Biogeosciences 16 3183ndash3195 2019httpsdoiorg105194bg-16-3183-2019copy Author(s) 2019 This work is distributed underthe Creative Commons Attribution 40 License

Identifying areas prone to coastal hypoxia ndash the role of topographyElina A Virtanen12 Alf Norkko34 Antonia Nystroumlm Sandman5 and Markku Viitasalo1

1Marine Research Centre Finnish Environment Institute Helsinki 00790 Finland2Department of Geosciences and Geography University of Helsinki Helsinki 00014 Finland3Tvaumlrminne Zoological Station University of Helsinki Hanko 10900 Finland4Baltic Sea Centre Stockholm University Stockholm 10691 Sweden5AquaBiota Water Research Stockholm 11550 Sweden

Correspondence Elina A Virtanen (elinaavirtanenenvironmentfi)

Received 10 April 2019 ndash Discussion started 16 April 2019Revised 25 June 2019 ndash Accepted 30 July 2019 ndash Published 27 August 2019

Abstract Hypoxia is an increasing problem in marineecosystems around the world While major advances havebeen made in our understanding of the drivers of hypoxiachallenges remain in describing oxygen dynamics in coastalregions The complexity of many coastal areas and lack ofdetailed in situ data have hindered the development of mod-els describing oxygen dynamics at a sufficient spatial res-olution for efficient management actions to take place Itis well known that the enclosed nature of seafloors and re-duced water mixing facilitates hypoxia formation but thedegree to which topography contributes to hypoxia forma-tion and small-scale variability of coastal hypoxia has notbeen previously quantified We developed simple proxiesof seafloor heterogeneity and modeled oxygen deficiency incomplex coastal areas in the northern Baltic Sea Accord-ing to our models topographical parameters alone explainedsim 80 of hypoxia occurrences The models also revealedthat less than 25 of the studied seascapes were prone tohypoxia during late summer (AugustndashSeptember) Howeverlarge variation existed in the spatial and temporal patterns ofhypoxia as certain areas were prone to occasional severe hy-poxia (O2 lt 2 mg Lminus1) while others were more susceptible torecurrent moderate hypoxia (O2 lt 46 mg Lminus1) Areas identi-fied as problematic in our study were characterized by lowexposure to wave forcing high topographic shelter from sur-rounding areas and isolation from the open sea all contribut-ing to longer water residence times in seabed depressionsDeviations from this topographical background are probablycaused by strong currents or high nutrient loading thus im-proving or worsening oxygen status respectively In someareas connectivity with adjacent deeper basins may also in-

fluence coastal oxygen dynamics Developed models couldboost the performance of biogeochemical models aid devel-oping nutrient abatement measures and pinpoint areas wheremanagement actions are most urgently needed

1 Introduction

Hypoxia is a key stressor of marine environments occurringin over 400 physically diverse marine ecosystems worldwide(Diaz and Rosenberg 1995b 2008 Conley et al 2009) De-clining oxygen levels have been recorded in fjords estuariesand in coastal and open-sea areas such as the ChesapeakeBay Gulf of Mexico Japan Sea Baltic Sea and the BlackSea (Gilbert et al 2010 Carstensen et al 2014) It is clearthat our oceans are losing their breath and recent projectionsindicate that anoxic zones devoid of higher life will be in-creasing in the forthcoming decades (Froumllicher et al 2009Meier et al 2011a 2012a) with severe consequences formarine ecosystems (Breitburg et al 2018)

The lack of oxygen alters the structure and functioning ofbenthic communities (Nilsson and Rosenberg 2000 Grayet al 2002 Karlson et al 2002 Valanko et al 2015)disrupts bioturbation activities (Timmermann et al 2012Villnas et al 2012 2013 Norkko et al 2015) changespredatorndashprey relationships (Eriksson et al 2005) and maylead to mass mortalities of benthic animals (Vaquer-Sunyerand Duarte 2008) Hypoxia does not only affect organismsof the seafloor but also influences biogeochemical cyclingand benthicndashpelagic coupling (Gammal et al 2017) Hy-poxia can increase releases of nutrients from the sediment

Published by Copernicus Publications on behalf of the European Geosciences Union

3184 E A Virtanen et al Identifying areas prone to coastal hypoxia

and thus promote planktonic primary production and sedi-mentation which in turn leads to enhanced microbial con-sumption of oxygen (Conley et al 2002 Kemp et al 2009Middelburg and Levin 2009) This creates a self-sustainingprocess often referred to as the ldquovicious circle of eutrophi-cationrdquo (Vahtera et al 2007) which may hamper the effectsof nutrient abatement measures

Biogeochemical processes contributing to hypoxia forma-tion are well known Factors affecting the development of hy-poxia are usually associated with the production of organicmatter level of microbial activity and physical conditionscreating stratification and limited exchange or mixing of wa-ter masses (Conley et al 2009 2011 Rabalais et al 2010Fennel and Testa 2019) Coastal hypoxia is common in areaswith moderate or high anthropogenic nutrient loading highprimary productivity and complex seabed topography limit-ing lateral movement of the water Shallow-water hypoxia isoften seasonal It is associated with warming water tempera-tures and enhanced microbial processes and oxygen demand(Buzzelli et al 2002 Conley et al 2011 Caballero-Alfonsoet al 2015 van Helmond et al 2017)

Projecting patterns and spatial and temporal variability ofhypoxia is necessary for developing effective managementactions Thus three-dimensional coupled hydrodynamicndashbiogeochemical models have been created for several sea ar-eas around the world such as the Gulf of Mexico (Fennel etal 2011 2016) Chesapeake Bay (Scully 2013 2016 Testaet al 2014) the North Sea (Hordoir 2018) and the BalticSea (Eilola et al 2009 2011 Meier et al 2011a 2012a b)These models simulate various oceanographic biogeochem-ical and biological processes using atmospheric and climaticforcing and information on nutrient loading from riversWhile such models are useful for studying processes at thescale of kilometers and aid in defining hypoxia abatementat the basin scale their horizontal resolution is too coarse(often 1ndash2 nmi nautical miles) for accurately describing pro-cesses in coastal areas Lack of detailed data on water depthcurrents nutrient loads stratification and local distribution offreshwater discharges (Breitburg et al 2018) (not to mentioncomputational limitations) usually prevents the applicationof biogeochemical models developed to large geographicalareas at finer horizontal resolutions (lt 100 m) Understand-ing spatial variability of hypoxia in topographically complexcoastal environments has therefore been impeded by the lackof useful methods and systematic good-quality data (Diazand Rosenberg 2008 Rabalais et al 2010 Stramma et al2012) Finding alternative ways to pinpoint areas prone tocoastal hypoxia could facilitate the management and deter-mination of efficient local eutrophication abatement mea-sures

It is widely recognized that the semienclosed nature of theseafloors and associated limited water exchange is a sig-nificant factor in the formation of hypoxia in coastal wa-ters (Diaz and Rosenberg 1995a Virtasalo et al 2005 Ra-balais et al 2010 Conley et al 2011) However to deter-

mine the degree to which seascape structure restricting wa-ter movement contributes to hypoxia formation has not beenquantified Analytical and theoretical frameworks developedspecifically for terrestrial environments such as landscapeheterogeneity or patchiness are analogous in marine envi-ronments and are equally useful for evaluating links betweenecological functions and spatial patterns in a marine contextWe tested how a large fraction of hypoxia occurrences couldbe explained only by the structural complexity of seascapeswithout knowledge on hydrographical or biogeochemical pa-rameters We adopted techniques and metrics from landscapeecology and transferred them to the marine environment andwe (1) examined if spatial patterns in seascapes can explainthe distribution of hypoxia (2) defined the relative contribu-tion of seascape structure to hypoxia formation and (3) es-timated the potential ranges of hypoxic seafloors in coastalareas To achieve this we concentrated on extremely het-erogeneous and complex archipelago areas in the northernBaltic Sea where coastal hypoxia is a common and an in-creasing problem (Conley et al 2011 Caballero-Alfonso etal 2015)

2 Data and methods

21 Study area

The studied area covers the central northern Baltic Seacoastal rim 23 500 km2 of Finnish territorial waters from theBothnian Bay to the eastern Gulf of Finland and 5100 km2

of Swedish territorial waters in the Stockholm archipelago inthe Baltic Proper Oxygen dynamics in the deeper areas ofthe Gulf of Finland and Stockholm archipelago are stronglyaffected by oceanography and biogeochemistry of the centralBaltic Proper not reflecting the dynamics of coastal hypoxia(Laine et al 1997) and were therefore excluded from thisstudy The outer archipelago of Finland is relatively exposedwith various sediment and bottom habitat types while theinner archipelago is more complex and shallower but main-tains a higher diversity of benthic habitats and sediment types(Valanko et al 2015) The inner archipelago of Stockholmis an equally complex archipelago area with a large numberof islands straits and coves Freshwater outflow from LakeMaumllaren creates an estuarine environment where freshwatermeets the more saline water in the Baltic Proper

In order to evaluate differences in oxygen deficiency be-tween coastal areas the study area was divided into fiveregions as defined by the EU Water Framework Directive(200060EC) (WFD 2000) the Archipelago Sea (AS) theeastern Gulf of Finland (EGoF) the Gulf of Bothnia (GoB)the Stockholm archipelago (SA) and the western Gulf of Fin-land (WGoF) (Fig 1) Small skerries and sheltered bays char-acterize AS EGoF and WGoF whereas the narrower band ofislands forms relatively exposed shores in GoB Deep elon-gated channels of bedrock fractures can reach depths of over

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3185

Figure 1 Study areas in Finland AS ndash Archipelago SeaWGoF ndash western Gulf of Finland EGoF ndash eastern Gulf ofFinland and GoB ndash Gulf of Bothnia Study area in Swe-den SA ndash Stockholm archipelago Orange dots represent mod-erately hypoxic sites (O2 lt 46 mg Lminus1) black dots severely hy-poxic sites (O2 lt 2 mg Lminus1) and white circles denote sites withO2 gt 46 mg Lminus1 Gray and black lines illustrate boundaries of Wa-ter Framework Directive areas

100 m in AS similarly to SA where narrow deep valleysseparate mosaics of islands and reefs Substrate in both ar-eas varies from organic-rich soft sediments in sheltered lo-cations to hard clay till and bedrock in exposed areas Atgreater depths soft sediments are common due to limitedwater movement As a whole the study area with rich topo-graphic heterogeneity forms one of the most diverse seabedareas in the world (Kaskela et al 2012) In many areas hy-poxia is a result of strong water stratification slow water ex-change and complex seabed topography creating pockets ofstagnant water (Conley et al 2011 Valanko et al 2015Jokinen et al 2018) Thus the area is ideal for testing hy-potheses of topographical controls for hypoxia formation

22 Hypoxia data

Bottom-water hypoxia is the main factor structuring benthiccommunities in the Baltic Sea (Villnas et al 2012 Norkkoet al 2015) A total of 2 mg Lminus1 of O2 is usually consid-ered a threshold where coastal organisms start to show symp-

toms of the lack of oxygen and this limit has been commonlyused in various global reviews (Diaz and Rosenberg 1995b2008) Some studies have however concluded that 2 mg Lminus1

is below the empirical sublethal and lethal oxygen limit formany species (Vaquer-Sunyer and Duarte 2008 Conley etal 2009) Here we define hypoxia based on two ecolog-ically meaningful limits moderately hypoxic lt 46 mg Lminus1

O2 ndash as this has been estimated to be a minimum safe limitfor species survival behavior and functioning in benthiccommunities (Norkko et al 2015) ndash and severely hypoxiclt 2 mg Lminus1 O2 which describes zones where larger marineorganisms suffer from severe mortality (Vaquer-Sunyer andDuarte 2008) As no reference values exist for severity ofhypoxia to marine organisms based on the frequency of hy-poxic events (Norkko et al 2012 2015 Villnas et al 2012)we here define a site to be prone to hypoxic events ieoccasionally hypoxic if it experienced hypoxia (O2 lt 2 andlt 46 mg Lminus1) at least once during the study period If hy-poxia was recorded in ge 20 of the visits it was categorizedas frequently hypoxic We consider this to be ecologicallyrelevant as species develop symptoms already from short ex-posures to hypoxia (Villnas et al 2012 Norkko et al 2015)This is also justified as our oxygen data are from sim 1 mabove the seafloor suggesting that the actual oxygen concen-trations at sediment where benthic species live are probablylower

Data from oxygen profiles were collated fromthe national monitoring environmental data portalsHertta (httpswwwsykefien-USOpen_informationOpen_web_servicesEnvironmental_data_API lastaccess 30 January 2019) and SHARK (httpswwwsmhisedataoceanografihavsmiljodata last access15 January 2019) Data were available from 808 monitoringsites Only the months of August and September 2000ndash2016were considered as hypoxia is usually a seasonal phe-nomenon occurring in late summer when water temperaturesare warmest (Conley et al 2011)

23 Predictors

For modeling hypoxia occurrences we developed five ge-omorphological metrics (1) bathymetric position indices(BPIs) with varying search radii (2) depth-attenuated waveexposure (SWM(d)) (3) topographic shelter index (TSI)(4) arc-chord rugosity (ACR) and (5) vector ruggedness mea-sure (VRM) BPI is a marine modification of the terrestrialversion of the topographic position index (TPI) originallydeveloped for terrestrial watersheds (Weiss 2001) BPI is ameasure of a bathymetric surface higher (positive values) orlower (negative values) than the overall seascape BPI valuesclose to zero are either flat areas or areas with constant slopeHere BPIs represent topographical depressions and crests atscales of 01 03 05 08 and 2 km calculated with BenthicTerrain Modeler (v30) (Walbridge et al 2018) SWM(d) es-timates dominant wave frequency at a given location with

