Incidental bycatch mortality and fishing restrictions: impacts on juvenile survival in the Endangered Saimaa ringed seal Pusa hispida saimensis

ENDANGERED SPECIES RESEARCH Endang Species Res Vol. 38: 91–99, 2019 h ps://doi.o g/10.3354/es 00939 Published Feb ua y 21 1. INTRODUCTION Fishe ies inciden al byca ch poses a signi ican h ea o many ma ine mammal popula ions, espe- cially wi h he use o passi e gea , such as gillne s, which ha e con ibu ed o he popula ion declines o se e al species (Ree es e al. 2013). Addi ionally, he Endange ed (Sipilä 2016) land-locked Saimaa inged seal Pusa hispida saimensis is h ea ened by a ious human-induced isk ac o s, among which inciden al by ca ch, oge he wi h clima e change, may be e - ga ded as he mos c i ical sou ces o cu en mo al- i y (Ko acs e al. 2012, Niemi e al. 2012, 2013, Au ila e al. 2014, Val onen e al. 2014, Lyy ikäinen e al. 2015, Kunnas an a e al. 2016, Liukkonen e al. 2017, 2018). In pa icula , ju eniles a e ulne able o © The au ho s 2019. Open Access unde C ea i e Commons by A ibu ion Licence. Use, dis ibu ion and ep oduc ion a e un - es ic ed. Au ho s and o iginal publica ion mus be c edi ed. Publishe : In e -Resea ch · www.in - es.com *Co esponding au ho : [email p o ec ed] Inciden al byca ch mo ali y and ishing es ic ions: impac s on ju enile su i al in he Endange ed Saimaa inged seal Pusa hispida saimensis Pekka Jounela1,*, Te o Sipilä2, Jouni Koskela2, Raisa Tiilikainen2, Miina Au ila2, Ma ja Niemi3, Me i Kunnas an a3,4 1Na u al Resou ces Ins i u e Finland, 20520 Tu ku, Finland 2Me sähalli us, Pa ks & Wildli e Finland, 57130 Sa onlinna, Finland 3Uni e si y o Eas e n Finland, Depa men o En i onmen al and Biological Sciences, 80101 Joensuu, Finland 4Na u al Resou ces Ins i u e Finland, 80110 Joensuu, Finland ABSTRACT: Inciden al byca ch, mos ly in gillne s used o ec ea ional ishing, is a c i ical mo al- i y ac o o he Endange ed Saimaa inged seal Pusa hispida saimensis. Byca ch pa icula ly a ec s ju eniles, and he e o e ishing has been es ic ed by banning gillne s du ing he mos c i ical ju enile dispe sal pe iod in sp ing. In addi ion, he mos ha m ul gea ypes o inged seals o all ages a e o bidden yea - ound. The sp ing ime es ic ion a eas ha e inc eased signi ican ly since he 1990s, now co e ing o e 90% o he pups’ bi h si es. Howe e , he impac s o ishing es ic ions on inciden al byca ch mo ali y ha e no p e iously been e alua ed s a is ically. The p esen es ima es o inciden al byca ch le els in ela ion o ishing es ic ions since he 1990s sug- ges ha he sp ing ime gillne ishing ban esul ed in an inc ease in popula ion size by 20% be ween 1991 and 2013 (60 indi iduals gi en an a e age es ima ed s ock size o 355 in 2013). In addi ion, he es ima ed c i ical pe iod o ju enile su i al in ela ion o ishing ope a ions appea ed o be he i s 15 mo, which is a much longe ime han p e iously expec ed, sugges ing ha s a iona y gillne ishing du ing la e win e mon hs may cause a seconda y peak in seal mo - ali y. One unwan ed es ima ed side e ec o he seasonal ban was a sligh ly inc eased inciden al es ima ed byca ch le el immedia ely a e he end o he sp ing ime ishing es ic ion pe iod. The es ima ed byca ch peaked in 2000−2005, wi h an a e age o 13.3 ju eniles annually and dec eased he ea e , mainly in 2009−2013, wi h an a e age o 6.3 ju eniles annually. KEY WORDS: En anglemen · Gillne s · Machine lea ning · Mo ali y a e · Pinnipeds O PEN PEN A CCESS CCESS Con ibu ion o he Special ‘Ma ine e eb a e byca ch: p oblems and solu ions’ Endang Species Res 38: 91–99, 2019 byca ch mo ali y caused mos ly by gillne s used in ec ea ional ishing. Va ious ishing es ic ions a e en o ced in Lake Saimaa: annual gillne ishing p o- hibi ions a e applied du ing he mos c i ical dispe - sal pe iod (15 Ap il–30 June) o imp o e he su i al o he ju eniles as hey dispe se, in addi ion o which he mos ha m ul gea ypes o seals o all ages ( ish- bai ed hooks, la ge ish aps wi h closed co e s, wi e baske ish aps wi h mou hs wide han 150 mm and a ied es ic ions ela ed o mesh size and line s eng h o gillne s) ha e been o bidden h oughou he