Spatio-temporal dynamics of a fish predator: Density-dependent and hydrographic effects on Baltic Sea cod population

RESEARCH ARTICLE Spa io- empo al dynamics o a ish p eda o : Densi y-dependen and hyd og aphic e ec s on Bal ic Sea cod popula ion Vale io Ba olino 1 , Huidong Tian 1 , Ul Be gs o ¨m 2 , Pekka Jounela 3 , Ee o A o 4 , Ch is ian Die e ich 5 , H. E. Ma kus Meie 5,6 , Massimiliano Ca dinale 1 , Ba ba a Bland 1 , Michele Casini 1 * 1Swedish Uni e si y o Ag icul u al Sciences, Depa men o Aqua ic Resou ces, Ins i u e o Ma ine Resea ch, Lysekil, Sweden, 2Swedish Uni e si y o Ag icul u al Sciences, Depa men o Aqua ic Resou ces, Ins i u e o Coas al Resea ch, O ¨ eg und, Sweden, 3Na u al Resou ces Ins i u e Finland (Luke), Tu ku, Finland, 4Puolipa ¨i a ¨nka u 4 A 6, Helsinki, Finland, 5Swedish Me eo ological and Hyd ological Ins i u e, Resea ch Depa men , No ko ¨ping, Sweden, 6Leibniz Ins i u e o Bal ic Sea Resea ch Wa nemu¨nde, Depa men o Physical Oceanog aphy and Ins umen a ion, Ros ock, Ge many *[email p o ec ed] Abs ac Unde s anding he mechanisms o spa ial popula ion dynamics is c ucial o he success ul managemen o exploi ed species and ecosys ems. Howe e , he unde lying mechanisms o spa ial dis ibu ion a e gene ally complex due o he concu en o cing o bo h densi y- dependen species in e ac ions and densi y-independen en i onmen al ac o s. Despi e he high economic alue and cen al ecological impo ance o cod in he Bal ic Sea, he d i e s o i s spa io- empo al popula ion dynamics ha e no been analy ically in es iga ed so a . In his pape , we used an ex ensi e awl su ey da ase in combina ion wi h en i onmen al da a o in es iga e he spa ial dynamics o he dis ibu ion o he Eas e n Bal ic cod du ing he pas h ee decades using Gene alized Addi i e Models. The esul s showed ha adul cod dis ibu ion was mainly a ec ed by cod popula ion size, and o a mino deg ee by small- scale hyd ological ac o s and he ex en o sui able ep oduc i e a eas. As popula ion size dec eases, he cod popula ion concen a es o he sou he n pa o he Bal ic Sea, whe e he p e e ed mo e ma ine en i onmen condi ions a e encoun e ed. Using he i ed models, we p edic ed he Bal ic cod dis ibu ion back o he 1970s and a empo al index o cod spa ial occupa ion was de eloped. Ou s udy will con ibu e o he managemen and conse a ion o his impo an esou ce and o he ecosys em whe e i occu s, by showing he o ces shaping i s spa ial dis ibu ion and he e o e he po en ial esponse o he popula ion o u u e exploi a ion and en i onmen al changes. In oduc ion Unde s anding he spa ial dynamics o animal popula ions and using his in o ma ion in bio- logical conse a ion and esou ce managemen ep esen s one o he new on ie s in ma ine PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 1 / 17 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Ci a ion: Ba olino V, Tian H, Be gs o¨m U, Jounela P, A o E, Die e ich C, e al. (2017) Spa io- empo al dynamics o a ish p eda o : Densi y-dependen and hyd og aphic e ec s on Bal ic Sea cod popula ion. PLoS ONE 12(2): e0172004. doi:10.1371/jou nal.pone.0172004 Edi o : B ian R. MacKenzie, Technical Uni e si y o Denma k, DENMARK Recei ed: May 23, 2016 Accep ed: Janua y 30, 2017 Published: Feb ua y 16, 2017 Copy igh : ©2017 Ba olino e al. This is an open access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License, which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal au ho and sou ce a e c edi ed. Da a A ailabili y S a emen : All da a used in he analyses a e a ailable D yad Digi al Reposi o y: h p://dx.doi.o g/10.5061/d yad.2pb08. Funding: This wo k was inanced by he BONUS INSPIRE p ojec suppo ed by he join Bal ic Sea esea ch and de elopmen p og amme BONUS (A 185), unded join ly by he EU and he Swedish Resea ch Council Fo mas. This s udy was also unded by he PLAN FISH p ojec unded by he Swedish En i onmen al P o ec ion Agency and he Swedish Agency o Ma ine and Wa e ecology. Changes in spa ial dis ibu ion and mig a ion pa e ns ha e been epo ed in aqua ic and e es ial ecosys ems, bo h a small and la ge spa ial scales [1,2]. These changes ha e been a ibu ed o clima e