StrathE2EPolar model
The StrathE2EPolar model builds on an existing temperate shelf-sea fisheries—food web model (StrathE2E2) developed as a package for the R statistical computing environment (www.r-project.org/about.html; https://CRAN.R-project.org/package=StrathE2E2; Heath et al. 2021). StrathE2EPolar is also available as an R-package (https://marineresourcemodelling.gitlab.io/sran/index.html)—version 2.0.0 was used in this study. The chemical and biological groups represented in the model are shown in Table 1. StrathE2E models are spatially resolved into an inshore and offshore zone, the former with a single water layer, the latter with two layers representing the euphotic and disphotic strata. The seabed in each zone is sub-divided into four sediment habitats (Fig. 1). Extensions of StrathE2E to create StrathE2EPolar were the inclusion of nutrient, ice algae and detritus dynamics in sea-ice, the effects of ice and snow on light penetration into the sea, and their effects on the feeding ecology, mortality and active migration rates of higher trophic levels. Documentation on the formulation of these extensions is also available at the package gitlab site. The model outputs data at daily intervals.
Table 1 Ecological guilds or classes of dead and living material included in the StrathE2EPolar and ECOSMO-Polar models. Terms marked * were added to the respective source models in order to create the polar versions. Detritus and bacteria were represented as a composite guild in both models ECOSMO-Polar model
The ECOSMO-Polar model builds on the existing ECOSMO-E2E model for the North and Baltic Seas (Daewel et al. 2019). The extensions of ECOSMO-E2E to create the ECOSMO-Polar was the implementation of a sympagic (sea-ice biogeochemistry) module, including ice algae, detritus and 4 nutrients groups (nitrate, ammonia, phosphate and silicate), and the sympagic interaction with the existing pelagic and benthic systems (Benkort et al. 2020; Table 1). In addition, state variables for chlorophyll were included to take account of the variable chlorophyll content in phytoplankton (Yumruktepe et al. unpubl.). ECOSMO-Polar was run in a 1-D vertical mode in this study with output at 30 min intervals (subsequently aggregated to daily averages), as in Benkort et al. (2020).
Input data for the models
StrathE2EPolar was driven by time varying physical and chemical data extracted from output of the 3-dimensional, quarter-degree latitude x longitude NEMO-MEDUSA earth system physical-biogeochemical model (Yool et al. 2015). NEMO-MEDUSA was run from 1980 to 2100 with atmospheric forcing assuming IPCC representative concentration pathway emissions scenario “RCP8.5” (Riahi et al. 2011). The driving data for StrathE2EPolar were assembled for the periods 2011–2019 (referred to here as the “2010s”; representing present-day conditions) and 2040–2049 (the “2040s”). The 2040s period was chosen to represent the future state as this is the onset of nearly year-round ice-free conditions in the Barents Sea, according to NEMO-MEDUSA (Fig. 2). Two types of data were extracted from NEMO-MEDUSA for input to StrathE2EPolar: (1) climatological (2010s and 2040s) annual cycles of monthly area or volume averaged values over each StrathE2EPolar water column zone and layer, and (2) climatological annual cycles of monthly averaged or integrated data at the external boundaries of the StrathE2EPolar domain. Data of the first type were water temperature, vertical diffusivity at the interface between vertical layers in the offshore zone, ice (and snow) extent, cover and thickness, and daily integrated incident irradiance. Data of the second type were daily integrated ocean and river water inflow volumes across the external boundaries, and the boundary concentrations of dissolved inorganic nitrogen (DIN), detritus and phytoplankton.
Other StrathE2EPolar driving data were significant wave height in the inshore zone (CERA-20C ‘Ocean Wave Synoptic Monthly Means’ product accessed through ECMWF); monthly averaged annual cycles of wet and dry atmospheric nutrient deposition rates assembled from the EMEP data centre (https://www.emep.int/mscw/mscw_moddata.html); riverine nutrient concentrations from Holmes et al. (2021), and suspended particulate matter (SPM) in the inshore zone and upper layer of the offshore zone from remote sensing data (Globcolour L3b; ftp://ftp.hermes.acri.fr/GLOB/merged/month/). Apart from wave height, we were unable to identify sources of future projections for these inputs, so we assumed that they will remain constant into the 2040s. Further details are provided in Appendix S3.
Atmospheric driving data for ECOSMO-Polar simulations were prescribed from the MERRA2 reanalysis (Gelaro et al. 2017), which is available with a 50 km horizontal resolution and hourly instantaneous output for the “2010s”. Atmospheric variables from HadGEM2-ES RCP8.5 (Jones et al. 2011; data accessed through www.isimip.org) were used for the “2040s”, consistent with driving conditions of NEMO-MEDUSA. The relevant forcing variables were air temperature, pressure and humidity, wind velocities and shortwave radiation.
Data to configure the fishing fleet model component of StrathE2EPolar were assembled for the period 2011–2019 from the Norwegian Directorate of Fisheries, EU STECF (https://stecf.jrc.ec.europa.eu/dd/fdi/spatial-land-map), Global Fishing Watch (Kroodsma et al. 2018), and regionally integrated landings by nation and species from the ICES/FAO data centre (FAO areas 27.1 and 27.2.b). Additional data on artisanal seal harvest, discards of fish by shrimp trawlers, catches by the recreational/tourism/subsistence fishers in Norway and Russia, and by-catches of seabirds, seals and cetaceans in coastal gillnet and longline fisheries, were assembled separately from a range of literature and data sources. Full documentation of the workflow to generate these input data to the fleet model is available separately (https://marineresourcemodelling.gitlab.io/sran/index.html), and a summary of the fishing gears represented in the model in Appendix S4.