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3184 E A Virtanen et al Identifying areas prone to coastal hypoxia

and thus promote planktonic primary production and sedi-mentation which in turn leads to enhanced microbial con-sumption of oxygen (Conley et al 2002 Kemp et al 2009Middelburg and Levin 2009) This creates a self-sustainingprocess often referred to as the ldquovicious circle of eutrophi-cationrdquo (Vahtera et al 2007) which may hamper the effectsof nutrient abatement measures

Biogeochemical processes contributing to hypoxia forma-tion are well known Factors affecting the development of hy-poxia are usually associated with the production of organicmatter level of microbial activity and physical conditionscreating stratification and limited exchange or mixing of wa-ter masses (Conley et al 2009 2011 Rabalais et al 2010Fennel and Testa 2019) Coastal hypoxia is common in areaswith moderate or high anthropogenic nutrient loading highprimary productivity and complex seabed topography limit-ing lateral movement of the water Shallow-water hypoxia isoften seasonal It is associated with warming water tempera-tures and enhanced microbial processes and oxygen demand(Buzzelli et al 2002 Conley et al 2011 Caballero-Alfonsoet al 2015 van Helmond et al 2017)

Projecting patterns and spatial and temporal variability ofhypoxia is necessary for developing effective managementactions Thus three-dimensional coupled hydrodynamicndashbiogeochemical models have been created for several sea ar-eas around the world such as the Gulf of Mexico (Fennel etal 2011 2016) Chesapeake Bay (Scully 2013 2016 Testaet al 2014) the North Sea (Hordoir 2018) and the BalticSea (Eilola et al 2009 2011 Meier et al 2011a 2012a b)These models simulate various oceanographic biogeochem-ical and biological processes using atmospheric and climaticforcing and information on nutrient loading from riversWhile such models are useful for studying processes at thescale of kilometers and aid in defining hypoxia abatementat the basin scale their horizontal resolution is too coarse(often 1ndash2 nmi nautical miles) for accurately describing pro-cesses in coastal areas Lack of detailed data on water depthcurrents nutrient loads stratification and local distribution offreshwater discharges (Breitburg et al 2018) (not to mentioncomputational limitations) usually prevents the applicationof biogeochemical models developed to large geographicalareas at finer horizontal resolutions (lt 100 m) Understand-ing spatial variability of hypoxia in topographically complexcoastal environments has therefore been impeded by the lackof useful methods and systematic good-quality data (Diazand Rosenberg 2008 Rabalais et al 2010 Stramma et al2012) Finding alternative ways to pinpoint areas prone tocoastal hypoxia could facilitate the management and deter-mination of efficient local eutrophication abatement mea-sures

It is widely recognized that the semienclosed nature of theseafloors and associated limited water exchange is a sig-nificant factor in the formation of hypoxia in coastal wa-ters (Diaz and Rosenberg 1995a Virtasalo et al 2005 Ra-balais et al 2010 Conley et al 2011) However to deter-

mine the degree to which seascape structure restricting wa-ter movement contributes to hypoxia formation has not beenquantified Analytical and theoretical frameworks developedspecifically for terrestrial environments such as landscapeheterogeneity or patchiness are analogous in marine envi-ronments and are equally useful for evaluating links betweenecological functions and spatial patterns in a marine contextWe tested how a large fraction of hypoxia occurrences couldbe explained only by the structural complexity of seascapeswithout knowledge on hydrographical or biogeochemical pa-rameters We adopted techniques and metrics from landscapeecology and transferred them to the marine environment andwe (1) examined if spatial patterns in seascapes can explainthe distribution of hypoxia (2) defined the relative contribu-tion of seascape structure to hypoxia formation and (3) es-timated the potential ranges of hypoxic seafloors in coastalareas To achieve this we concentrated on extremely het-erogeneous and complex archipelago areas in the northernBaltic Sea where coastal hypoxia is a common and an in-creasing problem (Conley et al 2011 Caballero-Alfonso etal 2015)

2 Data and methods

21 Study area

The studied area covers the central northern Baltic Seacoastal rim 23 500 km2 of Finnish territorial waters from theBothnian Bay to the eastern Gulf of Finland and 5100 km2

of Swedish territorial waters in the Stockholm archipelago inthe Baltic Proper Oxygen dynamics in the deeper areas ofthe Gulf of Finland and Stockholm archipelago are stronglyaffected by oceanography and biogeochemistry of the centralBaltic Proper not reflecting the dynamics of coastal hypoxia(Laine et al 1997) and were therefore excluded from thisstudy The outer archipelago of Finland is relatively exposedwith various sediment and bottom habitat types while theinner archipelago is more complex and shallower but main-tains a higher diversity of benthic habitats and sediment types(Valanko et al 2015) The inner archipelago of Stockholmis an equally complex archipelago area with a large numberof islands straits and coves Freshwater outflow from LakeMaumllaren creates an estuarine environment where freshwatermeets the more saline water in the Baltic Proper

In order to evaluate differences in oxygen deficiency be-tween coastal areas the study area was divided into fiveregions as defined by the EU Water Framework Directive(200060EC) (WFD 2000) the Archipelago Sea (AS) theeastern Gulf of Finland (EGoF) the Gulf of Bothnia (GoB)the Stockholm archipelago (SA) and the western Gulf of Fin-land (WGoF) (Fig 1) Small skerries and sheltered bays char-acterize AS EGoF and WGoF whereas the narrower band ofislands forms relatively exposed shores in GoB Deep elon-gated channels of bedrock fractures can reach depths of over

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3185

Figure 1 Study areas in Finland AS ndash Archipelago SeaWGoF ndash western Gulf of Finland EGoF ndash eastern Gulf ofFinland and GoB ndash Gulf of Bothnia Study area in Swe-den SA ndash Stockholm archipelago Orange dots represent mod-erately hypoxic sites (O2 lt 46 mg Lminus1) black dots severely hy-poxic sites (O2 lt 2 mg Lminus1) and white circles denote sites withO2 gt 46 mg Lminus1 Gray and black lines illustrate boundaries of Wa-ter Framework Directive areas

100 m in AS similarly to SA where narrow deep valleysseparate mosaics of islands and reefs Substrate in both ar-eas varies from organic-rich soft sediments in sheltered lo-cations to hard clay till and bedrock in exposed areas Atgreater depths soft sediments are common due to limitedwater movement As a whole the study area with rich topo-graphic heterogeneity forms one of the most diverse seabedareas in the world (Kaskela et al 2012) In many areas hy-poxia is a result of strong water stratification slow water ex-change and complex seabed topography creating pockets ofstagnant water (Conley et al 2011 Valanko et al 2015Jokinen et al 2018) Thus the area is ideal for testing hy-potheses of topographical controls for hypoxia formation

22 Hypoxia data

Bottom-water hypoxia is the main factor structuring benthiccommunities in the Baltic Sea (Villnas et al 2012 Norkkoet al 2015) A total of 2 mg Lminus1 of O2 is usually consid-ered a threshold where coastal organisms start to show symp-

toms of the lack of oxygen and this limit has been commonlyused in various global reviews (Diaz and Rosenberg 1995b2008) Some studies have however concluded that 2 mg Lminus1

is below the empirical sublethal and lethal oxygen limit formany species (Vaquer-Sunyer and Duarte 2008 Conley etal 2009) Here we define hypoxia based on two ecolog-ically meaningful limits moderately hypoxic lt 46 mg Lminus1

O2 ndash as this has been estimated to be a minimum safe limitfor species survival behavior and functioning in benthiccommunities (Norkko et al 2015) ndash and severely hypoxiclt 2 mg Lminus1 O2 which describes zones where larger marineorganisms suffer from severe mortality (Vaquer-Sunyer andDuarte 2008) As no reference values exist for severity ofhypoxia to marine organisms based on the frequency of hy-poxic events (Norkko et al 2012 2015 Villnas et al 2012)we here define a site to be prone to hypoxic events ieoccasionally hypoxic if it experienced hypoxia (O2 lt 2 andlt 46 mg Lminus1) at least once during the study period If hy-poxia was recorded in ge 20 of the visits it was categorizedas frequently hypoxic We consider this to be ecologicallyrelevant as species develop symptoms already from short ex-posures to hypoxia (Villnas et al 2012 Norkko et al 2015)This is also justified as our oxygen data are from sim 1 mabove the seafloor suggesting that the actual oxygen concen-trations at sediment where benthic species live are probablylower

Data from oxygen profiles were collated fromthe national monitoring environmental data portalsHertta (httpswwwsykefien-USOpen_informationOpen_web_servicesEnvironmental_data_API lastaccess 30 January 2019) and SHARK (httpswwwsmhisedataoceanografihavsmiljodata last access15 January 2019) Data were available from 808 monitoringsites Only the months of August and September 2000ndash2016were considered as hypoxia is usually a seasonal phe-nomenon occurring in late summer when water temperaturesare warmest (Conley et al 2011)

23 Predictors

For modeling hypoxia occurrences we developed five ge-omorphological metrics (1) bathymetric position indices(BPIs) with varying search radii (2) depth-attenuated waveexposure (SWM(d)) (3) topographic shelter index (TSI)(4) arc-chord rugosity (ACR) and (5) vector ruggedness mea-sure (VRM) BPI is a marine modification of the terrestrialversion of the topographic position index (TPI) originallydeveloped for terrestrial watersheds (Weiss 2001) BPI is ameasure of a bathymetric surface higher (positive values) orlower (negative values) than the overall seascape BPI valuesclose to zero are either flat areas or areas with constant slopeHere BPIs represent topographical depressions and crests atscales of 01 03 05 08 and 2 km calculated with BenthicTerrain Modeler (v30) (Walbridge et al 2018) SWM(d) es-timates dominant wave frequency at a given location with

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3185

Figure 1 Study areas in Finland AS ndash Archipelago SeaWGoF ndash western Gulf of Finland EGoF ndash eastern Gulf ofFinland and GoB ndash Gulf of Bothnia Study area in Swe-den SA ndash Stockholm archipelago Orange dots represent mod-erately hypoxic sites (O2 lt 46 mg Lminus1) black dots severely hy-poxic sites (O2 lt 2 mg Lminus1) and white circles denote sites withO2 gt 46 mg Lminus1 Gray and black lines illustrate boundaries of Wa-ter Framework Directive areas

100 m in AS similarly to SA where narrow deep valleysseparate mosaics of islands and reefs Substrate in both ar-eas varies from organic-rich soft sediments in sheltered lo-cations to hard clay till and bedrock in exposed areas Atgreater depths soft sediments are common due to limitedwater movement As a whole the study area with rich topo-graphic heterogeneity forms one of the most diverse seabedareas in the world (Kaskela et al 2012) In many areas hy-poxia is a result of strong water stratification slow water ex-change and complex seabed topography creating pockets ofstagnant water (Conley et al 2011 Valanko et al 2015Jokinen et al 2018) Thus the area is ideal for testing hy-potheses of topographical controls for hypoxia formation

22 Hypoxia data

Bottom-water hypoxia is the main factor structuring benthiccommunities in the Baltic Sea (Villnas et al 2012 Norkkoet al 2015) A total of 2 mg Lminus1 of O2 is usually consid-ered a threshold where coastal organisms start to show symp-

toms of the lack of oxygen and this limit has been commonlyused in various global reviews (Diaz and Rosenberg 1995b2008) Some studies have however concluded that 2 mg Lminus1

is below the empirical sublethal and lethal oxygen limit formany species (Vaquer-Sunyer and Duarte 2008 Conley etal 2009) Here we define hypoxia based on two ecolog-ically meaningful limits moderately hypoxic lt 46 mg Lminus1

O2 ndash as this has been estimated to be a minimum safe limitfor species survival behavior and functioning in benthiccommunities (Norkko et al 2015) ndash and severely hypoxiclt 2 mg Lminus1 O2 which describes zones where larger marineorganisms suffer from severe mortality (Vaquer-Sunyer andDuarte 2008) As no reference values exist for severity ofhypoxia to marine organisms based on the frequency of hy-poxic events (Norkko et al 2012 2015 Villnas et al 2012)we here define a site to be prone to hypoxic events ieoccasionally hypoxic if it experienced hypoxia (O2 lt 2 andlt 46 mg Lminus1) at least once during the study period If hy-poxia was recorded in ge 20 of the visits it was categorizedas frequently hypoxic We consider this to be ecologicallyrelevant as species develop symptoms already from short ex-posures to hypoxia (Villnas et al 2012 Norkko et al 2015)This is also justified as our oxygen data are from sim 1 mabove the seafloor suggesting that the actual oxygen concen-trations at sediment where benthic species live are probablylower

Data from oxygen profiles were collated fromthe national monitoring environmental data portalsHertta (httpswwwsykefien-USOpen_informationOpen_web_servicesEnvironmental_data_API lastaccess 30 January 2019) and SHARK (httpswwwsmhisedataoceanografihavsmiljodata last access15 January 2019) Data were available from 808 monitoringsites Only the months of August and September 2000ndash2016were considered as hypoxia is usually a seasonal phe-nomenon occurring in late summer when water temperaturesare warmest (Conley et al 2011)

