evaluation of reaching the targets of the water framework directive in the gulf of finland

9
Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland Jose A. Fernandes, , * Pirkko Kauppila, Laura Uusitalo, ,Vivi Fleming-Lehtinen, Sakari Kuikka, and Heikki Pitka ̈ nen Department of Environmental Sciences, University of Helsinki, Viikinkaari 1. FI-00014 Helsinki, Finland Finnish Environment Institute (SYKE), Mechelininkatu 34a. FI-00251 Helsinki, Finland ABSTRACT: This paper describes the development of the EU Water Framework Directive central water quality elements from 1970 to 2010 in the Gulf of Finland, a eutrophied sub- basin of the Baltic Sea. The likelihood of accomplishing the management objectives simultaneously is assessed using Bayesian networks. The objectives of good ecological status in winter-time total nitrogen and phosphorus, summer-time chlorophyll-a and summer-time Secchi depth have not been met yet. In addition, the results indicate that it is unlikely for them to be achieved in the near future, despite the decreasing trend in nutrient concentrations over the past few years. It was demonstrated that neither phosphorus nor nitrogen alone controls summertime plankton growth. Reaching good ecological status in nutrients does not necessarily lead to good ecological status of chlorophyll-a, even though a dependency between the parameters does exist. In addition, secchi-depth status is strongly related to chlorophyll-a status in three of the four study-areas. INTRODUCTION The Water Framework Directive (Directive 2000/60/EC of the European Parliament and of the Council establishing a framework for the Community action in the eld of water policy), with the EU Marine Framework Directive, is currently the most important legislative framework in the management of water quality in inland and coastal waters of the EU Member States. 1 They apply to both the coastal and open sea areas of the pan-European waters. 2-4 The Water Framework Directive (WFD 5 ) was adopted in the European Union in October 2000. The stated aim is to reach or maintain good ecological status in all waters by 2015. Ecological status is determined by assessment against dened reference levels in biological, physicochemical, and hydro- morphological quality elements. Type-specic reference con- ditions are established using historical data or modeling tools in order to form a baseline against which the changes can be measured. 6,7 Ecological status is divided into ve classes or levels (high, good, moderate, poor, and bad). The targeted status to be achieved being at least the boundary between good and moderate status. Due to the diversity of ecosystems, no absolute standards for the good status can be given across all European waters. 5 Therefore, the directive only gives guidelines for the denition of the reference condition. This denition must be implemented by each Member state. According to the normative guidelines of the WFD, good ecological status is achieved when biological communities present are close to those that would be present with minimal anthropogenic disturbance. The denition of the kind of community that would occur with minimal anthropogenic disturbance is not a trivial task. 8,9 Often there is a lack of data or the areas have specic properties. The Baltic Sea is a shallow, semienclosed brackish- water area characterized by a slow water renewal time and a four times larger drainage basin in relation to its surface water area. Eutrophication in the Baltic Sea 10-12 is the result of long human exploitation of natural resources. 13,14 Eutrophication has been identied as one of the key pressures on biological communities. For example, summertime cyanobacterial blooms are a regular phenomenon in the Baltic Sea. 15 In the Gulf of Finland eutrophication has triggered a self-sustaining vicious circleof internal nutrient loading, associated with anoxic bottom sediments and amounts of organic material that can reach a thickness of 10 cm. 16-18 Internal loading in the Gulf of Finland appears to counteract decreases in the external loads of phosphorus. 19 In Finnish coastal waters, ecological classication has mainly been based on summertime phytoplankton chlorophyll-a (chl-a) levels. Secchi depth and wintertime total nutrients have been used as supporting variables. In the classication Received: January 12, 2012 Revised: July 2, 2012 Accepted: July 3, 2012 Published: July 3, 2012 Article pubs.acs.org/est © 2012 American Chemical Society 8220 dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220-8228

Upload: heikki

Post on 20-Feb-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

Evaluation of Reaching the Targets of the Water FrameworkDirective in the Gulf of FinlandJose A. Fernandes,†,* Pirkko Kauppila,‡ Laura Uusitalo,†,‡ Vivi Fleming-Lehtinen,‡ Sakari Kuikka,†

and Heikki Pitkanen‡

†Department of Environmental Sciences, University of Helsinki, Viikinkaari 1. FI-00014 Helsinki, Finland‡Finnish Environment Institute (SYKE), Mechelininkatu 34a. FI-00251 Helsinki, Finland

ABSTRACT: This paper describes the development of theEU Water Framework Directive central water quality elementsfrom 1970 to 2010 in the Gulf of Finland, a eutrophied sub-basin of the Baltic Sea. The likelihood of accomplishing themanagement objectives simultaneously is assessed usingBayesian networks. The objectives of good ecological statusin winter-time total nitrogen and phosphorus, summer-timechlorophyll-a and summer-time Secchi depth have not beenmet yet. In addition, the results indicate that it is unlikely forthem to be achieved in the near future, despite the decreasingtrend in nutrient concentrations over the past few years. It wasdemonstrated that neither phosphorus nor nitrogen alonecontrols summertime plankton growth. Reaching goodecological status in nutrients does not necessarily lead to good ecological status of chlorophyll-a, even though a dependencybetween the parameters does exist. In addition, secchi-depth status is strongly related to chlorophyll-a status in three of the fourstudy-areas.

■ INTRODUCTION

The Water Framework Directive (“Directive 2000/60/EC ofthe European Parliament and of the Council establishing aframework for the Community action in the field of waterpolicy”), with the EU Marine Framework Directive, is currentlythe most important legislative framework in the management ofwater quality in inland and coastal waters of the EU MemberStates.1 They apply to both the coastal and open sea areas ofthe pan-European waters.2−4 The Water Framework Directive(WFD5) was adopted in the European Union in October 2000.The stated aim is to reach or maintain good ecological status inall waters by 2015.Ecological status is determined by assessment against defined

reference levels in biological, physicochemical, and hydro-morphological quality elements. Type-specific reference con-ditions are established using historical data or modeling tools inorder to form a baseline against which the changes can bemeasured.6,7 Ecological status is divided into five classes orlevels (high, good, moderate, poor, and bad). The targeted statusto be achieved being at least the boundary between good andmoderate status. Due to the diversity of ecosystems, no absolutestandards for the good status can be given across all Europeanwaters.5 Therefore, the directive only gives guidelines for thedefinition of the reference condition. This definition must beimplemented by each Member state. According to thenormative guidelines of the WFD, good ecological status isachieved when biological communities present are close to

those that would be present with minimal anthropogenicdisturbance.The definition of the kind of community that would occur

with minimal anthropogenic disturbance is not a trivial task.8,9

Often there is a lack of data or the areas have specificproperties. The Baltic Sea is a shallow, semienclosed brackish-water area characterized by a slow water renewal time and afour times larger drainage basin in relation to its surface waterarea. Eutrophication in the Baltic Sea10−12 is the result of longhuman exploitation of natural resources.13,14 Eutrophication hasbeen identified as one of the key pressures on biologicalcommunities. For example, summertime cyanobacterial bloomsare a regular phenomenon in the Baltic Sea.15 In the Gulf ofFinland eutrophication has triggered a self-sustaining “viciouscircle” of internal nutrient loading, associated with anoxicbottom sediments and amounts of organic material that canreach a thickness of 10 cm.16−18 Internal loading in the Gulf ofFinland appears to counteract decreases in the external loads ofphosphorus.19

