Estuaries and Coasts

, Volume 32, Issue 4, pp 621–641

Narragansett Bay Hypoxic Event Characteristics Based on Fixed-Site Monitoring Network Time Series: Intermittency, Geographic Distribution, Spatial Synchronicity, and Interannual Variability

Authors

    • Graduate School of OceanographyUniversity of Rhode Island
  • Heather E. Stoffel
    • Graduate School of OceanographyUniversity of Rhode Island
  • Christopher F. Deacutis
    • Narragansett Bay National Estuary ProgramUniversity of Rhode Island
  • Susan Kiernan
    • Department of Environmental ManagementOffice of Water Resources
  • Candace A. Oviatt
    • Graduate School of OceanographyUniversity of Rhode Island
Article

DOI: 10.1007/s12237-009-9165-9

Cite this article as:
Codiga, D.L., Stoffel, H.E., Deacutis, C.F. et al. Estuaries and Coasts (2009) 32: 621. doi:10.1007/s12237-009-9165-9

Abstract

Low dissolved oxygen events were characterized in Narragansett Bay (NB), a moderate-size (370 km2) temperate estuary with a complex passage/embayment geometry, using time series from 2001 to 2006 at nine fixed-site monitoring stations. Metrics for event intensity and severity were the event-mean deficit relative to a threshold (mg O2 l−1) and the deficit-duration (mg O2 l−1 day; product of deficit and duration [day]). Hypoxia (threshold 2.9 mg O2 l−1) typically occurred intermittently from late June through August at most stations, as multiple (two to five per season) events each 2 to 7 days long with deficit-duration 2 to 5 mg O2 l−1 day. Conditions were more severe to the north and west, a pattern attributed to a north–south nutrient/productivity gradient and east–west structure of residual circulation. Spatial patterns for suboxic and severely hypoxic events (thresholds 4.8 and 1.4 mg O2 l−1) were similar. The view that different processes govern event variability in different regions, each influenced by local hydrodynamics, is supported by both weak spatial synchronicity (quantified using overlap of event times at different sites) and multiple linear regressions of biological and physical parameters against event severity. Interannual changes were prominent and season-cumulative hypoxia severity correlated with June-mean river runoff and June-mean stratification. Benthic ecological implications for areas experiencing events include: NB hypoxia classifies as periodic/episodic on a near-annual basis; highest direct mortality risk is to sensitive and moderately sensitive sessile species in the northern West Passage and western Greenwich Bay, with some risk to Upper Bay; direct risk to mobile species may be ameliorated by weak spatial synchronicity; and indirect impacts, including reduced growth rates and shifts in predator–prey balances, are very likely throughout the sampled area due to observed suboxic and hypoxic conditions.

Keywords

HypoxiaSuboxicOxygen deficiencyTime seriesNarragansett BayWater quality

Introduction

One of the most widespread and deleterious anthropogenic impacts to estuarine and coastal waters is eutrophication-driven oxygen depletion or hypoxia (Diaz 2001; Diaz and Rosenberg 2008), which has been identified as a significant stressor on benthic communities including fish and invertebrates (Pihl et al. 1992; Gray et al. 2002; Diaz et al. 2004). Diaz and Rosenberg (2008) classified estuarine systems depending on whether hypoxia persists on timescales of weeks to months (“seasonal,” e.g., main stem of Chesapeake Bay; Hagy et al. 2004) or more episodically on timescales of days to weeks (“periodic,” e.g., Narragansett Bay (NB); Bergondo et al. 2005). In response to seasonal hypoxia, benthic communities appear to follow the Pearson–Rosenberg organic loading stress model (Pearson and Rosenberg 1978; Dauer et al. 1992; Dauer and Alden 1995; Diaz et al. 2004). Community responses to episodic/periodic hypoxia can be more complex and are often molded by factors such as the spatial extent, duration, frequency, and intensity of hypoxic events (Pihl et al. 1992; Diaz et al. 2004; Jewett et al. 2005). For example, less hypoxia-sensitive species may thrive in episodic moderately hypoxic situations, through decreased predation rates due to area avoidance by more sensitive predator species (Pihl et al. 1991; Sagasti et al. 2001; Jewett et al. 2005; Altieri and Witman 2006; Montagna and Ritter 2006; Altieri 2008). In many estuaries, hypoxia appears to have followed a phased impact progression, with conditions worsening over a number of years from episodic/periodic to seasonally persistent (Diaz and Rosenberg 2008).

Hypoxic events can be detrimental to benthic communities through both direct and indirect effects; the relative prevalence of direct and/or indirect effects and the nature of recovery are associated with the spatial extent, duration, frequency, and severity of hypoxia (Diaz and Rosenberg 1995, 2008; Wu 2002; Shimps et al. 2005; Montagna and Ritter 2006). Direct effects include mortality from lethal dissolved oxygen (DO) concentrations and changes in abundance and biomass due to combinations of mortality of sessile organisms and area avoidance by mobile species (Altieri and Witman 2006; Montagna and Ritter 2006). Taxonomic rankings indicate that fishes and crustaceans show greatest sensitivity, followed by echinoderms, while annelids and especially cnidarians and molluscs tend to show greatest tolerance (Diaz and Rosenberg 1995; Gray et al. 2002; Vaquer-Sunyer and Duarte 2008). The highest risk of direct mortality occurs at concentrations of 0.5 to 1.0 mg O2 l−1 for many moderately sensitive species and at 1.0 to 2.0 mg O2 l−1 for many sensitive species, with mortality often taking place within the first 4 to 7 h; for highly tolerant benthic infaunal species, risk requires more severe conditions (≤0.5 mg O2 l−1) of greater duration (days to weeks; Rosenberg et al. 1991; Sagasti et al. 2001; Wu 2002; Person-LeRuyet et al. 2003; Vaquer-Sunyer and Duarte 2008). Larvae are generally more acutely sensitive than juveniles and adults (USEPA 2000); Rhode Island (RI) water quality regulations, designed for protection of larvae, treat a 1-day exposure to 2.9 mg O2 l−1 as a violation (RIDEM 2006).

Indirect effects include significant decreases in growth rates and habitat compression that force species into areas of adequate DO but subject them to other stressors: lower prey density (Eby and Crowder 2002; Powers et al. 2005; Eby et al. 2005), nonpreferred thermal regimes (Craig and Crowder 2005; Niklitschek and Secor 2005), higher predator concentrations (Eggleston et al. 2005), and/or potential increased predation risk due to emergence of deep burrowers in sediments (Pihl et al. 1992; Taylor and Eggleston 2000; Wu 2002; Montagna and Ritter 2006). For example, significant decreases of >50% in growth rates were noted by (Eby et al. 2005) for juvenile demersal fish species due to decreased prey density in areas experiencing hypoxia. The spatial synchronicity of hypoxic events, or extent to which events in different subregions of an estuary are simultaneous, is a less-studied factor that may shape the nature of direct and indirect impacts on mobile species.