23 Predictors

For modeling hypoxia occurrences we developed five ge-omorphological metrics (1) bathymetric position indices(BPIs) with varying search radii (2) depth-attenuated waveexposure (SWM(d)) (3) topographic shelter index (TSI)(4) arc-chord rugosity (ACR) and (5) vector ruggedness mea-sure (VRM) BPI is a marine modification of the terrestrialversion of the topographic position index (TPI) originallydeveloped for terrestrial watersheds (Weiss 2001) BPI is ameasure of a bathymetric surface higher (positive values) orlower (negative values) than the overall seascape BPI valuesclose to zero are either flat areas or areas with constant slopeHere BPIs represent topographical depressions and crests atscales of 01 03 05 08 and 2 km calculated with BenthicTerrain Modeler (v30) (Walbridge et al 2018) SWM(d) es-timates dominant wave frequency at a given location with

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3186 E A Virtanen et al Identifying areas prone to coastal hypoxia

the decay of wave exposure with depth and takes into ac-count the diffraction and refraction of waves around islands(Bekkby et al 2008) SWM(d) characterizes areas wherewater movement is slower ie where water resides longerIn terrestrial realms landform influences on windthrow pat-terns ie exposure to winds have been noted in severalstudies eg Kramer (2001) and Ashcroft et al (2008) Herewe introduce an analogous version of windthrow-prone areasto the marine realm ie a wave-prone metric topographicshelter index (TSI) which differentiates wave directions andtakes into account the sheltering effects of islands (ie ex-posure above sea level) The identification of wave-proneareas was calculated for the azimuths multiple of 15 (0ndash345) and for altitudes (corresponding to the angle of lightsource) ranging from 0125 to 81 For each altitude the pro-duced wave-prone areas were combined to an index value foreach grid cell Surface roughness is a commonly used mea-sure of topographical complexity in terrestrial studies andhas been used in marine realms as well (see eg Dunn andHalpin 2009 for modeling habitats of hard substrates andWalker et al 2009 for complexity of coral reef habitats)Here we consider two approaches for estimating seascaperugosity arc-chord rugosity (ACR) and vector ruggednessmeasure (VRM) ACR is a landscape metrics which evalu-ates surface ruggedness using a ratio of contoured area (sur-face area) to the area of a plane of best fit which is a functionof the boundary data (Du Preez 2015) VRM on the otherhand is a more conservative measure of surface roughnessdeveloped for wildlife habitat models and is calculated usinga moving 3times3 window where a unit vector orthogonal to thecell is decomposed using the three-dimensional location ofthe cell center the local slope and aspect A resultant vec-tor is divided by the number of cells in the moving window(Sappington et al 2007) Both rugosity indices were usedhere to identify areas with complex marine geomorphologyDifferences of predictor variables are illustrated in Fig 2 Wealso included geographical study areas as predictors in orderto highlight the differences between WFD areas

24 Hypoxia models

Based on the ecologically meaningful limits of hypoxia (seeSect 22) we built four separate oxygen models based onfrequency and severity of hypoxia occasional with O2 lim-its lt 46 and lt 2 mg Lminus1 (hereafter referred to as OH46and OH2) and frequent hypoxia with O2 limits lt 46 andlt 2 mg Lminus1 (hereafter referred to as FH46 and FH2) We usedthe generalized boosted regression model (GBM) and its ex-tension boosted regression trees (BRTs) a method from sta-tistical and machine learning traditions (Dersquoath and Fabri-cius 2000 Hastie et al 2001 Schapire 2003) BRT opti-mizes predictive performance through integrated stochasticgradient boosting (Natekin and Knoll 2013) and forms thebest model for prediction based on several models

Ideal model tuning parameters (learning rate bag frac-tion tree complexity) for our hypoxia models were basedon optimizing the model performance ie minimizing theprediction error The learning rate was set to 0001 to de-termine the contribution of each successive tree to the finalmodel We varied the number of decision rules controllingmodel interaction levels ie tree complexities between 3and 6 We shuffled the hypoxia data randomly into 10 sub-sets for training (70 ) and testing (30 ) while preservingthe prevalence ratio of hypoxia occurrence Hypoxia mod-els were developed based on 10-fold cross-validation andthe resulting final best models leading to the smallest pre-dictive errors were chosen to predict the probability of de-tecting hypoxia across the study region at a resolution of20 m We believe such a high resolution is necessary due tothe complexity of the archipelagoes of Finland and SwedenModel predictions for the whole seascape were repeated 10times with different model fits from data subsets (40 separatemodel predictions) to identify areas where models agree onthe area to be potentially hypoxic Model performances wereestimated against the independent data (test data 30 ) notused in model fitting in order to evaluate the potential over-fitting of the models Analyses were performed in R 350(R Core Team 2018) with R libraries gbm (Greenwell et al2018) PresenceAbsence (Freeman and Moisen 2008) andrelevant functions from Elith et al (2008)

Variable selection in BRT is internal by including only rel-evant predictors when building models The importance ofpredictors is based on the time each predictor is chosen ineach split averaged over all trees Higher scores (summedup to 100 ) indicate that a predictor has a strong influenceon the response (Elith et al 2008) Although BRT is not sen-sitive to collinearity of predictors the ability to identify thestrongest predictors by decreasing the estimated importancescore of highly correlated ones is detrimental to the inter-pretability of models (Gregorutti et al 2017) Selecting onlyan optimal and a minimal set of variables for modeling iefinding all relevant predictors and keeping the number of pre-dictors as small as possible reduces the risk of overfittingand improves the model accuracy Here most of the predic-tors describe the seascape structure and are somehow relatedto the topography of the seabed We estimated the potentialto drop redundant predictors ie those that would lead tomarked improvement in model performance if left out fromthe model building For this we used internal backward fea-ture selection in BRT However we did not find marked dif-ferences in predictive performances and used all predictors

Estimation of model fits and predictive performances wasbased on the ability to discriminate a hypoxic site from anoxic one evaluated with the area under the curve (AUC)(Jimeacutenez-Valverde and Lobo 2007) and simply with the per-cent correctly classified (PCC) (Freeman and Moisen 2008)AUC is a measure of the detection accuracy of true positives(sensitivity) and true negatives (specificity) and AUC values

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3187

Figure 2 Predictor variables developed for hypoxia ensemble models (a) vector ruggedness measure (VRM) (b) depth-attenuated waveexposure (SWM(d)) (c) topographic shelter index (TSI) and (d) bathymetric position index (BPI) with a search radius of 2 km Red colorrepresents rugged seafloors (VRM) sheltered areas (SWM(d) TSI) and depressions (BPI2) Islands are shown as white

above 09 indicate excellent of 07ndash09 indicate good andbelow 07 indicate poor predictions

We transformed hypoxia probability predictions into bi-nary classes of presenceabsence and estimated the relativearea of potentially hypoxic waters due to topographical rea-sons Although dichotomization of probability predictionsflattens the information content it facilitates the interpreta-tion of results and is needed for management purposes Thepredicted range of hypoxia and the potential geographical ex-tent enables the identification of problematic areas and fa-cilitates management actions in a cost-effective way Thereare various approaches for determining thresholds which arebased on the confusion matrix ie how well the model cap-tures truefalse presences or truefalse absences Usually thethreshold is defined to maximize the agreement between ob-served and predicted distributions Widely used thresholdssuch as 05 can be arbitrary unless the threshold equalsprevalence of presences in the data ie the frequency ofoccurrences (how many presences of the total dataset) (Liu

et al 2005) Here we define thresholds objectively basedon an agreement between predicted and observed hypoxiaprevalence This approach underestimates areas potentiallyhypoxic (see Sect 34) and is expressed here as a conserva-tive estimate

3 Results

31 Hypoxia in complex coastal archipelagos

During 2000ndash2016 hypoxia was rather common through-out the whole study region In Finland hypoxia mostly oc-curred on the southern coast as in the Archipelago Sea (AS)eastern Gulf of Finland (EGoF) and western Gulf of Fin-land (WGoF) hypoxic events were recorded frequently Onlysim 30 of coastal monitoring sites in AS EGoF and SA werenot hypoxic ie oxygen concentrations were always above46 mg Lminus1 (see percentages in brackets in Fig 3) Coastalareas in the Stockholm archipelago (SA) were also regularly

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3188 E A Virtanen et al Identifying areas prone to coastal hypoxia

hypoxic with 70 of the sites moderately (O2 lt 46 mg Lminus1)and 53 severely (O2 lt 2 mg Lminus1) hypoxic Severe hypoxiawas quite a localized phenomenon in Finland as it wasrecorded at ca 30 of sites in AS and EGoF However inWGoF there are sites where severe hypoxia is rather persis-tent as every sampling event was recorded as hypoxic Thesame applies to SA as there are quite a few sites repeat-edly severely hypoxic In contrast in the northern study areaGulf of Bothnia (GoB) hypoxic events occurred rather infre-quently as in 98 of sites O2 was above 2 mg Lminus1 and in87 O2 was above 46 mg Lminus1 (Fig 3)

32 Importance of predictors

Models developed on 566 sites based on 10-fold cross-validation and 10 repeated predictions the most influentialpredictor (averaged across models) was SWM(d) with amean contribution of 33 (plusmn5 ) (Fig 4) The contribu-tion of SWM(d) was highest for frequent severe hypoxia(FH2) (41 plusmn 3 ) whereas for occasional moderate hypoxia(OH46) the influence was markedly lower (25 plusmn 5 ) Thissupports the hypothesis that in sheltered areas where watermovement is limited severe oxygen deficiency is likely todevelop It is also noteworthy that depth was not the most im-portant driver of hypoxia in coastal areas This suggests thatcoastal hypoxia is not directly dependent on depth but thatdepressions that are especially steep and isolated are moresheltered and become more easily hypoxic than smoother de-pressions

Across models BPIs identifying wider sinks (BPI2 andBPI08) were more influential than BPIs identifying smallersinks (BPIs 01 03 and 05) and terrain ruggedness mea-sures VRM and ACR were more important for frequentsevere hypoxia (FH2) than for moderate hypoxia (FH46)The relatively high contribution of topographic shelter (TSI7 plusmn 2 ) indicates that in areas where there are higher is-lands the basins between are prone to hypoxia formation

33 Model performance

Predictive ability of models to detect sites as hypoxic acrossmodels was good with a mean 10-fold cross-validated AUCof 085 (plusmn002) and mean AUC of 086 (plusmn003) when evalu-ated against independent test data for 242 sites (30 of sites)(Fig 5a) Models classified on average 88 of sites correctly(PCCcv in Fig 5b) and performed only slightly worse whenevaluated against independent data with 81 (plusmn3 ) cor-rectly classified (PCCin in Fig 5b) Models developed forfrequent hypoxia (FH2 and FH46) were better (mean AUCin088 plusmn 003) compared to occasional hypoxia models (meanAUCin 084 plusmn 004) This suggests that other factors beyondtopographical proxies contribute relatively more to the oc-currence of occasional hypoxia than for frequent hypoxia

34 Hypoxic areas

Although hypoxia was commonly recorded in all WFD ar-eas except in the Gulf of Bothnia the potential geographi-cal extent of hypoxic seafloors shows a rather different pat-tern Based on models topographically prone areas representonly a small part of the coastal areas with less than 25 affected (Fig 6) Frequent severe hypoxia (O2 lt 2 mg Lminus1)was most prominent in the Archipelago Sea and Stockholmarchipelago although representing only a small fraction ofthe total areas (on average 15 and 37 respectively)Problematic areas based on the models are the ArchipelagoSea Stockholm archipelago and western Gulf of FinlandThose areas seem to be topographically prone to oxygen de-ficiency Moreover around 10 of areas in the eastern Gulfof Finland are vulnerable to occasional moderate hypoxia butless to severe hypoxia Areas predicted as hypoxic in the Gulfof Bothnia were less than lt 2 which supports our hypothe-sis of the facilitating role that topography potential has Thereare fewer depressions (Supplement Fig S1) and the seaflooris topographically less complex than in the other study areas

4 Discussion

Hypoxia has been increasing steadily since the 1960s andanoxic areas are seizing the seafloor suffocating marineorganisms in the way (Diaz and Rosenberg 2008 Breit-burg et al 2018) Understanding the factors affecting theseverity and spatial extent of hypoxia is essential in or-der to estimate rates of deoxygenation and its consequencesto the marine ecosystems (Breitburg et al 2018) Earlierstudies have reported coastal hypoxia to be a global phe-nomenon (Diaz and Rosenberg 2008 Conley et al 2011)and is known to be widespread in the Baltic Sea (Conleyet al 2011) Our results confirmed this and showed thatcoastal hypoxia is perhaps a more common phenomenonthan previously anticipated According to our results over50 of sites in the complex archipelagoes of Finland andSweden experienced hypoxia that is ecologically signifi-cant (O2 lt 46 mg Lminus1) Especially alarming was the inten-sity of it For instance the Stockholm archipelago sufferedfrequently from severe hypoxia (O2 lt 2 mg Lminus1) as approx-imately half of the coastal monitoring sites were hypoxicacross our study period (Fig 3) This demonstrates that de-oxygenated seafloors are probably even more common incoastal environments than previously reported (Karlsson etal 2010 Conley et al 2011) It is notable that in areas abovethe permanent halocline hypoxia is in many areas seasonaland develops after the building of thermocline in late summer(Conley et al 2011) It is therefore probable that many ofthe areas we recognized as hypoxic may well be oxygenatedduring winter and spring This does not however reduce theseverity of the phenomenon Even a hypoxic event of shortduration eg a few days will reduce ecosystem resistance