In Finnish coastal waters, ecological classification has mainlybeen based on summertime phytoplankton chlorophyll-a(chl-a) levels. Secchi depth and wintertime total nutrientshave been used as supporting variables. In the classification

Received: January 12, 2012Revised: July 2, 2012Accepted: July 3, 2012Published: July 3, 2012

Article

pubs.acs.org/est

© 2012 American Chemical Society 8220 dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−8228

Page 2: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

schemes, the reference values of chl-a have been reconstructedempirically using historical Secchi observations from the early1900s and present-day monitoring data.20,21 Reference values oftotal phosphorus and total nitrogen have been estimated byfrequency distribution data. Basically, the high/good boundarieshave been defined as reference values plus 20%, and the good/moderate boundaries as high status plus 50%. Chl-a is chosen asa proxy to phytoplankton biomass because it is routinelymonitored (quick and easy to measure). Winter nutrients areused due to negligible activity of primary producers in winterseason.19,22 Winter nutrients control phytoplankton growthlater in spring and summer.23 The likelihood of wintertimenutrients affecting ecological status classification based on chl-amust be quantified for two reasons; (1) to help the planningprocess in assessing the required reductions of loading in orderto meet the environmental quality objectives of the directive;and (2) to give additional information for further indicatordevelopment and validation of the classification scheme.In this paper, the ecological status and the trends of

summertime chl-a and Secchi depth as well as the wintertimeconcentrations of total nitrogen and phosphorus in the Gulf ofFinland are examined (1970−2010). Using the high/good andgood/moderate boundary values defined by the water directiveas targets, the achievement of high and good ecological status indifferent coastal areas of the Gulf of Finland is studied. Finally,using probabilistic models, synchrony of the target levels set foreach variable is examined in the Gulf of Finland, that is, thelikelihood of reaching the target values of chl-a when the targetvalues of nutrients are met. In particular, the uncertainties ofthe relationships between levels of nutrients and the othervariables levels are evaluated. This uncertainty is crucial inbeing able to achieve a certain chl-a level by managing nutrientamounts in river basins by the WFD implementation actions.

■ MATERIALS AND METHODS

The Study Area. The study area (Figure 1) is based on thenational coastal areas of Finland. According to the WFD, theareas are characterized by descriptors such as salinity, mixing

conditions of water, exposure, water residence time andduration of ice cover.24,20 These coastal areas (Figure 1) werethe Gulf of Finland inner archipelago (“eastern inner”, EI), theGulf of Finland outer archipelago (“eastern outer”, EO), theSouthwestern inner archipelago (“western inner”, WI), and theSouthwestern outer archipelago (“western outer”, WO).The Gulf of Finland inner archipelago (EI) is shallow, less than

15 m deep. It is characterized by a mosaic of islands andnumerous enclosed bays and estuaries. These features make itsheltered and restrict the water exchange between the innercoastal waters and the open sea. The bottom consists ofsemienclosed basins limiting horizontal and vertical waterexchange. However, small trenches (20 m deep) extend to thisarea. The water is fully mixed or seasonally stratified. Thethermocline in midsummer conditions is at a depth of 10 m.The duration of ice cover varies from 90 to 150 days.The Gulf of Finland outer archipelago (EO) is mainly formed

by open water with small scattered islands. The water ismoderately exposed with depths ranging between 15 and 30 m.Some of the deeps of the open sea extend into this area. Salinityis below 5 psu and sinks eastwards. The water is seasonallystratified. Residence time is short, measured in days. Ice coverlasts on average 90 days.The Southwestern inner archipelago (WI) west of the Porkkala

Peninsula is characterized by (a) coastline with long shallowbays stretching deep into the mainland and large islands,sometimes connected to the mainland, intersected by narrowand shallow sounds. Deeps exceeding 30 m from the moresaline waters of the open Gulf of Finland extend as far as theinner archipelago. The morphometry is similar to the Gulf ofFinland inner archipelago, but the average salinity is over 5 psu.The Southwestern outer archipelago(WO) is much like the

eastern outer archipelago in terms of its depth, island cover andbottom topography. It comprises small islands surrounded bywide stretches of deep open water with occasional shallows anddeeper faults. The area is open and exposed. Salinity rangesbetween 5 and 6 psu.

Figure 1. Coastal areas in the Gulf of Finland considered in this work.

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288221

Page 3: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

Data. The selection of variables in this work was constrainedby the variables considered in the Water Framework Directive.Altogether, the data comprised 865 sampling stations in theGulf of Finland during 1970−2010. The data are maintained inthe database of the Finnish Environment Institute (SYKE). Thevariables are winter-time (January to March) concentrations oftotal phosphorus (tot-P) and nitrogen (tot-N), summertime(July−September) concentrations of chlorophyll-a (chl-a) andSecchi depths. Data on nutrient concentrations in surface waterlayers were averaged (samples taken from 0, 1, and 5 m). Bothnitrogen and phosphorus concentrations were spectrometricallyand colorimetrically determined from unfiltered samplesaccording to Finnish standard methods.20 Chl-a was takenfrom composite samples (twice the Secchi depth), representingthe productive water layer (0−10 m). Chl-a was analyzedaccording to Lorenzen’s methodology.25 Samples weregenerally extracted with acetone from 1980 to 1994, and withethanol thereafter. The data were prepared by applying the classboundaries of each of the above-mentioned variables used inFinnish ecological classification20,21 (Table I).The study area is divided into 54 water bodies located within

the four coastal areas, characterized according to Finnishnational typology. Water bodies are the basic classification unitsof the WFD,26 which are relatively uniform water areas withhydrographical characteristics that distinguish them from thesurrounding water bodies. The WFD assume that samplesrepresent the water quality in the water body from which theyhave been taken. The number of sampling stations in eachwater body varied from 5 to 70. We computed yearly averagesfor each of the 54 water bodies, and used this geographicallybalanced data set for the final analyses following the WFDprocedures.5,9 This procedures objective is to avoid biasinduced by the uneven data coverage in the water bodieswhen dealing with data discretized in categories. Thesummarizing data sets have one row for each year and waterbody, combining the winter-time total nutrients with thesummer chl-a and Secchi measurements.