e ec s [3–5], biological in e ac ions such as densi y-dependen esponses o p eda o -p ey dynamics [6–9] and human p essu es including ishe y o hun ing [10–13]. O en hese causes ac simul aneously, and mos likely in in e ac ion [14], o igge la ge e-loca ion o na u al popula ions. I is well demons a ed ha he abundance and dis ibu ion o ma ine ish popula ions is a ec ed by bo h densi y-dependen (o demog aphic) and densi y-independen (o en i on- men al) ac o s [15–22]. Thei e ec s in isola ion o h ough in e ac ions can also lead o la ge a ia ions in he spa ial dis ibu ion o ish popula ions [23–24]. Fo ins ance, expansion o he dis ibu ion owa ds ma ginal habi a s and, a low densi ies, con ac ion o he mos sui able habi a s a e known egula o y mechanisms in ish popula ions o a oid un a ou able condi- ions, elease compe i ion and op imise he use o esou ces and habi a s, as p edic ed by he ideal ee dis ibu ion heo y [25]. Beside i s ecological signi icance, unde s anding he p ocesses shaping he spa ial dis ibu- ion o ish popula ions is c ucial o hei conse a ion. In pa icula , he knowledge o wha de e mines he spa ial dis ibu ion o exploi ed popula ions could p o ide alid in o ma ion o egula e he ishe y [26] and p edic he e ec s o habi a loss and clima e change. Mo eo e , habi a occupa ion is a key indica o o species and communi y s a us and i s quan i ica ion is he e o e needed o a ull implemen a ion o an ecosys em app oach o he managemen o human ac i i ies [27]. The Bal ic Sea cod (Gadus mo hua) is he main pisci o ous ish o he open Bal ic Sea and, as such, plays a c ucial s uc u al and unc ional ole in he Bal ic ecosys em [28–30]. Du ing he pas h ee decades, in ac , la ge a ia ions in he abundance and dis ibu ion o he Eas e n Bal ic cod popula ion (he ea e e e ed o as Bal ic cod, Fig 1) ha e caused op-down d i en mul i-le el changes in bo h open sea and coas al ood webs [28,29,31,32]. Fishe ies s a is ics show la ge spa ial a ia ions in cod landings o e ime in he Bal ic Sea du ing he pas decades (Fig 2), sugges ing ha he dis ibu ion o he cod popula ion has also a ied conside ably. Al hough he causes o empo al a ia ion in he Bal ic cod popula ion ha e been s udied in dep h [33,34], no analy ical in es iga ion has been pe o med on he causes o he long- e m changes in cod spa ial dis ibu ion. He e we used, o he i s ime, a unique ex ensi e da ase o da a collec ed du ing in e na- ional bo om awl su eys in he las h ee decades by se e al coun ies bo de ing he Bal ic Sea. The aim o ou s udy was o in es iga e h ough s a is ical modelling he e ec s o densi y- dependen p ocesses (i.e. media ed by changes in popula ion abundance) and hyd ological condi ions on he spa io- empo al dynamics o cod in he Bal ic Sea. The model adop ed was also used o p edic he cod popula ion dis ibu ion in di e en pe iods, and an index o he a ea occupied was calcula ed as a no el indica o o he s a us o he cod popula ion. Ma e ials and me hods Biological and hyd ological da a We used cod s anda dized ca ch pe uni o e o (CPUE, ca ch in numbe s pe hou awling) collec ed du ing in e na ional awling su eys by he coun ies su ounding he Bal ic Sea be ween 1982–2012 (Fig 1,S1 Fig). The su eys ha e been coo dina ed as Bal ic In e na ional T awl Su ey (BITS) by he In e na ional Council o he Explo a ion o he Sea (ICES) since 1998 [35], bu a majo shi in he sampling gea occu ed in 2000 when na ional awls we e eplaced by a common s anda d TV-3 awl used by all coun ies. Con e sion ac o s among na ional and TV-3 awl ypes we e es ima ed o s anda dise CPUE by applica ion o a Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 2 / 17 Managemen ( o me ly he Swedish Boa d o Fishe ies). The high- esolu ion, h ee-dimensional model simula ion was pe o med wi hin he p ojec ECOSUPPORT (Ad anced modeling ool o scena ios o he Bal ic Sea ECOsys em o SUPPORT decision making) unded by he Eu opean Communi y’s Se en h F amewo k P og am (FP/2007-2013) unde g an ag eemen no. 217246 made wi h BONUS, he join Bal ic Sea esea ch and de elopmen p og am, and by he Swedish En i onmen al P o ec ion Agency (08/ 381). Addi ional suppo