LONG-TERM CHANGES AND IMPACTS OF HYPOXIA IN DANISH COASTAL WATERS

Transcription

1 Ecological Applications, 17(5) Supplement, 2007, pp. S165 S184 Ó 2007 by the Ecological Society of America LONG-TERM CHANGES AND IMPACTS OF HYPOXIA IN DANISH COASTAL WATERS DANIEL J. CONLEY, 1,2,4 JACOB CARSTENSEN, 1,3 GUNNI ÆRTEBJERG, 1 PETER BONDO CHRISTENSEN, 1 TAGE DALSGAARD, 1 JøRGEN L. S. HANSEN, 1 AND ALF B. JOSEFSON 1 1 Department of Marine Ecology, National Environmental Research Institute, P.O. Box 358, DK-4000 Roskilde, Denmark 2 Department of Marine Ecology, University of Aarhus, Finlandsgade 14, DK-8200 A rhus, Denmark 3 European Commission, Joint Research Centre, Institute for Environment and Sustainability, TP 280, I Ispra (VA), Italy Abstract. A 38-year record of bottom-water dissolved oxygen concentrations in coastal marine ecosystems around Denmark ( ) and a longer, partially reconstructed record of total nitrogen (TN) inputs ( ) were assembled with the purpose of describing longterm patterns in hypoxia and anoxia. In addition, interannual variations in bottom-water oxygen concentrations were analyzed in relation to various explanatory variables (bottom temperature, wind speed, advective transport, TN loading). Reconstructed TN loads peaked in the 1980s, with a gradual decline to the present, commensurate with a legislated nutrient reduction strategy. Mean bottom-water oxygen concentrations during summer have significantly declined in coastal marine ecosystems, decreasing substantially during the 1980s and were extremely variable thereafter. Despite decreasing TN loads, the worst hypoxic event ever recorded in open waters occurred in For estuaries and coastal areas, bottomwater oxygen concentrations were best described by TN input from land and wind speed in July September, explaining 52% of the interannual variation in concentrations. For open sea areas, bottom-water oxygen concentrations were also modulated by TN input from land; however, additional significant variables included advective transport of water and Skagerrak surface-water temperature and explained 49% of interannual variations in concentrations. Reductions in the number of benthic species and alpha diversity were significantly related to the duration of the 2002 hypoxic event. Gradual decreases in diversity measures (number of species and alpha diversity) over the first 2 4 weeks show that the benthic community undergoes significant changes before the duration of hypoxia is severe enough to cause the community to collapse. Enhanced sediment water fluxes of NH þ 4 and PO 3 4 occur with hypoxia, increasing nutrient concentrations in the water column and stimulating additional phytoplankton production. Repeated hypoxic events have changed the character of benthic communities and how organic matter is processed in sediments. Our data suggest that repeated hypoxic events lead to an increase in susceptibility of Danish waters to eutrophication and further hypoxia. Key words: anoxia; benthic communities; diversity; estuaries; hypoxia; nutrient enrichment; sediment water fluxes; Skive Fjord; thresholds. INTRODUCTION Hypoxia, defined here as,2 mg dissolved oxygen/l, is one of the common effects of eutrophication of coastal marine ecosystems worldwide (Nixon 1995). In Denmark, the first widespread severe hypoxia event was recorded in 1981 and large areas of the Kattegat, Sound, and Belt Seas had low oxygen concentrations and hydrogen sulfide (H 2 S) reported (Ehrhardt and Wenck 1984). Since that time, more or less widespread hypoxia has been observed nearly every autumn in Danish coastal waters. Extensive media coverage of the 1986 Manuscript received 16 May 2005; revised 16 February 2006; accepted 6 March 2006; final version received 18 August Corresponding Editor: A. R. Townsend. For reprints of this Special Issue, see footnote 1, p. S1. 4 Present address: Lund University, Department of Geology, GeoBiosphere Centre, So lvegatan 12, Lund SE Sweden. daniel.conley@geol.lu.se S165 oxygen depletion event lead to the implementation of a nationwide effort to reduce nutrient loads to the aquatic environment (Conley et al. 2002b). In addition, hypoxia has been commonly observed in Danish estuaries (Conley et al. 2000, Fallesen et al. 2000), with oxygendeficient water intermittently reported throughout the last century (Conley and Josefson 2001). The most severe and widespread hypoxia recorded in the open waters around Denmark occurred in late summer and autumn The event resulted from a combination of unusually high precipitation during winter, which washed nitrogen into the sea, and unusually calm and warm weather in late summer and autumn, limiting the supply of oxygen to the bottom water (HELCOM 2003). In km 2 (21% of the total bottom area) was exposed to oxygen concentrations,2 mg/l for up to 20 weeks, and km 2 (47%) were exposed to,4 mg/l (Fig. 1). By compar-

2 S166 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue FIG. 1. The coverage of hypoxia (defined as a level of dissolved oxygen,2 mg/l) measured in early October (A) in an average year (2001) and (B) in the most extreme year (2002). The gray areas represent the total coastal areas studied around Denmark, and the black areas are the hypoxic regions. ison, in 2001 bottom-water oxygen concentrations,2 mg/l only covered ;2900 km 2 (7%) and lasted 8 10 weeks, which is typical for Danish coastal waters (HELCOM 2003). Oxygen depletion events in Danish estuaries are not in phase with the open waters, with the most severe episode of estuarine hypoxia reported in 1997 when a 20-km stretch of the inner part of Mariager Fjord went completely anoxic (Fallesen et al. 2000). The causes of hypoxia are related to increasing nutrient loads, although physical factors influence the timing and extent of hypoxia (Diaz 2001, Scavia et al. 2003, Hagy et al. 2004). Nutrient loads to the marine environment in Denmark have been measured since 1989 (Conley et al. 2002b), and records of fertilizer use show rapid increases after World War II (Richardson 1996). Paleoecological reconstruction of total nitrogen (TN) loading in Roskilde Fjord, Denmark, show rapid increases in estuarine TN concentrations beginning in the 1950s, essentially doubling TN since the turn of the century (Clarke et al. 2003). Primary production in the Great Belt has doubled from the 1950s to the late 1970s (Ærtebjerg Nielsen and Gargas 1984) and more than doubled from the 1950s to the late 1980s in the Kattegat (Richardson and Heilmann 1995). These increases are greater than could be accounted for by climate variation. However, estimates of the actual nutrient load through time are not well documented prior to systematic monitoring that began with the creation of the Danish Nationwide Aquatic Monitoring and Assessment Program (DNAMAP) in 1988 (Kronvang et al. 1993, Conley et al. 2002b). The response of marine benthic macrofaunal communities to sporadic and seasonal hypoxia is a collapse or mass mortality when oxygen levels decline below some limit, i.e.,,2 mg/l (Josefson and Widbom 1988, reviews by Diaz and Rosenberg 1995, Rabalais et al. 2002). Yet it is well known from experimental work (e.g., Dries and Theede 1974) that different species have different tolerances against hypoxic conditions (see also review by Gray et al. 2002). While we should expect a progressive change in diversity and structure of benthic communities in response to decreasing oxygen concentrations, field documentation of changes in response to oxygen deficiency prior to the collapse of the community are few, in particular for the coupling between the temporal duration of hypoxic conditions and changes in community structure. In the marine environment mineralization of organic matter occurs both in the water column and in sediments. In Chesapeake Bay at water depths,5 m benthic respiration was significantly more important than pelagic respiration (Kemp et al. 1992). Nutrients regenerated by mineralization may be returned to the photic zone and be a source of new production (Rowe et al. 1975). This internal recycling of nutrients may be important, depending on season and loading from external sources, such as land and atmosphere. In a compilation of data from 10 different estuaries, Fisher et al. (1982) found that up to 79% and 75% of pelagic demand for nitrogen and phosphorus, respectively, was supplied from sediments. External loading for Danish waters is generally highest in winter and lowest in summer (Nielsen et al. 1995, 2001, Sundba ck et al. 2000, Dalsgaard 2003). In shallow estuaries, prevalent in Denmark (Conley et al. 2000), where light reaches the bottom, benthic microalgae may develop on the sediment surface. This microphytobenthic community may act as a barrier against transport of nutrients from

3 July 2007 HYPOXIA IN DENMARK S167 sediments to the water column. The degree to which nutrient release from sediments is regulated by microphytobenthos varies between locations as well as seasonally and diurnally (e.g., Nowicki and Nixon 1985, Sundbäck and Grane li 1988, Rysgaard et al. 1995, Cowan et al. 1996, Sundba ck et al. 2000). The objectives of this study were to document and analyze the changes in hypoxia in the open waters around Denmark and in Danish estuaries during the periods and , respectively, by analyzing the extensive data stored in the databases of DNAMAP (available online). 5 First, we determined long-term changes in TN load over the last century by combining present measurements of nutrient loads with estimates of surplus N from agriculture. Recognizing the potential of physical factors to modify the development of hypoxia in response to increased nutrient loading, we examined other explanatory factors that contribute to the variability observed in hypoxia. The ecological consequences of hypoxia were investigated for the benthic environment. We examined the impacts of severe hypoxic events on bottom communities in the open waters around Denmark following the severe hypoxic event in 2002 and investigated the response of sediment water fluxes to these events in a Danish estuary. METHODS Long-term trends in N loading Nitrogen surpluses from the Danish agricultural sector have been calculated back to the beginning of the 20th century (Kyllingsbæk 2000, Dalgaard and Kyllingsbæk 2003; A. Kyllingsbæk, unpublished data) on a yearly basis from July to the following June. Nitrogen surplus was calculated as nitrogen import, mainly as feed, fertilizers, atmospheric deposition, and nitrogen fixation, minus export of vegetable and animal products. Before 1920 nitrogen surplus did not include the southwestern part of Denmark (3978 km 2 ) that was part of Germany from 1864 to Therefore, nitrogen surpluses from 1900 to 1920 were scaled to represent the present area of Denmark. Nitrogen surplus data were compared to freshwater discharge for hydrological years (July through June) from Denmark to the Danish straits available since 1942 as well as the more recent (since 1989) comprehensive compilations of nutrient inputs from land established as part of DNAMAP (Conley et al. 2002b). Nutrient inputs were partitioned into point and diffuse sources. Flowweighted concentrations of TN were calculated as the diffuse terrestrial TN input divided by total freshwater discharge for each hydrological year. The dynamic relationship between agricultural nitrogen surplus and flow-weighted TN concentrations were analyzed within the framework of ARIMA time series models including prewhitening of the input and output series (Box and Jenkins 1976). The established relationship was extrapolated back in time to the hydrological year 1942/1943 using the freshwater discharge time series and further back to 1900 using the average freshwater discharge from 1942/1943 to 2002/2003. Before 1989 there were only sporadic assessments of TN loading from point sources. The Danish EPA (1996) estimated total Danish point source discharge to be ; kg for 1972, kg for 1977, and kg for Annual values for intermediate years were computed by linear interpolation. Biological nitrogen removal from municipal wastewater began in the late 1970s, whereas primary and secondary treatment started in the 1920s. Although implementing primary and secondary treatment may have removed a small nitrogen fraction, we estimate point sources from 1972 backwards to 1900 using population figures that doubled from ; to inhabitants (data available online). 