The Analysis Approach. The data analysis has two mainaims: (1) to describe the development of the WFD waterquality indicators from 1970 to 2010, and (2) to provide anestimate of the probability of reaching the quality classes ofgood or high status for chl-a in cases where the WFDrequirements related to nutrient concentrations are met (i.e.,total nutrient concentrations are at good or high levels). Thefirst aim plays an introductory role to the second. For the firstobjective, summarizing time series have been calculated foreach of the four coastal type areas. In order to follow the WFDpractices, the values were first calculated for each water body(Figure 2). The Mann-Kendall test27 was used to detectincreasing or decreasing trends in the time-series.28−30 For thesecond aim, Bayesian networks (Figure 3) were built using thedata set with one observation for each water body each year(with missing data in some water bodies in some years). In thissecond approach, the variables were discretized using the WFDlevels (high, good, moderate, poor, bad) before learning themodel in order to mimic the procedure that is being used in theimplementation of WFD activities of the Gulf of Finland. Thismodel allows estimation of the likelihood of reaching the WFDgoal, that is, good or better ecological status, if the nutrientconcentrations, or alternatively the chl-a concentrations, are atgood or high status (e.g., manipulated by activities in thedrainage area).

Bayesian Networks. A Bayesian network (BN) is a modelrepresented in graphical form. It describes the probabilisticdependencies (or lack thereof) between the modeledvariables.31−33 In the graphical representation, each variable isa node, and dependencies between the variables are representedby the presence (or absence) of arcs between them (Figure 3).These relationships are quantified by a set of parameters, in thepresent case discrete probability tables. Probabilistic modelshave been used over the last few decades in order to deal withuncertainty,34 whose graphical representation is suitable toprovide an intuitive interface to data35,36 as well as to produce

Table I. Definition of Levels or Classes for Each Quality Element and Each Areaa

vars. season depth unit ref high good mod. poor bad

WI area Chl-a VII−VIII 0−10 m μg/L 2 <2.4 2.4−3.7 3.7−10 10−20 >20Secchi VII−VIII 0−5 m m 5.5 >4.6 4.6−3.1 3.1−1.1 1.1−0.6 <0.6Tot-P I−III μg/L 19 <23 23−29 29−57 57−76 >76Tot-N I−III 0−5 m μg/L 325 <390 390−488 488−975 975−1300 >1300

WO area Chl-a VII−VIII 0−10 m μg/L 1.6 <1.9 1.9−2.9 2.9−8 8−16 >16Secchi VII−VIII 0−5 m m 8.9 >7.4 7.4−4.9 4.9−1.8 1.8−0.9 <0.9Tot-P I−III 0−5 m μg/L 18 <22 22−27 27−54 54−72 >72Tot-N I−III μg/L 230 <276 276−345 345−690 690−920 >920

EI area Chl-a VII−VIII 0−10 m μg/L 2.6 <3.2 3.2−4.7 4.7−13 13−26 >26Secchi VII−VIII 0−5 m m 5.4 >4.5 4.5−3 3−1.1 1.1−0.5 <0.5Tot-P I−III 0−5 m μg/LTot-N I−III μg/L 390 <468 468−585 585−585 1170−1560 >1560

EO area Chl-a VII−VIII 0−10 m μg/L 2.3 <2.8 2.8−4.1 4.1−12 12−23 >23Secchi VII−VIII 0−5 m m 6.7 >5.6 5.6−3.7 3.7−1.3 1.3−0.7 <0.7Tot-P I−III 0−5 m μg/L 19 <23 23−29 29−57 57−76 >76Tot-N I−III μg/L 340 <408 408−510 510−1020 1020−1360 >1360

aListed are the used seasons, measuring depths, reference values (ref) and the intervals for high, good and moderate, poor and bad status establishedfor each Water Framework Directive classification quality element (vars.). There are no boundaries defined for Tot-P in EI area due to the lack ofhistorical data.

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288222

Page 4: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

forecasts.37,38 BNs have been used for policy evaluation inrelation to water resources.39−41

The structure of a BN represents probabilistic relationshipsthat are not necessarily causal.42 However, in this work, thestructure and direction of arcs represent the causal relationshipsdefined by experts (Figure 3) in order to include the expertknowledge43,44 that has been obtained in limnology. The datawere discretized using the boundaries defined by the WFD,which were used to learn the parameters of a model for eacharea. Each node in the network is characterized by a state whichcan change depending on the state of other nodes andinformation about those states propagated through thenetwork.45 In Figure 3, an example of a BN is presentedwhere each node represents a variable with its associated

probability table (parameters). The complexity of theprobability tables depends on whether or not a variable hasparents (arc direction); that is, if it represents the probabilitiesin base to another variable (conditional probabilities). Forexample, the probability table of chl-a (Figure 3 and Table II)allows observation of the probability of a good level givenspecific levels of nutrients (tot-N and tot-P).

■ RESULTSThe time series analysis revealed that the levels of chl-a, Secchidepth and wintertime nutrients in the outer coastal areas hadsimilar patterns to the corresponding variables in the innercoastal areas. These tendencies exclude the peaks of totalphosphorus and chl-a recorded in the eastern inner archipelago

Figure 2. The time-series (5 years moving average) of Chl-a, Secchi, and nutrients for the four coastal areas. The targeted moderate/good boundaryfor each area is represented by the horizontal lines. Notice that for Secchi the target is to be above the boundary, whereas for the rest of variables thetarget is to be below. Currently, under the WFD, there is no targeted boundary for tot-P in the EI area (see Table I).

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288223

Page 5: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

(IE) in the late 1970s and the mid 1980s, respectively (Figure2). At that time, the level of chl-a in the eastern and westernouter archipelago (EO and WO) gave it a status close to good.Today, based on chl-a, all the areas are far from meeting thequality objectives of the WFD. Despite the levels of chl-adecreasing from around 2007, the general levels in the Gulf ofFinland have not reached the levels recorded in the mid 1990s.For most summers after the mid 1990s, the average

concentrations of chl-a in the western outer archipelago(WO) exceeded the values measured in the eastern outerarchipelago, which is affected by the eutrophied easternmostGulf of Finland.Based on Secchi depth, the eastern outer archipelago (EO)

achieved a good status only in the 1970s and early 1980s(Figure 2) and deteriorated afterward. Since the mid 1990s, thevalues in the western outer archipelago have approached thelevel of the eastern outer archipelago (WO). However, thetrends of Secchi depth and chl-a did not couple with nutrientlevels. Total nitrogen and chl-a in the eastern inner archipelago(IE) showed concurrent maxima in the early and mid 1980s,whereas the peaks of phosphorus in the 2000s coincided withthose of chl-a. Based on total phosphorus, none of the coastaltype areas have exhibited good status during the entiremonitoring period. In contrast, based on total nitrogen, theeastern outer archipelago (EO) reflected good status both in themid 1970s and since the mid 1990s. Similar to chl-a, thepresent levels of total phosphorus, especially in the westernouter archipelago (WO), are still above the levels recorded inthe 1990s.Nitrogen concentration shows a decreasing trend (Mann-