yea since 1999 (Sipilä 2003). The pa chy a eas subjec o he sp ing ime gillne ishing ban we e g adually enla ged du ing 1991−2013, especially in he cen al b eeding a eas o he seals (Minis y o he En i onmen 2011). The conse a ion e ec s o his ban we e e alua ed by compa ing p opo ions ( ela i e sizes) o he banned ishing a ea wi h he b eeding habi a . In 1991, he ban co e ed less han 20% o he seals’ bi h lai loca ions (Sipilä 2003), and i was assumed ha by en o cing new p o ec ion a eas, o e 90% o he bo n pups would be p o ec ed by ishing es ic ions (Minis y o he En i onmen 2011), al hough he a eas o conce n would s ill no co e all o he ju enile seals’ po en ial dispe sal pa - e ns (Niemi e al. 2012, 2013). Niemi e al. (2013) ound ha he home anges o Saimaa inged seal ju eniles we e much la ge han had ea lie been supposed, and hey e i ied ishing- induced inciden al dea hs o 40% o he adio- agged pups. This, and he combina ion o an inc eased numbe o pups being bo n bu he size o he adul Saimaa inged seal popula ion emaining small, boos ed a public deba e on he magni ude and causes o unobse ed dea hs inside and ou side he ishing es ic ion a eas. Some s akeholde s denied all unobse ed dea hs, while o he s sugges ed ha he unobse ed dea hs we e pe haps 3 imes g ea e in numbe han he obse ed dea hs (Minis y o he En i onmen 2011). Un o una ely, he e a e no o he es ima es o unobse ed dea hs han hose p esen ed by Niemi e al. (2013), and he causes o dea h o he addi ional obse ed s anded seals ha e emained unknown. A ew dead, s anded seals a e ound bo h inside and ou side he es ic ion a eas each yea , bu i is a ely possible o diagnose he causes o hose dea hs because o he bad condi ion o he ca casses. In ecen yea s, ield obse e s ha e also e i ied a ew inciden al bycaugh indi iduals wi hin he banned a eas, cas ing doub s on he eal impac o ishing es ic ions and he biological sus ainabili y o he popula ion in he ace o such unquan i iable inci- den al byca ch. In his wo k, we es ima ed he impac o ishing es ic ions on inciden al byca ch o Saimaa inged seals using s a is ical me hods. Ou ools a e pu ely s a is ical due o he absence o any conc e e es - ima es o he main pa ame e s equi ed o such an assessmen , such as causes o dea h, popula ion size and biological sus ainabili y. Ins ead, an a ay o machine lea ning echniques was used o com- pensa e o hese sho comings, wi h hei ad an- age being hei abili y o ecognize pa e ns (o dea hs) ha a e oo obscu e o be pe cep ible by human esea che s employing s anda d s a is ical ools. 2. MATERIALS AND METHODS 2.1. Da a Lake Saimaa (61°05−62° 36’ N, 27° 15−30° 00’ E) is a eshwa e basin ha is app oxima ely 180 km long and 140 km wide (Fig. 1); he a ea o in e es in his s udy co e s he cu en dis ibu ion o he Saimaa inged seal (see Niemi e al. 2012). The mo ali y and bi h lai da a om 1991−2013 used in he analyses we e ob ained om he Saimaa inged seal da abase o Me sähalli us Pa ks & Wildli e Finland (2018). The mo ali y da a, including cause, loca ion, yea and age, we e based on 422 Saimaa inged seal ca - casses. Since no eliable ime se ies exis ega ding o e all gillne loca ions and ishing e o s, only he da a based on he causes o seal mo ali y we e used he e. The causes o dea h we e ei he de e mined by a wildli e pa hologis o e i ied in he ield (as in he case o byca ch in gillne s). The causes o dea h we e classi ied as ollows: (1) s illbo n: pup dead a bi h, (2) lai dea hs: unweaned pups aged 0−2 mo, (3) ish- ing: bycaugh seals aged 0−35 y and (4) o he s: weaned seals olde han 2 mo ha died o na u al o unknown causes. The ages o he seals (excluding pups wi h lanugo hai ) we e de e mined by coun ing he cemen um laye s in he lowe canine ee h (S ew- a e al. 1996). The bi h a e da a om 1991−2013 included he numbe s o Saimaa inged seals bo n in he yea s conce ned (N = 1104). Since inged seals gi e bi h o a single pup in a subni ean lai , he numbe s o bi h lai s can be used o es ima e pup p oduc ion ( o de ails, see Sipilä e al. 1990, Sipilä 2003). Such da a we e collec ed du ing annual Saimaa inged seal snow lai censuses conduc ed h oughou he spe- cies’ dis ibu ion a ea in la e win e o ea ly sp ing om 1991−2013. 