NB is a medium-sized (370 km2) northeastern US estuary (Fig. 1) that has been experiencing hypoxic events for at least the last several decades (Olsen and Lee 1979; Oviatt et al. 1984; Bergondo et al. 2005; Deacutis et al. 2006; Melrose et al. 2007; Deacutis 2008; Saarman et al. 2008). We address areas south of the shallow northernmost portions of the Providence River estuary. Hypoxia severity generally follows the north–south gradient of nutrients, phytoplankton, and fresh water influence, decreasing in intensity with distance from the estuary head in the north; an exception is that hypoxia is severe in western Greenwich Bay, a shallow embayment located south of the region of peak nutrient enrichment (Oviatt et al. 2002; Prell et al. 2004; Melrose et al. 2007; Deacutis 2008; Oviatt 2008; Saarman et al. 2008).

Monitoring stations to collect continuous measurements of DO and associated water quality parameters in NB were first established by researchers in the mid-1990s. Over the next decade, stations were added (Fig. 1) by researchers and government entities, and the Narragansett Bay Fixed-Site Monitoring Network (NBFSMN) developed through interagency collaboration. Its aim is improved monitoring of water quality, in particular hypoxia, through sustained time series observations that include DO, chlorophyll, temperature, and salinity (NBFSMN 2007). Our analysis of temporal and spatial DO variability using multiple years of network observations complements previous descriptions based on synoptic spatial conductivity–temperature–depth–oxygen (CTDO) surveys (Deacutis 1999; Prell et al. 2004; Deacutis et al. 2006), an early subset of NBFSMN time series (Bergondo et al. 2005), towed-body surveys (Melrose et al. 2007), and combinations of the three datasets (Saarman et al. 2008).
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig1_HTML.gif
Fig. 1

Narragansett Bay region and bathymetry, with pertinent geographic features and primary river inflows labeled. The nine stations are: Bullock Reach (BR), Conimicut Point (CP), North Prudence Island (NP), Mount View (MV), Quonset Point (QP), Poppasquash Point (PP), T-Wharf (TW), Greenwich Bay Marina (GB), and Mount Hope Bay (MH). Providence and Newport sea level stations indicated by stars

The relative importance of numerous biological and physical processes that shape NB hypoxia is poorly understood. We hypothesize a strong influence of hydrodynamics local to each bay subregion. The monitoring network time series facilitate the exploratory analysis presented here. Seven parameters are most relevant, based on previous studies: river flow, chlorophyll, temperature, density stratification, tidal range, large-scale north–south nontidal sea level difference, and northeastward wind. River flow is both the dominant pathway for delivery of nutrients (e.g., Nixon et al. 2008) and a major influence on flushing rate (Pilson 1985) and stratification. Chlorophyll indicates abundance of phytoplankton, the main source of decaying material that fuels hypoxia, and temperature regulates the metabolic rate of DO consumption. Stratification reduces vertical exchange and aeration of deep water, and tidal range captures changes in spring- and neap-tide conditions, which regulate tidally driven flushing and vertical mixing and appear to influence hypoxic event timing (Bergondo et al. 2005). Large-scale north–south nontidal sea level difference is a proxy for circulation fluctuations driven by north–south winds, which regulate flushing of bottom waters (Bergondo 2004). Finally, detailed hydrodynamic modeling indicates that northeastward wind stalls the exchange of water among bay subregions (Rogers 2008).

The purpose of this study is to expand our understanding of a system subject to episodic/periodic hypoxia, through use of multiyear time series observations from a spatially distributed array of stations. Temporal intermittency, geographic patterns, spatial synchronicity, and interannual variability of low-DO events are characterized using three threshold DO concentrations to describe suboxic, hypoxic, and severely hypoxic conditions. New metrics designed specifically for time series (deficit-duration and fractional overlap, described below) are used, and comparison is made to a metric used by a state regulatory agency. Potential influences of biological and physical driving factors (specifically, the seven listed above) in shaping hypoxic event severity and interannual variability are explored. Finally, ecological implications of both direct and indirect effects of hypoxia on benthic communities are described; the state of NB with respect to phased hypoxia progression (Diaz and Rosenberg 2008) is assessed, and community shifts affecting commercially important shellfish are interpreted.

Methods

Monitoring Network Observations

The 33 time series analyzed are from nine sites (names and abbreviations given in Table 1; Fig. 1) sampled over the years analyzed, 2001 to 2006 (NBFSMN 2007). Most calculations are carried out on an individual-site basis. However, for some calculations and for convenience when presenting and summarizing results, it is helpful to use the following descriptive names for four groups: Upper Bay (UB) for BR, CP, and NP; West Passage (WP) for MV and QP; East Passage (EP) for PP and TW; and Embayments (EB) for GB and MH. Each site had (Table 1) a near-bottom (0.5 or 1.0 m above the seafloor) and a near-surface (0.5 m deep) Yellow Springs International sonde, sampling at 15-minute intervals. We used DO from the deep sonde, chlorophyll from the shallow sonde, and salinity and temperature from both. When referencing the data collection network, or more generally its products, we cite the network website (NBFSMN 2007) where data are distributed; individual years' products are described for 2001 and 2002 by Bergondo (2004), for 2003 by Stoffel (2003), and for remaining years by NBFSMN (2004, 2005, 2006).
Table 1

Station and sampling characteristics of fixed-site network

Group

Station

Latitude

Longitude

Water deptha [m]

Deep sensor depth [m]

Yearsb

Cumulative samplingc [day]

Upper Bay (UB)

Bullock Reach (BR)

41 44.434′

71 22.480′

6

5.5

2001

121

2002

100

2003

104

2004

102

2005

121

2006

110

Conimicut Point (CP)

41 42.828′

71 20.628′

7

6.5

2003

100

2005

101

2006

107

North Prudence (NP)

41 40.224′

71 21.283′

11

10.5

2001

81

2002

108

2003

110

2004

113

2005

119

2006

86

West Passage (WP)

Mt. View (MV)

41 38.304′

71 23.021′

7

6.5

2004

86

2005

86

2006

113

Quonset Point (QP)

41 35.288′

71 22.839′

7

6.5

2005

40

2006

82

East Passage (EP)

Poppasquash Point (PP)

41 39.807′

71 19.066′

8

7.5

2004

84

2005

56

2006

121

T-Wharf (TW)d

41 34.731′

71 19.287′

6

5

2003

91

2004

120

2005

105

2006

121

Embayments (EB)

Greenwich Bay Marina (GB)d

41 41.090′

71 26.762′

3

2.5

2003

96

2004

120

2005

113

2006

118

Mt. Hope Bay (MH)

41 40.808′

71 12.913′

5

4.5

2005

87

2006

121

All data available at http://www.dem.ri.gov/bart/stations.htm

aDepths relative to mean lower low water

bSampling coverage and gaps are shown in Fig. 4 for each year, from May to October

cJune 1 through September 30

dT-Wharf and Greenwich Bay Marina are dock-based stations; all others are buoy stations

Quality assurance measures include verification of calibrations and consistency among multiple instruments, corrections for sensor drift and biases due to biofouling, removal of outliers, and interpolation across selected intervals of missing data (RIDEM 2007). Protocols for calibration, field maintenance, and quality assurance and quality control (QA/QC) procedures are consistent with National Estuarine Research Reserve System-Wide Monitoring Program standard operating procedures (Small 2008). Stations are serviced by swapping the deployed instruments with newly calibrated instruments on a 2-week interval. Calibrations and sensor drift corrections are verified through a three-point comparison: data from the retrieved sonde are compared to the newly calibrated sonde, as well as an independent profiling sonde, all at the deployment depth. Outliers are removed based on exceeding two standard deviations or the 95th percentile, using monthly data for each station, in conjunction with inconsistencies in other parameters (RIDEM 2007). Gaps in coverage, affecting up to 6% of the record at an individual station in a given year (station year), are filled by linear interpolation following protocols detailed in the Quality Assurance Project Plan (RIDEM 2007). All data removed for QA/QC reasons is documented in the metadata documentation accompanying the data products (NBFSMN 2007).