6 Using the point source compilations, it was estimated that 76% of Danish point sources discharge directly to the Danish straits. Long-term trends in oxygen concentrations Conductivity, temperature, and depth (CTD) profiles of salinity, temperature, and oxygen concentrations, as well as oxygen concentrations from discrete water samples at specific depths, were analyzed using Winkler titration for coastal and open-water stations within the DNAMAP. Only stations with a minimum depth of 5 m were analyzed (Fig. 2). The density difference (sigma-t, Dr T ) from the surface to the bottom was calculated as an indicator of stratification. For coastal stations a density difference of Dr T. 0.5 was designated as being indicative of a stratified water column, whereas a density difference of Dr T. 1.0 was employed for open-water stations. If the water column was considered stratified the minimum oxygen concentration within the 2.5 m above the sediments was used. The depth of the minimum oxygen concentration was also recorded. Annual means of oxygen concentration for openwater stations as well as coastal stations were estimated from the minimum oxygen concentration near the bottom during July through November and under stratified conditions from a general linear model that included variations between stations, year, months, and depth of the minimum oxygen concentration. Oxygen concentration ¼ station þ year þ month þ sample depth: ð1þ Due to heterogeneity in the monitoring data, least squares marginal means were calculated (Searle et al. 1980) to derive annual means of oxygen concentration that were not biased by differences in stations, months, and depths sampled in different years. Oxygen data were 5 hhttp://mads.dmu.dk/i 6 hhttp://

4 S168 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue FIG. 2. Map of the Danish waters with the location of monitoring stations used in the study: DNAMAP monitoring stations in open waters, stations in Danish estuaries and coastal marine areas, stations sampled for benthic fauna, and stations in Skagerrak surface waters. occasionally available before regular monitoring programs were established, although the number of observations was low and consequently the yearly estimates uncertain. Therefore, least-squares oxygen concentrations with an associated standard error. 1 mg/l were not used. Trend analysis was carried out by weighted linear regression using reciprocal variances of the mean values as weights (best linear unbiased estimation). Long-term trends in temperature Measurements of temperature at coastal stations, at open-water stations located in the Danish straits, and at 19 stations located in the Skagerrak and the northern part of the Skagerrak-Kattegat front were analyzed to describe long-term trends in temperature. For coastal and open-water stations in the Danish straits, temperature in the bottom layer (,2.5 m from the bottom and deeper than 5 m) was investigated during the warm summer and autumn months (July September). The bottom water of the Danish straits originates from the Skagerrak where surface water penetrates into the Kattegat and the Belt Sea as a bottom current, typically in winter and early spring. Therefore, the temperature of Skagerrak surface water (top 10 m) was investigated during January April. The annual mean temperature was estimated by means of the following general linear model that described variations between stations, year, months: Temperature ¼ station þ year þ month: ð2þ Least squares marginal means of temperature were calculated (Searle et al. 1980) to account for differences in stations and months sampled during different years. For some of the earlier years there were only a few temperature observations, and yearly means were consequently uncertain. Thus, temperature levels with a standard error of the mean.18c were not used. Trend analysis was carried out by weighted linear regression using the reciprocal variances of mean values as weights (best linear unbiased estimation). Long-term trends in summer winds Observations of wind speed and direction were obtained from two separate and overlapping time series

5 July 2007 HYPOXIA IN DENMARK S169 at Sprogø located in the middle of the Great Belt (data source: Sund and Bælt Holding A/S) and Risø near Roskilde (data source: Department of Wind Energy, Risø National Laboratory). The Sprogø meteorology mast began monitoring in September 1977 and continued until September 1999, whereas data from the Risø meteorology mast were available from 1996 to Wind speed and direction was measured at 8 m height for the Sprogø time series and 7.7 m height for the Risø time series. Both time series had a temporal resolution of 10 min. The wind speed and direction was projected on a southeast northwest transect, since southeasterly winds enhance the outflow from the Baltic Sea into the Danish straits and northwesterly winds force water from the North Sea/Skagerrak into the Danish straits. For each of the two time series daily means of wind speed and the southeast northwest projection was calculated. These daily values were compared for the overlapping time periods of the two time series by calculating the mean difference and recalibrating the Risø data to the Sprogø time series. A combined time series was obtained by averaging over the Sprogø and the recalibrated Risø daily values. For this combined time series, yearly means of wind speed during July September and the projected southeast northwest wind speed during May September with data from 1977 excluded as the time series first began in September. These months were chosen for aggregation because oxygen depletion in coastal waters is at its highest during July September, and wind mixing has been shown to be important in disrupting oxygen depletion events in Danish waters (Møhlenberg 1999). In contrast, whereas open waters have almost permanent stratification with residence times of several months (Gustafsson 2000). Trend analysis was carried out by linear regression. Long-term trends in the advective transport Water transport in the bottom layer was calculated by the model of Gustafsson (2000), which is a hydrodynamic compartment model for the entire Baltic Sea with a 50-layer vertical resolution with depth. Layers contributing to surface and bottom transports were determined based on salinity