came om he Swedish Resea ch Council Fo mas wi hin he p ojec “Impac o accele a ed u u e global mean sea le el ise on he phospho us cycle in he Bal ic Sea”. Compe ing in e es s: The au ho s ha e decla ed ha no compe ing in e es s exis . machine lea ning algo i hm based on gene alized eg ession neu al ne wo ks [36]. The algo- i hm was ained on he 2000–2009 da ase and es ed on he 1998–1999 da ase o accoun o yea , mon h, coun y, geog aphical posi ion, and dep h ela ed di e ences in awls’ ca ch- ing e iciencies by cod age g oup (see S1 Tex o mo e de ails on awl s anda disa ion). Each eco d con ains he haul loca ion in o ma ion (la i ude and longi ude), awling dep h and he CPUE o cod by age-class (ages 1 o 10). We cons ained ou analyses o he pe iod 1982–2009 because o he lack o ine-scale hyd ological da a a e his yea (see below). We ocused on he adul componen o he cod popula ion (ages 3 and abo e) sampled be ween Janua y o Ma ch. The su eys co e he majo a ea o occu ence o he Eas e n Bal ic cod popula ion, i.e. he ICES Subdi isions (SDs) 25–28 (Fig 1). The ime se ies o cod popula ion size was om he la es accep ed analy ical s ock assessmen [37]. All he cod CPUE obse a ions in ime and space we e ma ched o modelled hyd ological da a. Mon hly mean oxygen and salini y da a we e in e pola ed om model esul s o he Swedish Coas al and Ocean Biogeochemical model coupled o he Rossby Cen e Ocean Fig 1. S udy a ea and loca ion o cod awl hauls. The map (made wi h Na u al Ea h) shows ICES Subdi isions (SDs, sepa a ed by solid black lines), and he ICES s a is ical ec angles ( hin do ed lines). Ba hyme y is shown in shades o blue. The Eas e n Bal ic cod s ock occu s in he a ea co e ed by SDs 25– 32 (SD 31, co e ing he Bo hnian Bay, is no shown). doi:10.1371/jou nal.pone.0172004.g001 Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 3 / 17 ci cula ion model (RCO-SCOBI) [38,39,S2 and S3 Figs]. RCO-SCOBI is a h ee-dimensional model o he Bal ic Sea wi h a ho izon al g id esolu ion o abou 3.7 km (2 nau ical miles) and wi h 83 e ical le els wi h laye hicknesses o 3 m. The physical and biogeochemical sub- models desc ibe he dynamics o wa e empe a u e, salini y, cu en s, sea le el, sea ice, bu also ni a e, ammonium, phospha e, phy oplank on, zooplank on, de i us and oxygen. RCO-SCOBI is o ced wi h h ee-hou ly a mosphe ic da a om a egionalized e-analysis sim- ula ion [40] and wi h mon hly i e uno and nu ien load da a [39] o he pe iod 1961– 2007. In ea lie s udies he model was success ully used o s udy pas clima e a iabili y (e.g. see [41]) and u u e p ojec ions [42]. The o iginal model ou pu was agg ega ed on he coa se ICES ec angles, bu he e ical esolu ion o 3 me e s emained unchanged. Each cod CPUE obse a ion was ma ched o he a e age hyd ological condi ions wi hin he same ICES ec an- gle, a he dep h and mon h o awling. We used salini y and oxygen because o hei ecog- nized impo ance as d i e o he Bal ic Sea bio a, especially in ela ion o cod [43]. We used a ela i ely coa se spa ial esolu ion o hyd ological condi ions compa ed o he ho izon al g id esolu ion o he hyd ological model because model esul s we e only alida ed a he sub- basin scale due o he lack o obse a ions a high esolu ion below he sea su ace. In addi ion, no assimila ion o obse a ions in o he model was applied and he chao ic beha io o he ocean’s mesoscale does no allow a one- o-one compa ison wi h obse a ions in any case. To es ima e how ep esen a i e he hyd ological pa ame e s salini y and oxygen a e aged wi hin each ICES ec angle a e o all he loca ions o he model g id wi hin each ICES ec angle, we calcula ed he spa ial a iance a each dep h o di e en ec angles ha a y be ween 2 imes 2 and 15 imes 15 model g id boxes. The spa ial a iances we e calcula ed e e y second day and hen a e aged o a 30-yea pe iod be ween 1966 and 1995. As expec ed, we ound a mode a e inc ease o he spa ial a iance a e aged o he en i e Bal ic Sea wi h he size o he ec angles Fig 2. Cod comme cial landings. (a) Comme cial landings in he ICES Subdi isions (SDs) 25–29 (excluding SD 28.1). The black line is he a io o landings in he no he n SDs (SD 27–29) o he o al landings (SDs 25–29, excluding SD 28.1). (b) Comme cial landings in SDs 28.1 (Gul o Riga), SD 30 (Gul o Finland) and SD 32 (Bo hnian Sea). Landings in SD 31 a e no p esen ed because o he e y small amoun . doi:10.1371/jou nal.pone.0172004.g002 Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 4 / 17 bu also an app oxima e sa u a ion o a iances o ec angles la ge han abou 8 imes 8 model g id boxes. Mo eo e , spa ial a iances a all spa ial scales wi hin ICES ec angles seem o be much smalle han empo al a iances a seasonal o decadal ime scales. Hence, we assume ha he e is no eal added alue o he modeling o cod dis ibu ions by using in o - ma ion om hyd ological pa ame e s a he high ho izon al esolu ion o he model g id. Fu - he in es iga ions o he impac o spa ial a iabili y a a ious scales on cod dis ibu ions a e ou o he scope o he p esen publica ion. To in es iga e he po en ial ela ion be ween cod dis ibu ion and spa io- empo al a ia- ions in ec ui men , we used ime-se ies o he cod ep oduc i e olume by SD. The ep o- duc i e olume (RV) is de ined as he olume o wa e wi h a salini y >11 psu and an oxygen concen a ion >2 mL L -1 , which a e ecognized as sui able hyd og aphic condi ions o he de elopmen o cod eggs [44]. The ep oduc i e olume is acknowledged o be a key d i e o cod ec ui men success in he Bal ic Sea [45]. In his s udy RV was used as p edic o o cod CPUE in each SD, wi h he hypo hesis ha cod adul s would end o emain close o he a eas whe e hey we e bo n, i.e. ha e a highe p obabili y o s ay in he same SD han o sp ead in o o he SDs. This accoun ed o he hypo hesis ha he spa ial expansion/con ac ions o he cod popula ion a e d i en by changes in he loca ion o sui able spawning a eas [46]. S a is ical modeling In ou model, h ee spa ial scales a e included, ackling di e en mechanisms po en ially capa- ble o shape he dis ibu ion o he adul ish: he small scale (hyd ological condi ions by ICES ec angle) indica es he habi a p e e ences/ equi emen s o ish, he basin scale ( ep oduc i e olume by SD) in es iga es he e ec o he ex en and loca ion o sui able spawning a ea on he u u e dis ibu ion o he adul s bo n in hose a eas, while he egional scale ( o al popula- ion size in he Bal ic Sea) analyses he densi y-dependen e ec in es iga ing he demog aphic e ec o popula ion size on he ish dis ibu ion o e he whole cen al Bal ic Sea. Reg ession me hods a e he main app oaches o analyze he ela ionships be ween species abundance (o o he esponse a iables) and hei en i onmen [47]. Conside ing he e sa il- i y and lexibili y in modeling he e ec s o densi y-dependen and densi y-independen a i- ables [5,17,20,22,48], we used Gene alized Addi i e Models (GAMs) [49,50]. GAMs assume ha he esponse a iable ollows some speci ic s a is ical dis ibu ion, such as Gaussian, Pois- son, e c. Howe e , biological da a o en iola e his assump ion, and issues like ze o-in la ion (mo e ze oes han he speci ied dis ibu ion sugges s) and o e -dispe sion ( a iance exceeds he mean) a e equen ly encoun e ed [51–53]. In ou case, mos o he awl da a we e col- lec ed du ing yea s o low cod popula ion size and o his eason, we used he quasi-Poisson dis ibu ion, which gi es highe weigh s o la ge CPUEs and a he same ime is able o handle ze o-in la ion. The a iance o a a iable Y ha ollows a quasi-Poisson dis ibu ion should be a linea unc ion o he mean (expec ed alue) o Y: Va (Y) = θE(Y), whe e θ(>1) is he dis- pe sion pa ame e [54]. The choice o p edic o s used in his s udy was based on in o ma ion om o icial ishe ies landings and acknowledged ecological p ocesses d i ing cod dynamics (see also Biological and hyd ological da a, abo e). Landing s a is ics ha e shown ha when cod popula ion size was high, as in he la e 1970s o he ea ly 1980, la ge amoun s o cod we e landed in he no he n a eas o he Bal ic Sea. On he con a y, du ing pe iods o low popula ion size, mos o he landings we e om he sou he n pa o he Bal ic Sea (Fig 2). This sugges s a geog aphical expansion and con ac ion o he popula ion a high espec i e low popula ion sizes. The e- o e, we used popula ion size om s ock assessmen es ima