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3189

Figure 3 Frequencies of hypoxic events at coastal monitoring sites across Water Framework Directive areas Archipelago Sea (AS) easternGulf of Finland (EGoF) Gulf of Bothnia (GoB) Stockholm archipelago (SA) and western Gulf of Finland (WGoF) Upper panels indicateO2 lt 46 mg Lminus1 and lower panels O2 lt 2 mg Lminus1 Numbers in brackets indicate the percentage of sites with O2gt 46 (upper panel) andO2 gt 2 mg Lminus1 (lower panel) Number of sites over 30 are not shown

Figure 4 Importance of predictors based on 10 prediction rounds Predictors are color-coded based on models FH2 frequent severehypoxia (O2 lt 2 mg Lminus1) FH46 frequent moderate hypoxia (O2 lt 46 mg Lminus1) OH2 occasional severe hypoxia (O2 lt 2 mg Lminus1) andOH46 occasional moderate hypoxia (O2 lt 46 mg Lminus1) Whiskers represent standard deviations

to further hypoxic perturbation and affect the overall ecosys-tem functioning (Villnas et al 2013)

As our study suggests topographically prone areas to de-oxygenation represent less than 25 of seascapes Howevermost of the underwater nature values in the Finnish sea areasare concentrated on relatively shallow areas where there ex-ist enough light and suitable substrates (Virtanen et al 2018Lappalainen et al 2019) Shallow areas also suffer from eu-trophication and rising temperatures due to changing climateand are most probably the ones that are particularly suscep-tible to hypoxia in the future (Breitburg et al 2018) Thissuggests that seasonal hypoxia may become a recurrent phe-nomenon in shallow areas above the thermocline in late sum-mer

Although extensive 3-D models have been developed forthe main basins of the Baltic Sea (Meier et al 2011b 2012a2014) the previous reports on the occurrence of coastal hy-poxia have mostly been based on point observations (Conleyet al 2009 2011) Due to the lack of data and computationallimitations no biogeochemical model has (yet) encompassedthe complex Baltic Sea archipelago with a resolution neededfor adjusting local management decisions This study pro-vides a novel methodology to predict and identify areas proneto coastal hypoxia without data on currents stratification orbiological variables and without complex biogeochemicalmodels Our approach is applicable to other low-energy andnontidal systems such as large shallow bays and semien-closed or enclosed sea areas The benefit of this approach isthat it requires far less computational power than a fine-scale

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3190 E A Virtanen et al Identifying areas prone to coastal hypoxia

Figure 5 Model performances based on (a) area under the curvevalues with 10-fold cross-validation (AUCcv) and against indepen-dent test data (30 of sites) (AUCin) as well as (b) percent of cor-rectly classified with 10-fold cross-validation (PCCcv) and againstindependent data (PCCin)

Figure 6 Percent () of areas potentially affected by hypoxia withvarying frequencies (occasional and frequent) and hypoxia sever-ities (O2 lt 46 and O2 lt 2 mg Lminus1) AS Archipelago Sea EGoFeastern Gulf of Finland SA Stockholm archipelago and WGoFwestern Gulf of Finland GoB is not reported as areas potentiallyaffected were below 2 across all models

3-D numerical modeling By using relatively simple prox-ies describing depressions of stagnant water we were able tocreate detailed hypoxia maps for the entire Finnish coastalarea (23 500 km2) and Stockholm archipelago (5100 km2)thus enabling a quick view of potentially hypoxic waters(Fig 7)

We quantified the facilitating role of seafloor complexityfor the formation of hypoxia Sheltered topographically het-

Figure 7 Modeled probability () of detecting frequent hypoxiaO2 (lt 46 mg Lminus1) due to topography in an example area off thesouthern coast of Finland Land shown as white color

erogeneous areas where water exchange is limited are moresusceptible to developing hypoxia according to our resultsThis statement is quite intuitive but it has not previouslybeen quantified It is also noteworthy that coastal hypoxiais not only related to depth deep seafloors are not automati-cally hypoxic or anoxic In our study hypoxia was commonin shallow and moderate depths of 10ndash45 m For instancein the Archipelago Sea deep (60ndash100 m) channels are notusually hypoxic as strong currents tend to keep them oxy-genated throughout the year (Virtasalo et al 2005) Shallowareas can be restricted by slow water movement due to to-pographical reasons thus creating opportunities for hypoxiaformation

We emphasize that our models only indicate where hy-poxia may occur simply due to restricted water exchangeAny deviations from this pattern are probably caused by ei-ther hydrographic factors which the hypoxia model basedon topography did not account for (such as strong currentsin elongated wide channels) or biogeochemical factors Es-pecially high external loading and local biogeochemical andbiological processes (nutrient cycling between sediment andthe water) obviously modifies the patterns and severity ofhypoxia also in the coastal areas Oxygen deficiency has beenprojected to increase faster in the coastal systems than in theopen sea (Gilbert et al 2010 Altieri and Gedan 2015) Suchcoastal areas are usually affected by external nutrient loadingfrom the watershed In larger basins and sea areas dominatedby large river systems such as the central Baltic Sea Gulfof Mexico and Chesapeake Bay large-scale oceanographicand biogeochemical processes or external loading governthe depth and extent of hypoxia This is the case also in theBaltic Sea In areas where our models underestimate oxygendeficiency major nutrient sources eg rivers cities or inten-sive agricultural areas probably contribute to hypoxia for-mation However in extremely complex archipelago areas

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3187

Figure 2 Predictor variables developed for hypoxia ensemble models (a) vector ruggedness measure (VRM) (b) depth-attenuated waveexposure (SWM(d)) (c) topographic shelter index (TSI) and (d) bathymetric position index (BPI) with a search radius of 2 km Red colorrepresents rugged seafloors (VRM) sheltered areas (SWM(d) TSI) and depressions (BPI2) Islands are shown as white

above 09 indicate excellent of 07ndash09 indicate good andbelow 07 indicate poor predictions

We transformed hypoxia probability predictions into bi-nary classes of presenceabsence and estimated the relativearea of potentially hypoxic waters due to topographical rea-sons Although dichotomization of probability predictionsflattens the information content it facilitates the interpreta-tion of results and is needed for management purposes Thepredicted range of hypoxia and the potential geographical ex-tent enables the identification of problematic areas and fa-cilitates management actions in a cost-effective way Thereare various approaches for determining thresholds which arebased on the confusion matrix ie how well the model cap-tures truefalse presences or truefalse absences Usually thethreshold is defined to maximize the agreement between ob-served and predicted distributions Widely used thresholdssuch as 05 can be arbitrary unless the threshold equalsprevalence of presences in the data ie the frequency ofoccurrences (how many presences of the total dataset) (Liu

et al 2005) Here we define thresholds objectively basedon an agreement between predicted and observed hypoxiaprevalence This approach underestimates areas potentiallyhypoxic (see Sect 34) and is expressed here as a conserva-tive estimate

3 Results

31 Hypoxia in complex coastal archipelagos

During 2000ndash2016 hypoxia was rather common through-out the whole study region In Finland hypoxia mostly oc-curred on the southern coast as in the Archipelago Sea (AS)eastern Gulf of Finland (EGoF) and western Gulf of Fin-land (WGoF) hypoxic events were recorded frequently Onlysim 30 of coastal monitoring sites in AS EGoF and SA werenot hypoxic ie oxygen concentrations were always above46 mg Lminus1 (see percentages in brackets in Fig 3) Coastalareas in the Stockholm archipelago (SA) were also regularly

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3188 E A Virtanen et al Identifying areas prone to coastal hypoxia

hypoxic with 70 of the sites moderately (O2 lt 46 mg Lminus1)and 53 severely (O2 lt 2 mg Lminus1) hypoxic Severe hypoxiawas quite a localized phenomenon in Finland as it wasrecorded at ca 30 of sites in AS and EGoF However inWGoF there are sites where severe hypoxia is rather persis-tent as every sampling event was recorded as hypoxic Thesame applies to SA as there are quite a few sites repeat-edly severely hypoxic In contrast in the northern study areaGulf of Bothnia (GoB) hypoxic events occurred rather infre-quently as in 98 of sites O2 was above 2 mg Lminus1 and in87 O2 was above 46 mg Lminus1 (Fig 3)

32 Importance of predictors

Models developed on 566 sites based on 10-fold cross-validation and 10 repeated predictions the most influentialpredictor (averaged across models) was SWM(d) with amean contribution of 33 (plusmn5 ) (Fig 4) The contribu-tion of SWM(d) was highest for frequent severe hypoxia(FH2) (41 plusmn 3 ) whereas for occasional moderate hypoxia(OH46) the influence was markedly lower (25 plusmn 5 ) Thissupports the hypothesis that in sheltered areas where watermovement is limited severe oxygen deficiency is likely todevelop It is also noteworthy that depth was not the most im-portant driver of hypoxia in coastal areas This suggests thatcoastal hypoxia is not directly dependent on depth but thatdepressions that are especially steep and isolated are moresheltered and become more easily hypoxic than smoother de-pressions

Across models BPIs identifying wider sinks (BPI2 andBPI08) were more influential than BPIs identifying smallersinks (BPIs 01 03 and 05) and terrain ruggedness mea-sures VRM and ACR were more important for frequentsevere hypoxia (FH2) than for moderate hypoxia (FH46)The relatively high contribution of topographic shelter (TSI7 plusmn 2 ) indicates that in areas where there are higher is-lands the basins between are prone to hypoxia formation

33 Model performance

Predictive ability of models to detect sites as hypoxic acrossmodels was good with a mean 10-fold cross-validated AUCof 085 (plusmn002) and mean AUC of 086 (plusmn003) when evalu-ated against independent test data for 242 sites (30 of sites)(Fig 5a) Models classified on average 88 of sites correctly(PCCcv in Fig 5b) and performed only slightly worse whenevaluated against independent data with 81 (plusmn3 ) cor-rectly classified (PCCin in Fig 5b) Models developed forfrequent hypoxia (FH2 and FH46) were better (mean AUCin088 plusmn 003) compared to occasional hypoxia models (meanAUCin 084 plusmn 004) This suggests that other factors beyondtopographical proxies contribute relatively more to the oc-currence of occasional hypoxia than for frequent hypoxia

34 Hypoxic areas

Although hypoxia was commonly recorded in all WFD ar-eas except in the Gulf of Bothnia the potential geographi-cal extent of hypoxic seafloors shows a rather different pat-tern Based on models topographically prone areas representonly a small part of the coastal areas with less than 25 affected (Fig 6) Frequent severe hypoxia (O2 lt 2 mg Lminus1)was most prominent in the Archipelago Sea and Stockholmarchipelago although representing only a small fraction ofthe total areas (on average 15 and 37 respectively)Problematic areas based on the models are the ArchipelagoSea Stockholm archipelago and western Gulf of FinlandThose areas seem to be topographically prone to oxygen de-ficiency Moreover around 10 of areas in the eastern Gulfof Finland are vulnerable to occasional moderate hypoxia butless to severe hypoxia Areas predicted as hypoxic in the Gulfof Bothnia were less than lt 2 which supports our hypothe-sis of the facilitating role that topography potential has Thereare fewer depressions (Supplement Fig S1) and the seaflooris topographically less complex than in the other study areas

4 Discussion

Hypoxia has been increasing steadily since the 1960s andanoxic areas are seizing the seafloor suffocating marineorganisms in the way (Diaz and Rosenberg 2008 Breit-burg et al 2018) Understanding the factors affecting theseverity and spatial extent of hypoxia is essential in or-der to estimate rates of deoxygenation and its consequencesto the marine ecosystems (Breitburg et al 2018) Earlierstudies have reported coastal hypoxia to be a global phe-nomenon (Diaz and Rosenberg 2008 Conley et al 2011)and is known to be widespread in the Baltic Sea (Conleyet al 2011) Our results confirmed this and showed thatcoastal hypoxia is perhaps a more common phenomenonthan previously anticipated According to our results over50 of sites in the complex archipelagoes of Finland andSweden experienced hypoxia that is ecologically signifi-cant (O2 lt 46 mg Lminus1) Especially alarming was the inten-sity of it For instance the Stockholm archipelago sufferedfrequently from severe hypoxia (O2 lt 2 mg Lminus1) as approx-imately half of the coastal monitoring sites were hypoxicacross our study period (Fig 3) This demonstrates that de-oxygenated seafloors are probably even more common incoastal environments than previously reported (Karlsson etal 2010 Conley et al 2011) It is notable that in areas abovethe permanent halocline hypoxia is in many areas seasonaland develops after the building of thermocline in late summer(Conley et al 2011) It is therefore probable that many ofthe areas we recognized as hypoxic may well be oxygenatedduring winter and spring This does not however reduce theseverity of the phenomenon Even a hypoxic event of shortduration eg a few days will reduce ecosystem resistance

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3189

Figure 3 Frequencies of hypoxic events at coastal monitoring sites across Water Framework Directive areas Archipelago Sea (AS) easternGulf of Finland (EGoF) Gulf of Bothnia (GoB) Stockholm archipelago (SA) and western Gulf of Finland (WGoF) Upper panels indicateO2 lt 46 mg Lminus1 and lower panels O2 lt 2 mg Lminus1 Numbers in brackets indicate the percentage of sites with O2gt 46 (upper panel) andO2 gt 2 mg Lminus1 (lower panel) Number of sites over 30 are not shown