levels of the different layers. The cumulative transport of water in the bottom layer was found for May September by aggregation for each year separately ( ). Trend analysis was carried out by linear regression. Interannual variations in oxygen concentrations Interannual variations in bottom-water oxygen concentrations during stratified conditions were analyzed in relation to various explanatory variables by means of weighted multiple regression with the following explanatory variables: (1) TN loading from land (July of one year to June the following year), (2) mean wind speed in July September, and (3) mean temperature in bottom water in July November. For bottom-water oxygen concentrations at open-water stations the following explanatory variables were examined: (1) TN loading from land (July of one year to June the following year), (2) southeast northwest wind projection, (3) bottomwater influx from the Kattegat to the Great Belt, (4) bottom-water influx from the Great Belt to the Fehmarn Belt, (5) surface-water temperature in the Skagerrak during January April, and (6) bottom-water temperature in the Danish straits during July November. A stepwise multiple linear regression was carried out with backward elimination, i.e., eliminating the least significant factor, until all explanatory variables were significant. The reciprocal variances of mean oxygen concentrations were used as weights (best linear unbiased estimation). The influence of individual observations on the multiple regression was investigated by calculating Cook s distance. Marginal relationships for each of the significant explanatory variables were found by inserting the mean value for the other explanatory variables into the regression equation. Benthic fauna As part of the DNAMAP benthic fauna were sampled in May 2002 and 2003 with a Haps corer from 24 locations (Fig. 2) in Danish estuaries, coastal areas, and open waters. From each location, individual samples were taken arranged in a grid covering an area between 0.2 and 290 km 2 in an area of uniform depth and homogenous sediment composition (Kaas and Markager 1998). All organisms retained on a 1-mm sieve were specified to species level and quantified in terms of abundance and biomass. From each fauna sampling grid area bottom-water oxygen concentrations and the duration of concentration levels in bottom waters were either directly measured or estimated with a depth area extrapolation model (HELCOM 2003) using oxygen measurements at a number of sampling points within the grid. Sediment water fluxes Sampling was carried out in the central part of Skive Fjord, Denmark, a shallow estuary (Conley et al. 2000) at a water depth of 5.5 m ( N; E). During the period from May until end of September 1994 depth profiles of oxygen in the water column were measured 29 times; nutrients, chlorophyll a concentrations, and pelagic primary production were measured 16 times. All variables were measured using standard methods of the DNAMAP (Kaas and Markager 1998). Briefly, oxygen was measured with a profiling electrode, nutrients were measured with standard colorimetric techniques, chlorophyll a was analyzed colorimetrically after ethanol extraction, and pelagic primary production was measured using short-time 14 C incubations. Intact sediment cores were sampled in plexiglass tubes (5.2 cm inner diameter, 30 cm long) by scuba diving. Sediment cores were transported at in situ temperature, sealed with rubber stoppers. After returning to the laboratory the upper stopper was removed, and a

6 S170 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue FIG. 3. Freshwater discharge from Denmark to the Danish straits for hydrological years (July June ). Teflon-coated magnetic stirring bar was placed 5 cm above the sediment surface and rotated at 60 rpm by a large external rotating magnet. Cores were submersed in bottom water from the station at in situ temperature and left overnight to recover from any sampling disturbances. The following day, sediment water fluxes of O 2, NO 3 þ NO 2,NH þ 4, and PO 3 4 were measured on five cores in light and five cores in darkness (Rysgaard et al. 1995). Sediment cores were closed with transparent lids and concentrations were measured at the start and end of the incubation. Samples for nutrient analysis were filtered through a GF/C filter into polypropylene vials and frozen for later analysis. Ammonia, NO 3 þ NO 2, and PO 3 4 were measured with standard techniques (Grasshoff et al. 1983). Oxygen concentrations were measured with a standard Winkler technique (Grasshoff et al. 1983). Incubation time was selected so the decrease in oxygen concentration during a dark incubation was between 10% and 20% of saturation. For the light incubations, light intensity was ;75 lmol photonsm 2 s 1, provided by an Osram Powerstar HQI-T fluorescent bulb (Osram, Munich, Germany). Benthic gross primary production was calculated as the sediment water oxygen flux in light minus that in darkness (flux out of the sediment is positive and uptake by the sediment is negative). Furthermore, an O 2 production to CO 2 fixation ratio of 1 was assumed. Diurnal sediment water fluxes of nutrients were calculated as the mean of light and dark fluxes assuming that the day and night period were of roughly equal duration. RESULTS Long-term trends in N loading Freshwater discharge from Denmark over the period of record ( ) averaged 8.54 km 3 /yr, although there were substantial interannual variations ranging from 4.77 km 3 /yr in 1995/1996 to 12.8 km 3 /yr in 1980/1981 (Fig. 3). On a decadal basis the 1970s were dry, the 1980s were wet, and the 1990s were variable with two out of the three driest years and two out of the fours years with highest rainfall. Large variations were also reflected in diffuse TN loading from land (Fig. 4) where TN inputs were more than three times larger in the 1993/1994 wet year compared to 1995/1996 dry year, demonstrating the importance of diffuse N loading with runoff. The TN point source contribution gradually declined from ; kg/yr to kg/yr. The nitrogen balance of Danish agriculture had a surplus of kg/yr in the beginning of the 1990s, which declined to kg/yr after The area-specific surplus similarly declined from kg Nha 1 yr 1 in