of information. Warmer colors show areas of higher density in these rugs (orange = 50%, yellow = 80%, light blue = 95%, and dark blue = 99%).

In contrast, smooth effects for other covariates showed distinctly different patterns between regions. For example, in the southern region, catch rates increased with increasing bottom temperature, but in the northern region catch rates reached a maximum at around 15°C with lower catches at higher and lower temperatures (Fig. 4). The impact of depth was also distinct between the regions, with catches in the southern region generally being greatest at the shallowest (<50 m) and deepest (>300 m) records, whereas in the northern region catches were highest at moderate depths (around 100 m). The partial effect of rugosity was also reversed, with the northern area seeing higher catches at moderate rugosity values and the southern area having higher catches at the lowest and highest rugosity values. One important note in evaluating these is that discrepancies between relationships are greater in regions of the covariates with fewer data point (areas shaded in blue in the rug plots), and more similarities are found in the shape of smooths where data is more available (areas shaded in yellow and orange).

Fixed effects showed some similarities across regions as well (Fig. S3). Specifically, the largest effects in both regions were related to targeting, with the largest catch rates coming unsurprisingly from trips targeting black sea bass. There were smaller effects for individual ports or vessel permits. The effect of the data source (observer or Study Fleet) was small in both regions but reversed in direction, with catch rates from observer records being slightly higher than Study Fleet records in the northern region and slightly lower in the southern region. For some relationships, such as that between catch and rugosity in the south, the level of uncertainty in the GAM derived relationship was quite high; suggesting that interpretation of these relationships should be approached with caution. Additional collection of covariate data at the haul-level would improve the precision and interpretability of the relationships between covariates and black sea bass catch.

CPUE trends

Area-weighted nominal CPUEs produced different trends between regions (Fig. 5). In the northern region, the nominal CPUE followed a generally increasing trend before plateauing, whereas in the southern region, the nominal CPUE fluctuated through time, with a period of slightly increasing abundance near the end of the study period (2015–2023). The standardized abundance estimates from the GAM models produced somewhat similar trends to the nominal CPUEs (Fig. 7). In the northern region, the GAM CPUE increased through time with a slight plateau near the end of the series (2013–2021). In the southern region, the GAM CPUE started at a higher level in the early 2000s, then dipped and again fluctuated around a slightly lower level (2005–2010 and then 2017–2021, Fig. 5). The correspondence between time series was greater in the north than in the south, with the southern series diverging near the end of the time period. Mohn’s rho calculated for the standardized series for each region suggested that the index for the northern region showed limited retrospective bias, with a rho of 0.08. The index for the southern region showed moderate levels of retrospective bias with a rho value of 0.31.

Figure 5.

Catch rate (CPUE) trends for the northern (left) and southern (right) regions through time for the nominal (red) and standardized (blue) methods. The ribbon associated with the standardized CPUE series approximates a 95% confidence interval. Time series are scaled to their means for comparison.

Comparison to other time series

The strength of correlations between commercial catch rates with other metrics of abundance varied by region (Fig. 6). In the northern region, both the fishery-independent survey and recreational catch time series were fairly well correlated with the GAM standardized index with coefficient of determination (R²) values ranging from 0.596 to 0.670 (Fig. 7a). In the southern region, while the GAM CPUE series follows similar trends as the survey and recreational catch series, correlations between them were weaker with R² values ranging from −0.006 to 0.204 (Fig. 7b). Correlations between the nominal CPUE and the survey and recreational catch series were higher in the north, with correlation coefficients as high as 0.550.

Figure 6.

The time series indices produced by each method showing the general trend from each analysis compared to other indices. Note that the GAM and nominal CPUEs represent a catch weight (biomass) per unit area, while the other indices represent numbers per tow (Survey) or number per angler (Rec CPUE). Indices have been scaled for comparison.

Figure 7.

Pearson correlations between the GAM standardization and other annual indices showing the general trend in each data set shown in Fig. 7. Generally, there was a greater correspondence among indices in the northern region (a) and less correspondence among all indices in the southern region (b).

Model validation

Despite using a proportion of the total data for validation, index values derived from reduced volumes of data produced similar patterns across replicates and to those seen in the full data set (Fig. S4). A limited set of iterations produced index trends that were more distinct, but these were more common in trials where a less complex model was used to generate the index. Across the many replicates, data sets and models fit, we observed a pattern of increasing explained deviance with increasing model complexity for both regions (Fig. S5a, Table S3). These increases plateau after and an initial steep increase, suggesting that the addition of variables has a more limited impact on model performance at a certain point. Increasing model complexity also led to lower estimates of relative error in index values (Fig. S5b). This pattern was more apparent in the trials for the northern region.