Figure 4 Importance of predictors based on 10 prediction rounds Predictors are color-coded based on models FH2 frequent severehypoxia (O2 lt 2 mg Lminus1) FH46 frequent moderate hypoxia (O2 lt 46 mg Lminus1) OH2 occasional severe hypoxia (O2 lt 2 mg Lminus1) andOH46 occasional moderate hypoxia (O2 lt 46 mg Lminus1) Whiskers represent standard deviations

to further hypoxic perturbation and affect the overall ecosys-tem functioning (Villnas et al 2013)

As our study suggests topographically prone areas to de-oxygenation represent less than 25 of seascapes Howevermost of the underwater nature values in the Finnish sea areasare concentrated on relatively shallow areas where there ex-ist enough light and suitable substrates (Virtanen et al 2018Lappalainen et al 2019) Shallow areas also suffer from eu-trophication and rising temperatures due to changing climateand are most probably the ones that are particularly suscep-tible to hypoxia in the future (Breitburg et al 2018) Thissuggests that seasonal hypoxia may become a recurrent phe-nomenon in shallow areas above the thermocline in late sum-mer

Although extensive 3-D models have been developed forthe main basins of the Baltic Sea (Meier et al 2011b 2012a2014) the previous reports on the occurrence of coastal hy-poxia have mostly been based on point observations (Conleyet al 2009 2011) Due to the lack of data and computationallimitations no biogeochemical model has (yet) encompassedthe complex Baltic Sea archipelago with a resolution neededfor adjusting local management decisions This study pro-vides a novel methodology to predict and identify areas proneto coastal hypoxia without data on currents stratification orbiological variables and without complex biogeochemicalmodels Our approach is applicable to other low-energy andnontidal systems such as large shallow bays and semien-closed or enclosed sea areas The benefit of this approach isthat it requires far less computational power than a fine-scale

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3190 E A Virtanen et al Identifying areas prone to coastal hypoxia

Figure 5 Model performances based on (a) area under the curvevalues with 10-fold cross-validation (AUCcv) and against indepen-dent test data (30 of sites) (AUCin) as well as (b) percent of cor-rectly classified with 10-fold cross-validation (PCCcv) and againstindependent data (PCCin)

Figure 6 Percent () of areas potentially affected by hypoxia withvarying frequencies (occasional and frequent) and hypoxia sever-ities (O2 lt 46 and O2 lt 2 mg Lminus1) AS Archipelago Sea EGoFeastern Gulf of Finland SA Stockholm archipelago and WGoFwestern Gulf of Finland GoB is not reported as areas potentiallyaffected were below 2 across all models

3-D numerical modeling By using relatively simple prox-ies describing depressions of stagnant water we were able tocreate detailed hypoxia maps for the entire Finnish coastalarea (23 500 km2) and Stockholm archipelago (5100 km2)thus enabling a quick view of potentially hypoxic waters(Fig 7)

We quantified the facilitating role of seafloor complexityfor the formation of hypoxia Sheltered topographically het-

Figure 7 Modeled probability () of detecting frequent hypoxiaO2 (lt 46 mg Lminus1) due to topography in an example area off thesouthern coast of Finland Land shown as white color

erogeneous areas where water exchange is limited are moresusceptible to developing hypoxia according to our resultsThis statement is quite intuitive but it has not previouslybeen quantified It is also noteworthy that coastal hypoxiais not only related to depth deep seafloors are not automati-cally hypoxic or anoxic In our study hypoxia was commonin shallow and moderate depths of 10ndash45 m For instancein the Archipelago Sea deep (60ndash100 m) channels are notusually hypoxic as strong currents tend to keep them oxy-genated throughout the year (Virtasalo et al 2005) Shallowareas can be restricted by slow water movement due to to-pographical reasons thus creating opportunities for hypoxiaformation

We emphasize that our models only indicate where hy-poxia may occur simply due to restricted water exchangeAny deviations from this pattern are probably caused by ei-ther hydrographic factors which the hypoxia model basedon topography did not account for (such as strong currentsin elongated wide channels) or biogeochemical factors Es-pecially high external loading and local biogeochemical andbiological processes (nutrient cycling between sediment andthe water) obviously modifies the patterns and severity ofhypoxia also in the coastal areas Oxygen deficiency has beenprojected to increase faster in the coastal systems than in theopen sea (Gilbert et al 2010 Altieri and Gedan 2015) Suchcoastal areas are usually affected by external nutrient loadingfrom the watershed In larger basins and sea areas dominatedby large river systems such as the central Baltic Sea Gulfof Mexico and Chesapeake Bay large-scale oceanographicand biogeochemical processes or external loading governthe depth and extent of hypoxia This is the case also in theBaltic Sea In areas where our models underestimate oxygendeficiency major nutrient sources eg rivers cities or inten-sive agricultural areas probably contribute to hypoxia for-mation However in extremely complex archipelago areas

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

3188 E A Virtanen et al Identifying areas prone to coastal hypoxia

hypoxic with 70 of the sites moderately (O2 lt 46 mg Lminus1)and 53 severely (O2 lt 2 mg Lminus1) hypoxic Severe hypoxiawas quite a localized phenomenon in Finland as it wasrecorded at ca 30 of sites in AS and EGoF However inWGoF there are sites where severe hypoxia is rather persis-tent as every sampling event was recorded as hypoxic Thesame applies to SA as there are quite a few sites repeat-edly severely hypoxic In contrast in the northern study areaGulf of Bothnia (GoB) hypoxic events occurred rather infre-quently as in 98 of sites O2 was above 2 mg Lminus1 and in87 O2 was above 46 mg Lminus1 (Fig 3)

32 Importance of predictors

Models developed on 566 sites based on 10-fold cross-validation and 10 repeated predictions the most influentialpredictor (averaged across models) was SWM(d) with amean contribution of 33 (plusmn5 ) (Fig 4) The contribu-tion of SWM(d) was highest for frequent severe hypoxia(FH2) (41 plusmn 3 ) whereas for occasional moderate hypoxia(OH46) the influence was markedly lower (25 plusmn 5 ) Thissupports the hypothesis that in sheltered areas where watermovement is limited severe oxygen deficiency is likely todevelop It is also noteworthy that depth was not the most im-portant driver of hypoxia in coastal areas This suggests thatcoastal hypoxia is not directly dependent on depth but thatdepressions that are especially steep and isolated are moresheltered and become more easily hypoxic than smoother de-pressions

Across models BPIs identifying wider sinks (BPI2 andBPI08) were more influential than BPIs identifying smallersinks (BPIs 01 03 and 05) and terrain ruggedness mea-sures VRM and ACR were more important for frequentsevere hypoxia (FH2) than for moderate hypoxia (FH46)The relatively high contribution of topographic shelter (TSI7 plusmn 2 ) indicates that in areas where there are higher is-lands the basins between are prone to hypoxia formation

33 Model performance

Predictive ability of models to detect sites as hypoxic acrossmodels was good with a mean 10-fold cross-validated AUCof 085 (plusmn002) and mean AUC of 086 (plusmn003) when evalu-ated against independent test data for 242 sites (30 of sites)(Fig 5a) Models classified on average 88 of sites correctly(PCCcv in Fig 5b) and performed only slightly worse whenevaluated against independent data with 81 (plusmn3 ) cor-rectly classified (PCCin in Fig 5b) Models developed forfrequent hypoxia (FH2 and FH46) were better (mean AUCin088 plusmn 003) compared to occasional hypoxia models (meanAUCin 084 plusmn 004) This suggests that other factors beyondtopographical proxies contribute relatively more to the oc-currence of occasional hypoxia than for frequent hypoxia

34 Hypoxic areas

Although hypoxia was commonly recorded in all WFD ar-eas except in the Gulf of Bothnia the potential geographi-cal extent of hypoxic seafloors shows a rather different pat-tern Based on models topographically prone areas representonly a small part of the coastal areas with less than 25 affected (Fig 6) Frequent severe hypoxia (O2 lt 2 mg Lminus1)was most prominent in the Archipelago Sea and Stockholmarchipelago although representing only a small fraction ofthe total areas (on average 15 and 37 respectively)Problematic areas based on the models are the ArchipelagoSea Stockholm archipelago and western Gulf of FinlandThose areas seem to be topographically prone to oxygen de-ficiency Moreover around 10 of areas in the eastern Gulfof Finland are vulnerable to occasional moderate hypoxia butless to severe hypoxia Areas predicted as hypoxic in the Gulfof Bothnia were less than lt 2 which supports our hypothe-sis of the facilitating role that topography potential has Thereare fewer depressions (Supplement Fig S1) and the seaflooris topographically less complex than in the other study areas

4 Discussion

Hypoxia has been increasing steadily since the 1960s andanoxic areas are seizing the seafloor suffocating marineorganisms in the way (Diaz and Rosenberg 2008 Breit-burg et al 2018) Understanding the factors affecting theseverity and spatial extent of hypoxia is essential in or-der to estimate rates of deoxygenation and its consequencesto the marine ecosystems (Breitburg et al 2018) Earlierstudies have reported coastal hypoxia to be a global phe-nomenon (Diaz and Rosenberg 2008 Conley et al 2011)and is known to be widespread in the Baltic Sea (Conleyet al 2011) Our results confirmed this and showed thatcoastal hypoxia is perhaps a more common phenomenonthan previously anticipated According to our results over50 of sites in the complex archipelagoes of Finland andSweden experienced hypoxia that is ecologically signifi-cant (O2 lt 46 mg Lminus1) Especially alarming was the inten-sity of it For instance the Stockholm archipelago sufferedfrequently from severe hypoxia (O2 lt 2 mg Lminus1) as approx-imately half of the coastal monitoring sites were hypoxicacross our study period (Fig 3) This demonstrates that de-oxygenated seafloors are probably even more common incoastal environments than previously reported (Karlsson etal 2010 Conley et al 2011) It is notable that in areas abovethe permanent halocline hypoxia is in many areas seasonaland develops after the building of thermocline in late summer(Conley et al 2011) It is therefore probable that many ofthe areas we recognized as hypoxic may well be oxygenatedduring winter and spring This does not however reduce theseverity of the phenomenon Even a hypoxic event of shortduration eg a few days will reduce ecosystem resistance

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3189

Figure 3 Frequencies of hypoxic events at coastal monitoring sites across Water Framework Directive areas Archipelago Sea (AS) easternGulf of Finland (EGoF) Gulf of Bothnia (GoB) Stockholm archipelago (SA) and western Gulf of Finland (WGoF) Upper panels indicateO2 lt 46 mg Lminus1 and lower panels O2 lt 2 mg Lminus1 Numbers in brackets indicate the percentage of sites with O2gt 46 (upper panel) andO2 gt 2 mg Lminus1 (lower panel) Number of sites over 30 are not shown

Figure 4 Importance of predictors based on 10 prediction rounds Predictors are color-coded based on models FH2 frequent severehypoxia (O2 lt 2 mg Lminus1) FH46 frequent moderate hypoxia (O2 lt 46 mg Lminus1) OH2 occasional severe hypoxia (O2 lt 2 mg Lminus1) andOH46 occasional moderate hypoxia (O2 lt 46 mg Lminus1) Whiskers represent standard deviations

to further hypoxic perturbation and affect the overall ecosys-tem functioning (Villnas et al 2013)

As our study suggests topographically prone areas to de-oxygenation represent less than 25 of seascapes Howevermost of the underwater nature values in the Finnish sea areasare concentrated on relatively shallow areas where there ex-ist enough light and suitable substrates (Virtanen et al 2018Lappalainen et al 2019) Shallow areas also suffer from eu-trophication and rising temperatures due to changing climateand are most probably the ones that are particularly suscep-tible to hypoxia in the future (Breitburg et al 2018) Thissuggests that seasonal hypoxia may become a recurrent phe-nomenon in shallow areas above the thermocline in late sum-mer

Although extensive 3-D models have been developed forthe main basins of the Baltic Sea (Meier et al 2011b 2012a2014) the previous reports on the occurrence of coastal hy-poxia have mostly been based on point observations (Conleyet al 2009 2011) Due to the lack of data and computationallimitations no biogeochemical model has (yet) encompassedthe complex Baltic Sea archipelago with a resolution neededfor adjusting local management decisions This study pro-vides a novel methodology to predict and identify areas proneto coastal hypoxia without data on currents stratification orbiological variables and without complex biogeochemicalmodels Our approach is applicable to other low-energy andnontidal systems such as large shallow bays and semien-closed or enclosed sea areas The benefit of this approach isthat it requires far less computational power than a fine-scale

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3190 E A Virtanen et al Identifying areas prone to coastal hypoxia

Figure 5 Model performances based on (a) area under the curvevalues with 10-fold cross-validation (AUCcv) and against indepen-dent test data (30 of sites) (AUCin) as well as (b) percent of cor-rectly classified with 10-fold cross-validation (PCCcv) and againstindependent data (PCCin)

Figure 6 Percent () of areas potentially affected by hypoxia withvarying frequencies (occasional and frequent) and hypoxia sever-ities (O2 lt 46 and O2 lt 2 mg Lminus1) AS Archipelago Sea EGoFeastern Gulf of Finland SA Stockholm archipelago and WGoFwestern Gulf of Finland GoB is not reported as areas potentiallyaffected were below 2 across all models