Kendall test, p < 0.01) in all the areas. In the outer coastal areas,phosphorus concentration trends showed a significant increasein both eastern (p < 0.01) and western (p < 0.05) parts of theGulf of Finland, although it has decreased in the past few years.In the inner eastern archipelago (EI), the results were stronglyaffected by the inclusion of data from Helsinki, one of the mostpolluted sea areas in Finland in the 1960s. Thus, phosphorusshowed a decrease (p < 0.01) in cases where the data ofHelsinki bays were included. This trend was reversed (p < 0.01)if (the) Helsinki data were excluded from the analyses.The biological variables chl-a concentration and Secchi depth

showed undesirable trends, with chl-a increasing (p < 0.001;when the Helsinki water body is excluded from the EI area) and

Figure 3. Example of Bayesian network structure and parameters(probability Tables) for the WO coastal area. The arcs showrelationships specified by experts.

Table II. Probabilistic Scenarios of the Quality Elements in Each Areaa

WI area WO area EI area EO area

basic nutr. Chl-a basic nutr. Chl-a basic nutr. Chl-a basic nutr. Chl-a

vars. levels state good good state good good state good good state good good

Tot-N high/good 0.22 1.0 0.28 0.02 1.0 0.09 0.24 1.0 0.36 0.65 1.0 0.59mod. 0.71 0.0 0.68 0.87 0.0 0.86 0.58 0.0 0.43 0.35 0.0 0.41poor/bad 0.07 0.0 0.04 0.11 0.0 0.05 0.18 0.0 0.21

Tot-P high/good 0.31 1.0 0.45 0.12 1.0 0.05 0.09 1.0 0.47mod. 0.67 0.0 0.53 0.78 0.0 0.93 0.88 0.0 0.29poor/bad 0.02 0.0 0.02 0.10 0.0 0.02 0.03 0.0 0.24

Chl-a high/good 0.14 0.18 1.0 0.12 0.33 1.0 0.06 0.09 1.0 0.04 0.13 1.0mod. 0.69 0.80 0.0 0.75 0.33 0.0 0.69 0.85 0.0 0.96 0.87 0.0poor/bad 0.17 0.02 0.0 0.14 0.33 0.0 0.25 0.06 0.0 0.0

Secchi good 0.16 0.65 0.64 0.19 0.16 0.99 0.07 0.09 0.23 0.3 0.31 0.5mod. 0.84 0.35 0.36 0.81 0.84 0.01 0.84 0.87 0.73 0.7 0.69 0.5poor/bad 0.09 0.04 0.04

aFirstly, the posterior probability distribution of the annual water body averages in three different status categories based on frequencies in data(“Basic state”) is shown. Secondly, the posterior distribution is presented based on the Bayesian network model under the scenario of both totalnitrogen and total phosphorus being at good or high state (“Nutr. Good”, bold). Finally, posterior distribution under the scenario of chl-a being atgood or high state (“Chl-a Good”, bold). The status categories are high or good status (high/good), moderate status (mod.) and poor or bad status(poor/bad). The empty cells mean no data were available. The results are presented for the four water areas (WI = Western inner archipelago, WO =Western outer archipelago, EI = Eastern inner archipelago, EO = Eastern outer archipelago).

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288224

Page 6: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

Secchi depth decreasing (p < 0.001). The high chl-a values inthe EI area early in the time series were influenced by theelevated Helsinki area levels. In the last 5 years of the timeseries, a reversal of this trend is observed, with decreasing chl-aand increasing Secchi depths. These results imply some successin the efforts to improve the ecological status of the Gulf ofFinland. However, it will be difficult to reach the targets set byWater Framework Directive by 2015 in these areas.According to the basic frequencies observed in the overall

historical water body data, the observations were most likely toindicate a moderate status in all of the water quality parameters.A high or good status was most probable only for total nitrogenin the eastern outer archipelago. For the nutrients, the leastprobable status was poor or bad in most of the areas, whereasfor chl-a poor or bad status was as probable as high or goodstatus. For Secchi depth, a bad status was observed only rarelyin the Eastern inner archipelago (Table II).Scenarios manipulating nutrients can be assessed using the

Bayesian networks (Table II, second column in each area). Theprobability of achieving high chl-a status increased slightly whenthe nutrient status was assumed as good or high. However, themost likely status in all water areas except the Western outerarchipelago remained moderate. The results of analysis of Secchidepth differed between the water areas: in the Western innerarchipelago good or high became the most common status,whereas in the other areas, moderate status remained the mostcommon. It is noteworthy that the probability of achievinggood or high Secchi depth status decreased in the Westernouter archipelago when nutrient status was set at good or highstatus (Table II).The model can be used also to calculate backward from

desired aims to the required inputs. In the Western inner andthe Western outer archipelago, the proportion of nutrientobservations expected to be found with good or high statusincreased when chl-a status was at good or high (Table II, thirdcolumn in each area). Nevertheless, moderate was still found tobe the most likely status of nutrients in most areas. Theexception was the Eastern outer archipelago, where the mostlikely state of nutrients was good or high.These results reinforce the need for reduction in both

phosphorus and nitrogen loading to fulfill the requirements ofthe WFD. It can also be seen that it is unlikely that all aims areachieved simultaneously. Even though phytoplankton biomassis affected by nutrient concentrations, reaching good ecologicalstatus in one does not naturally lead to good ecological status inthe other. Summertime Secchi depth is even less dependent onwinter nutrient concentration. On the other hand, a depend-ency between good chlorophyll-a and good Secchi depth statusis apparent, even though water clarity is affected by long-termchanges in other parameters besides algal biomass.

