Investigating the potential impacts of ocean warming

0 downloads 0 Views 3MB Size Report
Clarke, K.R., Warwick R.M., 2001. PRIMER ... Giron-Nava, A., James, C.C., Johnson, A.F., Dannecker, D., Kolody, B., Lee, A., Nagarkar, M., Pao, G.m., Ye,. 569.

NorBar EwE model: manuscript

Bentley et al., 2017

1 Investigating the potential impacts of ocean warming on the Norwegian and Barents Seas 2 ecosystem using a time-dynamic food-web model 3 Jacob W. Bentley a*, Natalia Serpetti a, Johanna Jacomina Heymans a 4

a

Scottish Association for Marine Science, Scottish Marine Institute, Oban, PA37 1QA, UK

5 ABSTRACT 6 Climate change in the Norwegian and Barents (NorBar) Seas is expected to generate major alterations to the 7 marine food-web and its associated uses. However, our current capacity to quantify the potential ecological impact 8 of physical change is hindered by a lack of fundamental knowledge regarding the forces and trophic interactions 9 which have driven historic ecosystem dynamics. Here we used a historic (1950) food web model (Ecopath with 10 Ecosim, EwE) of the NorBar Seas fitted to time series between 1950 and 2014 to simulate ecosystem response to 11 changes in ocean temperature over the next 85 years to 2100 under a range of temperature scenarios including a 12 large scale climate variability indices (Atlantic Multidecadal Oscillation, AMO). Fishing, top-down/bottom-up 13 trophic interactions, a primary production anomaly and annual ocean temperature were all found to be important 14 drivers of modelled ecosystem dynamics in the NorBar Seas from 1950-2014. Under projected temperature 15 scenarios, the biomass of pelagic species, such as mackerel and blue whiting, increased with rising ocean 16 temperature, whereas the biomass of boreal species, such as redfish, prawns and capelin, decreased. Whilst within 17 favourable temperature conditions, cod biomass is predicted to decrease under the warmest scenarios due to the 18 reduced availability of preferred prey and the increased pressure of pelagic predation upon juvenile cod. The 19 model produced by this study provides a useful baseline approximation of the 1950-2014 NorBar ecosystem, from 20 which future research can propagate, and offers valuable insight into the systems potential response to changing 21 ocean temperature. Such quantitative advancements are fundamental to achieve sustainable development in 22 rapidly changing marine ecosystems. 23 Keywords: Barents Sea, Norwegian Sea, Ocean warming, Ecopath with Ecosim 24 *Corresponding Author. Tel.: +44 7930082185 25 E-mail addresses: [email protected] 26 [email protected] (J.J. Heymans)

(J.W.

Bentley),

[email protected]

(N.

Serpetti),

27 1. Introduction 28 The Arctic has been labelled the most rapidly changing environment on the planet, warming at two to 29 three times the global rate (IPCC, 2014; Polyakov et al., 2010). Changes in sea ice, water temperature, 30 light penetration, nutrient cycling, pollution and ocean acidification act collectively upon the Arctic 31 ecosystem (Dalpadado et al., 2012), generating major, yet unknown, changes to the structure, stability 32 and efficiency of Arctic food-webs (Kedra et al., 2015). Given the rate of change, the severity of this 33 knowledge gap is particularly worrying as it threatens our ability to adapt marine management and 34 policy to achieve sustainable ecological and socio-economic development in the Arctic.