Thresholds: Suboxic, Hypoxic, and Severely Hypoxic

The threshold values T4.8 = 4.8 mg O2 l−1, T2.9 = 2.9 mg O2 l−1, and T1.4 = 1.4 mg O2 l−1 bounding ranges for suboxic, hypoxic, and severely hypoxic conditions (Tables 2 and 3) are used because they are explicitly incorporated in water quality regulations adopted by the state of Rhode Island (RIDEM 2006, p. 19) as survival-protective under chronic, 24-h, and 1-hr exposures, respectively. These regulations were developed based on Environmental Protection Agency (USEPA 2000) criteria, with incorporation of larval recruitment effects. Though the value 2.9 mg O2 l−1 does not appear explicitly in the text of USEPA (2000), it results from the equation in Table 6 on p. 37 of that document for an exposure interval duration of 1 day (as shown at left end point of curve in Fig. 7 of USEPA 2000) to determine the allowable minimum concentration protective of larvae, juveniles, and adults; the value 2.3 mg O2 l−1 (USEPA 2000) used in other studies is protective of juveniles and adults only. Our thresholds are in the ranges commonly implemented in the literature as reviewed by Vaquer-Sunyer and Duarte (2008).
Table 2

Synopsis of input parameters for the moving window trigger (MWT) algorithm; input parameters are provided to the algorithm together with a DO time series of arbitrary fixed time step and marked missing values

Parameter

Meaning

Valuea in this paper

Threshold

Events are identified as groups of values that are all or mostly (at least 50%) below the threshold

2.9, 4.8, 1.4 mg O2 l−1

Minimum event duration

Shorter events are ignored

1 day

Trigger duration

Duration of values below/above the threshold that causes (“triggers”) event start/end to be identified

9 h

For complete details, see Codiga (2008)

aWe use the terms “suboxic,” “hypoxic,” and “severely hypoxic” when referring to [O2] in the ranges 4.8 ≥ [O2] > 2.9, 2.9 ≥ [O2] > 1.4, and [O2] ≤ 1.4 mg O2 l−1, respectively

Event Characterization Using MWT Algorithm

An algorithm referred to as a “moving window trigger” (MWT; Codiga (2008)—“Electronic supplementary material”) was developed and applied to identify and characterize low-DO events from each time series in a systematic way. The MWT is designed to treat time series of arbitrary sampling resolution and duration and to handle gaps in sampling coverage that typically characterize such records (Table 2). MWT parameters used here (Tables 2 and 3) were based on attempting to match, as closely as possible, the state water quality criteria (RIDEM 2006). The 9-h trigger duration was motivated by our focus on events longer than tidal timescales and chosen to be (a) longer than half the period of the dominant M2 (12.42 h) tidal component of variability, in order that when DO variability was primarily tidal, as occurred during short portions of some records, alternating halves of a series of tidal cycles would not be identified as individual events, and (b) as different from a multiple of half the tidal period as possible, to improve the accuracy of the event start and end times assigned by the MWT algorithm (Codiga 2008). Results using trigger durations of 7 or 11 h differ in minor ways that do not affect our conclusions.
Table 3

Synopsis selected output parameters (metrics that characterize each individual event identified) for the moving window trigger (MWT) algorithm

Parameter

Meaning

Units

Event duration

Difference between event end and start times

days

Event-mean deficit

Average deficit (threshold less DO concentration) over event; higher value more intense hypoxia

mg O2 l−1

Deficit-duration

Integrated deficit during event; equivalently, product of event-mean deficit and duration

mg O2 l−1 day

For complete details, see Codiga (2008)

The MWT algorithm fills missing-data gaps shorter than the trigger duration by linear interpolation. The collective total duration of gaps interpolated was less than 50 h (0.04%) of the 33 time series, and the maximum collective interpolated duration for an individual station year was less than 20 h. Gaps longer than the trigger duration are not interpolated across.

Two key features of the MWT algorithm are demonstrated using T2.9 and the 2006 NP time series (Fig. 2), which exhibits characteristics that typify the data from other years and other sites. First, events are not terminated by above-threshold values that persist for less than the trigger duration. This permits events to be sensibly identified without each and every below-/above-threshold value causing an event to start/end, which would result in an unreasonably high number of events with little ecological relevance. Second, the algorithm identifies when an event starts/ends adjacent to a missing-data portion of the time series. Codiga (2008) includes a detailed catalog of plots and tables for all events, relative to T2.9, from all years and sites.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig2_HTML.gif
Fig. 2

Representative DO time series record (NP 2006) to demonstrate behavior and results of MWT algorithm applied to a typical record having missing-data gaps and scatter on timescales shorter than the trigger duration. Threshold value 2.9 mg O2 l−1, minimum event duration 1 day, and trigger duration 9 h. Horizontal dashed line is the threshold value. Asterisk indicates that event 1 begins following a missing-data gap

Metrics used to quantify event characteristics (defined in Table 3) include duration, event-mean deficit, and deficit-duration. Deficit-duration is a mixed measure of severity in the sense that it increases with both the duration of an event and its intensity, as measured by its event-mean deficit (the threshold less the event-mean DO concentration). Two events may have quite similar durations yet have very different deficit-durations; for example, in the NP 2006 record, events 1 and 2 have durations that differ by less than 20% but because event 2 DO values fall only slightly below the threshold, its deficit-duration is less than 25% that of event 1 (Fig. 2).

State Regulatory Metric

State of Rhode Island saltwater DO regulations (RIDEM 2006) are cast in terms of days exceedance over a chosen seasonal period. Here, time series data were evaluated using the RIDEM-adopted software application called Dissolved Oxygen Criteria Software for Rhode Island (DOCS-RI; SAIC 2006) to calculate season-cumulative (Jun. 1 to Sep. 30) days exceedance. The DOCS-RI algorithm is specific to NB because it is based on a constrained species suite or subset of those in the EPA criteria (USEPA 2000), appropriate for assessing impairments. DOCS-RI incorporates both a larval recruitment function and a time to death model that estimates larval mortality that occurs under fluctuating conditions within a day (USEPA 2000; SAIC 2006). DOCS-RI days exceedance is thus a similar metric to the MWT deficit-duration with respect to the fact that it reflects both duration and intensity of hypoxia. However, DOCS-RI differs from MWT in that it incorporates dependence on the nonlinear biological response of specific species to low-DO conditions.