■ DISCUSSIONThe coastal Gulf of Finland is far from reaching good ecologicalstatus by 2015 as required by the WFD. Based on the currenttarget values set for chl-a,21 the outer coastal waters (WO andWE) were close to good status in the late 1970s and early 1980s(Figure 2). However, even at that time the Gulf of Finlandsuffered from eutrophication.46 This eutrophication can beexplained by a combination of factors; weakening verticalstability and an increase in nitrogen concentrations at all waterdepths,47 as well as by the deep water or sediment nutrientreserves together with rather unstable physical stratification inthe eastern archipelago.48,49 In addition, the hydrodynamics of

the Gulf of Finland play an important role in nutrient dynamics.The currents and the vertical mixing of the water contribute tothe amounts of nutrients in the productive water layer,50

creating an additional element of variability. However, thisvariability included in our data was taken into account in thesimilar way to that implemented in the ecological classificationof Finland. Thus, our results describe the uncertainty that willbe faced when using the water quality data to assess the statusof the Gulf of Finland.In contrast to the general pattern, the level of chl-a has been

higher in the western outer archipelago (WO) than in theeastern archipelago (EO) since the mid 1990s (Figure 2). Thepoor or bad status in the western outer water bodies has beennoticed because the situation could not be attributed to externalnutrient loading. However, similarly to the eastern Gulf ofFinland, near-bottom oxygen conditions have been weak in thewestern part19,51 and the general eutrophication level elevatedin the whole Gulf,52,53 suggesting the role of the benthic releaseof nutrients. In the eastern archipelago, chl-a concentrationshave recently decreased (Figure 2), probably as a consequenceof a decrease in the total fluxes of phosphorus.54 The results arein accordance with previous work,55 where the effects ofloading scenarios have been studied using ecosystem models.Previous work also took into account the purification ofeffluents in the St. Petersburg region.The effect of water pollution control measures carried out in

Finland since the late 1970s could be detected in the levels ofwintertime nutrient concentrations and summertime chl-aconcentration (Figure 2). The greatest changes in water qualityoccurred in the eastern inner coastal waters (EI), which in thisstudy mostly represent the Helsinki sea area. The pronounceddecline of tot-P in the Helsinki sea area in the mid 1970sresulted from the introduction of the chemical removal ofphosphorus. Nitrogen concentrations started to decrease afterthe closing of local municipal treatment plants and redirectionof the purified waste waters to the outer archipelago.20

Consequently, chl-a concentrations did not start to decreasebefore the reduction of nitrogen in the water column (Figure2). This demonstrates the importance of reducing the loads ofboth nitrogen and phosphorus in order to affect eutrophicationstatus. However, increases and peaks of chl-a in all areas in the2000s seemed to follow total phosphorus concentration (Figure2). The elevated concentrations of wintertime total phospho-rus, in turn, could at least partly explained by increasedphosphorus reserves as a consequence of benthic release ofphosphorus. This phenomenon is recorded extensively in theGulf of Finland during the 1990s and 2000s.17,19 Bentic releaseof phosphorus also occurs in the inner archipelagos,20 whichexplains the increased phosphorus levels in EI in cases whereHelsinki was excluded from the analyses.We have demonstrated that, in the light of the data, the

probability to achieve a desired impact in chl-a levels given thedecreased nutrient level was low in most areas. The targetedGood/moderate (good/moderate) boundary did not exceed aprobability of 0.28, 0.09, 0.36, and 0.59 for the areas WI, WO,EI, and EO areas respectively. Neither phosphorus nor nitrogenalone controls summer-time plankton growth, which wasdemonstrated by neither nutrient being more consistently ingood status when good chl-a status was met (Table II). Theseconclusions were reinforced by the time-series analysis (Figure2). Nevertheless, the fact that the probability of good chl-astatus increased with increasing nutrient status in the Westernand Eastern inner Archipelago areas indicates a dependency

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288225

Page 7: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

between the metrics. The weakness of this dependency can beinterpreted as an unharmonised target-setting56 where thedependencies, variability and uncertainty have not been takeninto account.The Gulf of Finland has such variability that it is difficult to

promise a certain result by manipulating the nutrients. We takethe next step from the type of analysis conducted in previouswork,57 where the importance of various drivers has beenaddressed, but without assessing the probabilities linked tothem.The relationship between winter nutrient concentration and

summer algal biomass is not straightforward since it involvesinterlinked ecological processes. Most importantly, the develop-ment and fall of a light and nutrient initiated phytoplanktonspring bloom affects not only the early summer nutrientavailability58,59 but also upwelling of phosphorus andsubsequent phytoplankton increase later during the year. Thisaffects the sediment-nutrient processes after bloom decay andsedimentation,16,60 which causes inconsistency between winterand summer observations. There have been speculations onwhether an environmental threshold has been passed, inhibitingthe recovery from eutrophication.18 The so-called vicious cycle,including enhanced hypoxia due to added input of organicmaterial to the bottom causing internal loading of phosphorusalso to the coastal surface waters. This may result in acompletely new relationship between the levels of winternutrients and summer-time algal biomass. In addition, all formsof nutrients are not equally available to phytoplankton. Possiblelong-term changes in the proportional abundances of thedifferent forms is not taken into account in the presentclassification when using total nutrient concentrations as waterquality elements instead of, for example, the dissolved inorganicfraction.The probabilistic model found a strong dependency between

Secchi-depth and chl-a status (Table II, Figure 2). In three ofthe four areas, good ecological status in chl-a is related to goodecological status in Secchi depth. These metrics do not sufferfrom differences caused by an interannual time lag, since bothare measured during the summer period, most oftensimultaneously. Yet a complicated relationship does exist,since the water optics determining Secchi depth are affected byother factors besides phytoplankton abundance: other inorganicand organic matter, which in turn may be subject of long-termchange in loading, increase attenuation of light in the watercolumn.61,62 Taking this into account, the targets of chl-a andSecchi depth seem well harmonized except for in the EasternInner Archipelago area.In our work, the relationship between nutrients and Secchi

depth has been modeled assuming that increased nutrientconcentrations result in added algal biomass (represented byChl-a in our model), which, in turn, decreases Secchi depth.61

By disregarding any other pathways from nutrients to Secchi,the model implicitly assumes that this is the only relevantrelationship between them. However, phytoplankton is not theonly, or possibly not even the major factor behind decreasedSecchi depths.63 Some of the nonalgal matter increasing lightattenuation in the water column, such as dissolved humicsubstances, may contain nutrients.64 This matter, while difficultor impossible to utilize by the phytoplankton, would show inthe total nutrient concentrations. In the Western innerArchipelago area, good Secchi depth status seems to be linkedto good nutrient status as closely as it is to chl-a status. Thenagain, in the other areas, no dependency or a very weak one

exists. The model structure, the complex interactions and waterclarity explain why the level of Secchi depth relatescontroversially to the level of nutrients in our model. However,these uncertainties exist in nature, and they must be taken intoaccount when assessing how strong nutrient reductions areneeded in order to achieve the desired chl-a and Secchi depths.The probabilistic methodology employed is a flexible tool for

examining ecological data.65 In this work, the tool was usedsimply for creating an easy-to-understand interface forexploring the relationships of the variables in the data. Noother information, such as Bayesian priors,66 was used besidesthe data. Therefore, the model simply serves as a tool whichprovides an accessible interpretation of the data following theprocedures being taken in the WFD.One of the major challenges in applying Bayesian networks

to environmental data is that the BNs can handle continuousdistributions67 in a very limited manner, and the data oftenneeds to be discretized, divided into bins, to be examined in aBN.65,68 While discrete data are natural to some domains (e.g.,presence/absence), ecological and environmental data are mostoften numerical and continuous. Finding the breakpoints forthe bins is difficult and affects the results.37 In this case, thebreak points are found by the limits set for the ecologicalquality classes. In a sense, the Bayesian network approach canbe seen as conceptually close to the logic applied in thedirective: we are interested in which class bin the value fallsinto, not about whether it is close to the good or the bad side ofthat class. This makes BNs a suitable choice for modeling theWFD classification data and assessing the results of applying itsprocedures. BNs allow examining interrelations of variables,their strength and different scenarios of nutrients reduction.However, these models need to be developed and interpretedwith domain experts to avoid misinterpretations and there is aneed to utilize other information sources than the data only inorder to decrease uncertainties. These include ecosystemmodels, published papers, and expert knowledge.

