Three independent methods of time-series analysis were applied to a 37-year record of total phosphorus and chlorophyll-a data collected at four sampling stations representing the upper, middle and lower sections of the Bay of Quinte, NE Lake Ontario, Canada. The three methods were used to build consensus around the significance of the apparent declines in total phosphorus (TP) and chlorophyll-a (Chl-a) concentrations following two interventions: (1) point-source phosphorus loading reductions of about 50% to the upper bay in the winter of 1977–1978 and (2) the establishment of Zebra Mussels (Dreissena spp.) in the early to mid-1990's. The methods were applied to May to October monthly means and included: (1) nonparametric tests that accounted for persistence and seasonality and determined the statistical significance of step-trends, (2) ARIMA-Intervention modelling that produced forecasts into post-intervention time periods that were compared statistically to measured data and (3) regime shift detection for identifying the significance of persistent steps after removal of the seasonal components of the data series (modelled as periodic functions). Strong gradients in total phosphorus (TP) and Chl-a concentrations between the upper and lower Bay of Quinte still existed three decades after reductions in point-source loadings of TP to the upper bay, where, during 2000–2008, May-October average TP was 3.8 times higher and Chl-a, 2.9 times higher than in the lower bay. The Remedial Action Plan May-October TP objective of 0.030 mg l-1 for the upper Bay of Quinte was not achieved consistently in recent years. Concurrence among the data analysis methods suggests that the relative decreases in the upper bay TP (31%) and Chl-a (37%) after phosphorus loading reductions were greater than in the lower bay (0% and 10%, respectively). The relative decreases in the upper bay TP (0%) and Chl-a (29%) associated with Dreissena establishment were less than those measured in the lower bay (20% and 50%, respectively).

Introduction

The Bay of Quinte, a large (254 km2) multi-basin, Z-shaped embayment at the NE end of Lake Ontario, has had a long history of eutrophication (Warwick, 1980). The first comprehensive assessment of the nutrient status of the Bay was reported by Johnson and Owen (1971). Their assessment clearly identified the important role of high-concentration/low-flow sources of phosphorus (so-called “point” sources) originating from municipal sewage treatment facilities. Also at this time, municipal, provincial, state and national governments were enacting legislation to control the phosphate content in commercial laundry detergents (Prince and Bruce, 1972) and were upgrading sewage treatment plants and processes for reduction of phosphorus in treated wastewater (Anon, 1969; Hasler, 1969; Slater and Bangay, 1980).

The database for assessing the long-term ecosystem response to P loading reductions to the Bay of Quinte began in 1972 with the launch of “Project Quinte,” the preliminary results of which were published in Minns et al. (1986). That assessment continues in this article with an analysis of the apparent effects of two interventions: (1) an approximately 50% reduction in point-source P loading to the upper bay during the winter of 1977–1978 and (2) the establishment of Zebra and Quagga Mussels (Dreissena spp.) in the upper bay in the mid-1990's.

Environmental/biological time-series data present special challenges for detection and analysis of trends owing mainly to the characteristic serial dependencies of the data associated with seasonal, autocorrelative effects (so-called “red noise”) and other issues. A major purpose of this article is to apply three largely independent methods of trend detection to the 37-year record of phosphorus and chlorophyll data collected under Project Quinte in an effort to build consensus around the significance of the observed changes in total phosphorus and chlorophyll in the Bay of Quinte.

Methods

Field and laboratory methods for collection and laboratory analyses of phosphorus and chlorophyll samples were as described in Robinson (1986). Briefly, samples were collected as composites through the euphotic zone (twice the Secchi disc depth) at the four main sampling stations (B, N, HB and C) at weekly intervals from 1972 to 1982 and at biweekly intervals, 1983–2008. Chemical analyses (automated, colourimetric) were performed by the Laboratory Services Branch of the Ontario Ministry of the Environment (OME). A correction factor (+35%) was applied to chlorophyll a data generated prior to 1985 to account for a change in filtration/extraction methods (Nicholls and Hopkins, 1993). Chlorophyll a data reported here represent “total” chlorophyll a, uncorrected for phaeopigments.