the early 1990s to,80 kg Nha 1 yr 1 in the two most recent hydrological years (Fig. 5). Identification of an auto-regressive integrated moving average (ARIMA) model, a widely used tool for time series modeling, from the cross-correlation between flow-weighted TN concentrations and nitrogen surplus revealed a direct relationship, without any lagged or memory effects (although there was a small, but insignificant one-year lag effect). Furthermore, there was a significant relationship between nitrogen surplus and flow-weighted TN concentration (n ¼ 14, t ¼ 34.14, P, ; Fig. 5) that was used for extrapolating backwards in time. Reconstruction of the total Danish TN input to the Danish straits showed that point sources gradually increased up to the mid-1970s (Fig. 6) with population increases. Biological nutrient removal of municipal and industrial wastewater reduced the point source contribution by ;80% from the 1970s to recent levels of ; kg/yr. Diffuse and point source TN inputs were of similar magnitude up to the mid-1960s when the diffuse contribution exploded. The diffuse TN input increased by more than a factor of four from the early 1900s to the 1980s, when it reached its highest rates (Fig. 6). In the two dry years of 1995/1996 and 1996/1997, diffuse TN

7 July 2007 HYPOXIA IN DENMARK S171 FIG. 4. Total nitrogen input from land to the Danish straits for hydrological years (July June) partitioned into diffuse and point sources. input was comparable to average levels in the 1950s. Total nitrogen input from land has decreased significantly from 1980 to 2003 (F 1,21 ¼ 30.45, P, ) by kg/yr. Subsequent significant declines in TN concentrations have also been noted in estuaries and coastal areas (Carstensen et al. 2006). Long-term trends in oxygen concentrations Mean bottom-water oxygen concentrations are integrated over large spatial scales and at stations with and without hypoxia (Fig. 2), and therefore mean concentrations are quite high, e.g., mg/l (Fig. 7). Mean bottom-water dissolved oxygen concentrations during stratified conditions (July November) at open-water stations were relatively constant from the mid-1960s to the mid-1980s, when it began to decline (Fig. 7A). Although bottom-water dissolved oxygen concentrations appeared to improve in with relatively high oxygen levels during dry years (Rask et al. 1999), the last five years of the study period were dominated by low bottom-water dissolved oxygen concentrations, culminating with a large oxygen deficiency in 2002 (HELCOM 2003). Over the entire period, mean dissolved oxygen concentrations under stratified conditions decreased significantly (F 1,36 ¼ 13.03, P ¼ ) by mgl 1 yr 1, corresponding to a decline of 1.4 mg/l over the 38 years used in the regression ( ). Data on bottom-water dissolved oxygen concentrations from coastal and estuarine stations were available only from 1981 onwards (Fig. 7B). No significant trends through time were observed over the period of record (F 1,21 ¼ 2.64, P ¼ ), although the decline over the 23 years had a magnitude comparable to that of the decline observed in open waters ( mgl 1 yr 1 ). In particular, the last five years of the study period had significantly lower dissolved oxygen concentrations ( 0.53 mg/l) than all the previous years (F 1,21 ¼ 10.75, P ¼ ). High bottom-water dissolved oxygen concentrations were observed in 1981, 1986, 1993, 1996, and 1998, and the two lowest mean bottom-water dissolved oxygen concentrations were observed in 1984 and Long-term trends in temperature Temperature generally increased in both the bottom layer during July September and in the Skagerrak surface layer during January April (Fig. 8A); however, the trend was significant for the coastal bottom layer only ( , F 1,25 ¼ 7.66, P ¼ ), whereas it was not significant for the open-water bottom layer ( , F 1,36 ¼ 3.24, P ¼ ) nor for the Skagerrak surface layer ( , F 1,35 ¼ 3.74, P ¼ FIG. 5. The diffuse total nitrogen (TN) input from land relative to freshwater discharge vs. the nitrogen surplus from the agricultural sector divided by the total agricultural area of Denmark ( km 2 in 1995). Data points are from individual hydrologic years determined from total freshwater inputs from all rivers during July June from 1989/1990 to 2002/2003.

8 S172 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue FIG. 6. Reconstruction of total nitrogen inputs to the Danish straits. Extensive loading compilations from DNAMAP are available from 1989 onward only. Interannual variations in the freshwater discharge were included after 1942/ ). Interannual variations in coastal and openwater bottom temperature were correlated (t test, n ¼ 27, r ¼ 0.50, P ¼ ), as were the Skagerrak surface temperature and open-water bottom temperature (t test, n ¼ 37, r ¼ 0.41, P ¼ ), whereas the Skagerrak surface temperature did not correlate with coastal bottom temperature (t test, n ¼ 27, r ¼ 0.23, P ¼ ). From the mid-1970s until 1987, the temperature FIG. 7. The mean dissolved oxygen concentration (July November) near the bottom during stratified conditions for (A) open-water stations and (B) coastal and estuarine stations in DNAMAP. Error bars show the 95% confidence intervals of the mean oxygen concentrations. was relatively low, increasing by 1.318C, 1.068C, and 0.688C for Skagerrak surface water, open-water bottom layer, and coastal bottom layer, respectively. Increases in the Skagerrak temperatures are consistent with oceanic scale circulation changes, in which advection of warmer water along the western edge of the European shelf into the North Sea and Skagerrak increased, beginning in 1987 (Reid et al. 1998, 2001), coincident with a trend for positive anomalies in the North American Oscillation (NAO). Increases in the Skagerrak surface-layer temperature correspond with a decrease in oxygen saturation concentration from mg/l to mg/l, a relative decrease of 3.2%. Long-term trends in summer winds The Risø meteorology mast had a mean wind speed 0.88 m/s higher than the Sprogø data. This also affected the southeast northwest projected wind speed that was 0.61 m/s higher at Risø. The mean wind speed during the hypoxia-prone months (July September) was typically ;6 m/s (Fig. 8B), although it decreased significantly from 1978 to 2003 (F 1,25 ¼ 16.09, P ¼ ). The trend was ms 1 yr 1, corresponding to a change of 0.93 m/s in the 27 years of wind data. Mean wind on the southeast northwest projection changed more markedly from years with dominating winds from southeast to years with more northwesterly winds (Fig. 8B). There was no trend in the wind transport on the southeast northwest projection (F 1,25 ¼ 1.19, P ¼ ). Moreover, annual mean wind speed and the southeast northwest projection were not correlated (t test, n ¼ 27, r ¼ 0.24, P ¼ ). Long-term trends in advective transport Mean bottom influx of water (May September) from the Kattegat to the Great Belt was 194 km 3 as opposed