■ AUTHOR INFORMATIONCorresponding Author*E-mail:[email protected] AddressUniversity of East Anglia, Norwich Research Park, Norwich,NR4 7TJ, UK Phone: +44 (0)1603591375.NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSThis research is supported by a fellowship of the by SpanishMinistry of Education for European PhD mention and acontract by University of Helsinki in the FEM group (IBAMproject). “Integrated Bayesian risk analysis of ecosystemmanagement in the Gulf of Finland” (IBAM) is funded bythe European Community’s Seventh Framework Programunder Grant Agreement No. 217246. IBAN is made with thejoint Baltic Sea research and development program BONUSand the Academy of Finland. We are grateful to Paivi Korpinenfor help with data and Marco Nurmi for his help with GIS map.

■ REFERENCES(1) Borja, A.; Galparsoro, I.; Irigoien, X.; Iriondoa, A.; Menchaca, I.;Muxika, I.; Pascual, M.; Quincoces, I.; Revilla, M.; Rodríguez, J. G.;Santurtun, M.; Solaun, O.; Uriarte, A.; Valencia, V.; Zorita, I.

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288226

Page 8: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

Implementation of the European Marine Strategy FrameworkDirective: A methodological approach for the assessment ofenvironmental status, from the Basque Country (Bay of Biscay).Mar. Pollut. Bull. 2011, 62 (5), 889−904.(2) Burke, M. Unified European water quality regulatory systemproposed. Environ. Sci. Technol. 1997, 31 (5), 223A.(3) Borja, A.; Elliott, M.; Carstensen, J.; Heiskanen, A.; van de Bund,W. Marine managementTowards an integrated implementation ofthe European Marine Strategy Framework and the Water FrameworkDirectives. Mar. Pollut. Bull. 2010, 60 (12), 2175−2186.(4) Van Hoey, G.; Borja, A.; Birchenough, S.; Buhl-Mortensen, L.;Degraer, S.; Fleischer, D.; Kerckhof, F.; Magni, P.; Muxika, I.; Reiss,H.; Schroder, A.; Zettler, M. L. The use of benthic indicators inEurope: From the Water Framework Directive to the Marine StrategyFramework Directive. Mar. Pollut. Bull. 2010, 60 (12), 2187−2196.(5) European Commission. Directive 2000/60/EC of the EuropeanParliament and of the Council of 23 October 2000 establishing aframework for Community action in the field of water policy. OfficialJournal 22 December 2000 L 327/1. Brussels.(6) Nielsen, K.; Somod, B.; Ellegaard, C.; Krause-Jensen, D.Assessing reference conditions according to the European WaterFramework Directive using modelling and analysis of historical data:an example from Randers Fjord, Denmark. Ambio 2003, 32 (4), 287−294.(7) Kauppila, P.; Weckstrom, K.; Vaalgamaa, S.; Korhola, A.;Pitkanen, H.; Reuss, N.; Drew, S. Tracing pollution and recovery usingsediments in an urban estuary, northern Baltic Sea: are we far f romecological reference conditions. Mar. Ecol.: Prog. Ser. 2005, 290, 35−53.(8) Devlin, M.; Painting, S.; Best, M. Setting nutrients thresholds tosupport an ecological assessment based on nutrient enrichment,potential primary production and undesirable disturbance. Mar. Pollut.Bull. 2007, 55 (1−6), 65−73.(9) Garmendia, M.; Revilla, M.; Bald, J.; Franco, J.; Laza-Martínez, A.;Orive, E.; Seoane, S.; Valencia, V.; Borja, A. Phytoplanktoncommunities and biomass size structure (fractionated chlorophyll‘‘a’’), along trophic gradients of the Basque coast (northern Spain).Biogeochemistry 2010, 106 (2), 243−263.(10) Andren, E.; Andren, T.; Kunzendorf, H. Holocene history of theBaltic Sea as a background for assessing records of human impact inthe sediments of the Gotland Basin. Holocene 2000, 10 (6), 687−702.(11) Poutanen, E. L.; Nikkila, K. Carotenoid pigments as tracers ofcyanobacterial blooms in recent and post-glacial sediments of theBaltic Sea. Ambio 2001, 30 (4), 179−183.(12) Ranft, S.; Pesch, R.; Schroder, W.; Boedeker, D.; Paulomaki, H.;Fagerli, H. Eutrophication assessment of the Baltic Sea ProtectedAreas by available data and GIS technologies. Mar. Pollut. Bull. 2011,63 (5−12), 209−214.(13) Voss, M.; Emeis, K. C.; Hille, S.; Neumann, T.; Dippner, J. W.Nitrogen cycle of the Baltic Sea from an isotopic perspective. GlobalBiogeochem. Cycles 2005, 19 (3), 1−15.(14) Golubkov, S.; Alimov, A. Ecosystem changes in the NevaEstuary (Baltic Sea): Natural dynamics or response to anthropogenicimpacts? Mar. Pollut. Bull. 2010, 61 (4−6), 198−204.(15) Kahru, M.; Horstmann, U.; Rud, O. Satellite detection ofincreased cyanobacteria blooms in the Baltic Sea: Natural fluctuationsor ecosystem change? Ambio 1994, 23 (8), 469−472.(16) Lehtoranta, J. Dynamics of sediment phosphorus in the brackishGulf of Finland. Monogr. Boreal Environ. Res. 2003, 24, 1−58.(17) Conley, D. J.; Stockenberg, A.; Carman, R.; Johnstone, R. W.;Rahm, L.; Wulff, F. Sediment-water nutrient fluxes in the Gulf ofFinland, Baltic Sea. Estuarine Coastal Shelf Sci. 1997, 45 (5), 591−598.(18) Vahtera, E.; Conley, D. J.; Gustafsson, B. G.; Kuosa, H.;Pitkanen, H.; Savchuk, O. P.; Tamminen, T.; Viitasalo, M.; Voss, M.;Wasmund, N.; Wulff, F. Internal ecosystem feedbacks enhancenitrogen-fixing cyanobacteria blooms and complicate management inthe Baltic Sea. Ambio 2007, 36 (2), 186−194.(19) Pitkanen, H.; Lehtoranta, J.; Raike, A. Internal nutrient fluxescounteract decreases in external load: the case of the estuarial easternGulf of Finland, Baltic Sea. Ambio 2001, 30 (4), 195−201.