1

NorBar EwE model: manuscript

Bentley et al., 2017

35 The Barents Sea and the Norwegian Sea represent two ecoregions subject to physical change within the 36 Arctic. Rather than through direct warming, it is the increased northern penetration and residence time 37 of warm, saline Atlantic waters which drives warming in the Norwegian and Barents (NorBar) Seas 38 (Polyakov et al., 2005; Wassmann et al., 2011). As well as being hydrographically linked, the multitude 39 of ontogenetic and seasonal migrations undertaken by biota between the Norwegian and Barents Seas 40 (Yaragina and Dolgov, 2009) make them hard to disentangle from a fisheries management (ICES, 2015) 41 and food-web perspective (Dommasnes et al., 2001), with the direct and indirect consequences of 42 overfishing historically transcending ecoregion boundaries (Hamre, 1994). 43 Arctic species in the NorBar Seas, especially those which are ice-dependent, are predicted to be the 44 most adversely impacted by rising temperatures (Kedra et al., 2015). Boreal species, such as Atlantic 45 cod (Gadus morhua), have started to migrate north with rising temperatures (Drinkwater, 2009, 2011), 46 whilst the distributions of pelagic species, such as blue whiting (Micromesistius poutassou) (Hátún et 47 al., 2009) and mackerel (Scomber scombrus) (Astthorsson et al., 2012), are expanding north and east, 48 benefiting from warmer waters and the increased primary production associated with reduced seasonal 49 ice cover (Loeng et al., 2005). The impact of these changes on the NorBar food-web are difficult to 50 quantify with our current knowledge. To credibly forecast the potential impact of ocean warming on 51 the NorBar food-web our focus must therefore be to increase our relatively limited quantitative 52 understanding of the ecosystem. 53 The development of new and existing food-web models provides a means to evaluate the ecosystem54 level consequences of ocean warming. A popular tool for exploring food-web dynamics is Ecopath with 55 Ecosim (EwE). Within Ecopath the user constructs a mass-balanced snapshot of their ecosystem 56 (Christensen et al., 1992), which is then used as a baseline to forward-project time-dynamic simulations 57 in Ecosim (Christensen et al., 2014). To date, few Ecosim models employ the use of time series data 58 for model calibration (Heymans et al., 2014) and even fewer include ocean temperature as an 59 environmental driving force. Instead, existing EwE models of the NorBar Seas have focused on the 60 impacts of fishing (Dommasnes et al., 2001) and the interactions between fisheries and marine 61 mammals (Blanchard et al., 2002) using mass-balanced, uncalibrated ecosystem snapshots. However, 62 recent studies have shown that, incorporating temperature as a driver in a food-web model can increase 63 that model’s capacity to reproduce observed trends and project the potential impact of ocean warming 64 (Serpetti et al., in review). 65 The aim of this study is to therefore advance the fundamental and quantified understanding needed to 66 generate projections of the impacts of future change in the NorBar Seas. We provide a quantitative 67 insight into the interactions and drivers which governed past ecosystem dynamics using an updated, 68 time series-calibrated version of the Dommasnes et al., (2001) NorBar EwE model. Through the 69 incorporation of functional responses, we explore the historic role of temperature and how, when 2

NorBar EwE model: manuscript

Bentley et al., 2017

70 coupled with trophic dynamics, the food-web may behave under a series of projected temperature 71 scenarios. 72 2. Materials and methods 73

2.1. Study area

74 The study area covers International Council for the Exploration of the Sea (ICES) areas I (Barents Sea) 75 and II (Norwegian Sea and a small portion of the Barents Sea) north to approximately 81°N, as 76 previously defined by Dommasnes et al., (2001) and encapsulates a surface area of approximately 77 3,116,000 km2 (ICES area I ≈ 1,006,100 km2; ICES area II ≈ 2,109,900 km2) (Fig. 1). 78