es [37] as densi y-dependen p edic o o cod CPUEs. Salini y and oxygen a e he majo hyd ological ac o s known o Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 5 / 17 in luence cod abundance and dis ibu ion [55], wi h high alues being physiologically a o - able o his ma ine species [43]. Mo eo e , a ia ions in cod ep oduc i e olume (RV) ha e been acknowledged as majo d i ing o ces o cod ec ui men [44]. T awl haul dep h was also included in he models in acco dance o p e ious s udies showing he p e e ence o cod o ce ain dep hs [55]. To es di e en hypo heses o densi y-dependence and densi y-indepen- dence, we buil he ollowing models in es iga ing he spa io- empo al dynamics o he adul componen o he cod popula ion (CPUE a ages 3+): mod1 : ðCPUEÞ ¼aþs1ðlon;la Þþs2ðlon;la Þpopula ion þs3ðoxy;salÞþs4ðdep hÞþs5ðRVÞþε mod2 : ðCPUEÞ¼ aþs1ðlon;la Þþbðpopula ionÞþs3ðoxy;salÞþs4ðdep hÞþs5ðRVÞþε mod3 : ðCPUEÞ¼ aþs1ðlon;la Þþs2ðlon;la Þpopula ion þs3ðdep hÞþs4ðRVÞþε mod4 : ðCPUEÞ¼ aþs1ðlon;la Þþs2ðlon;la Þpopula ion þs3ðoxy;salÞþs4ðdep hÞþε mod5 : ðCPUEÞ¼ aþs1ðlon;la Þþs2ðlon;la Þpopula ion þs3ðdep hÞþε mod6 : ðCPUEÞ¼ aþs1ðlon;la Þþs2ðoxy;salÞþs3ðdep hÞþε mod7 : ðCPUEÞ ¼ aþs1ðlon;la Þþs2ðoxyÞþs3ðsalÞþs4ðdep hÞþ ε whe e is he log-link unc ion o quasi-Poisson dis ibu ion, α he model in e cep , β he pa ame ic coe icien and s 1 -s 5 he smoo hing unc ions wi h no p io cons ain on he maxi- mum numbe o kno s. lon,la ,oxy,sal and dep h a e he longi ude, la i ude, oxygen, salini y and dep h a each sampling loca ion, espec i ely, RV is he ep oduc i e olume by SD (lagged 3 yea s back o ma ch wi h he mos abundan age-class included in he adul CPUE, i.e. age 3) and popula ion he o al cod popula ion size (ages 3+) a a speci ic yea . εis he e o e m whose a iance should ha e, in he case o quasi-Poisson dis ibu ion, a posi i e linea ela- ionship wi h CPUE. The hypo hesis o densi y-dependency is es ed wi h he wo al e na i e o mula ions p e- sen ed by model.1 and model.2 whe e he linea e ec o cod popula ion size is assumed o be spa ially- a iable by he e m s(lon,la ) popula ion [20,22] and spa ially-in a ian by he e m βpopula ion, espec i ely. Models.3-5 es o he inclusion o he di e en en i onmen al a - iables in o models wi h spa ial densi y-dependence. Model.6 and model.7 assume ha only densi y-independen en i onmen al a iables a ec cod spa ial dynamics and hey a e used o es o he in e ac ing and addi i e e ec s o oxygen and salini y, espec i ely. All models assume ha he esponse a iable ollows a quasi-Poisson dis ibu ion, which means he a i- ance o he model esiduals should ha e a linea ela ionship wi h he i ed alues. Model selec ion was based on he analysis o de iance explained and minimiza ion o he gene alized c oss alida ion (GCV, [56]). In addi ion, he p edic i e powe o each model was compa ed using a boo s ap app oach based on he genuine c oss alida ion sco e (gCV). Each model was i ed on 80% o he da a, andomly selec ed, and he a e age squa ed p edic ion e o was calcula ed on he da a poin s emo ed. This p ocedu e was epea ed 1000 imes o each model and he gCV was calcula ed as he mean o his e o s a is ic [57]. The bes model was checked o iola ion om he main model assump ions including spa ial au o-co ela ions in he model esiduals (S5 Fig). All analyses we e conduc ed in R (www. -p ojec .o g) using he mgc lib a y [50]. Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 6 / 17 Resul s The model selec ion p ocedu e showed ha model.1 pe o med be e han he o he models, indica ing ha cod spa io- empo al dynamics du ing he pas hi y yea s we e d i en by bo h spa ial densi y-dependence and hyd og aphic condi ions (Table 1). The combina ion o bo h be e s a is ics in compa ison o model.2 and es ima ion o a s a is ically signi ican spa ially- a iable e ec o popula ion size (Fig 3b) suppo he hypo hesis ha a spa ially he e ogeneous Table 1. Model selec ion esul s. In e cep (A), linea coe icien (β)and es ima ed deg ees o eedom o smoo hing e ms a e shown o all he models. The de iance explained (de .expl), he gene alized c oss alida ion (GCV) and genuine c oss alida ion (gCV) sco es a e also indica ed o each model. All he e ms o each model a e signi ican (p- alue <0.001). Model P edic o s de .expl GCV gCV A s(lon,la ) s(lon, la ) popula ion β(popula ion) s(oxy, sal) s(oxy) + s(sal) s(dep h) s(RV) Model.1 4.2 26.1 18.4 21.9 6.3 8.5 44.0 252.6 356.1 Model.2 4.3 28.1 2.4 25.4 6.1 8.8 41.0 264.5 362.3 Model.3 4.2 26.1 21.2 6.2 8.5 41.7 260.4 359.0 Model.4 4.3 26.1 18.6 21.2 6.3 42.7 257.2 357.7 Model.5 4.2 26.0 21.1 6.2 40.6 264.1 359.0 Model.6 5.0 28.0 26.8 6.3 27.3 324.7 390.9 Model.7 5.0 28.2 6.5; 8.9 6.6 25.3 331.5 394.0 doi:10.1371/jou nal.pone.0172004. 