92 Jounela e al.: Byca ch mo ali y o Saimaa inged seals 2.2. S a is ical me hods 2.2.1. Model s uc u e Es ima ion o he o al numbe o dea hs began wi h he o al numbe o annually bo n pups and ended when he su i o s eached 30 y o age. The e was only 1 e y old seal (>25 y , max. 35 y ). The es ima ion o he o al numbe o dea hs by cause o dea h in yea yis di ided as ollows: (1) es ima ion o he o al mo ali y a e om age 0 o age 30 y in yea yand (2) es ima ion o he causes o dea h a gi en ages in yea y. In ciden al byca ch is assumed o occu only in a eas wi hou ishing es ic ions, while o he causes o dea h (s ill bo n, lai dea hs, o he s) a e assumed o occu bo h inside and ou side he ishing es ic ion a ea. 2.2.2. Causes o dea h es ima ion The causes o dea h pa e ns we e es i- ma ed using a andom o es (RF) model (B eiman 2001). RF consis s o building an ensemble ( o es ) classi ie o decision ees p oduced om bagging (boo s ap agg ega ed; see B eiman 1996) and a an- domized a ian o he ee induc ion algo- i hm. In ecological s udies, RF has been applied o species dis ibu ion models (SDMs), e.g. o species conse a ion and biodi e si y managemen pu poses, and in he con ex o clima e change (Guo e al. 2015). One gene al eason o using en - semble SDMs has been o educe unce - ain y and s abili y in p edic ions, espe- cially when compa ed wi h a single SDM, such as a gene alized linea model o a gene alized addi i e model (Ma mion e al. 2009a,b, G enouille e al. 2011, Guo e al. 2015). Die e ich (2000) iden i ied s a is- ical, compu a ional and ep esen a ional easons, explaining why ensembles o en wo k be e han a single model. Fi s , when he lea ning se is oo small (as in ou s udy), a lea ning algo i hm can ypically ind se e al models ( unc ions) in he hypo hesis space and, by a e aging se - e al models, can educe he isk o choos- ing he w ong hypo hesis. Second, an en - semble made o indi idual models buil om many di e en s a ing poin s may p o ide a be e app oxima ion o he ue unknown unc ion han one using any o he single models. By combining se e al models in an ensemble, i may be possible o expand he space o ep esen able unc ions and ob ain a be e model o he ue unc ion. Consequen ly, among 179 classi ie s a ising om 17 amilies (dis- c iminan analysis, Bayesian, neu al ne wo ks, sup- po ec o machines, decision ees, ule-based clas- si ie s, boos ing, bagging, s acking, RFs and o he ensembles, gene alized linea models, nea es neigh- bou s, pa ial leas squa es and p incipal componen eg ession, logis ic and mul inomial eg ession, mul- iple adap i e eg ession splines and o he me hods), he bes classi ie s we e di e en e sions o RF (Fe - nandez-Delgado e al. 2014). The p edic o s o cause o dea h dis ibu ions we e la i ude, longi ude, seal age (in mon hs) and yea . The pa ame e s o he RF model (speci ically, numbe o ees: 10−1500; con idence le el used o pes- 93 Fig. 1. A ea o Lake Saimaa, Finland, inhabi ed by Saimaa inged seals, wi h loca ions o hei bi h lai s (main b eeding a eas) and co e age o sp ing ime ishing es ic ions in 2013 Endang Species Res 38: 91–99, 2019 simis ic e o calcula ion o p uning, i.e. emo al o ee sec ions o imp o e p edic i e accu acy by he educ ion o o e i ing: 0.05−0.5; maximal dep h: 5−30; and c i e ion o selec ing a ibu es and nume - ical spli s: accu acy, gain a io, Gini index, in o ma- ion gain) we e op imized using a sequen ial ( e - pea ed) g id sea ch. Tha is, he op imiza ion was epea ed nume ous imes using di e en sub- anges o he numbe o ees and weaked using he e - maining pa ame e s un il a pla eau was ound in he s a is ical pe o mance (accu acy) beyond which mo e ees did no make any di e ence, and he pe - o mance sco e e en showed a small dip. The s a is ical pe o mance o he RF model was e alua ed using 10- old c oss- alida ion (Koha i 1995), which means ha he da a we e andomly spli in o k(10) olds using s a i ied sampling, and each ime k−1 olds we e used o aining and he e - maining 1 old o es ing. The model wi h ce ain pa ame e alues ha es ed bes o e he 10 non- o