2.2. Construction of a 1950 mass balance Ecopath model

79 The NorBar model was constructed using the EwE v.6.5.1 software package. EwE, established in 1984 80 by Polovina (1984) and subsequently updated (Christensen and Pauly, 1992; Walters et al., 1997), was 81 primarily developed to overcome the fisheries management issues associated with modelling species in 82 isolation (Pauly et al., 2000). EwE has three main components: Ecopath, Ecosim and Ecospace 83 (http://www.ecopath.org/) (Christensen and Walters, 2004). Ecopath was used to create a static mass84 balanced snapshot of the trophically linked resources in the NorBar food-web and their interactions in 85 1950 (Christensen, 2013) (Supplemental information A). 86 The trophic structure of the model represents the principle organisms in the NorBar food-web: an 87 updated version of the EwE trophic structure proposed by Dommasnes et al., (2001) and includes 88 important commercial species such as cod, herring (Clupea harengus), haddock (Melanogrammus 89 aeglefinus) and capelin (Mallotus villosus) as well as primary producers such as phytoplankton and top 90 predators such as mammals and seabirds. In addition to these commercially important species, other 91 abundant fish species include Polar cod (Boreogadus saida), mesoplangic fish, mackerel and blue 92 whiting. More information regarding the ecology and importance of functional groups can be found in 93 the model descriptions of Dommasnes et al., (2001) and Blanchard et al., (2002). Where the Dommasnes 94 model had 30 functional groups, the model constructed in this study has 34 functional groups. 95 Dependent on the ecological or commercial importance of a species, their inclusion in the model is 96 either as a single species group or aggregated into groups with other species that occupy a similar 97 ecological niche (Supplemental Table B1). This models structure predominantly diverts from the 98 Dommasnes structure in the lower trophic levels, where groups have been re-adjusted and a microbial 99 loop has been added. These changes address the assessments of recent studies which stress the high 100 importance of lower trophic organisms, warranting a more detailed representation of these groups in 101 the model design (Dalpadado et al., 2012; Kvile et al., 2016; Slagstad et al., 2011). 102 As species often undergo ontogenetic shifts in diet and habitat during their lifetime being able to model 103 different stages of species enables a better understanding of the changes in the ecosystem. 3

NorBar EwE model: manuscript

Bentley et al., 2017

104 Unfortunately, not enough information is available for all species however data for commercially 105 important species tend to be of greater quality and quantity, affording us the opportunity to create data 106 based stanzas for these groups only. Cod, herring, haddock and capelin, were split into two life history 107 stages (stanzas), juveniles (immature) and adults (mature). Splitting groups enables us to acknowledge 108 the ontogenetic differences between adults and juveniles within the model structure. Adults are more 109 important to the fisheries and are thus subject to higher catches and fishing mortalities whilst juveniles 110 tend to experience higher natural mortality, have different consumption rates, diets and predation 111 pressures. Splitting these groups to capture ontogenetic differences brings the model closer to reality by 112 permitting the dynamics of juveniles and adults to directly influence each other (Christensen et al., 113 2004) (Supplemental Table B2). 114

2.2.1. Ecopath parameterisation

115 The Ecopath model was parameterised using data from fisheries surveys, stock assessments, literature, 116 FishBase (Froese et al., 2017) and existing ecosystem models. The development of this model relied 117 heavily upon data from ICES which records collective catches and assessments dating back to 1950 118 (http://ices.dk/marine-data). The main parameters required by Ecopath to construct a mass-balanced 119 ecosystem model are biomass (B), productivity/biomass (P/B), consumption/biomass (Q/B) and 120 ecotrophic efficiency (EE). Biomass estimates for single species groups were preferentially taken from 121 recent stock assessments. For groups with multiple species, their biomasses were summed and then 122 divided by the model area. Fishing fleets were incorporated into the model to account for the fisheries 123 catch and effort in the NorBar Seas. Fleets were generalised into the four predominantly used gear 124 types: pelagic gear, demersal trawl, shrimp trawl and other gear (whalers and sealers) (Dommasnes et 125 al., 2001). 126 Diets for trophic groups in 1950 were constructed based on a qualitative understanding of who eats 127 whom in the NorBar Seas (Bogstad et al., 2000; Planque et al., 2014), taking quantitative reference 128 from previous NorBar models (Blanchard et al., 2002; Dommasnes et al., 2001). There is no bulk 129 stomach data available for the NorBar ecosystem in 1950, therefore the diet matrix was compiled using 130 reference data from existing NorBar models and adapted based on the ecological assumption that, whilst 131 trying to ensure consumer preferences remain the same, changes in prey biomass could reflect change 132 in the diets of predators (Heymans et al., 2016). 133