3-D numerical modeling By using relatively simple prox-ies describing depressions of stagnant water we were able tocreate detailed hypoxia maps for the entire Finnish coastalarea (23 500 km2) and Stockholm archipelago (5100 km2)thus enabling a quick view of potentially hypoxic waters(Fig 7)

We quantified the facilitating role of seafloor complexityfor the formation of hypoxia Sheltered topographically het-

Figure 7 Modeled probability () of detecting frequent hypoxiaO2 (lt 46 mg Lminus1) due to topography in an example area off thesouthern coast of Finland Land shown as white color

erogeneous areas where water exchange is limited are moresusceptible to developing hypoxia according to our resultsThis statement is quite intuitive but it has not previouslybeen quantified It is also noteworthy that coastal hypoxiais not only related to depth deep seafloors are not automati-cally hypoxic or anoxic In our study hypoxia was commonin shallow and moderate depths of 10ndash45 m For instancein the Archipelago Sea deep (60ndash100 m) channels are notusually hypoxic as strong currents tend to keep them oxy-genated throughout the year (Virtasalo et al 2005) Shallowareas can be restricted by slow water movement due to to-pographical reasons thus creating opportunities for hypoxiaformation

We emphasize that our models only indicate where hy-poxia may occur simply due to restricted water exchangeAny deviations from this pattern are probably caused by ei-ther hydrographic factors which the hypoxia model basedon topography did not account for (such as strong currentsin elongated wide channels) or biogeochemical factors Es-pecially high external loading and local biogeochemical andbiological processes (nutrient cycling between sediment andthe water) obviously modifies the patterns and severity ofhypoxia also in the coastal areas Oxygen deficiency has beenprojected to increase faster in the coastal systems than in theopen sea (Gilbert et al 2010 Altieri and Gedan 2015) Suchcoastal areas are usually affected by external nutrient loadingfrom the watershed In larger basins and sea areas dominatedby large river systems such as the central Baltic Sea Gulfof Mexico and Chesapeake Bay large-scale oceanographicand biogeochemical processes or external loading governthe depth and extent of hypoxia This is the case also in theBaltic Sea In areas where our models underestimate oxygendeficiency major nutrient sources eg rivers cities or inten-sive agricultural areas probably contribute to hypoxia for-mation However in extremely complex archipelago areas

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3189

Figure 3 Frequencies of hypoxic events at coastal monitoring sites across Water Framework Directive areas Archipelago Sea (AS) easternGulf of Finland (EGoF) Gulf of Bothnia (GoB) Stockholm archipelago (SA) and western Gulf of Finland (WGoF) Upper panels indicateO2 lt 46 mg Lminus1 and lower panels O2 lt 2 mg Lminus1 Numbers in brackets indicate the percentage of sites with O2gt 46 (upper panel) andO2 gt 2 mg Lminus1 (lower panel) Number of sites over 30 are not shown

Figure 4 Importance of predictors based on 10 prediction rounds Predictors are color-coded based on models FH2 frequent severehypoxia (O2 lt 2 mg Lminus1) FH46 frequent moderate hypoxia (O2 lt 46 mg Lminus1) OH2 occasional severe hypoxia (O2 lt 2 mg Lminus1) andOH46 occasional moderate hypoxia (O2 lt 46 mg Lminus1) Whiskers represent standard deviations

to further hypoxic perturbation and affect the overall ecosys-tem functioning (Villnas et al 2013)

As our study suggests topographically prone areas to de-oxygenation represent less than 25 of seascapes Howevermost of the underwater nature values in the Finnish sea areasare concentrated on relatively shallow areas where there ex-ist enough light and suitable substrates (Virtanen et al 2018Lappalainen et al 2019) Shallow areas also suffer from eu-trophication and rising temperatures due to changing climateand are most probably the ones that are particularly suscep-tible to hypoxia in the future (Breitburg et al 2018) Thissuggests that seasonal hypoxia may become a recurrent phe-nomenon in shallow areas above the thermocline in late sum-mer

Although extensive 3-D models have been developed forthe main basins of the Baltic Sea (Meier et al 2011b 2012a2014) the previous reports on the occurrence of coastal hy-poxia have mostly been based on point observations (Conleyet al 2009 2011) Due to the lack of data and computationallimitations no biogeochemical model has (yet) encompassedthe complex Baltic Sea archipelago with a resolution neededfor adjusting local management decisions This study pro-vides a novel methodology to predict and identify areas proneto coastal hypoxia without data on currents stratification orbiological variables and without complex biogeochemicalmodels Our approach is applicable to other low-energy andnontidal systems such as large shallow bays and semien-closed or enclosed sea areas The benefit of this approach isthat it requires far less computational power than a fine-scale

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3190 E A Virtanen et al Identifying areas prone to coastal hypoxia

Figure 5 Model performances based on (a) area under the curvevalues with 10-fold cross-validation (AUCcv) and against indepen-dent test data (30 of sites) (AUCin) as well as (b) percent of cor-rectly classified with 10-fold cross-validation (PCCcv) and againstindependent data (PCCin)

Figure 6 Percent () of areas potentially affected by hypoxia withvarying frequencies (occasional and frequent) and hypoxia sever-ities (O2 lt 46 and O2 lt 2 mg Lminus1) AS Archipelago Sea EGoFeastern Gulf of Finland SA Stockholm archipelago and WGoFwestern Gulf of Finland GoB is not reported as areas potentiallyaffected were below 2 across all models

3-D numerical modeling By using relatively simple prox-ies describing depressions of stagnant water we were able tocreate detailed hypoxia maps for the entire Finnish coastalarea (23 500 km2) and Stockholm archipelago (5100 km2)thus enabling a quick view of potentially hypoxic waters(Fig 7)

We quantified the facilitating role of seafloor complexityfor the formation of hypoxia Sheltered topographically het-

Figure 7 Modeled probability () of detecting frequent hypoxiaO2 (lt 46 mg Lminus1) due to topography in an example area off thesouthern coast of Finland Land shown as white color

erogeneous areas where water exchange is limited are moresusceptible to developing hypoxia according to our resultsThis statement is quite intuitive but it has not previouslybeen quantified It is also noteworthy that coastal hypoxiais not only related to depth deep seafloors are not automati-cally hypoxic or anoxic In our study hypoxia was commonin shallow and moderate depths of 10ndash45 m For instancein the Archipelago Sea deep (60ndash100 m) channels are notusually hypoxic as strong currents tend to keep them oxy-genated throughout the year (Virtasalo et al 2005) Shallowareas can be restricted by slow water movement due to to-pographical reasons thus creating opportunities for hypoxiaformation

We emphasize that our models only indicate where hy-poxia may occur simply due to restricted water exchangeAny deviations from this pattern are probably caused by ei-ther hydrographic factors which the hypoxia model basedon topography did not account for (such as strong currentsin elongated wide channels) or biogeochemical factors Es-pecially high external loading and local biogeochemical andbiological processes (nutrient cycling between sediment andthe water) obviously modifies the patterns and severity ofhypoxia also in the coastal areas Oxygen deficiency has beenprojected to increase faster in the coastal systems than in theopen sea (Gilbert et al 2010 Altieri and Gedan 2015) Suchcoastal areas are usually affected by external nutrient loadingfrom the watershed In larger basins and sea areas dominatedby large river systems such as the central Baltic Sea Gulfof Mexico and Chesapeake Bay large-scale oceanographicand biogeochemical processes or external loading governthe depth and extent of hypoxia This is the case also in theBaltic Sea In areas where our models underestimate oxygendeficiency major nutrient sources eg rivers cities or inten-sive agricultural areas probably contribute to hypoxia for-mation However in extremely complex archipelago areas

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

3190 E A Virtanen et al Identifying areas prone to coastal hypoxia

Figure 5 Model performances based on (a) area under the curvevalues with 10-fold cross-validation (AUCcv) and against indepen-dent test data (30 of sites) (AUCin) as well as (b) percent of cor-rectly classified with 10-fold cross-validation (PCCcv) and againstindependent data (PCCin)

Figure 6 Percent () of areas potentially affected by hypoxia withvarying frequencies (occasional and frequent) and hypoxia sever-ities (O2 lt 46 and O2 lt 2 mg Lminus1) AS Archipelago Sea EGoFeastern Gulf of Finland SA Stockholm archipelago and WGoFwestern Gulf of Finland GoB is not reported as areas potentiallyaffected were below 2 across all models

3-D numerical modeling By using relatively simple prox-ies describing depressions of stagnant water we were able tocreate detailed hypoxia maps for the entire Finnish coastalarea (23 500 km2) and Stockholm archipelago (5100 km2)thus enabling a quick view of potentially hypoxic waters(Fig 7)

We quantified the facilitating role of seafloor complexityfor the formation of hypoxia Sheltered topographically het-

Figure 7 Modeled probability () of detecting frequent hypoxiaO2 (lt 46 mg Lminus1) due to topography in an example area off thesouthern coast of Finland Land shown as white color

erogeneous areas where water exchange is limited are moresusceptible to developing hypoxia according to our resultsThis statement is quite intuitive but it has not previouslybeen quantified It is also noteworthy that coastal hypoxiais not only related to depth deep seafloors are not automati-cally hypoxic or anoxic In our study hypoxia was commonin shallow and moderate depths of 10ndash45 m For instancein the Archipelago Sea deep (60ndash100 m) channels are notusually hypoxic as strong currents tend to keep them oxy-genated throughout the year (Virtasalo et al 2005) Shallowareas can be restricted by slow water movement due to to-pographical reasons thus creating opportunities for hypoxiaformation

We emphasize that our models only indicate where hy-poxia may occur simply due to restricted water exchangeAny deviations from this pattern are probably caused by ei-ther hydrographic factors which the hypoxia model basedon topography did not account for (such as strong currentsin elongated wide channels) or biogeochemical factors Es-pecially high external loading and local biogeochemical andbiological processes (nutrient cycling between sediment andthe water) obviously modifies the patterns and severity ofhypoxia also in the coastal areas Oxygen deficiency has beenprojected to increase faster in the coastal systems than in theopen sea (Gilbert et al 2010 Altieri and Gedan 2015) Suchcoastal areas are usually affected by external nutrient loadingfrom the watershed In larger basins and sea areas dominatedby large river systems such as the central Baltic Sea Gulfof Mexico and Chesapeake Bay large-scale oceanographicand biogeochemical processes or external loading governthe depth and extent of hypoxia This is the case also in theBaltic Sea In areas where our models underestimate oxygendeficiency major nutrient sources eg rivers cities or inten-sive agricultural areas probably contribute to hypoxia for-mation However in extremely complex archipelago areas

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3191

such as the Finnish and Swedish archipelago physical fac-tors limiting lateral and vertical movement of water probablyfacilitate and in some areas even dictate the development ofhypoxia

There were spatial differences in the frequency and sever-ity of hypoxia that can be explained by topographical char-acteristics of the areas external loading and interaction withthe adjacent deeper basins For instance in the Stockholmarchipelago severe hypoxia covered the largest percentage ofseascapes of all study areas The Stockholm archipelago ispart of a joint valley landscape with deep steep areas also inthe inner parts where wave forcing is exceptionally low anddisconnected from the open sea making it very susceptible tohypoxia which was also confirmed by the model In Finlandthe inner archipelago is mostly shallow with steep but widerchannels occurring only in the Archipelago Sea These elon-gated channels are connected to adjacent open-sea areas andthus well-ventilated as opposed to the narrow channels ofthe Stockholm archipelago Geographically hypoxia in Fin-land was most prominent in the Archipelago Sea and the Gulfof Finland where the inner archipelago is isolated from theopen sea and the complex topography results in overall poorwater exchange in the existing depressions Both the Stock-holm archipelago and the Archipelago Sea suffer from exter-nal loading from the associated watersheds and internal load-ing from sediments which probably contributes to the pooroxygen status of these areas (Puttonen et al 2014 Walveet al 2018) Biogeochemical factors were however not ac-counted for by our analysis and cannot be used in explainingthe observed spatial differences

In the Gulf of Finland eutrophication increases in theopen sea from west to east which has traditionally been ex-plained by nutrient discharges from the Neva River (HEL-COM 2018) In our data there was however no clear gradi-ent of coastal hypoxia increasing towards east In contrastfrequent hypoxia was more common in the Archipelago Seaand western Gulf of Finland than in the eastern Gulf of Fin-land where hypoxia occurred only occasionally This sug-gests that the coastal hypoxia is more dependent on local pro-cesses ie internal loading and external loading from nearbyareas whereas open-sea hypoxia is governed by basin-scaledynamics However the occasional nature of hypoxia in theeastern Gulf of Finland may be at least partly caused by thedependency on the deep waters of the open parts of the Gulfof Finland The Gulf of Finland is an embayment 400 kmlong and 50ndash120 km wide which has an open western bound-ary to the Baltic Proper A tongue of anoxic water usuallyextends from the central Baltic Sea into the Gulf of Finlandalong its deepest parts Basin-scale oceanographic and atmo-spheric processes influence how far east this tongue proceedsinto the Gulf of Finland each year (Alenius et al 2016) It ispossible that when this anoxic tongue extends close to east-ern Gulf of Finland it also worsens the oxygen situation ofthe EGoF archipelago