Spatial Synchronicity

Spatial synchronicity is the extent to which hypoxic events at one location occur simultaneously to hypoxic events at a separate location. The index used to quantify spatial synchronicity was the fractional overlap (FO) of time intervals during which events occur at two stations. FO is the unitless ratio between (a) the cumulative duration when events (MWT relative to T2.9) occupied both stations A and B simultaneously and (b) the cumulative duration of events at station A or the cumulative duration of events at station B, whichever is smaller. The upper limit for FO is 1, representing complete or full spatial synchronicity, when events at one station only occur during events at the other station; the lower limit for FO is 0, a completely asynchronous condition for which events at one station never occur during events at the other station (Fig. 3). The extent to which FO is less than 1 is primarily a measure of reduced overlap in event timing, rather than indicating the two stations have events of different durations that nonetheless overlap. This is demonstrated by the fact that, even for two stations having markedly different event durations, for example because at one station events begin later and end earlier than at the other (B1 in Fig. 3), there is full spatial synchronicity (FO = 1). FO was calculated using multiple years’ observations, as necessitated to ensure an adequately high number of events at both stations. All station pairs were treated that met the data availability criterion: at least three summers of coincident sampling, during which a total of at least five events occurred at each station.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig3_HTML.gif
Fig. 3

Schematic to illustrate definition of fractional overlap index (FO) used to quantify spatial synchronicity. Each rectangle indicates a hypoxic event. Events at stations B1, B2, and B3 are shaded where they overlap with events at station A. Between station A and station B1, the fractional overlap takes its maximum value FO = 1.0, indicating complete/full spatial synchronicity for this station pair. For station A and station B2, FO = 0.5. There is a complete lack of spatial synchronicity (FO = 0) between station A and station B3

Exploratory Analysis of Factors Driving Event Variability

Multiple linear regression (MLR) was used to assess the relative importance of a range of biological and physical parameters in accounting for variability in hypoxic event severity. The dependent (output/response) variable was the MWT deficit-duration of fully sampled (not begun nor ended by a missing-data gap) hypoxic events relative to T2.9, natural-log transformed to normalize its skewed distribution. The MLR was carried out once using all events from all years at the three sites (BR, CP, and NP) in the UB group of stations collectively, in order to have an adequately high number of events (n = 39). It was carried out a second time using all events from all years at the GB site (n = 27) because characteristics of events there differ appreciably from those at other sites, as described below. Other stations had too few events to be treated.

Independent (input/predictor) variables were calculated from seven parameters (see “Introduction”): river flow, chlorophyll concentration, temperature, density stratification, tidal range, low-passed Providence–Newport sea level difference, and northeastward wind. River flow was the sum of daily values (e.g., USGS 2004) from the five main rivers (Blackstone, Pawtuxet, Ten Mile, Moshassuck, and Woonasquatucket; Fig. 1). Chlorophyll, temperature, and stratification (defined as deep density less shallow density) were from the fixed-site 15-min resolution records (NBFSMN 2007; near surface for chlorophyll, near bottom for temperature, and both for stratification) at the station where each hypoxic event occurred. Sea level data were hourly observations from the National Atmospheric and Oceanic Administration (NOAA) stations (tidesandcurrents.noaa.gov) at Providence and Newport (stars, Fig. 1). Tidal range was calculated as the daily average of the differences between higher high tide and the succeeding lower low tide using the Newport station, which for this purpose is representative of bay-wide conditions. The Providence–Newport sea level difference was calculated after application of the inverse-barometer correction (using hourly surface air pressure measured at each station) and removal of the tidal component with a low-pass (25-h half-width triangle-weight running mean) to each individual record. Winds are 3-h resolution North American Regional Reanalysis data-assimilative operational meteorological model hindcasts (Mesinger et al. 2006), which compare well with local wind records (tidesandcurrents.noaa.gov) that were not used due to temporal coverage gaps.

Event characteristics can be influenced by the driving parameters either during the event or prior to it by a lead time of hours to several days. To incorporate the potential influence of lead times that span this range, three independent variables for the MLR were calculated from each of the above seven parameter time series by averaging available values over different time intervals: during the hypoxic event (“zero lead”) and during the 2-day (“2-day lead”) and 5-day (“5-day lead”) intervals immediately preceding the hypoxic event start time.

The forward stepwise approach was used (Matlab R2008b) to maximize adjusted R2 and exclude independent variables with p > 0.05. Relative importance of independent variables included in the model was based on rank of standard partial regression coefficient magnitudes.

Interannual Variability

The relationship between interannual variability in hypoxia and in the seven parameters treated by the MLR was investigated. The calculation was limited to stations for which all years’ (n = 6) sampling were available, BR and NP, and hence applies only to hypoxia in the Upper Bay region. To gauge the annual severity of hypoxia, we used an index for seasonal hypoxia ISH (mg O2 l−1) defined as the season-cumulative deficit-duration relative to T2.9, normalized by the days sampled June to September, averaged across the two stations. Higher values of the index correspond to more severe hypoxia. For each of the seven parameters, interannual variations were first quantified using the mean of all available values during the entire hypoxia season (June to September). Next, the means were calculated using individual months (May to September), in recognition that season-cumulative hypoxia severity may depend primarily on late spring or early summer conditions; this effectively includes leads of one or more months, as appropriate in contrast to the 2- and 5-day leads in the MLR analysis (described above) for individual-event variability. Kendall’s tau correlations with p < 0.05 were used to identify significant association of the seasonal hypoxia index with the entire-season mean of each of the seven parameters and with the individual-month means.

Results

Hypoxic Event Characteristics

We first describe events determined using MWT relative to T2.9, which include hypoxic conditions ([O2] < T2.9) or both severely hypoxic ([O2] < T1.4) and hypoxic conditions. Each individual event from all stations and all years is depicted on a timeline as a rectangle (Fig. 4): the deficit-duration corresponds to the rectangle area, with the duration and event-mean deficit represented by the rectangle width and height respectively. Table 4 lists event statistics for each station year: the minimum, mean, and maximum values of duration, event-mean deficit, and deficit-duration. A group of bar charts arranged geographically by station (Fig. 5) illuminates spatial patterns by presenting deficit-duration on a season-cumulative basis (total bar lengths), as well as contributions of individual events (divisions within bars), for all years. Events not fully sampled because they start/end adjacent to missing data are included in Fig. 4 (marked by asterisks) and Fig. 5; as appropriate for statistical compilation, they are omitted from Table 4.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig4_HTML.gif
Fig. 4

Timeline showing MWT events (including their durations, event mean deficits, and deficit durations) relative to 2.9 mg O2 l−1, and sampling coverage, for all stations and all years. Each hypoxic event is represented as a rectangle: deficit duration is represented as area; event duration is represented as width (horizontal scale at bottom); event mean deficit is represented as height (vertical scale at top). Heavy black lines indicate sampling coverage and show missing-data gaps; asterisks indicate where an event started or ended during a period of missing data (events in Fig. 2 correspond to NP 2006 row here)

Table 4

Statistics of MWT hypoxic events (threshold T2.9 = 2.9 mg O2 l−1, minimum event duration 1 day, trigger duration 9 h)

Year

Station

Number of events

Duration [day]

Event-mean deficit [mg O2 l−1]

Deficit-duration [mg O2 l−1 day]

Min.