Data for model optimization and validation
Observational data on the state of the Barents Sea ecosystem during the 2010s modelling period was assembled from literature sources. These data formed the target for a computational optimization of the StrathE2EPolar parameters by simulated annealing (Kirkpatrick et al. 1983; Heath et al. 2021). The data and their sources are listed in Appendix S5. The outcome of the optimization was a parameter set which produced the best fit of the model to the observations (Fig. S3).
Independent validation of models was carried out by comparison with satellite data on chlorophyll concentrations, and acoustic survey data on fish and zooplankton distributions. Ocean colour sensing data, calibrated as chlorophyll concentrations, were downloaded at 1 km resolution for 2011–2019 (https://resources.marine.copernicus.eu). The 25th, 50th, and 75th percentiles of monthly aggregated pixel chlorophyll concentrations were calculated for the inshore and offshore zones of StrathE2EPolar (Fig. 1).
Raw Simrad EK60 echosounder data (18, 38 and 120 kHz; calibrated as per Demer et al. 2015) collected annually in August and September between 2011 and 2016 as part of the Barents Sea Ecosystem Survey (Fig. S4; Eriksen et al. 2018) were obtained from the Norwegian Marine Data Centre. Details of the processing to generate Nautical-Area Scattering Coefficients (NASC, m2 nmi−2: average received echo energy over a given depth range scaled up to a square nautical mile) are provided as Appendix S6. Mean NASC values for both fish and macro-zooplankton were computed for both the inshore and offshore zones of the StrathE2EPolar model domain. Bootstrap sampling was used to calculate confidence intervals. NASC values were also binned into a 0.5 by 0.5 degree grid and averaged to map the spatial distribution of fish and macro-zooplankton.
Strategy for using the models
In all cases the models were run to a steady state (until they produced repeating annual cycles of outputs with repeating annual cycles of input driving data; 60–100 years for StrathE2EPolar depending on scenario conditions). Under these conditions the results from the models were completely independent of their initial conditions. The models were used to conduct 3 types of experiments as described below.
Experiment 1: Consistency between models and independent observational data. StrathE2EPolar has a relatively parsimonious representation of physics and biogeochemistry. We assessed its effectiveness at simulating the base of the food web by comparison with ECOSMO-Polar which has a more elaborate representation of physics, microbes and autotrophs in terms of vertical resolution, nutrient and guild diversity.
ECOSMO-Polar was run at 20 locations covering the inshore and offshore zones of the StrathE2EPolar model with 2010s driving data. Within each zone, between-site variations in phytoplankton chlorophyll concentrations were averaged over the upper 60 m and summarised by the mean, median and quantiles of the results across all sites within each zone. These distributional properties were then compared with credible intervals of equivalent outputs for each zone from StrathE2EPolar generated by a likelihood-weighted Monte Carlo analysis of parameter uncertainty (full details of the methodology available from the R-package gitlab site), and with mean, median and quantiles of the independent observational data for the 2010s period which were not used in the optimization processes for either model (annual cycles of satellite remote sensing data on chlorophyll).
In addition, August and September data on inshore and offshore macro-zooplankton and fish biomass from StrathE2EPolar (these variables were not available from ECOSMO-Polar) were compared with August/September acoustic survey data. Macro-zooplankton biomass from StrathE2EPolar was the sum of the carnivorous zooplankton and fish larvae guilds; fish biomass was the sum of planktivorous, migratory and demersal fish guilds.
Experiment 2: Projection of future ecosystem state in the 2040s. Both StrathE2EPolar and the 20 ECOSMO-Polar site models were run to a steady state with 2040s external driving data from NEMO-MEDUSA. For StrathE2EPolar, inputs to the 2040’s fishing fleet model were assumed to be identical to the 2010s. Annual mean masses of each of the state variables in each model were then expressed as a percentage change relative to the corresponding properties of the 2010s model. ECOSMO-Polar state variable outputs were aggregated across functional guilds so as to correspond with the coarser guild resolution of StrathE2EPolar.
Experiment 3: Ecosystem sensitivity to fishing. For StrathE2EPolar only (ECOSMO-Polar did not include any representation of fish or fishing), the 2010s and 2040s models were run to a steady state for each member of a sequence of increasing values of harvest rate on planktivorous and demersal fish. Here, harvest rate was the daily fish mortality rate due to fishing—related to the proportion of fish biomass captured per day. For both periods, the rates were expressed as multiples of 2010s mean values as determined from the observational data. Proportions of total effort attributable to each gear type and their spatial distributions (inshore-offshore), discard rates and seabed abrasion, were held constant across all runs. For each decadal period, two sets of runs were carried out: (a) increments of fishing mortality on planktivorous fish with demersal fishing mortality held constant at the 2010s value, and (b) increments of fishing mortality on demersal fish with planktivorous fishing mortality held constant at the 2010s value.
Simulated catches in each of the fishing runs mapped out the standard dome-shaped “yield curves'' for planktivorous and demersal fish (fishing mortality rate vs catch) and the corresponding fishing mortality rate vs biomass curves, which form the basis for setting fisheries management plan reference points, i.e. the biomass and fishing mortality rate at maximum sustainable yield (BMSY and FMSY). The present status of a fishery is often expressed by the ratios Bcurrent/BMSY and Fcurrent/FMSY. The credibility of the 2010s fishing sensitivity results was assessed by comparing these indicators derived from the simulations with the guild-level value aggregated from individual species stock assessments produced by the ICES Arctic Fisheries Working Group (ICES 2020).
In addition to yield curve outputs, the results were used to assess the sensitivity of annual average fish, seabird, pinniped, cetacean and maritime mammal (polar bear) biomasses to each fishing scenario.