Projected impacts of warming seas on commercially fished species at a biogeographic boundary of the European continental shelf

1. Projecting the future effects of climate change on marine fished populations can help prepare the fishing industry and management systems for resulting ecologi -cal, social and economic changes. Generating projections using multiple climate scenarios can provide valuable insights for fisheries stakeholders regarding uncer tainty arising from future climate data. 2. Using a range of climate projections based on the Intergovernmental Panel on Climate Change A1B, RCP4.5 and RCP8.5 climate scenarios, we modelled abundance of eight commercially important bottom dwelling fish species across the Celtic Sea, English Channel and southern North Sea through the 21st century. This region spans a faunal boundary between cooler northern waters and warmer southern waters, where mean sea surface temperatures are projected to rise by 2 to 4°C by 2098. 3. For each species, Generalized Additive Models were trained on spatially explicit abundance data from six surveys between 2001 and 2010. Annual and seasonal temperatures were key drivers of species abundance patterns. Models were used to project species abundance for each decade through to 2090. 4. Projections suggest important future changes in the availability and catchability of fish species, with projected increases in abundance of red mullet


| INTRODUC TI ON
Climate change has affected the abundance, dynamics and distribution of marine fish populations and their associated fisheries, resulting in substantive social and economic consequences (Barange et al., 2018;Brander, 2007;Cheung, Dunne, Sarmiento, & Pauly, 2011;Perry, Low, Ellis, & Reynolds, 2005). Projections of climate change impacts on marine systems provide important insights into future species responses to increased temperatures, as well as the future availability, productivity and catchability of stocks for dependent fisheries (Barange et al., 2014;Blanchard et al., 2012;Cheung et al., 2011).
Spatially explicit projections capturing the effects of future climate change on the abundance and distribution of fish populations underpin assessments of fisheries consequences. For the Northeast Atlantic region, Species Distribution Models (SDMs) have generally projected further poleward movements and/or deeper distributions of those species with preference for cooler waters (Cheung et al., 2011;Jones et al., 2013). However, a study using Generalized Additive Models (GAMs) suggested that many bottom dwelling (demersal) species in the North Sea could not move further polewards because they were constrained by availability of habitat at suitable depth (Rutterford et al., 2015). Differences between model projections highlight the need to examine the uncertainty associated with future projections to guide future model development and decision-making.
Identifying and disclosing the sources and extent of uncertainty associated with modelled projections provides insight into their strengths and weaknesses and can guide appropriate responses to, and treatment of, projections during decision-making Freer, Partridge, Tarling, Collins, & Genner, 2018;Payne et al., 2015). Research comparing model approaches and performances or using different climate scenarios can help to quantify levels of uncertainty arising from these different potential sources Freer et al., 2018).
Recent attempts to consider uncertainty within the marine literature include exploring future long-term responses of individual species (Gårdmark et al., 2013), habitat suitability , ocean marine animal biomass (Lotze et al., 2019) and fisheries catch potentials and revenues Lam, Cheung, Reygondeau, & Sumaila, 2016). While progress has been made in this area, there remains a significant lack of ecological research exploring the effects of uncertainty arising within and across climate scenarios, which could have important consequences for interpretation of resulting modelled projections (Freer et al., 2018;Payne et al., 2015).
The Celtic Sea, English Channel and southern North Sea sections of the European continental shelf comprise a faunal boundary between cooler northern waters and warmer southern waters (Hinz, Capasso, Lilley, Frost, & Jenkins, 2011). Sea temperatures in this region have warmed 0.17-0.45°C per decade between 1985 and 2014 (Hughes et al., 2017), and climate projections suggest further sea warming of 2-4°C by 2098 (Tinker, Lowe, Pardarns, Holt, & Barciela, 2016). The region is fished by many countries including the UK, Ireland, France and the Netherlands, with major fishing ports based along the coastline including Newlyn, Brixham, IJmuiden and Le Havre. Given the significance of this region for fisheries (STECF, 2017), and extent of projected climate change within the region (Tinker et al., 2016), we aimed to project responses of commercially important species while incorporating climate uncertainty. Specifically, we trained GAMs based upon multiple downscaled climate projections for the north-west European shelf seas alongside extensive fisheries survey data, and used these models with climate projections to estimate changes in the abundance of eight demersal fish species through the 21st century.

| Study area
The region of study included all marine areas from 47-53°N and 12°W-3°E, represented in our analyses as 72 1°× 1° sea grid cells. and adaptive approaches that reduce climate impacts on species while also supporting industry adaptation are required.