In the Gulf of Bothnia hypoxia was markedly less fre-quent and severe than in the other study areas GoB has a rel-atively open coastline with only few depressions (see Fig S1)and strong wave forcing which probably enhances the mix-ing of water in the coastal areas Moreover as the open-seaareas of the Gulf of Bothnia are well oxygenated due to alack of halocline and topographical isolation of GoB from theBaltic Proper (by the sill between these basins) (Leppaumlrantaand Myrberg 2009) hypoxic water is not advected from theopen sea to the coastal areas

Such observations suggest that formation of coastal hy-poxia is not totally independent from basin-scale oxygen dy-namics While we suggest that coastal hypoxia can be formedentirely based on local morphology and local biogeochemi-cal processes the relatively low occurrence of hypoxia in theGulf of Bothnia and differences in frequency of hypoxia indifferent parts of the Gulf of Finland both highlight the inter-action of these coastal areas with the Baltic Proper

While our results confirm that hypoxia in most study ar-eas is a frequently occurring phenomenon they also showthat areas affected by hypoxia are geographically still lim-ited Our modeling results indicate that overall less than25 of the studied sea areas were afflicted by some formof hypoxia (be it recurrent or occasional) and less than 6 of seascapes were plagued by frequent severe hypoxia Therelatively small spatial extent of coastal hypoxia does notmean that it is not a harmful phenomenon In the Stockholmarchipelago severe hypoxia is a pervasive and persistent phe-nomenon and also in Finland many local depressions areoften hypoxic Anoxic local depressions probably act as lo-cal nutrient sources releasing especially phosphorus to thewater column which further enhances pelagic primary pro-duction Such a vicious circle tends to worsen the eutrophica-tion and maintain the environment in a poor state (Pitkaumlnenet al 2001 Vahtera et al 2007) In this way even small-sized anoxic depressions especially if they are many mayaffect the ecological status of the whole coastal area More-over as climate change has been projected to increase watertemperatures and worsen hypoxia in the Baltic Sea (Meieret al 2011a) shallow archipelago areas that typically havehigh productivity warm up quickly and are topographicallyprone to hypoxia may be especially vulnerable to the nega-tive effects of climate change

In order to establish reference conditions and implementnecessary and cost-efficient measures to reach the goals of in-ternational agreements such as the EU Water Framework Di-rective (WFD 200060EC) the Marine Strategy FrameworkDirective (MSFD 200856EC) and the HELCOM BalticSea Action Plan (BSAP) in-depth knowledge of ecologicalfunctions and processes as well as natural preconditions isneeded Although eutrophication is a problem for the wholeBaltic Sea nutrient abatement measures are taken locallyWe therefore need to know where the environmental bene-fits are maximized and where natural conditions are likelyto counteract any measures taken As some places are natu-

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

3192 E A Virtanen et al Identifying areas prone to coastal hypoxia

rally prone to hypoxia our model could aid directing mea-sures to places where they are most likely to be efficient aswell as explain why in some areas implemented measures donot have the desired effect Our approach could be used todevelop an early-warning system for identification of areasprone to oxygen loss and to indicate where eutrophicationmitigation actions are most urgently needed

5 Conclusions

While biogeochemical 3-D models have been able to accu-rately project basin-scale oxygen dynamics describing spa-tial variation of hypoxia in coastal areas has remained a chal-lenge Recognizing that the enclosed nature of seafloors con-tributes to hypoxia formation we used simple topographicalparameters to model the occurrence of hypoxia in the com-plex Finnish and Swedish archipelagoes We found that asurprisingly large fraction (sim 80 ) of hypoxia occurrencescould be explained by topographical parameters alone Mod-eling results also suggested that less than 25 of the studiedseascapes were prone to hypoxia during late summer Largevariation existed in the spatial and temporal patterns of hy-poxia however with certain areas being prone to occasionalsevere hypoxia (O2 lt 2 mg Lminus1) while others were more sus-ceptible to recurrent moderate hypoxia (O2 lt 46 mg Lminus1)Sheltered topographically heterogeneous areas with limitedwater exchange were susceptible to developing hypoxia incontrast to less sheltered areas with high wave forcing Insome areas oxygen conditions were either better or worsethan predicted by the model We assume that these devi-ations from the topographical background were caused byprocesses not accounted for by the model such as hydro-graphical processes eg strong currents causing improvedmixing or by high external or internal nutrient loading in-ducing high local oxygen consumption We conclude thatformation of coastal hypoxia is probably primarily dictatedby local processes and can be quite accurately projectedusing simple topographical parameters but that interactionwith the associated watershed and the adjacent deeper basinsof the Baltic Sea can also influence local oxygen dynamicsin many areas Our approach gives a practical baseline forvarious types of hypoxia-related studies and consequentlydecision-making Identifying areas prone to hypoxia helps tofocus research management and conservation actions in acost-effective way

Data availability Hypoxia point data and scripts for reproducingthe models and plots are available at the Dryad data repositoryhttpsdoiorg105061dryadcn5kh5m (Virtanen et al 2019)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-16-3183-2019-supplement

Author contributions EAV and MV designed the study EAV andANS performed all analyses EAV wrote the main text and all au-thors contributed to the writing and editing of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements Elina A Virtanen and Markku Viitasalo ac-knowledge the SmartSea project (grant nos 292985 and 314225)funded by the Strategic Research Council of the Academy of Fin-land and the Finnish Inventory Programme for the Underwater Ma-rine Environment VELMU funded by the Ministry of the Envi-ronment Alf Norkko acknowledges the support of the Academy ofFinland (no 294853) and the Sophie von Julins Foundation Anto-nia Nystroumlm Sandman acknowledges the IMAGINE project (grantno 15247) funded by the Swedish EPA We wish to thank twoanonymous reviewers for insightful comments that significantly im-proved the manuscript

Financial support This research has been supported by the Strate-gic Research Council of the Academy of Finland (SmartSea project)(grant nos 292985 and 314225) the Ministry of the Environment(the Finnish Inventory Programme for the Underwater Marine En-vironment VELMU) the Academy of Finland (grant no 294853)and the Swedish EPA (IMAGINE project) (grant no 15247)

Review statement This paper was edited by Carol Robinson andreviewed by two anonymous referees

References

Alenius P Myrberg K Roiha P Lips U Tuomi L Petters-son H and Raateoja M Gulf of Finland Physics in The Gulfof Finland assessment edited by Raateoja M and Setaumllauml OHelsinki Finnish Environment Institute Reports of the FinnishEnvironment Institute 27 42ndash57 2016

Altieri A H and Gedan K B Climate change anddead zones Glob Change Biol 21 1395ndash1406httpsdoiorg101111gcb12754 2015

Ashcroft M B Chisholm L A and French K O The ef-fect of exposure on landscape scale soil surface temperaturesand species distribution models Landscape Ecol 23 211ndash225httpsdoiorg101007s10980-007-9181-8 2008

Bekkby T Isachsen P E Isaeus M and Bakkestuen VGIS modeling of wave exposure at the seabed A depth-attenuated wave exposure model Mar Geod 31 117ndash127httpsdoiorg10108001490410802053674 2008

Greenwell B Boehmke B Cunningham J and GBM De-velopers gbm Generalized Boosted Regression Models Rpackage version 214 available at httpsCRANR-projectorgpackage=gbm (last access 11 March 2019) 2018

Breitburg D Levin L A Oschlies A Greacutegoire M Chavez FP Conley D J Garccedilon V Gilbert D Gutieacuterrez D Isensee

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3193

K Jacinto G S Limburg K E Montes I Naqvi S W APitcher G C Rabalais N N Roman M R Rose K ASeibel B A Telszewski M Yasuhara M and Zhang J De-clining oxygen in the global ocean and coastal waters Science359 1ndash11 httpsdoiorg101126scienceaam7240 2018

Buzzelli C P Luettich Jr R A Powers S P Peterson C HMcNinch J E Pinckney J L and Paerl H W Estimating thespatial extent of bottom-water hypoxia and habitat degradationin a shallow estuary Mar Ecol Prog Ser 230 103ndash112 2002

Caballero-Alfonso A M Carstensen J and Con-ley D J Biogeochemical and environmental driversof coastal hypoxia J Mar Syst 141 190ndash199httpsdoiorg101016jjmarsys201404008 2015

Carstensen J Andersen J H Gustafsson B G and Con-ley D J Deoxygenation of the Baltic Sea during thelast century P Natl Acad Sci USA 111 5628ndash5633httpsdoiorg101073pnas1323156111 2014

Conley D J Humborg C Rahm L Savchuk O P and WulffF Hypoxia in the Baltic Sea and basin-scale changes in phos-phorus biogeochemistry Environ Sci Technol 36 5315ndash5320httpsdoiorg101021es025763w 2002

Conley D J Bjoumlrck S Bonsdorff E Carstensen J DestouniG Gustafsson B G Hietanen S Kortekaas M Kuosa HMarkus Meier H E Muumlller-Karulis B Nordberg K NorkkoA Nuumlrnberg G Pitkaumlnen H Rabalais N N Rosenberg RSavchuk O P Slomp C P Voss M Wulff F and Zilleacuten LHypoxia-Related Processes in the Baltic Sea Environ Sci Tech-nol 43 3412ndash3420 httpsdoiorg101021es802762a 2009

Conley D J Carstensen J Aigars J Axe P Bonsdorff EEremina T Haahti B M Humborg C Jonsson P KottaJ Lannegren C Larsson U Maximov A Medina M RLysiak-Pastuszak E Remeikaite-Nikiene N Walve J Wil-helms S and Zillen L Hypoxia Is Increasing in the CoastalZone of the Baltic Sea Environ Sci Technol 45 6777ndash6783httpsdoiorg101021es201212r 2011

Dersquoath G and Fabricius K E Classification and regressiontrees A powerful yet simple technique for ecological dataanalysis Ecology 81 3178ndash3192 httpsdoiorg1018900012-9658(2000)081[3178CARTAP]20CO2 2000

Diaz R and Rosenberg R Marine benthic hypoxia A review of itsecological effects and the behavioural response of benthic macro-fauna Oceanogr Mar Biol 33 245ndash303 1995a

Diaz R J and Rosenberg R Marine benthic hypoxia A reviewof its ecological effects and the behavioural responses of benthicmacrofauna in Oceanography and Marine Biology ndash an AnnualReview Vol 33 edited by Ansell A D Gibson R N andBarnes M Oceanography and Marine Biology U C L PressLtd London 245ndash303 1995b

Diaz R J and Rosenberg R Spreading dead zones and con-sequences for marine ecosystems Science 321 926ndash929httpsdoiorg101126science1156401 2008

Dunn D C and Halpin P N Rugosity-based regional modelingof hard-bottom habitat Mar Ecol Prog Ser 377 1ndash11 2009

Du Preez C A new arcndashchord ratio (ACR) rugosity index forquantifying three-dimensional landscape structural complexityLandscape Ecol 30 181ndash192 httpsdoiorg101007s10980-014-0118-8 2015

Eilola K Meier H E M and Almroth E On the dy-namics of oxygen phosphorus and cyanobacteria in the

Baltic Sea A model study J Mar Syst 75 163ndash184httpsdoiorg101016jjmarsys200808009 2009

Eilola K Gustafsson B G Kuznetsov I Meier H E MNeumann T and Savchuk O P Evaluation of biogeochem-ical cycles in an ensemble of three state-of-the-art numeri-cal models of the Baltic Sea J Mar Syst 88 267ndash284httpsdoiorg101016jjmarsys201105004 2011

Elith J Leathwick J R and Hastie T A working guideto boosted regression trees J Anim Ecol 77 802ndash813httpsdoiorg101111j1365-2656200801390x 2008

Eriksson S P Wennhage H Norkko J and Norkko AEpisodic disturbance events modify predator-prey interactionsin soft sediments Estuar Coast Shelf S 64 289ndash294httpsdoiorg101016jecss200502022 2005

Fennel K and Testa J M Biogeochemical Controls onCoastal Hypoxia Annu Rev Mar Sci 11 105ndash130httpsdoiorg101146annurev-marine-010318-095138 2019

Fennel K Hetland R Feng Y and DiMarco S A cou-pled physical-biological model of the Northern Gulf ofMexico shelf model description validation and analysisof phytoplankton variability Biogeosciences 8 1881ndash1899httpsdoiorg105194bg-8-1881-2011 2011

Fennel K Laurent A Hetland R Justic D Ko D S LehrterJ Murrell M Wang L X Yu L Q and Zhang W X Effectsof model physics on hypoxia simulations for the northern Gulfof Mexico A model intercomparison J Geophys Res-Ocean121 5731ndash5750 httpsdoiorg1010022015jc011577 2016

Freeman E A and Moisen G PresenceAbsence An R Packagefor Presence-Absence Model Analysis J Stat Softw 23 1ndash312008

Froumllicher T L Joos F Plattner G K Steinacher Mand Doney S C Natural variability and anthropogenictrends in oceanic oxygen in a coupled carbon cyclendashclimate model ensemble Global Biogeochem Cy 23 1ndash15httpsdoiorg1010292008GB003316 2009

Gammal J Norkko J Pilditch C A and Norkko A CoastalHypoxia and the Importance of Benthic Macrofauna Commu-nities for Ecosystem Functioning Estuar Coast 40 457ndash468httpsdoiorg101007s12237-016-0152-7 2017

Gilbert D Rabalais N N Diacuteaz R J and Zhang JEvidence for greater oxygen decline rates in the coastalocean than in the open ocean Biogeosciences 7 2283ndash2296httpsdoiorg105194bg-7-2283-2010 2010