2.2.2. Model balancing

134 EwE’s popularity and accessibility has grown considerably in recent years and as a result concerns were 135 raised over the ecological coherence and quality of the large quantity of models being produced (Link, 136 2010). These concerns warranted the publication of reports which propose ecological rules for marine 137 Ecopath models (Link, 2010) and best practice methods (Heymans et al., 2016). Therefore pre-balance 138 diagnostics, commonly referred to as PREBAL diagnostics (Link, 2010), were implemented following 4

NorBar EwE model: manuscript

Bentley et al., 2017

139 the initial construction of the NorBar Ecopath model as a means to highlight areas where ecological 140 laws may not be being met. For ecological coherence in a marine ecosystem, group biomass (on a log141 scale) should span 5-7 orders of magnitude and decline by 5-10% with increasing trophic level. P/B and 142 Q/B vital ratios (log-scales) should also show a general decline with increasing trophic level, whilst 143 production/consumption (P/Q) ratios mostly fall between 0.1 and 0.3, under the assumption that a group 144 is unable to produce more than a fraction of what is consumed (Heymans et al., 2016). 145 Energy ‘imbalance’ is determined by examining each group’s EE. If a group’s EE exceeds 1, the model 146 is indicating that the energy demand placed upon that group exceeds its production, and should therefore 147 be reduced. Using insight gained from PREBAL diagnostics, model estimates were systematically 148 balanced. 149

2.2.3. Development of temporal simulations using Ecosim

150 Ecosim (Christensen et al 2004) was used to create time-dynamic simulations of the NorBar food-web, 151 as initially represented by the static mass-balanced Ecopath model, in response to observed data (1950152 2014) and ocean warming scenarios (2015-2100) (Supplemental info A: Ecosim). 153 Water depth integrated (0-200 m) temperature time series data from 1950-2014 for the NorBar Seas 154 was taken from ICES Report on Ocean Climate (http://ocean.ices.dk/iroc/). The time series was 155 generated by averaging the annual depth integrated temperatures from three oceanographic stations 156 within the modelled area: Gismoy, Bear Island and Kola Section (Fig. 1). From 1950-1976 data is only 157 available from the Kola Section, therefore the average NorBar temperature for this time period was 158 estimated by multiplying the 1950-1976 Kola data by its average NorBar conversion factor for the years 159 1976-2014 (average conversion factor = 1.325) (Fig. 2a).Water depth integrated temperature time series 160 was entered into EwE as a temporal forcing function. 161 Optimal temperatures and tolerances were defined for each functional group using estimates for 162 minimum and maximum preferable and survivable temperatures from AquaMaps (Kaschner et al., 163 2016) (Fig. 3) (Supplemental Table B3). AquaMaps uses estimates of environmental tolerance to 164 provide model-generated species distribution maps which show strong agreement for many well-studied 165 species (O’Hara et al., 2017). Optimum temperatures were calculated by averaging 10th and 90th 166 preferable temperature quantiles. For groups with multiple species, temperature parameters were 167 averaged and weighted by biomass or, in cases where biomass data was unavailable, by catch. In EwE, 168 optimal temperatures and tolerances were included as species temperature Gaussian response functions 169 (Fig. 3). The Gaussian function has been used widely throughout optimality studies of thermal 170 performance curves (Angilletta, 2006) and has been demonstrated to provide the best statistical fit for 171 the temperature survival curves of North Atlantic fish eggs (Tsoukail et al., 2016). The other available 172 response functions in EwE, such as linear, exponential, sigmoid and hyperbolic, do not accurately 173 capture the characteristic shape of a thermal performance curve. The intercept between each specific 5

NorBar EwE model: manuscript

Bentley et al., 2017

174 response function and the annual water temperature were used to calculate a factor to modify the 175 predator consumption rates with a maximum multiplier of 1 which declines as the water temperature 176 deviates from the optimum at a rate determined by the function standard deviations (Serpetti et al., 177 2016, in review). Within Ecosim, consumption rates (Qij) are calculated based on the “foraging arena” 178 concept and a group’s vulnerability to top-down/bottom-up trophic interactions (Eq.(1)) (Christensen 179 and Walters, 2004; Walters and Christensen, 2007). 180 𝑄𝑖𝑗 =