e lapping es olds was hen selec ed. Finally, he bes model wi h he bes pa ame e s was applied o all da a. In gene al, he aim o 10- old c oss- alida- ion is o a oid o e - i ing and o choose a model ha is nei he oo simple no oo complex. I gua ds agains es ing hypo heses sugges ed by he da a (called ‘Type III e o s’) in cases whe e u he sam- ples a e cos ly o impossible o collec . A model ha is oo simple usually has high bias and low a iance, while a model ha is oo complex usually has low bias and high a iance. Tha is, inding he bes model amoun s o making an app op ia e ade-o be ween bias and a iance, i.e. inding he bes model ollows he ‘no ee lunch’ heo em (Wolpe & Mac eady 1997). 2.2.3. Mo ali y a e es ima ion The sample size o ju enile seals (aged 0−15 mo) was much highe (5−15 dea hs annually) han ha o he olde adul seals (0−5 dea hs annually), so i was possible o es ima e an annual mo - ali y a e o he ju eniles, bu no o he olde seals. The annual mo ali y pa e ns we e ecognized using a gene - alized eg ession neu al ne wo k (GRNN; Spech 1991). Fi s , he alues o he p e dic o s o mo ali y (age and yea ) we e s anda dized be ween 0 and 1. Then, log(ln + 1) ans o med numbe s (dead ju eniles) in each age class (in mon hs) in yea ywe e assessed. Finally, mon hly mo ali y a es we e es ima ed om he (nega i e) eg ession slope pa ame e s o he GRNN eg ession es ima e. The mo ali y a e o he adul seals aged 1−30 y was es ima ed using log- ans o med age equency dis ibu ions (also e e ed o as he ‘ca ch cu e’ me hod; Hilbo n & Wal e s 1992). Ini ially, he num- be o seals in each age class was de e mined, and hen he numbe s we e log(ln + 1) ans o med; he ea e , o al mo ali y was calcula ed as he neg- a i e slope o he eg ession. Finally, he e o o he mo ali y es ima es was calcula ed as he e o o he slope o he eg ession. In he adul mo ali y a e es ima ion, we used Bayesian linea eg ession wi h unin o ma i e p io s, i.e. we applied equal p obabili y o each possible alue o he p io mo ali y a e, which means ha we did no assume any pa icula (unknown) mo ali y le el a p io i be o e pe o ming he mo ali y analy- sis. Fu he mo e, we assumed a cons an annual mo - ali y a e o he adul seal age g oups o e he yea s because he annual sample size o adul seals was ei he 0 o ex emely small (0−5 annually). 3. RESULTS 3.1. Inciden al byca ch and he impac s o ishing es ic ion a eas The sp ing ime (15 Ap il−30 June) ishing es ic- ion a eas ha e inc eased signi ican ly in size and co e age since hey we e es ablished in 1982, and 94 Fig. 2. Annual sizes o he sp ing ime ishing es ic ion a eas and es ima ed p opo ions (%) o Saimaa inged seal bi h si es loca ed wi hin hem in 1991−2013 Jounela e al.: Byca ch mo ali y o Saimaa inged seals he inc ease has had an e ec on ju enile su i al. The a eas co e ed 18% o he bi h si es in 1991, whe eas in 2013, he co e age was o e 90% (Fig. 2). Al hough he i s es ic ions imposed o limi by - ca ch mo ali y we e es ablished as ea ly as 1982 (Sipilä e al. 1990), sys ema ic es ic ions ha e been implemen ed on a la ge scale (applying o bo h he season o he yea and he gea ype) only since 1999. In 2010, a majo inc ease in he sp ing ime o al ish- ing es ic ion a ea was en o ced in he main b eed- ing a eas o he seals, pa ly due o an inc ease in he dis ibu ion o he ju eniles and o he inged seal popula ion in gene al (Fig. 2). In 1991, he sp ing ime ishing e - s ic ion a eas had only a small posi- i e in luence on ju enile su i al, whe e as in 2013, he in luence o he enla ged o al es ic ion a ea was mo e signi ican (Fig. 3). Howe e , an es ima ed side e ec o he es ic ion a eas in 2013 was a sligh inc ease in es ima ed inciden al byca ch le els a e ishing began (on 1 July), when he ju eniles we e 4 mo old (Fig. 3). Tha is, a highe es ima ed numbe o su i o s on 30 June led o a highe es ima ed inciden al byca ch a e ishing began on 1 July. The es ima ed by ca ch peaked in 2000− 2005, wi h an a e age o 13.3 ju eniles annually, and was highes in 2004. The en - la ged ishing es ic ion a eas in 2010− 2013 e ec i ely e duced he by - ca ch o seals 0−4 mo old (Fig. 4). The