9 July 2007 HYPOXIA IN DENMARK S173 to 146 km 3 from the Great Belt to the Fehmarn Belt. The two fluxes were significantly correlated (t test, n ¼ 27, r ¼ 0.414, P ¼ ), although fluxes appeared to diverge with time (Fig. 8C). This pattern was also clear from the trend analysis, in which there was no trend in influx of water from the Kattegat to the Great Belt (F 1,25 ¼ 0.02, P ¼ ), whereas there was a significant decreasing trend in influx of water from the Great Belt to the Fehmarn Belt (F 1,25 ¼ 4.42, P ¼ ). The trend was 1.39 km 3 /yr, corresponding to a total change of 37.5 km 3 over the 27 years modeled. Interannual variations in oxygen concentrations In Danish estuaries and coastal areas, bottom-water temperature did not explain interannual variations in bottom-water dissolved oxygen concentrations (F 1,19 ¼ 1.17, P ¼ ) and was eliminated as an explanatory variable, although data did show a negative gradient with respect to temperature ( mgl 1 8C 1 ). Subsequently, TN input from land (F 1,20 ¼ 4.40, P ¼ ) and mean wind speed (F 1,20 ¼ 21.47, P ¼ ) explained 52% of interannual variations in bottomwater dissolved oxygen concentrations in estuaries and coastal areas (Fig. 9). The most significant explanatory factor in the multilinear regression was mean wind speed in which the range in yearly values could explain a difference in dissolved oxygen concentration of 1.19 mg/l. Interannual variations in TN input from land, which were large in this period, explained a 0.77 mg/l difference in dissolved oxygen concentration. The largest Cook s distance was 0.26, suggesting that there was no particular observation with a high influence on the regression. For open-water stations backward elimination reduced the number of explanatory variables from six to three, as the southeast northwest wind projection was first eliminated (F 1,17 ¼ 0.08, P ¼ ), followed by bottom-water influx from the Kattegat to the Great Belt (F 1,21 ¼ 0.21, P ¼ ), and, finally, bottom-water temperature in the Danish straits (F 1,22 ¼ 1.11, P ¼ ). Eliminating wind projection as an explanatory variable allowed for the inclusion of three additional years of data ( ). The three significant explanatory variables were TN input from land (F 1,23 ¼ 8.55, P ¼ ), bottom-water influx from the Great Belt to the Fehmarn Belt (F 1,23 ¼ 5.85, P ¼ ), and Skagerrak surface-water temperature (F 1,23 ¼ 8.06, P ¼ ), which, when combined, explained 49% of interannual variations in bottom-water dissolved oxygen concentrations (Fig. 10). The largest Cook s distance was 0.48, and thus, there was no particular observation with a high influence on the regression. Actual ranges in TN input from land, bottom-water influx from the Great Belt to the Fehmarn Belt, and Skagerrak surface temperature ( ) corresponded to differences of 1.10 mg/l, 0.76 mg/l, and 1.00 mg/l in predicted dissolved oxygen concentrations, respectively. FIG. 8. Time series of temperature, wind speed, and water exchange in Danish coastal waters. (A) Yearly mean temperatures for the bottom layer (July September) for coastal stations and open-water stations in the Danish straits as well as yearly mean temperatures for Skagerrak surface water in January through April. (B) Mean wind speed (July September) and wind projection on the southeast northwest (SE NW) direction (positive values indicate a cumulative wind transport from northwest to southeast) (data sources: Department of Wind Energy, Risø National Laboratory, and Sund and Bælt Holding A/S). (C) Estimated bottom flux over the intersections between the Kattegat and the Great Belt and between the Great Belt and the Fehmarn Belt (May September). The fluxes were calculated by the model of Gustafsson (2000). Effects of 2002 oxygen depletion on benthic fauna In the autumn of 2002 the most severe and widespread hypoxia to date occurred in marine waters around Denmark. More than 21% of the total bottom area (9000 km 2 ) experienced bottom-water dissolved oxygen concentrations,2 mg/l (Fig. 1). In some places, bottom-water hypoxia lasted for.20 weeks, significantly longer than seasonal hypoxia (8 10 weeks) typically persists in bottom waters around Denmark in late summer and early fall (HELCOM 2003). Significant changes in benthic communities were observed following this severe hypoxic event, especially in areas that are not

10 S174 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue FIG. 9. Marginal relationships for bottom-water dissolved oxygen concentrations in estuaries and coastal areas to (A) total nitrogen loading and (B) wind speed ( ). Error bars show 95% confidence intervals of mean oxygen concentrations. often exposed to seasonal hypoxia. Sampling of benthic macrofauna immediately after the hypoxia event revealed that the benthic community was decimated in areas exposed to four to six weeks of hypoxia. Macroorganisms were partly or entirely eliminated in a km 2 area in the western Baltic and around the island of Fyn (Hansen et al. 2003). The total loss of macrofauna in 2002 was estimated to be Mg wet mass within the hypoxic area. Benthic community data collected prior to the hypoxic event as part of the annual sampling for DNAMAP was compared with data collected after the event in areas where the hypoxic event did not last long enough to cause a complete collapse of the community (Table 1, Fig. 11). A significantly higher species turnover/change in species composition measured by Bray-Curtis dissimilarity index was found in areas where bottom-water dissolved oxygen concentrations had fallen below 2 mg/l in autumn 2002 (P, 0.05). There was a