𝑎𝑖𝑗 ∗𝑣𝑖𝑗 ∗𝐵𝑖 ∗𝑃𝑗 ∗𝑇𝑖 ∗𝑇𝑗 ∗𝑀𝑖𝑗 ⁄𝐷𝑗 𝑣𝑖𝑗 +𝑣𝑖𝑗 ∗𝑇𝑖 ∗𝑀𝑖𝑗 +𝑎𝑖𝑗 ∗𝑀𝑖𝑗 ∗𝑃𝑖 ∗𝑇𝑗 ⁄𝐷𝑗

∗ 𝑓(𝑇𝑒𝑚𝑝𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 , 𝑡) (1)

181 where aij is the effective search rate for predator j feeding on a prey i, vij is vulnerability expressing the 182 rate with which prey i move between being vulnerable and not-vulnerable, Bi is prey biomass, Pj is 183 predator abundance, Ti represents prey relative feeding time, Tj is predator relative feeding time, Mij is 184 mediation forcing effects, and Dj represents handling time as a limit to consumption rate (Christensen 185 et al., 2004; Ahrens et al., 2012). 186

2.2.4. Ecosim parameterisation

187 Relative biomass, catch, fishing mortality and fishing effort time series were imported to calibrate the 188 model. Relative biomass and catch acted as validation time series, whilst fishing mortality and fishing 189 effort time series were used to drive the model. Time series data for relative biomass and catch were 190 generally sourced from ICES, with some taken from other stock assessments or literature (see 191 Supplemental Table B4). Fishing mortality data was either taken from ICES stock assessments or 192 calculated using biomass and catch data. Fishing effort data was only available for the ‘Pelagic Trawl’ 193 fleet (ICES, 2015). In cases where groups were not fished by the ‘Pelagic Trawl’ fleet, and fishing 194 mortality data was absent, the catch time series was used to force the model over time, converting catch 195 to a biomass driving force in lieu of any other alternative. In total, 54 time series were compiled and 196 input into Ecosim: 14 relative biomasses, 14 relative catches, 12 forced catches, 13 fishing mortalities 197 and 1 fishing effort (Supplemental Table B5). 198

2.2.5. Ecosim calibration

199 Inter-specific interactions are manifest in Ecosim as ‘vulnerabilities’ under the foraging arena theory 200 (Ahrens et al., 2012), wherein spatial and temporal limitations have the capability to increase or reduce 201 predator/prey interactions through top down and bottom up dynamics (Supplemental info A). An 202 automated stepwise fitting procedure, recently developed by Scott et al., (2016), was used to calibrate 203 the model and increase the statistical fit of predicted temporal trends against observed trends. The 204 automated system constructed a series of model permutations to determine which combination of 205 ecological mechanisms provided the best statistical fit, as determined by sum of squares and Akaike’s 206 Information Criterion (Eq.(2)) (Akaike, 1974):

6

NorBar EwE model: manuscript 207 AIC = N × ln (SS/N) + 2 × K

Bentley et al., 2017

(2)

208 where N is the number of observations, SS is the residual sum of squares and K is the number of 209 parameters estimated. Ecological fitting mechanisms include fishing driving forces, predator/prey 210 vulnerabilities and an anomaly on primary production (PP) with multiple spline points (Scott et al., 211 2016). In this instance we estimated vulnerabilities by predator/prey interaction. The stepwise fitting 212 system estimated up to 27 parameters (K) given that a total of 28 catch and biomass time series were 213 provided. Parameters included predator/prey vulnerabilities and spline points (up to a maximum of 5) 214 estimated for the PP anomaly. Estimating one parameter less than the number of validation time series 215 provided ensures statistical strength is maintained. 216