conse a ion impac o he sp ing ime ishing es ic ions in - c eased signi ican ly h oughou he pe iod conce ned (Fig. 5). In he pe iod 1991− 2013, he annual impac o he es ic ion a eas peaked a a ju enile age o 2.5− 3 mo. The conse - a ion impac o he ishing es ic ion a eas in e ms o he inciden al by - ca ch was smalle a age 3.5 mo han a 2.5−3 mo, and was minimal a 0− 1 mo (Fig. 5). I he sp ing ime ishing es ic ion a eas co e ed all o he inged seals’ bi h si es h oughou he pe iod con- ce ned, i is es ima ed ha he num - be o ju eniles su i ing as a esul o ishing es ic ion a eas would ha e inc eased o an a e age o ~7.8 in - di iduals annually (a e age slope pa ame e , Fig. 6) compa ed o a si ua ion wi h no es ic ion a eas a all. As he es ic ion co e age was less han 100% (Fig. 2), only an a e age o 2.1 ju eniles we e es ima ed o ha e been sa ed annually by his man- agemen measu e. Du ing 1990−1995, he es ima ed a e age annual byca ch o ju eniles (0−15 mo old) was 7.9 ( ange 6.4−10.6) indi iduals, and du ing 2009−2013, i was 6.3 ( ange 4.4−8) indi iduals. The es ima ed numbe o seals inciden ally bycaugh du ing 2009−2013 was app oxima ely hal o ha in he ea ly 2000s. How- 95 Fig. 3. Es ima ed annual ins an aneous byca ch o Saimaa inged seals (indi- iduals) wi h and wi hou ishing es ic ion a eas in (A) 1991 and (B) 2013. When ishing began in 2013, i.e. when he ju enile seals we e 4 mo old, he g ey line is highe han he black do ed line, indica ing a sligh ly inc eased byca ch Fig. 4. Es ima ed byca ch o ju enile Saimaa inged seals aged 0−15 mo in 1991, 2004, 2010 and 2013. No e he peak in 2004, a e which he enla ged ishing es ic ion a eas e ec i ely educed he byca ch o seals aged 0−4 mo Endang Species Res 38: 91–99, 2019 e e , he obse ed byca ch mo ali y does no show he same pa e n as he es ima ed mo ali y, as i has emained a app oxima ely he same le el h ough- ou he in e al 1991−2013 (Fig. 7). 3.2. Popula ion size The ishing es ic ion a eas induced an inc ease in he popula ion o he Saimaa inged seal. The cumu- la i e inc easing e ec o he sp ing ime ishing es ic ions on he popula ion size o he Saimaa inged seal du ing 1991−2013 was 47 ju eniles + 13 adul s (20% inc ease in popula ion size om 295 seals in 1991 o 355 seals in 2013, Fig. 8). The es i- ma ed popula ion size is an unce ain pa ame e ha a ies be ween 2 ex emes, he es ima ed minimum and maximum dis ibu ions (Fig. 9). 4. DISCUSSION Ou esul s show he e ec i eness o he enla ged sp ing ime ishing es ic ion a eas in educing by - ca ch numbe s o Saimaa inged seal ju eniles. Ou analyses show ha he es ima ed byca ch numbe s in ecen yea s we e app oxima ely hal hose in he ea ly 2000s, whe eas he obse ed byca ch numbe s o ju eniles (Fig. 7) emained app oxima ely he same, and i appea s ha he es ic ions did no ha e any impac on he su i al o ju eniles. Howe e , i is also he unobse ed (no epo ed, i.e. es ima ed minus obse ed, Fig. 7) byca ch numbe s ha ha e dec eased in ecen yea s. This shows he e ec i e- ness o es ic ions in sa egua ding ju eniles (Fig. 5), which nea ly doubled he popula ion size (seal num- be s be o e bi h) du ing 1991−2013. I can be ex - pec ed ha wi hou he es ic ions, he inciden al byca ch in ecen yea s would ha e been app oxi- ma ely double. Simila ly, implemen a ion o he ish- ing es ic ions in 1991− 2013 o e an a ea co e ing all o he bi h si es could po en ially ha e inc eased ju enile su i al, wi h an a e age o ~7.8 indi iduals es ima ed o ha e been sa ed annually. In p ac ice, howe e , he e se e co e age was mode a e, and he e o e he popula ion o 355 seals a he end o he pe iod s udied he e included an es ima ed 60 indi- iduals mo e han he e would ha e been wi hou he ishing es ic ions (295 seals, Fig. 9). Ou esul s con i m ea lie obse a ions o he mos c i ical pe iod o ju enile su i al (Sipilä 2003, Min- is y o he En i onmen 2011, Niemi e al. 2013, Au - ila e al. 2014) and show ha he impac o ishing es ic ions on ju enile su i al begins a he age o 1.5 mo and is less ma ked a age 3.5 mo han a 2.5− 3 mo, which oughly co esponds o he ime in e al 15 May−15 July. Howe e , he da a also indica e