Gray J S Wu R S S and Or Y Y Effects of hypoxia and or-ganic enrichment on the coastal marine environment Mar EcolProg Ser 238 249ndash279 httpsdoiorg103354meps2382492002

Gregorutti B Michel B and Saint-Pierre P Correlation and vari-able importance in random forests Stat Comput 27 659ndash678httpsdoiorg101007s11222-016-9646-1 2017

Hastie T Tibshirani R and Friedman J H The Elements ofStatistical Learning Data Mining Inference and PredictionSpringer-Verlag New York 2001

HELCOM Sources and pathways of nutrients to the Baltic SeaBaltic Sea Environment Proceedings No 153 1ndash48 2018

Hordoir R Axell L Houmlglund A Dieterich C Fransner FGroumlger M Liu Y Pemberton P Schimanke S AnderssonH Ljungemyr P Nygren P Falahat S Nord A JoumlnssonA Lake I Doumloumls K Hieronymus M Dietze H Loumlptien U

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019

3194 E A Virtanen et al Identifying areas prone to coastal hypoxia

Kuznetsov I Westerlund A Tuomi L and Haapala J Nemo-Nordic 10 a NEMO-based ocean model for the Baltic and Northseas ndash research and operational applications Geosci Model Dev12 363ndash386 httpsdoiorg105194gmd-12-363-2019 2019

Jimeacutenez-Valverde A and Lobo J M Threshold crite-ria for conversion of probability of species presence toeitherndashor presencendashabsence Acta Oecol 31 361ndash369httpsdoiorg101016jactao200702001 2007

Jokinen S A Virtasalo J J Jilbert T Kaiser J DellwigO Arz H W Haumlnninen J Arppe L Collander M andSaarinen T A 1500-year multiproxy record of coastal hypoxiafrom the northern Baltic Sea indicates unprecedented deoxy-genation over the 20th century Biogeosciences 15 3975ndash4001httpsdoiorg105194bg-15-3975-2018 2018

Karlson K Rosenberg R and Bonsdorff E Temporal and spatiallarge-scale effects of eutrophication and oxygen deficiency onbenthic fauna in Scandinavian and Baltic waters ndash A review inOceanography and Marine Biology Vol 40 edited by GibsonR N Barnes M and Atkinson R J A Oceanography andMarine Biology Taylor amp Francis Ltd London 427ndash489 2002

Karlsson O M Jonsson P O Lindgren D Malmaeus JM and Stehn A Indications of Recovery from Hypoxiain the Inner Stockholm Archipelago Ambio 39 486ndash495httpsdoiorg101007s13280-010-0079-3 2010

Kaskela A M Kotilainen A T Al-Hamdani Z Leth JO and Reker J Seabed geomorphic features in a glaciatedshelf of the Baltic Sea Estuar Coast Shelf S 100 150ndash161httpsdoiorg101016jecss201201008 2012

Kemp W M Testa J M Conley D J Gilbert D andHagy J D Temporal responses of coastal hypoxia to nutrientloading and physical controls Biogeosciences 6 2985ndash3008httpsdoiorg105194bg-6-2985-2009 2009

Kramer M G Hansen A J Taper M L and Kissinger E JAbiotic controls on long-term windthrow disturbance and tem-perate rain forest dynamics in Southeast Alaska Ecology 102749ndash2768 2001

Laine A O Sandler H Andersin A-B and Stigzelius JLong-term changes of macrozoobenthos in the Eastern Got-land Basin and the Gulf of Finland (Baltic Sea) in rela-tion to the hydrographical regime J Sea Res 38 135ndash159httpsdoiorg101016S1385-1101(97)00034-8 1997

Lappalainen J Virtanen E A Kallio K Junttila S and Vi-itasalo M Substrate limitation of a habitat-forming genusFucus under different water clarity scenarios in the north-ern Baltic Sea Estuarine Coast Shelf Sci 218 31ndash38httpsdoiorg101016jecss201811010 2019

Leppaumlranta M and Myrberg K Physical Oceanography of theBaltic Sea Springer-Verlag Berlin-Heidelberg-New York 131ndash187 2009

Liu C Berry P M Dawson T P and Pearson R G Select-ing thresholds of occurrence in the prediction of species distri-butions Ecography 28 385ndash393 httpsdoiorg101111j0906-7590200503957x 2005

Meier H E M Andersson H C Eilola K GustafssonB G Kuznetsov I Muumlller-Karulis B Neumann T andSavchuk O P Hypoxia in future climates A model ensem-ble study for the Baltic Sea Geophys Res Lett 38 1ndash6httpsdoiorg1010292011GL049929 2011a

Meier H E M Eilola K and Almroth E Climate-relatedchanges in marine ecosystems simulated with a 3-dimensionalcoupled physicalbiogeochemical model of the Baltic Sea ClimRes 48 31ndash55 2011b

Meier H E M Hordoir R Andersson H C Dieterich CEilola K Gustafsson B G Hoglund A and Schimanke SModeling the combined impact of changing climate and chang-ing nutrient loads on the Baltic Sea environment in an ensem-ble of transient simulations for 1961ndash2099 Clim Dynam 392421ndash2441 httpsdoiorg101007s00382-012-1339-7 2012a

Meier H E M Muller-Karulis B Andersson H C Di-eterich C Eilola K Gustafsson B G Hoglund A Hor-doir R Kuznetsov I Neumann T Ranjbar Z Savchuk OP and Schimanke S Impact of Climate Change on Ecologi-cal Quality Indicators and Biogeochemical Fluxes in the BalticSea A Multi-Model Ensemble Study Ambio 41 558ndash573httpsdoiorg101007s13280-012-0320-3 2012b

Meier H E M Andersson H C Arheimer B Donnelly CEilola K Gustafsson B G Kotwicki L Neset T-S Niira-nen S Piwowarczyk J Savchuk O P Schenk F WesławskiJ M and Zorita E Ensemble Modeling of the Baltic SeaEcosystem to Provide Scenarios for Management Ambio 4337ndash48 httpsdoiorg101007s13280-013-0475-6 2014

Middelburg J J and Levin L A Coastal hypoxia andsediment biogeochemistry Biogeosciences 6 1273ndash1293httpsdoiorg105194bg-6-1273-2009 2009

Natekin A and Knoll A Gradient boosting ma-chines a tutorial Front Neurorobotics 7 1ndash21httpsdoiorg103389fnbot201300021 2013

Nilsson H C and Rosenberg R Succession in marine benthichabitats and fauna in response to oxygen deficiency analysed bysediment profile-imaging and by grab samples Mar Ecol ProgSer 197 139ndash149 httpsdoiorg103354meps197139 2000

Norkko J Reed D C Timmermann K Norkko A Gustafs-son B G Bonsdorff E Slomp C P Carstensen J andConley D J A welcome can of worms Hypoxia mitiga-tion by an invasive species Glob Change Biol 18 422ndash434httpsdoiorg101111j1365-2486201102513x 2012

Norkko J Gammal J Hewitt J E Josefson A BCarstensen J and Norkko A Seafloor Ecosystem Func-tion Relationships In Situ Patterns of Change Across Gradi-ents of Increasing Hypoxic Stress Ecosystems 18 1424ndash1439httpsdoiorg101007s10021-015-9909-2 2015

Pitkaumlnen H Lehtoranta J and Raumlike A Internal Nutrient FluxesCounteract Decreases in External Load The Case of the EstuarialEastern Gulf of Finland Baltic Sea 30 195ndash201 2001

Puttonen I Mattila J Jonsson P Karlsson O M Kohonen TKotilainen A Lukkari K Malmaeus J M and Rydin E Dis-tribution and estimated release of sediment phosphorus in thenorthern Baltic Sea archipelagos Estuarine Coast Shelf Sci145 9ndash21 httpsdoiorg101016jecss201404010 2014

R Core Team R A language and environment for statisticalcomputing R foundation for Statisticl a Computing ViennaAustria available at httpswwwR-projectorg (last access7 April 2019) 2018

Rabalais N N Diacuteaz R J Levin L A Turner R EGilbert D and Zhang J Dynamics and distribution of nat-ural and human-caused hypoxia Biogeosciences 7 585ndash619httpsdoiorg105194bg-7-585-2010 2010

Biogeosciences 16 3183ndash3195 2019 wwwbiogeosciencesnet1631832019

E A Virtanen et al Identifying areas prone to coastal hypoxia 3195

Sappington J M Longshore K M and Thompson D B Quan-tifying Landscape Ruggedness for Animal Habitat Analysis ACase Study Using Bighorn Sheep in the Mojave Desert SPIE 8pp 2007

Schapire R The boosting approach to machine learning ndash anoverview in MSRI Workshop on Nonlinear Estimation andClassification 2002 edited by Denison D Hansen M HHolmes C B M and B Y Springer New York 2003

Scully M E Physical controls on hypoxia in Chesapeake Bay Anumerical modeling study J Geophys Res-Ocean 118 1239ndash1256 httpsdoiorg101002jgrc20138 2013

Scully M E The contribution of physical processes tointer-annual variations of hypoxia in Chesapeake Bay A30-yr modeling study Limnol Oceanogr 61 2243ndash2260httpsdoiorg101002lno10372 2016

Stramma L Oschlies A and Schmidtko S Mismatch be-tween observed and modeled trends in dissolved upper-oceanoxygen over the last 50 yr Biogeosciences 9 4045ndash4057httpsdoiorg105194bg-9-4045-2012 2012

Testa J M Li Y Lee Y J Li M Brady D C DiToro D M Kemp W M and Fitzpatrick J J Quan-tifying the effects of nutrient loading on dissolved O-2cycling and hypoxia in Chesapeake Bay using a coupledhydrodynamic-biogeochemical model J Mar Syst 139 139ndash158 httpsdoiorg101016jjmarsys201405018 2014

Timmermann K Norkko J Janas U Norkko A Gustafsson BG and Bonsdorff E Modelling macrofaunal biomass in rela-tion to hypoxia and nutrient loading J Mar Syst 105 60ndash69httpsdoiorg101016jjmarsys201206001 2012

Vahtera E Conley D J Gustafsson B G Kuosa HPitkanen H Savchuk O P Tamminen T ViitasaloM Voss M Wasmund N and Wulff F Internalecosystem feedbacks enhance nitrogen-fixing cyanobac-teria blooms and complicate management in the BalticSea Ambio 36 186ndash194 httpsdoiorg1015790044-7447(2007)36[186iefenc]20co2 2007

Valanko S Heino J Westerbom M Viitasalo M and NorkkoA Complex metacommunity structure for benthic invertebratesin a low-diversity coastal system Ecol Evol 5 5203ndash5215httpsdoiorg101002ece31767 2015

van Helmond N A Krupinski N B Q Lougheed B CObrochta S P Andreacuten T and Slomp C P Seasonal hypoxiawas a natural feature of the coastal zone in the Little Belt Den-mark during the past 8 ka Mar Geol 387 45ndash57 2017

Vaquer-Sunyer R and Duarte C M Thresholds of hypoxia formarine biodiversity P Natl Acad Sci USA 105 15452ndash15457httpsdoiorg101073pnas0803833105 2008

Villnas A Norkko J Lukkari K Hewitt J and Norkko AConsequences of Increasing Hypoxic Disturbance on BenthicCommunities and Ecosystem Functioning Plos One 7 1ndash12httpsdoiorg101371journalpone0044920 2012

Villnas A Norkko J Hietanen S Josefson A B LukkariK and Norkko A The role of recurrent disturbancesfor ecosystem multifunctionality Ecology 94 2275ndash2287httpsdoiorg10189012-17161 2013

Virtanen E A Viitasalo M Lappalainen J and MoilanenA Evaluation Gap Analysis and Potential Expansion of theFinnish Marine Protected Area Network Front Mar Sci 5 1ndash19 httpsdoiorg103389fmars201800402 2018

Virtanen E A Norkko A Nystroumlm Sandman A and Vi-itasalo M Data from Identifying areas prone to coastal hy-poxia ndash the role of topography Dryad Digital Repositoryhttpsdoiorg105061dryadcn5kh5m 2019

Virtasalo J J Kohonen T Vuorinen I and HuttulaT Sea bottom anoxia in the Archipelago Sea north-ern Baltic Sea ndash Implications for phosphorus remineraliza-tion at the sediment surface Mar Geol 224 103ndash122httpsdoiorg101016jmargeo200507010 2005

Walbridge S Slocum N Pobuda M and Wright D Unified Ge-omorphological Analysis Workflows with Benthic Terrain Mod-eler Geosciences 8 1ndash24 2018

Walker B K Jordan L K B and Spieler R E Relationship ofReef Fish Assemblages and Topographic Complexity on South-eastern Florida Coral Reef Habitats SPIE 10 pp 2009

Walve J Sandberg M Larsson U and Laumlnnergren C A BalticSea estuary as a phosphorus source and sink after drastic load re-duction seasonal and long-term mass balances for the Stockholminner archipelago for 1968ndash2015 Biogeosciences 15 3003ndash3025 httpsdoiorg105194bg-15-3003-2018 2018

Weiss A D Topographic position and landforms analy-sis In Poster presented at the Esri User ConferenceSan Diego CA USA httpwwwjennessentcomdownloadstpi-poster-tnc_18x22pdf (last access 17 January 2019) 2001

WFD Water Framework Directive Common Implementation2000

wwwbiogeosciencesnet1631832019 Biogeosciences 16 3183ndash3195 2019