K E Y W O R D S
Celtic Sea, climate change, English Channel, fish, fisheries, North Sea, regional projections, uncertainty Collectively, the region spans the English Channel, Celtic Sea, the Bristol Channel and parts of the southern North Sea (Figure 1).  Table S1). Whole-sediment median grain size for each 1 × 1° grid cell was generated using data from Wilson, Spiers, Sabatino, and Heath (2018; Figure 1d).

| Temperature and salinity
In all, 13 climate projections for the region were obtained from the UK Met Office Hadley Centre. In total, 11 projections were generated from a project that dynamically downscaled, using the see Tinker et al., 2016). The Tinker et al. (2016) projections are the only set of north-west European shelf sea projections (to date) that systematically and extensively consider this aspect of climate uncertainty (Tinker & Howes, 2020). These projections' annual and ensemble mean sea surface temperature and salinity absolute biases were <0.2°C and 0.2 psu when averaged over the shelf compared to observed data over the same time period and area: (1986-2006: Roberts-Jones, Fiedler, & Martin, 20121960-2001: Ingleby & Huddleston, 2007. Using this SRES ensemble enabled exploration of aspects of uncertainty within a single climate scenario through the use of different ensemble-members and examining their relative effects on projected species responses. A further two climate projections were obtained to compare uncertainty on projected species responses across climate scenarios. These climate projections represented Relative Concentration Pathway (RCP) 4.5 and RCP 8.5 scenarios, which were developed from a regional shelf sea model (AMM7) that dynamically downscaled long-term simulations of two CMIP5 Global Climate Models (HadGEM2-ES and MPI-ESM-LR) for the north-west European shelf (Hermans et al., 2020).

| Fish abundance data
Eight demersal fish species with a range of biogeographic affinities were selected, based on prior assessment of landing statistics and their social and economic importance to fisheries within the region (  Figure 1a).
Surveys only partially overlap in space, differ in temporal coverage and use different sampling methods and gears (Table S2).  Tables S3-S5).   (Rutterford et al., 2015). While some variables were collinear (Table S7)

| GAM performance
Least-square mean analysis identified that for most model statistics Model A, the model with all variables included, provided the optimal combination of variables to model patterns of species abundance ( Figure 3; Table S9). This was the most parsimonious model owing to low AICc and GCV scores ( Figure 3) and a high mean Pearson correlation score of 0.92 between observed and projected abundances across predictive ability compared to other models ( Figure 3; Table S9).

| Projected abundance
Region-wide declines in abundance were projected towards the 2080s for anglerfish, Atlantic cod and megrim ( Figure 4; Table 2).  Red mullet was projected to show a region-wide increase in abundance from the 2000s to 2040s, particularly in the English Channel. Dover sole and John dory were also projected to increase in most parts of the study area. Anglerfish (with the exception of RCP 4.5) and megrim were projected to decline across their range, with some localized increases for megrim to the west. Lemon sole was projected to decline in the northern extent of the region but increase towards the south. European plaice showed increases to the east of the region but decreases to the west.

| D ISCUSS I ON
For the first time for this region, we provide projections of species abundance in response to future warming that are based on multiple climate projections and use survey information to identify variables influencing these responses. On average, projections towards 2090 suggest increased abundance for warm-water-associated species (e.g. red mullet and John dory) and declines for those associated with cooler waters (e.g. Atlantic cod and anglerfish). Projections presented here can help to inform the fishing industry and management systems about the potential social, economic and ecological risks and opportunities resulting from these changes.
Our projections suggest that fisheries within the region could benefit from projected increases in Lusitanian species such as Dover sole, John dory and red mullet. Projected increases and expansions across the region for John dory and red mullet continue a trend that has been documented since the mid-1990s within the North Sea and Celtic Sea (Beare et al., 2004;Simpson et al., 2011;ter Hofstede, Hiddink, & Rijnsdorp, 2010). Such expansions may provide new fishing opportunities, as seen with other species in the region such as boarfish Capros aper off the south of Ireland .
However, the extent to which these opportunities can be realized depends on factors including fishers' capacity to modify fishing practices and the development of consumer demand and emergence of export markets (Jennings et al., 2016;Perry, Barange, & Ommer, 2010;Pinsky & Mantua, 2014). John dory and red mullet are currently subject to no fishing regulations or quota. A lack of TA B L E 2 Mean CPUE index of abundance (back-transformed CPUE; catch per hour) for each time period and climate projection. Bold numbers represent mean abundance across all ensemble-members (SRES only) and grid cells. Standard deviation is in brackets Species SRES RCP 4.5 RCP 8.5