A conservative approach to “far outlier” detection (median ± 5*st. dev.) was used for the data reported here. Outliers so identified were evaluated in a case-by-case approach that took into account the concentration of other variables in the same sample (e.g. total Kjeldahl nitrogen) and the same variables on nearby dates or nearby sampling stations (within Project Quinte or the OME's water treatment plant intake monitoring program at the city of Belleville). For the most part, outliers were replaced by linear interpolation between adjacent values (before and after). All data were reduced to monthly means. For a few of the earliest years of the project, there were no October samples, in which case mean values from adjacent years were substituted.

Time-series analyses were based on three independent approaches:

  • (1) nonparametric statistical tests for step-trends (Cluis et al., 1989), in recognition that the upper bay theoretically is flushed by river flows several times per year and that both interventions occurred over short time intervals. Responses to the interventions were, therefore, more likely to be seen as “steps” rather than monotonic changes over several years.

  • (2) ARIMA/intervention (A/I) analysis (Box and Tiao, 1975) based on Box and Jenkins (1970) autoregressive integrated moving average (ARIMA) models. For each data set, an ARIMA function describing the pre-intervention data was used to forecast for the post-intervention period. The forecast was compared with measured values (permutation test) to assess statistical difference. The PC software used (Human Edge Software, 1992) afforded full ARIMA capabilities, including inclusion of an iteratively optimized set of ARIMA parameters drawn from both seasonal and non-seasonal ARIMA models of the classic p, d, q type.

  • (3) Regime Shift Detection (RSD; Rodionov, 2004, 2006), based on sequential statistics, was applied to residual data sets after modelling the seasonal component of each data set as a periodic function (Cohort Software, 1995).

The three procedures outlined above were applied systematically to all four stations for both total phosphorus (TP) and chlorophyll a (Chl-a), and produced copious output that can best be presented in charts and plots, most of which could not be presented here for space reasons (see the online appendix for examples of some of this output*).

Results

Simple plots of monthly mean TP and Chl-a (Figure 1) suggest the initiation of important negative step-trends in the late 1970's, especially at the upper bay stations (B and N). Total P concentrations in the middle bay (Station HB) appeared to follow a negative monotonic trend until about the early 1990's (Figure 1c). The apparent response of TP to point-source loading reductions in 1977–1978 is rendered less clear by increases in concentrations in the early 1980's at stations B and HB (also incompletely evident at Station N). Chlorophyll trends generally agreed with the TP trends; steep declines were most evident at Stations B and N beginning in 1978. Clearly, more detailed testing is required to explore the possible significance of these changes.

Figure 1.

May to October means (±1 st. dev) of total phosphorus and total chlorophyll a (uncorrected for phaeopigments) based on monthly means for each of the four main sampling stations in the Bay of Quinte, B, N, HB and C, 1972–2008. Station N was not sampled during 1983–1989.

Figure 1.

May to October means (±1 st. dev) of total phosphorus and total chlorophyll a (uncorrected for phaeopigments) based on monthly means for each of the four main sampling stations in the Bay of Quinte, B, N, HB and C, 1972–2008. Station N was not sampled during 1983–1989.

Nonparametric step-trend tests

The nonparametric step-trend detection tests revealed a total of 13 statistically significant negative steps associated with both the TP loading intervention and the Dreissena intervention out of a possible 16 cases (2 variables, 4 stations, 2 interventions; Table 1). The apparent impact of point-source TP loading reductions on TP concentrations in the upper bay was greater than the reductions associated with the establishment of Dreissena. A similar conclusion can be drawn from the results of the same tests on TP data at Station N, revealing a highly significant negative step in 1978 and a nonsignificant step in 1996 (Table 1). The reverse was found for Station C in the lower bay, where a statistically significant negative step was associated with the Dreissena establishment but not with the upper bay TP loading reduction. Station HB in the middle bay was intermediate in its apparent response; there was a significant negative step in 1978 (though less so than at Station B), and, as at Station C, a significant negative step in 1996.

Table 1.