larger reduction of number of species in areas where hypoxia occurred and, in particular, in areas where hypoxia was of a long duration. Diversity on the scale of the sampling device (centimeters), alpha diversity, was significantly reduced where bottom-water dissolved oxygen concentrations were,2 mg/l. Reductions in both number of species and alpha diversity were significantly correlated with duration of the hypoxic event (Fig. 11). Gradual decreases in diversity measures (number of species and alpha diversity) during the first two to four weeks of exposure to hypoxia show that the benthic community undergoes significant changes before hypoxia causes the community to collapse. These changes could only be detected, however, because the prehistory of the macrofauna community was well characterized through time by regular sampling of the benthic community as part of the DNAMAP. Effect of oxygen depletion on benthic fluxes in Skive Fjord Oxygen depletion events in Danish estuaries occur on different frequencies than those in open-water areas (Conley and Josefson 2001). In 1994, Skive Fjord in the Limfjorden experienced a severe hypoxic event during the height of summer stratification. Bottom-water oxygen concentrations were close to saturation until the latest part of June, and from 29 June until 11 August there was severe oxygen depletion in bottom waters (Fig. 12A). Sporadic small increases in bottom-water dissolved oxygen concentrations occurred during the first half of this period (21 days) and the second, most severe part of the hypoxia event (20 July until 11 August) lasted 22 days. After 11 August bottom-water dissolved oxygen concentrations increased and were never below 36% of the surface value. Water column PO 3 4 and NH þ 4 concentrations increased significantly when hypoxia became severe (Fig. 12B, C). After re-oxygenation of bottom waters PO 3 4 and NH þ 4 concentrations only dropped slowly. Surface-water PO 3 4 concentrations followed bottomwater concentrations with a time lag, but remained slightly lower. Ammonium concentrations in surface waters did not increase during the hypoxic period. Concentrations of NO 3 þ NO 2 were very similar in bottom and surface waters (Fig. 12D), decreased until the beginning of the hypoxic period, and remained low until the end of September. Chl a concentrations were similar in bottom and surface waters until the beginning of August, when surface concentrations were consistently higher (Fig. 12E). Pelagic primary production was reduced during the early part of hypoxia (Fig. 13A) with chl a in surface waters following the same pattern. Benthic primary production reached its maximum value on 14 May, comprising 44% of pelagic primary production and then decreased to low values during and after hypoxia (Fig. 13B). Some degree of mixing occurred between bottom and surface waters, as indicated by increases in PO 3 4 in surface waters (Fig. 12C), triggering a phytoplankton bloom in the photic zone during the later stages of hypoxia (Figs. 12E and 13A). Hypoxia caused large changes in benthic fluxes of þ nutrients (Fig. 14). Before the hypoxic event, both NH 4 and PO 3 4 fluxes were low and into the sediment (Fig. 14A, B). However, after the onset of low bottom-

11 July 2007 HYPOXIA IN DENMARK S175 water dissolved oxygen concentrations, there was a large release of NH 4 þ and PO 4 3. The ratios of NH 4 þ to PO 4 3 release, on the three sampling occasions during hypoxia, were 0.8, 0.7, and 6.3, respectively. After re-oxygenation of bottom waters PO 4 3 was still released, though at a lower rate, whereas NH 4 þ was again taken up by the sediment. Nitrate was taken up by the sediment throughout the period but at much lower rates during hypoxia (Fig. 14C). Both denitrification of NO 3 from the water column (D w ) and denitrification of nitrate from nitrification (D n ) decreased significantly during hypoxia and increased again when bottom-water dissolved oxygen concentrations increased (Fig. 14D). However, D n was still low at the first measurement after hypoxia, but recovered to earlier values in September. Before hypoxia visual inspection found there was a coherent brown diatom mat on the sediment surface. This mat had already disappeared in the beginning of July. As of 18 July the brown mat could be reestablished after exposure to one day in light. On 2 August, however, this was no longer possible, and it is assumed that the benthic diatoms were no longer viable. DISCUSSION Trends in nutrient loading Determining the load of nutrients through time is essential to the evaluation of the consequences of nutrient enrichment on coastal marine ecosystems (Nixon 1995, Billen and Garnier 1997, Conley 1999). Although it is recognized that anthropogenic activities have increased nutrient loading, the time scales and magnitude of those increases are poorly known. Few monitoring programs pre-date the 1980s, making longterm assessments of the nutrient load problematic. Different approaches can be taken to determine past nutrient loading, including simple nutrient budgets (Boynton et al. 1995) to sophisticated catchment-scale modeling of nutrient sources and sinks in watersheds (Billen and Garnier 1997) and paleoecological methodologies such as the use of diatom-based transfer functions (Clarke et al. 2003, 2006). We show here that the diffuse TN load to Danish waters during the period was strongly related to variations in surplus nitrogen from Danish agriculture superimposed on variations in runoff (Fig. 5), allowing us to reconstruct long-term changes in TN loading through time. Surplus nitrogen from agriculture, i.e., the difference between the amount of nitrogen imported to agriculture and exported as agricultural products, can be readily calculated from available agricultural statistics (Dalgaard and Kyllingsbæk 2003) and, combined with measured TN loads, may be an important way forward in the empirical reconstruction of TN loading through time in other ecosystems that lack monitoring data. Our reconstructed TN load from surplus nitrogen and runoff is a reasonable approximation of the development of the TN load from Denmark, although we may be overestimating diffuse source inputs during the FIG. 10. Marginal relationships for bottom-water dissolved oxygen concentrations in open waters to (A) total nitrogen loading, (B) bottom exchange, and (C) Skagerrak surface-water temperatures ( ). Error bars show the 95% confidence intervals of the mean oxygen concentrations. beginning of the 20th century when wetlands were more extensive and drainage practices were not as intense as they are today. Comparison with reconstructed values of TN concentrations in a Danish estuary, Roskilde Fjord, using paleoecological diatom-based transfer functions (Clarke et al. 2003, 2006) shows similar increases in the timing and magnitude of TN loading. The time scale of increases in nitrate load to Chesapeake Bay (Hagy et al. 2004) are similar, with a doubling in nitrate concentrations observed in the 1970s, reflecting the growth of fertilizer usage, the industrialization of agriculture, and increases in population. Comparison of several coastal