2.3. Monte Carlo routine

217 Ecosim simulations are also sensitive to change in basic input parameters (B, P/B, Q/B and EE). The 218 Monte Carlo routine in Ecosim was therefore used to perform sensitivity analysis against the best fitting 219 model, as determined by the stepwise fitting process. This routine tests the sensitivity of Ecosim’s 220 output to basic estimates by drawing new basic estimates from a uniform distribution centered on the 221 base Ecopath value with the coefficient of variation (CV) set to 0.1 for all groups (Christensen et al., 222 2008). 500 simulations and 10,000 Ecopath runs were carried out with the aim of determining if altered 223 basic estimates could increase the statistical fit of the model (based on SS) or alter the temporal 224 dynamics and trends. 225

2.4. Predicting the impact of ocean warming

226 To predict the impact of ocean warming on the NorBar food-web the final, best fitting Ecosim model 227 was extended from 2015 to 2100 under predicted temperature scenarios (Fig. 2b). Depth integrated 228 temperature time series were developed using the existing 1950-2014 linear trend. Extrapolating the 229 constant historical rising rate culminated in a temperature rise of 1.05°C by the end of the century (from 230 5.96°C in 2014 to 7.002°C in 2100). Further scenarios were developed with altered rates of temperature 231 increase to produce scenarios wherein temperature reaches 6.002°C (-1°C scenario), 8.002°C (+1°C 232 scenario) and 9.002°C (+2°C scenario) by 2100. The extrapolated 1950-2016 trend falls within the 233 projected boundaries of the best case IPCC scenario (RCP2.6), the +1°C scenario falls within the range 234 of an intermediate IPCC scenario (RCP4.5) and the +2°C scenario falls within the range of the IPCC 235 worst case scenario (RCP8.5) (IPCC, 2014). Whilst the -1°C scenario does not fall within any of the 236 IPCC projections, it was included to enhance our understanding of the ecosystems response to 237 temperature change. Random variation was added to the temperature predictions to avoid straight line 238 projections. The potential range for random variation was set as 1°C above or below the straight line 239 projection to match the range of historic annual variation. The random variation time series was 240 smoothed using a two-year moving average to match historic variation. Recent warming in the NorBar 241 Seas has also been linked to Atlantic Multidecadal Oscillation (AMO) (Levitus et al., 2009; Drinkwater, 7

NorBar EwE model: manuscript

Bentley et al., 2017

242 2011) which experiences positive and negative phases with a periodicity of 60-80 years (Kerr, 2000). 243 Therefore a smoothed AMO anomaly, representing the sea surface temperature (SST) anomaly 244 normalised by the mean value of SST, was also added to temperature projections. The AMO approached 245 the end of its positive ‘warming’ phase by 2012 and was expected to decline into its negative ‘cooling’ 246 phase (Kotenev et al., 2012). AMO trends were therefore projected to 2100 by mirroring the trend from 247 2011 back to 1930. 248 This study has projected ecosystem simulations under the assumption of constant fishing mortality and 249 effort, as of 2014. Resultant projections for species biomass are thus driven by change in ocean 250 temperature and inter-specific dynamics, not fishing pressure. A multivariate analysis was carried out 251 to assess Bray-Curtis similarity within scenario biomass compositions using PRIMER (v.6.1.16). 252 Cluster analysis, similarity profile tests (SIMPROF) and analysis of similarity test (ANOSIM) were 253 used to evaluate the significant differences within the clusters (Clarke et al., 2001). A similarity 254 percentage test (SIMPER) was carried out to evaluate the role of individual functional groups in 255 contributing towards composition differences (Clarke et al., 2001). 256 3. Results 257

3.1. Final model

258 The final mass balance food-web model for the NorBar ecosystem is presented in Fig. 4 and the model 259 parameters (initial and balanced) in Table 1 (final diet matrix is provided in Supplemental Table B6). 260 When evaluated against PREBAL diagnostics, biomass was found to span 5 orders of magnitude and 261 decline with increasing trophic level (Supplemental Fig. C1). P/B and Q/B vital ratios generally 262 decreased with increasing trophic level. Q/B ratios for planktonic micro-organisms, benthic micro263 organisms and seabirds were notably higher than the ecological trend line. P/Q ratios for all 264 homeotherms fell below 0.1, whilst herring (1-3) was the only poikilotherm with a P/Q

Suggest Documents