Mean

Max.

Min.

Mean

Max

Min.

Mean

Max.

2001

BR

7

1.6

4.8

17.1

0.2

0.5

1.0

0.3

3.5

16.4

NP

3

1.3

3.4

5.8

0.3

0.6

0.9

0.4

2.7

5.4

2002

BR

3

2.5

3.9

6.0

0.3

0.7

1.5

0.8

3.5

8.9

NP

2

5.6

6.2

6.9

0.7

0.9

1.0

4.0

5.4

6.8

2003

BR

3

1.9

2.9

4.2

0.3

0.3

0.5

0.7

0.9

1.1

CP

1

3.4

3.4

3.4

0.5

0.5

0.5

1.7

1.7

1.7

NP

6

1.3

4.6

9.1

0.4

0.8

1.4

0.6

4.5

12.4

GB

2

1.1

6.9

12.7

2.1

2.2

2.3

2.5

14.8

27.1

TW: no events

2004

GB

5

1.2

2.3

4.2

0.7

1.0

1.4

1.2

2.3

4.8

BR, NP, MV, PP, TW: no events

2005

BR

1

3.3

3.3

3.3

0.3

0.3

0.3

0.9

0.9

0.9

NP

3

1.3

1.6

1.8

0.4

0.5

0.6

0.6

0.8

1.0

MV

2

3.9

4.5

5.1

0.3

0.5

0.7

1.5

2.0

2.6

GB

9

1.6

2.8

5.1

0.5

1.1

2.2

0.8

3.1

4.8

CP, QP, PP, TW, MH: no events

2006

BR

4

1.5

7.2

11.3

0.5

0.7

0.9

0.8

5.3

9.6

CP

4

1.1

3.3

5.4

0.5

0.6

0.9

0.7

2.3

4.7

NP

2

1.4

3.2

5.0

0.3

0.4

0.4

0.5

1.3

2.2

MV

2

1.4

10.7

20.0

0.5

0.9

1.3

0.7

13.6

26.5

QP

3

1.6

2.5

4.1

0.5

0.7

0.8

0.9

1.8

3.4

PP

4

1.1

2.0

2.8

0.7

1.0

1.3

0.8

2.1

3.4

GB

11

1.2

2.2

4.6

0.6

1.4

2.4

1.2

3.4

7.7

MH

2

2.3

2.5

2.7

0.4

0.5

0.6

1.0

1.3

1.7

TW: no events

Events that start/end adjacent to missing data (asterisks, Fig. 4) have been excluded

https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig5_HTML.gif
Fig. 5

Deficit durations relative to 2.9 mg O2 l−1 from all stations all years, displayed to demonstrate season-cumulative results and geographic distributions. Each vertical section of each bar corresponds to an individual event (see Fig. 4). Values shown at tops of bars indicate number of days sampled

A prominent feature of NB hypoxia is temporal intermittency at any given station, within any given year. Events last from a day to a few weeks, conditions denoted periodic by Diaz and Rosenberg (2008) as discussed above. The number of fully sampled events at an individual station ranged from 0 to 11 in a single season (Table 4). During the most recent 3 years, events at GB were the most numerous, the shortest (typical duration 2 to 3 days) and the most intense (typical event-mean deficit about 1 to 2 mg O2 l−1), with typical mean deficit-duration of about 2 to 4 mg O2 l−1 day.

Stations BR, CP, NP (the UB group), and MV (northern WP) experienced the highest numbers of events following GB. Based on 6 years of sampling, BR and NP typically had three to four events per summer, each with a duration of about 3 to 5 days, with event-mean deficit in the range of 0.3 to 0.9 mg O2 l−1, and deficit-durations of typically 2 to 4 mg O2 l−1 d. Events at CP (located between BR and NP) based on 3 years’ sampling were comparable; relative to BR and NP, the number of events was in the lower range; durations were similar, and event-mean deficits and deficit-durations were in the lower range. At MV, hypoxia was severe relative to BR and NP; events there were typically longer (mean duration of 10.7 days in 2006) and more intense (event-mean deficit of typically about 0.5 to 1.5 mg O2 l−1) with correspondingly higher deficit-durations (a 2006 event at MV had the highest deficit-duration, 26.5 mg O2 l−1 day, in any year at any station other than GB).

Stations farther south and east (QP in WP; PP and TW in EP; MH), sampled in the later 2 to 3 years only (Table 1), experienced fewer events (typically zero to two per summer). No events were observed at TW in any of the 4 years sampled there.

Season-cumulative deficit-durations (Fig. 5) were highest at GB (reaching 40 to 50 mg O2 l−1 day), followed by MV (up to 39 mg O2 l−1 day) and BR and NP (up to 20 to 30 mg O2 l−1 day). Interannual variability was prominent (as discussed in more detail below) and generally consistent across all sites, with the exception of GB where it was less pronounced.

Suboxic Event Characteristics

Events that include suboxic ([O2] < T4.8) conditions, in addition to hypoxic and severely hypoxic conditions, were determined by MWT relative to T4.8 (Fig. 6, Table 5). Multiple events occurred at every station in every sampled year except for TW in 2004 and 2005, and QP in 2004. Due to the higher threshold, events including suboxic conditions were more frequent and longer and had higher deficit-duration compared to events relative to T2.9 as described in the previous subsection. Higher durations were particularly pronounced at the UB group of stations, with peak values (mean 15.3 days; maximum 43.2 days) seen at BR in 2001. General patterns of intermittency and geographic variability paralleled those described in the previous subsection. For example, events at GB were most numerous (12 to 18 per year) and had typical durations (about 3 to 4 days) near the low end of the observed ranges. However, season-cumulative deficit durations relative to T4.8 at GB did not exceed those in the UB group to the same degree as was true for events relative to T2.9.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig6_HTML.gif
Fig. 6

Same as Fig. 5 except for deficit duration relative to 4.8 mg O2 l−1

Table 5

Same as Table 4 except for threshold T4.8 = 4.8 mg O2 l−1

Year

Station

Number of events

Duration [day]

Event-mean deficit [mg O2 l−1]

Deficit-duration [mg O2 l−1 day]

Min.

Mean

Max.

Min.

Mean

Max

Min.

Mean

Max.