ha he es ima ed c i ical pe iod o ju enile su i al, as a as ishing p ac ices a e conce ned, ex ends o e he i s 15 mo, as seen in he seconda y peak in annual byca ch mo ali y a e he end o he sp ing- ime ishing es ic ion pe iod. Ou analyses clea ly poin o his seconda y peak in mo ali y and show 96 Fig. 5. Es ima ed ins an aneous sp ing ime ishing es ic- ion-induced Saimaa inged seal su i o s (a e age numbe o indi iduals) a gi en ages (in mon hs) in 1991−2013. The numbe o su i o s peaked in 2012. The impac o he ish- ing es ic ions on he su i al o he ju eniles s a ed o in- c ease a he age o 1.5 mo Fig. 6. Sp ing ime ishing es ic ion-induced Saimaa inged seal su i o s (aged 0−4 mo) in ela ion o he co e age o bi h si es (%) in 1991−2013. The eg ession slope pa ame- e (7.8258) sugges s ha 100% es ic ion co e age s. 0% co e age could po en ially inc ease he numbe o su i o s by ~7.8 indi iduals (a e age annual es ima e) Jounela e al.: Byca ch mo ali y o Saimaa inged seals ha he es ima ed byca ch peaked in 2004 and de - c eased he ea e o e he in e al 2009−2013 o an a e age o 6.3 indi iduals annually. Ou es ima es a e also suppo ed by ecen eco ds, which indica e inc eased byca ch mo ali y du ing ha ime (Me - sähalli us Pa ks and Wildli e Finland 2018). This sec- ond mo ali y peak is ypically caused by gillne ish- ing du ing he au umn and win e . Al hough ou esul s indica e ela i ely low byca ch numbe s a he end o he pe iod conce ned (2009−2013, see Fig. 7), his could be an unde es ima e due o noise o con- cep d i (a change in s a is ical p ope ies o he a - ge a iable). Some algo i hms may o e eac o noise, e oneously in e p e ing i as concep d i , while o he s may be highly obus wi h ega d o noise, adjus ing o he changes oo slowly (Tsymbal 2004). The po en ially ele a ed byca ch numbe s ha e also been obse ed in mo ali y s a is ics in ecen yea s (Me sähalli us Pa ks and Wildli e Fin- land 2018). This is he i s s udy o show he hidden byca ch mo ali y (see Fig. 7) in he Saimaa inged seal popu- la ion, as specula ed ea lie (Sipilä 2003, Minis y o he En i onmen 2011). One o he main assump ions he e is ha he cause o dea h assessmen assumes ha he p obabili y o epo ing is equal be ween all causes, which may no hold ue. In July, o example, he e a e nume ous people p esen (summe co age owne s and ec ea ional ishe men) who a e likely o epo inciden al byca ch dea hs mo e o en han in win e , when he ha sh condi ions a ac ewe ish- e men. Howe e , he numbe o ne s can also be con- side able du ing ha ime because gillne ishing in Lake Saimaa is ocussed especially on he pike pe ch Sande luciope ca, which is one o he mos aluable species bo h comme cially and ec ea ionally. In addi ion, he willingness o epo inciden al byca ch dea hs may a y be ween seasons, and hence he es ima ed inciden al byca ch pa e ns p esen ed he e 97 Fig. 7. Obse ed and es ima ed byca ch mo ali y o ju enile Saimaa inged seals a age 0−15 mo. Obse ed and es i- ma ed bycaugh indi iduals and hei p opo ions (%) o pups bo n annually in 1991−2013 Fig. 8. Es ima ed a e age Saimaa inged seal popula ion sizes in 1991−2013, wi h cumula i e ishing es ic ion im- pac s ( ), wi h non-cumula i e (single-yea ) es ic ion e - ec s (m) and wi h no ishing es ic ion impac s (j) Fig. 9. Two es ima ed Saimaa inged seal popula ion size p obabili y dis ibu ions in 2013, wi h 2 unce ain impac s o he ishing es ic ion a eas. The maximum dis ibu ion (black solid line) assumes emo al o all (obse ed and hid- den) inciden al byca ch wi hin he ishing es ic ion a eas in all yea s, and he minimum dis ibu ion (black dashed line) assumes ha he ishing es ic ion a eas had a ze o impac on ishing byca ch inside ishing es ic ion a eas Endang Species Res 38: 91–99, 2019 should be conside ed ‘bes -guess’ es ima es (app ox- ima ions) o he inciden al byca ch mo ali y. The e has also been deba e abou he accep abili y o ish- ing es ic ions o local people (Ra amäki & Salmi 2015), which may also ha e caused he willingness o ishe men o epo inciden al byca ch o a y be - ween he yea s o e he whole pe iod 1991−2013. The esul s and ends o he p esen s