001 Fig 3. Resul s o he bes model. (a) e ec o spa ial loca ion. (b) spa ial e ec o cod popula ion size. (c) in e ac i e e ec s o salini y and oxygen. (d) e ec o dep h. (e) e ec o he ep oduc i e olume. ( ) es ima ed a iance- o-mean ela ionship (solid line, slope equals o 268). The ci cles a e a e aged squa ed esiduals in each ca ego y (e.g. 0 <E(Y) <2, 2 <E(Y) <4 and so on). doi:10.1371/jou nal.pone.0172004.g003 Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 7 / 17 densi y-dependen p ocess con ibu e o explain cod CPUE. Mo eo e , all he model s a is ics conside ed a ou model.6 in compa ison o model.7 sugges ing an in e ac ing a he han addi i e e ec o oxygen and salini y on he cod local densi ies. The esul s o model.1 (bes model) a e shown in Fig 3. The e m s(lon,la ) shows he iso- la ed e ec o spa ial loca ion on cod CPUE. High CPUEs appea ed in he sou h (especially SD 25) and low CPUEs in he no h o he s udy a ea (Fig 3a). The e ec o popula ion size has a p onounced posi i e spa ial g adien owa ds he no heas o he s udy a ea. In p ac ice, when he cod popula ion size dec eases, he local CPUEs in he no heas e n dec ease mo e han in he es o he Bal ic Sea. Acco dingly, local CPUEs end o inc ease mo e in he no heas e n pa o he dis ibu ion when he cod popula ion size inc eases (Fig 3b). The in e ac ion e m be ween oxygen and salini y shows ha when he salini y is low, he e ec o oxygen is weak, and salini y is he limi ing ac o ; con e sely, when he concen a ion o oxygen is low, he e ec o salini y is weak, and oxygen is he limi ing ac o (Fig 3c). Dep h shows a clea non- linea e ec on cod CPUE: he e ec inc eased apidly un il a ound 50 me e s, peaked a a ound 70 me e s, and sligh ly dec eased he ea e (Fig 3d). The e ec o he ep oduc i e ol- ume (RV) is sligh ly con ounded by a endency o o e i he da a, bu sensi i i y analysis on he le el o smoo hing con i ms he posi i e ela ionship wi h he cod CPUE h ee yea s la e (Fig 3e). The e ec d ops o RV >270 km 3 bu his is d i en by he las high alue o RV. By emo ing one co a ia e a a ime om he bes model and calcula ing he ela i e dec ease in he de iance explained, we e alua ed he con ibu ion o each a iable o explain he local CPUE o cod in he bes model (S1 Table). The spa ial densi y-dependen e m is by a he mos ele an ac o wi h a d op o almos 32% in he de iance explained by he model exclud- ing his e m. The second and hi d e ms by ele ance a e he geog aphical coo dina es and dep h wi h a con ibu ion o he a iance o only 8% and 7% espec i ely. The e ms ha included he hyd og aphic a iables appea as he less in luen ial wi h a con ibu ion o 5% o less. The es ima ed dispe sion pa ame e (θ) = 268 sugges s ha solid o e dispe sion exis s in ou da a. We also plo ed a e aged squa ed esiduals and i ed alues o ca ego ies 0 <E(Y) <2,2<E(Y) <4and so on, o diagnose he linea ela ionship be ween a iance and mean, ypical o a quasi-Poisson dis ibu ion (Fig 3 ). The diame e s o he ci cles a e p opo ional o he numbe o samples o each ca ego y, and all ca ego ies ha e a leas 10 samples. The a io- g am o he esiduals showed no spa ial au o-co ela ion in he esiduals o he bes model (da a no shown). Using he i ed bes model (model.1), we p edic ed cod dis ibu ion o i e di e en pe i- ods cha ac e is ic o he popula ion dynamics obse ed in he las 30 yea s (Fig 4). In he ea ly 1980s, du ing a pe iod o high popula ion size (Fig 4b), ou model p edic ed gene ally highe local densi ies and a wide spa ial dis ibu ion ha ex ended ac oss mos o he s udy a ea. In he mid 1980s he popula ion size