2001

BR

6

1.8

15.3

43.2

0.7

1.3

2.0

1.7

25.8

87.1

NP

8

1.2

3.8

13.2

0.3

0.8

1.9

0.4

4.7

24.9

2002

BR

5

1.6

6.4

15.9

0.4

0.8

1.9

0.6

7.6

29.8

NP

5

2.9

8.2

14.6

0.3

1.0

1.8

1.0

10.5

25.9

2003

BR

4

4.5

10.4

21.5

0.2

0.8

1.6

1.1

11.1

34.4

CP

1

3.0

3.0

3.0

0.2

0.2

0.2

0.7

0.7

0.7

NP

2

1.6

3.0

4.4

0.3

0.4

0.4

0.6

1.0

1.4

TW

1

3.6

3.6

3.6

0.7

0.7

0.7

2.6

2.6

2.6

GB

12

1.1

3.2

13.3

0.8

1.9

4.0

1.4

7.8

51.5

2004

BR

8

1.2

3.8

8.6

0.2

0.6

1.2

0.6

2.6

10.3

NP

10

1.3

2.5

4.1

0.4

0.7

1.3

0.6

2.0

4.6

MV

4

1.5

1.9

2.6

0.4

0.7

0.9

0.9

1.4

2.2

PP

5

1.3

2.1

3.3

0.5

0.7

1.0

0.7

1.7

3.2

GB

17

1.3

3.8

9.7

0.6

1.5

2.8

1.5

6.1

19.4

TW: no events

2005

BR

12

1.3

5.6

20.4

0.2

0.7

1.3

0.4

4.9

26.6

CP

8

1.4

7.2

19.5

0.3

0.8

1.3

0.4

6.9

25.4

NP

7

1.7

7.5

20.5

0.5

1.0

1.5

0.8

9.5

29.9

MV

6

1.6

2.9

5.2

0.1

0.7

1.2

0.2

2.4

6.1

PP

8

1.3

3.0

5.5

0.4

0.7

1.2

0.8

2.3

4.5

GB

16

1.4

4.4

13.6

0.5

1.7

3.6

0.9

8.2

29.6

MH

3

2.1

4.2

7.3

0.4

0.6

1.0

1.0

2.5

3.4

QP, TW: no events

2006

BR

6

1.7

11.7

24.1

0.7

1.4

2.3

1.8

20.2

56.3

CP

5

1.8

12.3

32.5

0.2

0.9

1.5

0.5

17.1

50.3

NP

3

1.2

2.6

4.6

0.8

1.2

1.4

1.5

2.7

3.8

MV

5

2.0

8.4

24.0

0.6

1.4

2.9

1.5

17.7

70.1

QP

4

1.1

4.2

10.3

0.5

0.8

1.8

0.6

5.5

18.7

PP

11

1.2

4.2

12.3

0.6

1.2

2.0

0.8

6.0

19.7

TW

2

2.3

6.3

10.3

0.3

0.7

1.0

0.8

5.5

10.2

GB

18

1.4

4.0

11.6

0.5

1.7

3.4

1.0

7.9

30.0

MH

7

1.7

5.2

11.9

0.3

0.7

1.5

0.5

4.7

18.2

Severely Hypoxic Event Characteristics

Severely hypoxic event ([O2] < T1.4) characteristics were calculated by MWT as above but with T1.4 (Fig. 7, Table 6). Events were rare except at GB where there were four, one, three, and five events in years 2003 to 2006, respectively. There were a total of three events between BR and NP in all 6 years, and none in the 3 years’ sampling at CP. In 2006, the MV site had a relatively large number of events and a high season-cumulative deficit duration, each comparable to conditions at GB. At PP, there was one short event in 2006 with low event-mean deficit and low deficit-duration, and at southern and eastern stations (QP, TW, and MH) there were no events in any year sampled.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig7_HTML.gif
Fig. 7

Same as Fig. 5 except for deficit duration relative to 1.4 mg O2 l−1

Table 6

Same as Table 4 except for threshold T1.4 = 1.4 mg O2 l−1

Year

Station

Number of events

Duration [day]

Event-mean deficit [mg O2 l−1]

Deficit-duration [mg O2 l−1 day]

Min.

Mean

Max.

Min.

Mean

Max

Min.

Mean

Max.

2001

BR, NP: no events

2002

BR

1

4.1

4.1

4.1

0.4

0.4

0.4

1.6

1.6

1.6

NP: no events

2003

NP

1

4.0

4.0

4.0

0.7

0.7

0.7

2.8

2.8

2.8

GB

4

1.8

4.1

6.5

1.0

1.1

1.4

1.8

4.3

6.5

BR, CP, TW: no events

2004

GB

1

1.7

1.7

1.7

0.4

0.4

0.4

0.7

0.7

0.7

BR, NP, NV, PP, TW: no events

2005

GB

3

1.1

1.3

1.4

0.6

0.7

0.9

0.6

0.9

1.3

BR, CP, NP, MV, QP, PP, TW, MH: no events

2006

NP

1

4.1

4.1

4.1

0.7

0.7

0.7

2.8

2.8

2.8

MV

1

10.6

10.6

10.6

0.6

0.6

0.6

6.1

6.1

6.1

PP

1

1.1

1.1

1.1

0.5

0.5

0.5

0.6

0.6

0.6

GB

5

1.3

2.1

3.1

0.6

0.9

1.2

1.0

2.0

2.9

BR, CP, QP, TW, MH: no events

Days Exceedence of State Regulations

In DOCS-RI results (Fig. 8), GB had the highest season-cumulative exceedences (from 43 to 55 days), followed by MV, BR, and NP (up to 30 to 40 days). Sites in WP had more days exceedence than EP sites, with no days exceedence at TW in any year. Interannual variability was generally similar across sites except for GB where it was weaker. DOCS-RI results thus parallel most closely those described above for season-cumulative MWT deficit-duration relative to T2.9 (Fig. 5). These similarities suggest that the nonlinear biological response component of DOCS-RI, not included in the MWT algorithm, is not of primary importance for characterizing events in NB at these stations. DOCS-RI is, however, integral to the approach taken by state regulators to assess whether seasonal ambient water quality conditions in NB are in compliance with the saltwater DO criterion.
https://static-content.springer.com/image/art%3A10.1007%2Fs12237-009-9165-9/MediaObjects/12237_2009_9165_Fig8_HTML.gif
Fig. 8

Season-cumulative DOCS-RI days exceedance, all stations all years. Values shown at tops of bars indicate number of days sampled

Spatial Synchronicity

The timelines (Fig. 4) make clear that NB hypoxic events do not occur with geographically widespread synchronous onset and retreat. This deviation from full spatial synchronicity (FO = 1) is quantified by FO values (Table 7) that range from 0.39 to 0.93 and average 0.58. The meaning of FO = 0.58 between a station pair is that 42% of the time when conditions are hypoxic at the station with the shorter cumulative duration of hypoxia, the other station is not hypoxic. The highest fractional overlap values were between MV and the three stations in the Upper Bay group (NP, CP, and BR), with a comparably high value between BR and CP. The three lowest fractional overlap values are for station pairs that include GB, and a comparably low FO value applies between stations BR and NP.
Table 7