(20) Kauppila, P. Phytoplankton quantity as an indicator ofeutrophication in Finnish coastal watersApplications within theWater Framework Directive. Monogr. Boreal Environ. Res. 2007, 31, 1−60.(21) Vuori, K.-M.; Back, S.; Hellsten, S.; Holopainen, L. L.; Jarvinen,M.; Kauppila, P.; Kuoppala, M.; Lax, H. G.; Lepisto, L.; Marttunen,M.; Mitikka, S.; Mykra, H.; Niemi, J.; Olin, M.; Perus, J.; Pilke, A.;Rask, M.; Ruuskanen, A.; Vehanen, T.; Westberg, V. Vertailuolot jaluokan maarittaminen. Vertailuolot ja luokan maarittaminen. InPintavesien ekologisen tilan luokittelu Ympa ristohallinnon ohjeita 3/2009; Vuori, K.-M., Mitikka, S., Vuoristo, H., Eds.; FinnishEnvironment Institute: Ymparistohallinnon ohjeita 3/2009: Helsinki,2009; (in Finnish); pp 9−80.(22) Pawlak, J. F.; Laamanen, M.; Andersen, J. H. Eutrophication inthe Baltic Sea - An integrated thematic assessment of the effects ofnutrient enrichment and eutrophication in the Baltic Sea region. BalticSea Environ. Proc. No 115B, Helsinki Commission, Baltic MarineEnvironment Protection Commission, 2009.(23) Jørgensen, B. B.; Richardson, K., Eds. Eutrophication: definition,history and effects. In Eutrophication in Coastal Marine Ecosystems;Coastal Estuarine Studies 52, American Geophysical Union:Washington, DC, 1996; pp 1−19.(24) Suggestions for a Typology of Coastal Waters for the Finnish CoastAccording to the European Union Water Framework Directive (2000/60/EC); Kangas, P., Back, S., Kauppila, P., Eds.; Mimeograph Series ofFinnish Environment Institute, 2003; Vol. 284, p 51.(25) Lorenzen, C. J. Determination of chl and pheopigments:spectrophotometric equations. Limnol. Oceanogr. 1967, 12 (2), 343−346.(26) Borja, A. The European Water Framework Directive: achallenge for nearshore, coastal and continental shelf research. Cont.Shelf Res. 2005, 25 (14), 1768−1783.(27) Mann, H. B. Nonparametric test against trend. Econometrica1945, 13, 245−259.(28) Wasmund, N.; Uhlig, S. Phytoplankton trends in the Baltic Sea.ICES. J. Mar. Sci. 2003, 60 (2), 177−186.(29) Suikkanen, S.; Laamanen, M.; Huttunen, M. Long-term changesin summer phytoplankton communities of the open northern BalticSea. Estuar. Coast. Shelf Sci. 2006, 71 (3−4), 580−592.(30) Danielsson, Å.; Papush, L.; Rahm, L. A. Alterations in nutrientlimitations - Scenarios of a changing Baltic Sea. J. Mar. Syst. 2008, 73(3), 263−283.(31) Heckerman, D.; Geiger, D.; Chickering, D. Learning Bayesiannetworks: The combination of knowledge and statistical data. Mach.Learn. 1995, 20 (3), 197−243.(32) Jensen, F.; Nielsen, T. Bayesian Networks and Decision Graphs;Springer-Verlang: New York, 2001.(33) Neapolitan, R. Learning Bayesian Networks; Pearson PracticeHall: Upper Saddle River, NJ, 2003.(34) Carringer, J. F.; Barron, G. Minimizing Risks from Spilled Oil toEcosystem Services Using Influence Diagrams: The DeepwaterHorizon Spill Response. Environ. Sci. Technol. 2011, 45 (18), 7631−7639.(35) Wiesner, M. R.; Lowry, G. V.; Jones, K. L.; Hochella, M. F.; DiGiulio, R. T.; Casman, E.; Bernhardt, E. Decreasing uncertainties inassessing environmental exposure, risk, and ecological implications ofnanomaterials. Environ. Sci. Technol. 2009, 43 (17), 6458−6462.(36) Stelzenmuller, V.; Lee, J.; Garnacho, E.; Rogers, S. I. Assessmentof a Bayesian Belief Network−GIS framework as a practical tool tosupport marine planning. Mar. Pollut. Bull. 2010, 60 (10), 143−1754.(37) Fernandes, J. A.; Irigoien, X.; Goikoetxea, N.; Lozano, J. A.; Inza,I.; Perez, A.; Bode, A. Fish recruitment prediction, using robustsupervised classification methods. Ecol. Model. 2010, 221 (2), 338−352.(38) Andonegi, E.; Fernandes, J. A.; Quincoces, I.; Uriarte, A.; Perez,A.; Howell, D.; Stefansson, G. The potential use of a Gadget model topredict stock responses to climate change in combination withBayesian Networks: the case of the Bay of Biscay anchovy. ICES. J.Mar. Sci. 2011, 68 (6), 1257−1269.

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288227

Page 9: Evaluation of Reaching the Targets of the Water Framework Directive in the Gulf of Finland

(39) Varis, O.; Kettunen, J.; Sirvio, H. Bayesian influence diagramapproach to complex environmental management including observa-tional design. Comput. Stat. Data An. 1990, 9 (1), 77−91.(40) Varis, O.; Kajander, T.; Lemmela, R. Climate and water: On thesearch for improved links from climate models to water resourcesmanagement and vice versa. Clim. Change 2004, 66 (3), 321−344.(41) Varis, O.; Keskinen, M. Policy Analysis for the Tonle Sap Lake,Cambodia - A Bayesian network model approach. Int. J. Water Resour.Dev. 2006, 22 (3), 417−431.(42) Pearl, J. Graphs, structural models, and causality. InComputation, Causation and Discovery; Glymour, C, Cooper, G. F.,Eds.; AAAI Press: Menlo Park, CA, 1999; pp 95−138.(43) Stiber, N. A.; Pantazidou, M.; Small, M. J. Expert systemmethodology for evaluating reductive dechlorination at TCE sites.Environ. Sci. Technol. 1999, 33 (17), 3012−3020.(44) Seto, C. J.; McRae, G. J. Reducing risk in basin scale CO2