12 S176 DANIEL J. CONLEY ET AL. Ecological Applications Special Issue TABLE 1. Student s t test results for differences in changes in community structure, , between areas affected or not affected by hypoxia for 1 week in autumn 2002 in Denmark. Parameter No hypoxia,.2 mg DO/L Hypoxia,2 mg DO/L P Bray-Curtis dissimilarity index (%) No. species/area (%) Alpha diversity (%) Notes: Mean values are given for three parameters, where the Bray-Curtis dissimilarity index is calculated on square-root transformed data and alpha diversity is the number of species per core. Hypoxia is defined by the level of dissolved oxygen, DO. The Bray-Curtis dissimilarity is a measure of differences in the species composition between communities and ranges between 0% for two identical communities and 100% when two communities have no species in common. The probability of a true H o, that there are no differences in the magnitude of change between unaffected and affected areas, is given by P. For all parameters n 1 ¼ 13 areas not affected and n 2 ¼ 11 areas affected by hypoxia. marine areas have shown increases in TN loading from ca to the 1990s of ; times (Conley 1999), consistent with the fourfold increases during the peak of TN loading in the 1980s found here. Prior to nutrient reductions, area-specific TN loads from Denmark in the 1980s and 1990s were some of the highest rates in Europe, primarily due to the intensity of animal husbandry and fertilizer use (Conley et al. 2000). Trends in bottom-water dissolved oxygen concentrations around Denmark The results of this study show that mean bottomwater dissolved oxygen concentrations have significantly declined in coastal marine ecosystems around Denmark. For estuaries and coastal areas, this long-term trend was modulated by TN loading and wind speed in July September, explaining 52% of interannual variations in bottom-water dissolved oxygen concentrations. For open sea areas this long-term trend was also modulated by TN loading; however, additional important variables included advective transport of water and Skagerrak surface-water temperature, which combined to explain 49% of interannual variations in bottom-water dissolved oxygen concentrations. The mechanisms governing the oxygen consumption rate in bottom waters should be the same for coastal and open-water stations, i.e., the organic export from the surface to the bottom layer is assumed to be related to external TN input from land and a temperature effect. However, the mechanisms for oxygen transport to the bottom layer are different between estuaries and open-water stations. The Danish estuaries and coastal areas are mostly shallow, with a stratified water column during calm periods in summer and mixed during more windy conditions (Møhlenberg 1999). The open waters of the Danish straits, on the contrary, are permanently stratified during summer (Andersson and Rydberg 1988), and bottom water originates as surface-water inflow in winter and early spring from the Skagerrak and penetrates into the Danish straits as a bottom current. Combining the long-term trend analyses of bottomwater dissolved oxygen concentrations with results from the multiple regression indicate the importance of different explanatory factors to describe observed increases in hypoxia in Danish waters. The TN loading decrease of kg/yr ( ) should have improved the oxygen concentration by mgl 1 yr 1, according to the multiple regression. The decreasing trend in wind speed during the summer period of ms 1 yr 1 ( ) counteracted this by lowering the oxygen concentration by mgl 1 yr 1. Thus, the measures taken to reduce TN loading (Kronvang et al. 1993, Conley et al. 2002b) may be counteracted by unfavorable meteorological conditions. Increasing temperatures also may have contributed to increasing hypoxia, although it appears to be less important than wind conditions. In Chesapeake Bay, long-term increases in average river flow in the 1990s contributed to upward trends in hypoxic volume, perhaps offsetting the effects of nutrient reductions by delivering more diffuse TN to the Bay (Hagy et al. 2004). Mean wind speed during July September is only a proxy indicator for the development and duration of stratification, and thus the quantitative effect of decreasing average wind speed is uncertain. Hypoxia can develop in Danish estuaries provided that stratified conditions prevail for longer periods (Møhlenberg 1999). This was, for instance, the case in August 1997 when a long-lasting high pressure over Denmark resulted in very calm winds for four weeks and widespread hypoxia in Danish estuaries (Fallesen et al. 2000, Conley and Josefson 2001). The shallow estuaries and coastal areas in Denmark (Conley et al. 2000) are susceptible to hypoxia developing under such meteorological conditions, since the pycnocline often is located close to the bottom and consequently the oxygen in the thin bottom layer rapidly becomes depleted (Møhlenberg 1999). Increasing susceptibility to hypoxia The repeated occurrence of hypoxia has been hypothesized to lead to a persistent increase in the susceptibility of coastal marine ecosystems to eutrophication and hypoxia (Hagy et al. 2004). Results from the multiple linear regression suggest that decreases in TN input ( ) should have potentially improved mean bottom-water dissolved oxygen concentrations by 0.030