Fractional overlap (FO) results

BR

CP

NP

MV

GB

0.45

0.60

0.39

0.43

MV

0.78

a

0.93

 

NP

0.47

0.51

  

CP

0.69

   

aData availability criteria listed in “Methods” section not met

Event Timescale Variability: MLR with Driving Factors

The MLR model for the UB group of stations (Table 8) captured a small amount of variance (16.7%; p = 0.027) in hypoxic event deficit-duration. The 5-day lead river flow and zero-lead low-pass sea level difference were the only two independent variables meriting inclusion in the model (p < 0.05). River flow was the most important independent variable and covaried positively such that, for higher river flow, hypoxic events were more severe. The sea level difference covaried negatively such that, for increased Providence sea level relative to Newport sea level, hypoxic events were less severe. At the GB site, the MLR model (Table 9) captured more variance than for the UB group though still a relatively small amount (34.9%; p = 0.003); 2-day lead chlorophyll and 5-day lead northeastward wind were the only two variables meriting inclusion, with chlorophyll more important, and each covaried positively with deficit-duration.
Table 8

Same as Table 7 except for GB site

Independent variable

Standard partial regression coefficient

t statistic

p valuea

Chlorophyll, 2-day lead

0.347

2.72

0.012

Northeastward wind, 5-day lead

0.286

2.24

0.035

aOverall model: n = 27; adjusted R2 = 0.349; F = 7.70; p = 0.003

Table 9

Same as Table 8 except for GB site

Independent variable

Standard partial regression coefficient

t statistic

p valuea

Chlorophyll, 2-day lead

0.347

2.72

0.012

Northeastward wind, 5-day lead

0.286

2.24

0.035

aOverall model: n = 27; adjusted R2 = 0.349; F = 7.70; p = 0.003

Interannual Variability: Characteristics and Driving Factors

Interannual variability in hypoxia was prominent, with end points illustrated by comparison of 2006 and 2004 (MWT events relative to T2.9: Figs. 3 and 5, Table 4). In 2006, all sites experience multiple events, except TW where no events were observed in any year; in 2004, no events occurred except at GB. Interannual variability at GB was the least pronounced. In 2003 and later years, when multiple stations were sampled across several different regions of the bay, a general pattern of interannual variations was shared bay-wide: season-cumulative deficit-durations were higher in 2001, 2003, and 2006 than in 2004 and 2005. The index for season-cumulative hypoxia, based on the BR and NP stations, was ISH = 0.15, 0.11, 0.15, 0.00, 0.01, and 0.17 mg O2 l−1 in years 2001 to 2006, respectively; the corresponding ranking of years from most to least hypoxic was 2006, then 2001 and 2003 tied, followed by 2002, 2005, and 2004. Two variables correlated significantly (p < 0.05) with ISH: June-mean river flow (p = 0.006, τ = 0.966) and June-mean stratification (p = 0.006, τ = 0.966).

Discussion

Geographic Patterns, Intermittency, Spatial Synchronicity

Observed geographic variations in suboxic, hypoxic, and severely hypoxic events are similar and can be understood in terms of the spatial distribution of nutrient loads and productivity together with the long-term average bay circulation pattern. Nutrient loading and productivity both peak to the north in the urbanized Providence River estuary where delivery of nutrients and freshwater rivers (Fig. 1) is concentrated (e.g., Oviatt 2008). The typical pattern of residual nontidal circulation through the bay (Rogers 2008) consists predominantly of oceanic water moving northward at depth in the East Passage, entering the Providence River estuary, then mixing to shallower depths and reversing its course southward to exit the bay largely through the West Passage; as seen in other estuaries (e.g., Codiga and Aurin 2007), the lateral (east–west) asymmetry is due to the Coriolis effect. This combination of patterns in productivity and circulation favors hypoxia at stations in the UB group and northern WP, as compared to sites farther south and/or in EP, as accounts for the observations. Infrequent events observed at MH in an eastern embayment, based on 2 years’ sampling (2005 and 2006), are consistent with the bay-wide pattern and with the relatively rapid flushing of Mt. Hope Bay (MacDonald 2006). Severe hypoxia at the GB site appears to be associated with the long flushing time of the western embayment, due to its shallow depth, lack of river inputs, and weak exchange with the WP. The possibility exists that the reason the northern WP site (MV) is subject to severe hypoxia is that it is influenced by waters originating from both the Upper Bay and Greenwich Bay.

We attribute the temporal intermittency and weak spatial synchronicity of hypoxic events to the complex bay geometry and the likelihood that hypoxic events in different regions of the bay are shaped by different influences. For example, characteristics of hypoxic events at GB (in particular, high frequency, short durations, and less-pronounced interannual variability) suggest that the site is influenced by a set of processes distinct from those active at upper bay sites. A potentially important driver of geographic variations in processes influencing hypoxia is local hydrodynamics, which substantially modifies the long-term mean circulation described above on timescales similar to hypoxic events and is shaped by river flow events, wind fluctuations, and spring–neap cycles in ways that each vary strongly from site to site.

Event Variability and Driving Factors

In the exploratory MLR analyses, variance in event severity captured by the model was low (16.7% for UB group) to moderate (34.9% for GB site). It is possible that processes governing event variability are not represented well by the independent variables treated, despite use of numerous biological and physical parameters known to influence NB hypoxia, each with an appropriate range of temporal leads. In addition, relationships of hypoxic events to parameters may be more complex than can be identified well by the MLR method. While further analysis beyond these introductory calculations is clearly warranted, the MLR-selected independent variables support self-consistent interpretations in terms of expected processes, as follows.

For the UB group of sites, MLR model inclusion of 5-day lead river flow with positive coefficient is consistent with the dual role of rivers in delivering both the nutrients that fuel algal growth and freshwater that sustains density stratification. Higher chlorophyll levels and stronger stratification are closely associated with hypoxic events (e.g., Bergondo et al. 2005), but the fact that neither merited inclusion in the MLR model indicates that neither covaries as closely with event deficit-duration as do river flow and sea level difference. MLR model inclusion of zero-lag north–south sea level difference with negative coefficient is consistent with wind-driven variations in circulation influencing hypoxic events: northward wind stalls the estuarine exchange flow, increases the north–south sea level difference on timescales of hours (Rogers 2008), and enhances flushing of deep water thus reducing hypoxia severity (Bergondo 2004).

For the GB site, MLR model inclusion of 2-day lead chlorophyll with positive coefficient is consistent with event deficit-duration increasing with availability of algal material to decay. The GB site is in a western embayment distant from the influence of the major rivers. MLR model inclusion of 5-day lead northeastward wind with positive coefficient is consistent with the pattern of decreased embayment flushing in response to these wind conditions, due to gyre-like flows in Greenwich Bay that apparently lessen its exchange with the rest of the bay (Rogers 2008).