Results of nonparametric tests (Mann-Whitney; Mann-Whitney seasonal and Mann-Whitney seasonal-Lettenmaier tests; Cluis et al. 1989) for step-trends in May to October means of total phosphorus, chlorophyll a, and chlorophyll a-to-total phosphorus ratios at each of the four main Project Quinte sampling stations, B, N, HB and C. Two time periods were analyzed to determine changes associated with two interventions: (1) phosphorus removal at sewage treatment plants initiated in the winter of 1977–1978 (step-trend tests: 1972–1977 vs. 1978–1984) and (2) the establishment of Zebra Mussels in 1995 (step-trend tests: 1985–1994 vs. 1996–2008, with 1995 data having been removed as a likely transitional year). Pre- and post-step means and the root mean square error (RMSE [in mg l−1 = the precision of the step model for the defined time period]) are given for each intervention. Note: Where step-trends were not significant (ns), pre-step mean = post step mean (i.e. the “global mean”); for Station N, there were no samples for 1983–1989, so the pre- and post-step time periods are different from those used for the other stations.

1972–1984 (step in 1978)1985–2008 (step in 1995)
StationPre-stepPost-stepRMSESignificancePre-stepPost-stepRMSESignificance
a. Total P 
 B 0.078 0.053 0.024 P<0.001 0.038 0.038 0.015 P = 0.059 (ns) 
 N 0.071 0.050 0.020 P = 0.001 0.033 0.033 0.012 P = 0.165 (ns) 
 HB 0.054 0.046 0.017 P = 0.004 0.034 0.030 0.013 P = 0.005 
 C 0.021 0.021 0.006 P = 0.203 (ns) 0.015 0.012 0.005 P = 0.004 
b. Chl-a 
 B 0.040 0.022 0.020 P = 0.006 0.018 0.014 0.010 P = 0.016 
 N 0.041 0.029 0.022 P = 0.047 0.017 0.011 0.009 P = 0.002 
 HB 0.031 0.025 0.015 P = 0.033 0.017 0.010 0.009 P<0.001 
 C 0.010 0.009 0.004 P = 0.022 0.006 0.003 0.003 P<0.001 
c. Chl-a-to-TP ratio 
 B 0.498 0.412 0.194 P = 0.026 0.491 0.360 0.260 P = 0.001 
 N 0.560 0.560 0.243 P = 0.465 (ns) 0.480 0.330 0.185 P = 0.003 
 HB 0.560 0.560 0.230 P = 0.434 (ns) 0.500 0.340 0.210 P<0.001 
 C 0.510 0.510 0.350 P = 0.150 (ns) 0.490 0.300 0.250 P<0.001 
1972–1984 (step in 1978)1985–2008 (step in 1995)
StationPre-stepPost-stepRMSESignificancePre-stepPost-stepRMSESignificance
a. Total P 
 B 0.078 0.053 0.024 P<0.001 0.038 0.038 0.015 P = 0.059 (ns) 
 N 0.071 0.050 0.020 P = 0.001 0.033 0.033 0.012 P = 0.165 (ns) 
 HB 0.054 0.046 0.017 P = 0.004 0.034 0.030 0.013 P = 0.005 
 C 0.021 0.021 0.006 P = 0.203 (ns) 0.015 0.012 0.005 P = 0.004 
b. Chl-a 
 B 0.040 0.022 0.020 P = 0.006 0.018 0.014 0.010 P = 0.016 
 N 0.041 0.029 0.022 P = 0.047 0.017 0.011 0.009 P = 0.002 
 HB 0.031 0.025 0.015 P = 0.033 0.017 0.010 0.009 P<0.001 
 C 0.010 0.009 0.004 P = 0.022 0.006 0.003 0.003 P<0.001 
c. Chl-a-to-TP ratio 
 B 0.498 0.412 0.194 P = 0.026 0.491 0.360 0.260 P = 0.001 
 N 0.560 0.560 0.243 P = 0.465 (ns) 0.480 0.330 0.185 P = 0.003 
 HB 0.560 0.560 0.230 P = 0.434 (ns) 0.500 0.340 0.210 P<0.001 
 C 0.510 0.510 0.350 P = 0.150 (ns) 0.490 0.300 0.250 P<0.001 