had an ab up dec ease o le els compa able wi h hose obse ed in he ea ly 1970s (Fig 4a and 4c), and he model p edic ed an o e all educ ion o local densi ies, bu wi h a majo dec ease obse ed in he no heas (i.e. SD 28). In he ea ly 1990s (Fig 4d) he cod popula ion size eached a his o ical minimum, and he model p edic ed a u he con ac ion in he dis ibu ion mainly owa ds he sou hwes e n pa o he s udy a ea (i.e. SD 25). Du ing he mos ecen yea s (Fig 4e), a mode a e inc ease in he popula ion size p oduced mino expansion o he dis ibu ion owa ds he sou heas (i.e. SD 26), bu no in he no he n pa o he s udy a ea (i.e. SDs 27 and 28). We calcula ed a”habi a occupancy index” as he minimum a ea con aining 95% o he es i- ma ed abundance by yea . The esul s showed ha when cod popula ion size was high, he occupancy index was abo e 70%, while when he popula ion dec eased he habi a occupied was educed o 57% (Fig 5). A e he 2005, a sligh inc ease in he a ea occupied has occu ed. Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 8 / 17 Fig 4. P edic ions o cod spa io- empo al popula ion dis ibu ion a di e en popula ion le els. (a) 1970–1972 (hindcas ), (b) 1982–1984, (c) 1986–1988, (d) 1991–1992, (e) 2006–2007. ( ) Bal ic cod popula ion size (ages 3+). Red and blue colo s indica e high and low p edic ed CPUEs, espec i ely. doi:10.1371/jou nal.pone.0172004.g004 Fig 5. Habi a occupancy index. Minimum a ea con aining 95% o he es ima ed cod abundance by yea . doi:10.1371/jou nal.pone.0172004.g005 Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 9 / 17 33. Ko ¨s e FW, Mo ¨llmann C, Hin ichsen H-H, Wieland K, Tomkiewicz J, K aus G, e al. Bal ic cod ec ui - men – he impac o clima e a iabili y on key p ocesses. ICES J Ma Sci. 2005; 62: 1408–1425. 34. Ee o M., MacKenzie BR, Kos e FW, Gislason H. Mul i-decadal esponses o a cod (Gadus mo hua) popula ion o human-induced ophic changes, ishing, and clima e. Ecol Appl. 2011; 21: 214–226. PMID: 21516899 35. ICES. Repo o he Bal ic In e na ional Fish Su ey Wo king G oup (WGBIFS), 24–28 Ma ch 2014, Gdynia, Poland. ICES CM 2014/SSGESST:13. 527 pp. 2014. 36. Spech DF. A gene alized eg ession neu al ne wo k. IEEE T ansac ions on Neu al Ne wo ks 1991; 2: 568–76. PMID: 18282872 37. ICES. Repo o he Bal ic Fishe ies Assessmen Wo king G oup (WGBFAS), 10–17 Ap il 2013, ICES Headqua e s, Copenhagen. ICES CM 2013/ACOM:10. 747 pp. 2013. 38. Meie HEM, Do ¨sche R, Faxe ´n T. A mul ip ocesso coupled ice-ocean model o he Bal ic Sea: applica- ion o sal in low. J Geophys Res. 2003; 108:3273. 39. Eilola K, Meie HEM, Alm o h E. On he dynamics o oxygen, phospho us and cyanobac e ia in he Bal- ic Sea; A model s udy. J Ma Sys . 2009; 75: 163–184. 40. Samuelsson P, Jones CG, Wille ´n U, Ulle s ig A, Gol ik S, Hansson U, e al. The Rossby Cen e egional clima e model RCA3: model desc ip ion and pe o mance. Tellus. 2011; 63A: 4–23. 41. Meie HEM. Modeling he pa hways and ages o in lowing sal - and eshwa e in he Bal ic Sea. Es ua . Coas Shel Sci. 2007; 74:717–734. 42. Meie HEM, Ho doi R, Ande sson HC, Die e ich C, Eilola K, Gus a sson BG, e al. Modeling he com- bined impac o changing clima e and changing nu ien loads on he Bal ic Sea en i onmen in an ensemble o ansien simula ions o 1961–2099. Clim Dynam. 2012; 39: 2421–2441. 43. Schabe M, Hin ichsen H-H, G o ¨ge J. Seasonal changes in e ical dis ibu ion pa e ns o cod (Gadus mo hua) in he Bo nholm Basin, cen al Bal ic Sea. Fish Oceanog . 2012; 21: 33–43. 44. MacKenzie BR, Hin ichsen HH, Plikshs M, Wieland K, Zeze a AS. Quan i ying en i onmen al he e oge- nei y: habi a size necessa y o success ul de elopmen o cod Gadus mo hua eggs in he Bal ic Sea. Ma Ecol P og Se . 2000; 193: 143–156. 45. Ma gonski P, Hansson S, Tomczak M, G zebielec R. Clima e in luence on Bal ic cod, sp a , and he ing s ock– ec ui men ela ionships. P og Ocenog . 2010; 87: 277–288. 46. Ee o M, Hjelm J, Beh ens J, Buchmann K, Ca dinale M, Casini M. e al. Eas e n Bal ic cod in dis ess: biological changes and challenges o s ock assessmen . ICES J Ma Sci. 2015; 72: 2180–2186. 47. Ciannelli L, Fauchald P, Chan KS, Agos ini VN, Dingso GE. Spa ial ishe ies ecology: ecen p og ess and u u e p ospec s. J Ma Sys . 