Mean CPUE 2080s
Anglerfish 2.08 (2.02) Projected future declines in the boreal species anglerfish, Atlantic cod and megrim appear to be highly likely given that the majority of climate projections indicate declines. The projected trends add to wider literature demonstrating poleward (northward) shifts and/or deepening of these species in response to warming (Dulvy et al., 2008;Perry et al., 2005;van Hal et al., 2016 is therefore crucial to allow fishers to adapt to future changes (Holsman et al., 2019;Pinsky & Mantua, 2014).
Alterations in future fishing effort or management decisions were not incorporated in our projections, but such changes are likely (Haynie & Pfeiffer, 2012;Melnychuk, Banobi, & Hilborn, 2014). Projected abundance decreases would likely be linked to requirements to reduce fishing mortality and fishing effort to meet lower reference points for spawning stock biomass (Brander, 2007;Holsman et al., 2019). If abundances increased, the converse would occur. Increases in stock distribution owing to distributionabundance relationships (Fisher & Frank, 2004) may also lead to fishing opportunities in new areas and effort spreading more widely, depending on management restrictions. Future research could explore the effects of different distributions of fishing activity and mortality for particular vessels and gears, resulting from prescribed or climate responsive management regimes.
All parameters included in the full model used have been demonstrated in wider research to affect species abundance and distributions, such as the role of depth in providing thermal relief for species or associations of demersal fish with benthic habitat type (Dulvy et al., 2008;Johnson, Jenkins, Hiddink, & Hinz, 2013). Crucially however, iterative removal of individual parameters across the model set suggested that the mean effects of temperature are important in driving species responses. Varying depth and climatic conditions have already shaped marine species assemblages within the region (Hinz et al., 2011;ter Hofstede et al., 2010). Given the importance of temperature affecting species responses, incorporating temperature driven effects within projections from stock assessments is crucial for anticipating future climate effects on stock dynamics and informing management decisions to help consider broader ecosystem effects (Sguotti et al., 2019;Skern-Mauritzen et al., 2015). Yet, for many stocks within this region, incorporating such environmental variability within assessments is still lacking.
Assessing model performance and understanding the uncertainty associated with projections is important from both scientific and fisheries management perspectives Freer et al., 2018). Using GAMs provided a data-driven, statistical approach, but differences in the design, coverage and duration of fisheries surveys used limited training data duration and precluded in-depth testing of GAMs using iteratively partitioned time series.
However, we have relatively high confidence in our approach for making projections through to 2040, and potentially beyond, as model statistics and comparisons for the tested time period indicated good model fits and there were relatively consistent patterns among climate projections. The performance of the approach has been systematically tested in the North Sea where annual surveys have been conducted in a relatively standardized way since the early 1980s (Rutterford et al., 2015). This study showed that projections using GAMs trained on data from the early part of the time series provided relatively reliable predictions of distribution and abundance for eight of 10 demersal species over a period of 30 years. Additional uncertainty about mid-to long-term projections and their consequences may result from factors we do not address directly such as changes in interspecific interactions (e.g. predator-prey dynamics) or fishing activity.
Using multiple climate projections provides greater transparency regarding the confidence in resulting modelled projections (Freer et al., 2018). An SRES ensemble allowed exploration of the consequences on projected species responses of climate-model parameter uncertainty within a single climate scenario (Tinker et al., 2015(Tinker et al., , 2016. Resulting uncertainties, captured by the spread of ensemble projections, implied that we should have high confidence in the direction of changes for most species, but the magnitude of change was more uncertain. Agreement among abundance projections was especially strong for cold adapted species with narrow thermal ranges, such as anglerfish and megrim, indicating high confidence that the fishing industry will have to adapt to declining catching opportunities. For other species, there was greater variability among projections that also increased towards 2090, which may be due to these species having wider thermal ranges or tolerances. Using different climate scenarios did not result in substantially different species responses, with RCP-based projections often within the range of those produced from the SRES ensemble. There has been little change in the median temperature (and its uncertainty range) between CMIP3 (generated SRES projections) and CMIP5 (generated RCP projections) global climate projections (e.g. Kumar, Kodra, & Ganguly, 2014), and Tinker et al. (2016) showed that the median of their SRES ensemble is likely to be consistent with an ensemble downscaled from RCP8.5.
Consequently, downscaled shelf sea climate projections are relatively similar, helping explain why limited differences between SRES-and RCP-based species projections were found. Given the uncertainty in future emission trajectories, projecting species responses using multiple climate projections provides the opportunity to examine the spread of all possible future outcomes, which are critical for allowing fisheries stakeholders to make climate-informed decisions.
In summary, our analyses suggest that climate change will continue to modify the abundance and distribution of commercially important fishes in the Celtic Sea, English Channel and southern North Sea. For species likely constrained by the coolest and warmest conditions, the projected directions of change in abundance are consistent among climate projections. Results suggest implications not only for the wider ecosystem (e.g. predator-prey dynamics or community composition) but also that the fishing industry and management systems will likely have to adjust their operations to address changes in availability, catchability and composition of catches. For declining species, fisheries managers may need to consider options that can reduce the vulnerability of stocks to warming temperatures, such as reducing fishing mortality rates or imposing stricter catch limits. For species not currently regulated, as a first step, species may need to be closely monitored for increasing fishing pressure, with future regulations or measures such as quotas potentially necessary. Fishers 'on-the-ground' experiences should be incorporated with scientific information to inform future management decisions to enable sustainable exploitation while supporting fishers' adaptation to changes in species' relative abundance.

ACK N OWLED G EM ENTS
We thank Simon Jennings for his invaluable advice and comments throughout this work, and for assistance sourcing survey data. This

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are available from the Dryad Digital Repository https://doi. org/10.5061/dryad.8cz8w 9gmz .