Though not completely conclusive, taken as a whole, the MLR results lend support to the interpretations that (a) circulation-related conditions (river flow, sea level difference, northeastward wind) play prominent roles in regulating hypoxic event characteristics and (b) characteristics of hypoxia in different regions of the bay are shaped by different processes, as is consistent with the reduced spatial synchronicity. Finally, the fact that neither MLR model included tidal range suggests that spring–neap cycles are not prominent in variability of event severity. The relation of spring–neap cycles to event timing described previously (Bergondo et al. 2005) was either not detected by the MLR or possibly only held during the limited number of years examined in their early analysis.

Interannual Variability

Correlations of June-mean river flow and June-mean stratification with the index for season-cumulative hypoxia severity were significant, though based on a small number of years. The similar results for river flow and stratification are an indication that on monthly timescales stratification variations at BR and NP are closely regulated by river flow variations. Given that hypoxia peaks in July and August, likely in association with peak annual temperatures that enhance respiration rates, the correlation with June-mean conditions implies a 1- to 2-month response timescale for hypoxia. This is loosely consistent with 10- to 40-day estimates of water residence time bay-wide (Pilson 1985). A similar relation between late spring runoff and summer hypoxia severity has been identified in Chesapeake Bay (Hagy et al. 2004). Although our hypoxia index was calculated using the BR and NP stations so it does not reflect variability in bay-wide spatial extent of hypoxia, independent datasets with broader geographic coverage (CTDO surveys, e.g., Deacutis et al. (2006); towed-body surveys, e.g., Melrose et al. (2007)) indicate that variability in spatial extent parallels that of the index (e.g., high spatial extent in 2006 and low spatial extent in 2004). On this basis, June-mean river flow appears to be a potentially useful index for the severity of subsequent July–August hypoxia as well as its spatial extent.

Implications for Benthic Ecology

Prior studies of the estuarine Providence and Seekonk Rivers at the head of NB document heavily degraded water quality, including common severe hypoxia, occurrence of anoxic conditions in the Seekonk River, and vulnerability to seasonally persistent hypoxia (Turner 1997). Our findings apply outside this area to the south and indicate that, in large areas of NB, where hypoxic events are commonly observed (Table 4), episodic/periodic hypoxia lasts days to weeks during summer months on a near-annual basis. Such conditions represent the third phase in the four-phase progression towards seasonally persistent hypoxia observed in many other systems as described by Diaz and Rosenberg (2008).

Locations at highest risk for direct mortality in benthic communities include western Greenwich Bay, deeper portions of Upper West Passage, and the southern part of the Upper Bay. This conclusion is based on results presented above using T2.9 (Fig. 5) and T1.4 (Fig. 7), as well as results of the MWT algorithm applied to all sites and all years using lower threshold and shorter minimum event duration values (0.5 mg O2 l−1 and 4 h; trigger duration 2 h) as justified in the “Introduction” (Rosenberg et al. 1991; Sagasti et al. 2001; Wu 2002; Person-LeRuyet et al. 2003; Vaquer-Sunyer and Duarte 2008). Such more intense and shorter events occurred only at GB (18, 3, and 11 events in 2003, 2005, and 2006, respectively), MV (eight events in 2006), and NP (two and one event, in 2003 and 2006, respectively). Our interpretations are conservative in the sense that they are based on measurements at least 0.5 m above the sediment–water interface, where the DO gradient can be steep, so they probably overestimate DO concentrations experienced by benthic species.

In these regions (containing the GB, MV, and NP sites), lethal levels for most benthic organisms are almost certainly reached for at least 4-h durations, and for much of the season concentrations likely cycle between lethal and just above lethal concentrations. Moderately sensitive sessile organisms have the greatest risk of mortality (Vaquer-Sunyer and Duarte 2008); highly tolerant species would likely require DO to fall below 0.5 mg O2 l−1 for longer durations (Sagasti et al. 2001). For many fish, macrocrustaceans, and echinoderms, such conditions are sufficient to cause mortality; however, due to their mobility and the reduced spatial synchronicity we have documented (Table 7), the likely impact is a combination of mortality and behavioral exclusion (Wannamaker and Rice 2000; Eby and Crowder 2002; Bell and Eggleston 2005; Hazen et al. 2006; Diaz and Rosenberg 2008). While it is recognized that, for many reasons, fish kills are not particularly good indicators of hypoxia impacts, it is relevant to note that the GB site is within 100 m of the central “kill zone” of the largest fish kill in over 50 years in August 2003 (RIDEM 2003).

The Greenwich Bay and Mount View areas have significant densities (e.g., RIDEM 2008) of the highly tolerant hard-shelled clam (Mercenaria mercenaria), a commercially important resource. The success of these populations may be partially due to predation refuge since oxygen conditions reach levels that exclude typical bivalve predators for significant periods of the summer (Altieri 2008 and Table 4). However, other important filter feeders within this functional group, such as blue mussels (Mytilus edulis), have experienced significant population losses in this area (Altieri and Witman 2006; Altieri 2008), suggesting that shifts in communities are sensitive to relative tolerance of species.

At all remaining stations, while there is some risk of direct mortality from rarer severe events, based on our results for hypoxic (Fig. 5) and suboxic (Fig. 6) conditions, we expect the dominant influence of hypoxia to be indirect effects. These include suboptimal growth or functionality, habitat compression, and shifts in predator–prey balances. Hypoxic conditions in the Upper Bay impair filter feeding by blue mussels (Altieri and Witman 2006), and juvenile flounder species common to NB (Pseudopleuronectes americanus and Paralichthys dentatus) exhibit significant decreases in growth at oxygen levels common to NB (Meng et al. 2002, 2008; Eby et al. 2005; Stierhoff et al. 2006). An example of habitat compression is movement to shallow areas during events by mobile species (Deacutis et al. 2006; Deacutis 2008; Diaz and Rosenberg 2008). Shifts in predator–prey balances associated with such habitat compression are likely, as the shallow peripheries of the Upper Bay and southern Providence River areas are known to be important habitat for juvenile fish (Meng and Powell 1999; Meng et al. 2005). Suboxic events, with DO levels that while not hypoxic nonetheless commonly cause indirect effects (Vaquer-Sunyer and Duarte 2008), were observed at every site in every sampled year with few exceptions. We conclude that indirect effects are very likely causing significant impacts across broad areas of the bay.

Acknowledgements

Funding from the NOAA Coastal Hypoxia Research Program (CHRP; Grant NA05NOS4781201) and from RIDEM-OWR is gratefully acknowledged. This is CHRP Contribution #106. The monitoring network is funded in part by the NOAA Bay Window Program, EPA Clean Water Act (sections 319 and 106), Narragansett Bay Commission (NBC), NOAA National Estuary Program, and State of Rhode Island; all that contribute are appreciated. We thank Sherry Poucher for guidance on DOCS-RI; NBC for Bullock Reach data; and Edwin Requintina and Elizabeth Crockford for their help collecting and processing data. Finally, we pay special tribute to the late Dana Kester, for his leadership and vision in fostering development of the fixed-site monitoring network.

Supplementary material

12237_2009_9165_MOESM1_ESM.pdf (2.1 mb)
Electronic supplementary material(PDF 2128 kb)

Copyright information

© Coastal and Estuarine Research Federation 2009