Discussion

Fishery-dependent data, like other preferentially sampled information (e.g. citizen science) is increasingly being applied to develop scientific products (Bonney 2021, Alglave et al. 2022, Braun et al. 2023). This study provides a novel example of how long-running and high-resolution fishery-dependent catch data can be used to understand the abundance of species from the northeast US continental shelf ecosystem. We highlight the potential utility of this information for exploring potential patterns of a stock’s relative abundance and the value in combining data sources when possible. This type of work is likely a valuable addition to assessments in that it should help either bolster support for trends seen in more traditional surveys or lead to constructive discussions of the reasons for discrepancies between surveys and the catch-rates of harvesters. More broadly, this work exemplifies how high-resolution fishery-dependent data sets can be used to develop indices of abundance for stock assessments, paralleling similar work in other regions (Gerritsen and Lordan 2011, Cheng et al. 2023).

The process of generating a data set for this analysis highlighted the benefits of thoughtfully combining available sources of fishery-dependent data. For example, while estimates of net wing spread were only available in the NEFSC Observer data set, these data enabled a more precise estimation of fishing effort across the entire data set and corroborated other information from the literature. Similarly, empirical bottom temperature measurements collected with catch information as part of the Study Fleet program allowed us to assess if modeled bottom temperatures were acceptable for applying to the larger data set. Moreover, subtly different distributions of sampling effort in both space and time makes these two sources of fishery dependent data complementary. Based on this experience, we suggest that combining compatible sources of high resolution fishery dependent data is a beneficial step for CPUE analyses, as it maximizes the spatial and temporal extent of the analysis, addresses concerns related to representativeness of the larger fishing fleet, and provides opportunities to integrate additional covariates. It should be noted, however, that combining fishery dependent data sets can be an onerous and time consuming task.

When assessing trends through time or changes in distribution, researchers often worry that a subset of available effort captured in high-resolution sampling programs may not be representative of a full fishery or stock. Certainly, this is something each analysis must consider, but for model-based standardization, we highlight a number of benefits of using this type of information. First, the high-resolution nature of the catch and effort data allows for coupling with high-resolution oceanographic and ecological variables. These included modeled covariates that are likely to be ecologically important, such as the bottom temperature and bottom rugosity. These variables typically capture information about processes occurring on a fine scale and are difficult to effectively pair with traditional, coarser scale fishery-dependent data. Additionally, having information on both kept and discarded catch (as opposed to only landings of kept catch) helps to build a more complete picture of a species and the associated fisheries dynamics and allows for the inclusion of records that might have otherwise been omitted. This work demonstrates that analyses using fishery-dependent data (like citizen science data), especially when coupled with qualitative expertise of the fishing community, increases the data’s value and identifies limitations that help researchers understand best practices for analysis and interpretation. Ocean ecosystems are experiencing rapid change at space and time scales that are impossible to monitor using traditional scientific methods (Hyder et al. 2015, Saba et al. 2023). Thus, maximizing the utility of fishery-dependent data is essential to understanding changes in marine ecosystems and developing effective management practices.

The northeast US continental shelf ecosystem is relatively data rich, having long time-series of both fishery-dependent and fishery-independent information (Cadrin et al. 2020, Saba et al. 2023). These multiple sources of information allow us to test the utility of fishery-dependent data for tracking trends of relative abundance. Our aim in generating these series for black sea bass was to test the suitability of the fishery data sets for assessing trends in abundance in commercially important species. To that end, we constructed a CPUE standardization following published best practices (Maunder and Punt 2004, Hoyle et al. 2024). The resulting indices do appear to track abundance of black sea bass, especially in the northern region, as evidenced by high correspondence to fishery-independent series, low retrospective bias, and large sample sizes. Moreover, model validation suggested that these models increased in skill as they increased on complexity (Fig. S5). In the southern region, the trend resulting from the GAM was less closely associated with other indices. This discrepancy was reduced in some of the more spatially balanced subsets included in the model validation (Fig. S4), and therefore it is possible that clustering of effort could be a driver of these divergent trends. Despite some differences, these time series are still quite useful as a qualitative tool to elucidate stock dynamics; however, in order to quantitatively incorporate the CPUE indices into the black sea bass stock assessment would require additional work to translate catch rates to catch-at-age estimates. Further, because commercial fishery catches tend to be older and larger fish than those caught in the fishery independent surveys, differences in selectivity are important to consider.

When analyzing CPUE series in an assessment context, it is important to consider nonlinearities in the relationship between catch rates and abundance (sensu Kulka and Rose 1999). Specifically, changes in catchability and the hyperstability of catch rates have been implicated in the collapse of stocks such as Atlantic Cod in maritime Canada (Hutchings and Myers 1994), although the relative contribution of different scientific and managerial factors is debated (Shelton 2005). Simulations that explored the impacts of changing catchability to statistical catch at age models (Wilberg and Bence 2006), did note the potential value of fishery catch rate data provided that CPUEs are appropriately standardized and assessment models allow for changes in catchability through time. This research directly addressed these recommendations, and thus, we believe that the work presented here supports quantitative application of fishery-dependent data and CPUE indices in stock assessments.