udy, howe e , do no signi ican ly de ia e om hose epo ed ea - lie , and hence, he o e all eliabili y o he assess- men was conside ed plausible. The na ionally accep ed biological e e ence poin o he size o he Saimaa inged seal popula ion is ha i should each 400 adul seals by 2025 (Minis y o he En i onmen 2011). This has been o icially de - ined as an ‘in e media e goal’ and would be achie ed by allowing he popula ion o g ow by app oxima ely 3−4% y −1. This e e ence poin in eg a es he knowl- edge and opinions o a ious s akeholde g oups, bu i emains an open ques ion as o which compu a- ional e e ence poin me hod would be mos app o- p ia e o he Saimaa inged seal. Consequen ly, he e ha e been di icul ies in jus i ying he main e- nance o easonably conse a i e ishing es ic ions ha bo h p o ec he Saimaa inged seals and enable local small-scale subsis ence ishing (Ra amäki & Salmi 2015). In addi ion, along wi h inciden al by - ca ch mo ali y, an inc easing h ea o popula ion eco e y om he mid-2000s onwa d has been he high pe ina al mo ali y caused by he lack o ade- qua e snow co e du ing he b eeding season in Feb ua y−Ma ch (Au ila e al. 2014). Addi ionally, a sligh ly less acknowledged bu po en ially se e e h ea has been he apid inc ease in he numbe o summe co ages and associa ed ac i i ies, which has s eadily educed he po en ial b eeding g ounds along he sho es o he Saimaa lake sys em (Liukko- nen e al. 2017). Coping wi h hese h ea s may p o e o be a ela i ely complica ed ma e . Ne e heless, he indings p esen ed he e do show ha educing inciden al byca ch mo ali y by ex ending he ishing es ic ions has had subs an ial long- e m e ec s on he g ow h a e o he Endange ed Saimaa inged seal popula ion. Ju enile su i al o he Saimaa inged seal is a c i - ical elemen o success ul conse a ion. Pe ina al mo ali y ypically a ies be ween 10 and 20% (Au - ila 2015), and a e weaning, byca ch is he domi- nan mo ali y ac o o ju eniles (Kokko e al. 1999, Sipilä 2003, Niemi e al. 2013). In he u u e, pe ina al mo ali y is likely o inc ease due o human dis u - bance (Liukkonen e al. 2017) and changing clima e (Au ila 2015); he e o e, mi iga ion o byca ch, a mos acu e mo ali y ac o , is highligh ed in conse - a ion s a egies, and ishing closu es ha e been es ablished. Howe e , o be e ec i e, ishing es ic- ions should be p olonged a leas o he end o July o educe he byca ch mo ali y peak in summe . In addi ion, ou esul s on hidden mo ali y indica e ha he obse ance o ishing egula ions s ill equi es bo h educa ion and pa olling. Acknowledgemen s. We hank Ma ja Isomu su o he Finnish Food Sa e y Au ho i y o pe o ming he seal ca cass nec op- sies. This wo k was unded by he EU Li e + Sa egua ding Saimaa inged seal p ojec LIFE12NAT/ FI/ 000367. LITERATURE CITED Au ila M (2015) The endange ed Saimaa inged seal in a changing clima e—challenges o conse a ion and moni o ing. PhD disse a ion, Uni e si y o Eas e n Fin- land, Joensuu Au ila M, Niemi M, Sk zypczak T, Viljanen M, Kunnas an a M (2014) Es ima ing and mi iga ing pe ina al mo ali y o he endange ed Saimaa inged seal (Phoca hispida saimensis) in a changing clima e. Ann Zool Fenn 51: 526−534 B eiman L (1996) Bagging p edic o s. Mach Lea n 24: 123−140 B eiman L (2001) Random o es s. Mach Lea n 45: 5−32 Die e ich TG (2000) Ensemble me hods in machine lea n- ing. In: Mul iple classi ie sys ems. MCS 2000. Lec u e No es in Compu e Science, Vol 1857. Sp inge , Be lin, p 1−15 Fe nandez-Delgado M, Ce nadas E, Ba o S, Amo im D (2014) Do we need hund eds o classi ie s o sol e eal wo ld classi ica ion p oblems? J Mach Lea n Res 15: 3133−3181 G enouille G, Buisson L, Casajus N, Lek S (2011) Ensemble modelling o species dis ibu ion: he e ec s o geo- g aphical and en i onmen al anges. Ecog aphy 34: 9−17 Guo C, Lek S, Ye S, Li W, Liu J, Li Z (2015) Unce ain y in ensemble modelling o la ge-scale species dis ibu ion: e ec s om species cha ac e is ics and model ech- niques. Ecol Model 306: 67−75 Hilbo n R, Wal e s CJ (1992) Quan i a i e ish s ock assess- men . Choice, dynamics and unce ain y. Chapman & Hall, New Yo k, NY Koha i R (1995) A s udy o c