sequestration: A framework for integrated monitoring design. Environ.Sci. Technol. 2011, 45 (3), 845−859.(45) Tucker, A.; Duplisea, D. Bioinformatics tools in predictiveecology: Applications to fisheries. Phil. Trans. R. Soc. B 2012, 367(1586), 279−290.(46) Pitkanen, H.; Kangas, P.; Miettinen, V.; Ekholm, P. the Estate ofthe Finnish Coastal Waters in 1979−1983; National Board of Watersand the Environment. Publications of the Water and EnvironmentAdministration, 1987; No. 8, pp 167.(47) Perttila, M.; Savchuck, O.; Sphaer, I. Gulf of Finland,Hydrochemistry. HELCOM, 1996. Third periodic assessment of thestate of the marine environment of the Baltic Sea, 1989−1993;Background document. Baltic Sea Environ. Proc. 1996, 64B, 48−51.(48) Pitkanen, H. Nutrient dynamics and trophic conditions in theEastern Gulf of Finland: The regulatory role of the Neva estuary. AquaFenn. 1991, 21 (2), 105−115.(49) Pitkanen, H.; Tamminen, T.; Kangas, P.; Huttula, T.; Kivi, K.;Kuosa, H.; Sarkkula, J.; Eloheimo, K.; Kauppila, P.; Skakalsky, B. Latesummer trophic conditions in the north-east Gulf of Finland and theRiver Neva Estuary, Baltic Sea. Estuarine Coastal. Shelf Sci. 1993, 37(5), 453−474.(50) Pitkanen, H.; Lehtoranta, J.; Peltonen, H. The Gulf of Finland.In Ecology of Baltic coastal waters; Schiewer,U., Eds.; Ecological Studies,Springer: Berlin Heidelberg, 2008; Chapter 13, Vol. 197, pp 285−308.(51) Lehtoranta, J.; Ekholm, P.; Pitkanen, H. Eutrophication-drivensediment microbial processes can explain the regional variation inphosphorus concentrations between Baltic Sea sub-basins. J. Mar. Syst.2008, 74 (1−2), 495−504.(52) Pitkanen, H.; Kauppila, P.; Poutanen, E. L. Muutoksetravinnetilassa ja rehevyydessa. H. Pitkanen (toim Rannikko- jaavomerialueiden tila vuosituhannen vaihteessa. Suomen Itamerensuojeluohjelman taustaselvitykset. Suomen ympa risto 2004, 669, 104,(In Finnish).(53) Fleming-Lehtinen, V.; Laamanen, M.; Kuosa, H.; Haahti, H.;Olsonen, R. Long-term develpment of inorganic nutrients andchlorophyll α in the northern Baltic Sea. Ambio. 2008, 37 (2), 86−92.(54) HELCOM. The Fifth Baltic Sea Pollution Load Compilation(PLC-5). Baltic Sea Environ. Proc. No. 128. Helsinki Commission,2011, 217.(55) Pitkanen, H.; Kiirikki, M.; Savchuk, O.; Raike, A.; Korpinen, P.;Wulff, F. Searching efficient protection strategies for the eutrophiedGulf of Finland: The combined use of 1D and 3D modelling inassessing long-term state scenarios with high spatial resolution. Ambio2007, 36 (2), 272−279.(56) Borja, A.; Rodríguez, J. G. Problems associated with the ’one-out, all-out principle, when using multiple ecosystem components inassessing the ecological status of marine waters. Mar. Pollut. Bull. 2011,60 (8), 1143−1146.(57) Borja, A.; Galparsoro, I.; Solaun, O.; Muxika, I.; Tello, E. M.;Uriarte, A.; Valencia, V. The European Water Framework Directiveand the DPSIR, a methodological approach to assess the risk of failingto achieve good ecological status. Estuarine Coastal Shelf Sci. 2006, 66(1−2), 84−96.

(58) Hallfors, G.; Niemi, Å.; Ackerfors, H.; Lassig, J; Leppakoski, E.Biological oceanography. In The Baltic Sea; Voipio, A., Ed.; ElsevierOceanography Series, 1981; Vol. 30, pp 219−274.(59) Wasmund, N.; Nausch, G.; Matthaus, W. Phytoplankton springblooms in the southern Baltic Sea - spatio-temporal development andlong-term trends. J. Plankton Res. 1998, 20 (6), 1099−1117.(60) Vahtera, E.; Laanemets, J.; Pavelson, J.; Huttunen, M.; Kononen,K. Effect of upwelling on the pelagic environment and bloom-formingcyanobacteria in the western Gulf of Finland, Baltic Sea. J. Mar. Syst.2005, 58 (1−2), 67−82.(61) Preisendorfer, R. W. Secchi depth science: Visual optics ofnatural waters. Limnol. Oceanogr. 1986, 31 (5), 909−926.(62) Fleming-Lehtinen, V., Laamanen, M. Long-term changes inSecchi depth and the role of phytoplankton in explaining lightattenuation in the Baltic Sea. Estuarine Coastal Shelf Sci. 2012,doi:10.1016/j.ecss.2012.02.01(63) Irigoien, X.; Castel, J. Light limitation and distribution ofchlorophyll pigments in a highly turbid estuary: The Gironde (SWFrance). Estuarine Coastal Shelf Sci. 1997, 44 (4), 507−517.(64) Ubner, M.; Treuman, M.; Viitak, A.; Lopp, M. Properties ofhumic substances from the Baltic Sea and Lake Ermistu mud. J. Soil.Sed. 2004, 4 (1), 24−29.(65) Uusitalo, L. Advantages and challenges of Bayesian networks inenvironmental modelling. Ecol. Model. 2007, 203 (3−4), 312−318.(66) Humphries, R.; Jenkins, C.; Leuning, R.; Zegelin, S.; Griffith, D.W. T.; Caldow, C.; Berko, H.; Feitz, A. Atmospheric Tomography: ABayesian inversion technique for determining the rate and location offugitive emissions. Environ. Sci. Technol. 2012, 46 (3), 1739−1746.(67) Perez, A.; Larranaga, P.; Inza, I. Bayesian classifiers based onkernel density estimation: Flexible classifiers. Int. J. ApproximateReasoning 2009, 50 (2), 341−362.(68) Huang, D. B.; Scholz, R. W.; Gujer, W.; Chitwood, D. E.;Loukopoulos, P.; Schertenleib, R.; Siegrist, H. Discrete eventsimulation for exploring strategies: An urban water managementcase. Environ. Sci. Technol. 2007, 41 (3), 915−921.

Environmental Science & Technology Article

dx.doi.org/10.1021/es300126b | Environ. Sci. Technol. 2012, 46, 8220−82288228