2008; 71: 223–236. 48. Ciannelli L, Chan KS, Bailey KM, S ense h NC. Nonaddi i e e ec s o he en i onmen on he su i al o a la ge ma ine ish popula ion. Ecology. 2004; 85: 3418–3427. 49. Has ie TJ, Tibshi ani RJ. Gene alized Addi i e Models, Chapman and Hall/CRC. 1990. 50. Wood S. Gene alized addi i e models: an in oduc ion wi h R. Chapman and Hall/CRC. 2006. 51. Ba y SC, Welsh AH. Gene alized addi i e modelling and ze o in la ed coun da a. Ecol Model. 2002; 157: 179–188. 52. Po s JM, Eli h J. Compa ing species abundance models. Ecol Model. 2006; 199: 153–163. 53. Lewin WC, F eyho J, Hucks o V, Mehne T, Wol e C. When no ca ches ma e : Coping wi h ze os in en i onmen al assessmen s. Ecol Ind. 2010; 10: 572–583. 54. Hoe JMV, Bo eng PL. Quasi-poisson s. nega i e binomial eg ession: How should we model o e dis- pe sed coun da a? Ecology. 2007; 88: 2766–2772. PMID: 18051645 55. Tomkiewicz J, Lehmann KM, S John MA. Oceanog aphic in luences on he dis ibu ion o Bal ic cod, Gadus mo hua, du ing spawning in he Bo nholm Basin o he Bal ic Sea. Fish Oceanog . 1998; 7: 48– 62. 56. G een PJ, Sil e man BW. Nonpa ame ic eg ession and gene alized linea models: a oughness pen- al y app oach. Chapman & Hall; 1994. 57. Ciannelli L, Dingsø GE, Bogs ad B, O e sen G, Chan KS, Gjøsae e H, e al. Spa ial ana omy o spe- cies su i al: e ec s o p eda ion and clima e-d i en en i onmen al a iabili y. Ecology. 2007; 88: 635– 646. PMID: 17503592 58. ICES. Repo o he S udy G oup on Spa ial Analyses o he Bal ic Sea (SGSPATIAL), 5–8 No embe 2013, Riga, La ia. ICES CM 2013/SSGRSP:11. 60 pp. 2013. 59. Rosenzweig ML. Habi a selec ion and popula ion in e ac ions: he sea ch o mechanism. Am Na . 1991; 137: S5–S28. Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 16 / 17 60. MacCall AD. Dynamic geog aphy o ma ine ish popula ions. Uni e si y o Washing on P ess, Sea le, WA; 1990. 61. Swain DP. Changes in he dis ibu ion o A lan ic cod (Gadus mo hua) in he sou he n Gul o S . Law ence: e ec s o en i onmen al change o change in en i onmen al p ocesses? Fish Oceanog . 1999; 8: 1–17. 62. Gill JA, No is K, Po s PM, Gunna sson TG, A kinson PW, Su he land WJ. The bu e e ec and la ge- scale popula ion egula ion in mig a o y bi ds. Na u e. 2001; 412: 436–438. doi: 10.1038/35086568 PMID: 11473317 63. Balbon in J, Fe e M. Densi y-dependence by habi a he e ogenei y: indi idual quali y e sus e i o y quali y. Oikos. 2008; 117: 1111–1114. 64. Mo is DW. Habi a ma ching: al e na i es and implica ions o popula ions and communi ies. E ol Ecol. 1994; 8: 387–406. 65. Ma shall CT, F ank KT. The e ec o in e annual a ia ion in g ow h and condi ion on haddock ec ui - men . Can J Fish Aqua Sci. 1999; 56: 347–355. 66. Ca dinale M, A henius F. The in luence o s ock s uc u e and en i onmen al condi ions on he ec ui - men p ocess o Bal ic cod es ima ed using a gene alized addi i e model. Can J Fish Aqua Sci. 2000; 57: 2402–2409. 67. ICES. Repo o he Bal ic Fishe ies Assessmen Wo king G oup (WGBFAS), 14–21 Ap il 2015, ICES HQ, Copenhagen, Denma k. ICES CM 2015/ACOM:10. 806 pp. 2015. 68. Hansson M, Ande sson L. Oxygen Su ey in he Bal ic Sea 2013—Ex en o Anoxia and Hypoxia, 1960–2013. Repo Oceanog aphy No. 49, Swedish Me eo ological and Hyd ological Ins i u e, Go hen- bu g, Sweden. 2013. 69. Slo e A. E ec s o ish leng h and condi ion on spawning mig a ion in No wegian sp ing spawning he - ing (Clupea ha engus L.). Sa sia. 1999; 84: 111–127. 70. Jø gensen C, Dunlop ES, Opdal AF, Fiksen O. The e olu ion o spawning mig a ions: s a e dependence and ishing-induced changes. Ecology. 2008; 89: 3436–48. PMID: 19137949 71. Lindeg en M, O ¨s man O ¨, Gå dma k A. In e ac ing ophic o cing and he popula ion dynamics o he - ing. Ecology. 2011; 92: 1407–1413. PMID: 21870614 72. Be gs o ¨m U, Olsson J, Casini M, E iksson BK, F ed iksson R, Wennhage H, e al. S ickleback inc ease in he Bal ic Sea–a ho ny issue o coas al p eda o y ish? Es Coas Shel Sci. 2015; 163: 134–142. Spa io- empo al dynamics o Bal ic Sea cod PLOS ONE | DOI:10.1371/jou nal.pone.0172004 Feb ua y 16, 2017 17 / 17