oss- alida ion and boo s ap o accu acy es ima ion and model selec ion. In: Mellish CS (ed) P oceedings o he 14 h in e na ional join con e - ence on a i icial in elligence, Mon eal, Quebec. Mo - gan Kau mann Publishe s, San Ma eo, CA, p 1137−1143 Kokko H, Helle E, Linds öm J, Ran a E, Sipilä T, Cou - champ F (1999) Backcas ing popula ion sizes o inged and g ey seals in he Bal ic and Lake Saimaa du ing he 20 h cen u y. Ann Zool Fenn 36:65–73 Ko acs KM, Aguila A, Au ioles D, Bu kano V and o he s (2012) Global h ea s o pinnipeds. Ma Mamm Sci 28: 414−436 Kunnas an a M, Niemi M, Au ila M (2016) Conse a ion bi- ology o he Saimaa inged seal: om esea ch o ac ions. Suom Riis a 62: 71−82 (in Finnish wi h English summa y) 98 Jounela e al.: Byca ch mo ali y o Saimaa inged seals Liukkonen L, Rau io A, Sipilä T, Niemi M, Au ila M, Koskela J, Kunnas an a M (2017) Long- e m e ec s o land use on pe ina al mo ali y in he Endange ed Saimaa inged seal popula ion. Endang Species Res 34: 283−291 Liukkonen L, Ayllón D, Kunnas an a M, Niemi M, Nabe- Nielsen J, G imm V, Nyman AM (2018) Modelling mo e- men s o Saimaa inged seal using an indi idual-based app oach. Ecol Model 368: 321−335 Lyy ikäinen M, Pä ynen J, Hy ä inen H, Sipilä T, Kunnas- an a M (2015) Me cu y and selenium balance in endan- ge ed Saimaa inged seal depend on age and sex. En i- on Sci Technol 49: 11808−11816 Ma mion M, Pa iainen M, Luo o M, Heikkinen RK, Thuille W (2009a) E alua ion o consensus me hods in p edic i e species dis ibu ion modelling. Di e s Dis ib 15: 59−69 Ma mion M, Luo o M, Heikkinen RK, Thuille W (2009b) The pe o mance o s a e-o - he-a modelling ech- niques depends on geog aphical dis ibu ion o species. Ecol Model 220: 3512−3520 Me sähalli us Pa ks & Wildli e Finland (2018) Hyljekan a 2018 (Saimaa inged seal popula ion). www. me sa. i/ saimaano ppa/hyljekan a2018 (in Finnish; accessed 8 Ma ch 2018) Minis y o he En i onmen (2011) A conse a ion s a egy and ac ion plan o he Saimaa inged seal. Ympä is ö - minis e iö, Helsinki (in Finnish) Niemi M, Au ila M, Viljanen M, Kunnas an a M (2012) Mo emen da a and hei applica ion o assessing he cu en dis ibu ion and conse a ion needs o he endange ed Saimaa inged seal. Endang Species Res 19: 99−108 Niemi M, Au ila M, Viljanen M, Kunnas an a M (2013) Home ange, su i al and dispe sal o endange ed Saimaa inged seal pups: implica ions o conse a ion. Ma Mamm Sci 29: 1−13 Ra amäki O, Salmi P (2015) The mos con es ed in Finland: la ge ca ni o es and he Saimaa inged seal − challenges o socio-ecological hy hms and hei p ac ical implica- ions. Eu Coun ys 7: 1−15 Ree es RR, McClellan K, We ne TB (2013) Ma ine mammal byca ch in gillne and o he en angling ne ishe ies, 1990 o 2011. Endang Species Res 20: 71−97 Sipilä T (2003) Conse a ion biology o Saimaa inged seal (Phoca hispida saimensis) wi h e e ence o o he Eu o- pean seal popula ions. PhD disse a ion, Uni e si y o Helsinki Sipilä T (2016) Pusa hispida ssp. saimensis. The IUCN Red Lis o Th ea ened Species 2016: e.T41675A66991678. h p:// dx.doi.o g/10.2305/IUCN.UK.2016-1.RLTS. T41675 A 66991678.en (accessed 7 Feb ua y 2019) Sipilä T, Helle E, Hy ä inen H (1990) Dis ibu ion, popula- ion size and ep oduc i i y o he Saimaa inged seal (Phoca hispida saimensis No dq.) in Finland, 1980–84. Finn Game Res 47: 3−10 Spech DF (1991) A gene alized eg ession neu al ne wo k. IEEE T ans Neu al Ne w 2: 568−576 S ewa REA, S ewa BE, S i ling I, S ee E (1996) Coun s o g ow h laye g oups in cemen um and den ine in inged seals (Phoca hispida). Ma Mamm Sci 12: 383−401 Tsymbal A (2004) The p oblem o concep d i : de ini ions and ela ed wo k. Tech Rep TCD-CS-2004-15. T ini y College, Dublin Val onen M, Palo JU, Aspi J, Ruokonen M, Kunnas an a M, Nyman T (2014) Causes and consequences o ine-scale popula ion s uc u e in a c i ically endange ed esh- wa e seal. BMC Ecol 14: 22 Wolpe DH, Mac eady WG (1997) No ee lunch heo ems o op imiza ion. IEEE T ans E ol Compu 1:67–82 99 Edi o ial esponsibili y: Rebecca Lewison, San Diego, Cali o nia, USA Submi ed: Ap il 25, 2018; Accep ed: Janua y 7, 2019 P oo s ecei ed om au ho (s): Feb ua y 7, 2019