The guidance and input from fishers, along with the existing scientific literature on the species, played a pivotal role in shaping and refining this analysis. Through structured discussions, fishers highlighted several variables that could be impactful on catch rates, such as market price and bottom rugosity, while scientific literature, including works by Miller et al. (2016) and Cullen and Guida (2021), provided additional insights on ecosystem drivers. When utilizing these data sets for developing CPUE indices or species distribution models, it is crucial for researchers to engage with the fishing community early on. This allows sufficient time to document the shared insights and identify potential applications for modeling, such as determining relevant covariates. Furthermore, once the products are developed, creating opportunities for constructive feedback from fishers can significantly contribute to their improvement (Steins et al. 2020, Mercer et al. 2023). Recognizing that fishers are the data collectors and possess valuable observations that may not be captured in the data sets, it is essential to foster an open exchange of ideas. This can be achieved through activities such as scheduling presentations to share the results of the study or circulating drafts of analyses and preliminary findings. Such activities facilitate the flow of information and maximize the opportunities for fishers (or other stakeholders) to contribute their experiences and perspectives. Additionally, the consideration and evaluation of indices of relative abundance from multiple sources should help enhance constructive dialogue, potentially generate constructive areas for future collaboration, and potentially enhance the perceived validity of an assessment.

The techniques employed in this study hold promise for application in other regions or with different data sets to construct catch rate time series. Where compatible data sets are available, a very similar approach could be pursued. In regions where data sets are more varied, more work may be needed to find commonalities that allow data sets to be incorporated into a single model. For example, conflicts in the resolution of reporting may require high-resolution data to be aggregated. Or, effort and gear metrics may need careful translation to make catch rates useful. As additional modeling tools such as VAST and sdmTMB are developed (Thorson 2019, Anderson et al. 2024), the integration of fishery data sets into stock assessments may become easier and reduce patterns in the residuals observed here. Exploration of these techniques for fishery-independent data sets shows great promise (e.g. Thompson 2023, Nephin et al. 2023), and may help improve model performance.

The growing interest in and availability of preferentially sampled data have expanded opportunities for their utilization (Fletcher et al. 2019, Pennino et al. 2019, Alglave et al. 2022). Worked examples such as the analysis presented here are particularly valuable when coming from regions with extensive fishery-independent information, as these examples can provide insights into how well these approaches capture the true dynamics of a population and make large volumes of existing data more useful. Once constructed, CPUE time series can serve as valuable resources in the assessment process whether they are leveraged for quantitative or qualitative uses (Cadrin et al. 2020, Steins et al. 2020, Cheng et al. 2023). For example, even when only evaluated quantitatively in the assessment process, similarities between the results of these analyses and the fishery-independent time series provide stakeholders with confidence that survey indices adequately represent the true abundance of the target species. In conclusion, the increasing availability of high-resolution preferentially sampled data presents new opportunities, and worked examples are invaluable to help suggest fruitful approaches for analysts. Taken together, our analyses suggest that if given careful consideration, CPUE time series can serve as powerful tools, offering insights into both population and fishery dynamics, while also serving as a foundation for more advanced modeling approaches and enhanced stakeholder engagement.

Acknowledgments

We express our sincere gratitude to the NEFSC Study Fleet participants and Northeast observers, whose dedicated efforts have made this work possible. Generating the data required for this study is a demanding and often underappreciated task. We would also like to extend our appreciation to the diligent staff members of the Study Fleet and Observer programs, whose meticulous editing, auditing, and data management work have been instrumental in facilitating this research. We are thankful for the valuable suggestions and contributions made by J. Deroba, J. Brust, G. Fay, and C. Demarest, whose insights significantly enhanced both the analysis and written portions of this manuscript. Finally, this manuscript was greatly improved by the practical suggestions of three thoughtful reviewers.

Author contributions

Conceptualization: A.W.J. and A.J.M. Investigation: all authors. Visualization: A.W.J. Writing—original draft: A.W.J., A.J.M. Writing—review and editing: all authors.

Conflict of interest

None declared.

Data availability

Data for this study come from a variety of sources. Some are available from public repositories (e.g. modeled bottom temperature products), while others are confidential in nature and must be requested by an authorized user (e.g. haul-level catch and effort information). For publicly available bottom temperature data, please see EU Copernicus Marine Service information at 10.48670/moi-00016. To request confidential data, inquiries can be sent to the Northeast Fisheries Observer Programs and the Cooperative Research Branch. Both of which are components of the United States National Marine Fisheries Service Northeast Fisheries Science Center (https://www.fisheries.noaa.gov/about/northeast-fisheries-science-center). Code used to conduct this study can be made available on request from the corresponding author.