Chlorophyll a revealed negative steps at all four stations associated with both interventions (Table 1). The largest reductions associated with the TP loading intervention were seen in the upper bay at Stations B (45%) and N (29%), while Station N, as well as the middle and lower stations HB and C, experienced reductions of 36%, 41%, and 44%, respectively, in association with the Dreissena intervention. Decreases in the Chl-a-to-TP ratio were clearly much more strongly associated with the Dreissena establishment than with the TP load reductions at all stations. The statistical significance levels (P values in Table 1) for the inferred Dreissena-related 1996 negative step were 0.001, 0.003, <0.001, and <0.001 for Stations B, N, HB and C, respectively. In contrast, the corresponding P values for the 1978 negative steps associated with the TP load reductions were 0.026, 0.456, 0.434 and 0.150 (Table 1). The reduction in the ratio Chl-a/TP was clearly identified by regression analysis (Figure 5) showing significantly different relationships pre- and post-Dreissena. This was also demonstrated in the results from a multiple regression analysis of the entire1972–2008 data set, where the Zebra Mussel influence (ZM) was included as a dummy variable (+1 for 1995–2008 and zero for 1972–1994): LogChl-a = 1.127*LogTP - 0.138ZM - 0.525 (adjusted multiple R2 = 0.882; P < 0.0001).

Figure 5.

Shift in the relationship between May-October total phosphorus and chlorophyll-a at the four main Bay of Quinte stations, B, N, HB and C, pre-Dreissena and post-Dreissena (closed circles and open squares, respectively).

Figure 5.

Shift in the relationship between May-October total phosphorus and chlorophyll-a at the four main Bay of Quinte stations, B, N, HB and C, pre-Dreissena and post-Dreissena (closed circles and open squares, respectively).

ARIMA/intervention (A/I) analysis

ARIMA modelling was completed on all stations and variables that were included previously in the nonparametric tests (above), but to save space, only one detailed example (Figure 2) is included here. In no cases (four stations, two variables) did ARIMA modelling produce predictions for post-intervention time periods that were exclusively outside the 95% confidence band of the prediction. In many cases (e.g. Stn B, chl-a [1978]; Stn B, TP [1978]; Stn N, TP [1978]; Stn N, Chl-a [1978]; Stn N, Chl-a [1996]; Stn HB, Chl-a [1996]; and Stn C, Chl-a [1996]), the majority of the measured values in the post-intervention period were well below the prediction but still within the 95% confidence boundaries of the ARIMA prediction. In all of these cases, however, permutation tests showed that measured mean TP and Chl-a values were significantly below the ARIMA predictions (P < 0.001).

Figure 2.

One example illustrating an ARIMA model prediction compared with measured data pre-and post-TP loading reductions at Station B in the Bay of Quinte.

Figure 2.

One example illustrating an ARIMA model prediction compared with measured data pre-and post-TP loading reductions at Station B in the Bay of Quinte.

Regime shift detection (RSD)

Seasonality was a strong element of all 16 time-series investigated (4 stations, 2 variables, 2 intervention time periods), yielding periodic functions with correlation coefficients ranging from 0.750 to 0.991. In the TP time-series for Station B, for example (Figure 3a), initiation of phosphorus loading reductions in the winter of 1977–1978 was very evident in the time-series of “de-seasonalized” residuals (Figure 3b). Clearly, the strong autocorrelation effects of seasonal periodicity in the original data (Figure 3c) were removed by subtraction of the periodic function given in Figure 3a.

Figure 3.

May to October monthly mean total phosphorus concentrations at Station B, 1972–1984 (pre- and post-TP loading intervention; evenly spaced, with the unsampled November to April periods left out): (a) monthly means with a sinusoidal function representing the monthly means, (b) residuals (monthly means minus the sinusoid in part (a)) and (c) the autocorrelation coefficients between 0 and n/2 (= 38) lags, showing strong autocorrelation in the original data and no significant autocorrelation in the residuals.

Figure 3.

May to October monthly mean total phosphorus concentrations at Station B, 1972–1984 (pre- and post-TP loading intervention; evenly spaced, with the unsampled November to April periods left out): (a) monthly means with a sinusoidal function representing the monthly means, (b) residuals (monthly means minus the sinusoid in part (a)) and (c) the autocorrelation coefficients between 0 and n/2 (= 38) lags, showing strong autocorrelation in the original data and no significant autocorrelation in the residuals.

Regime shifts were detected over a wide range of statistical probabilities, but only those shifts that were revealed at levels of P < 0.2 were defined here. The most significant negative regime shifts (those with the lowest P values) included TP at Stations B and N in 1978 and Chl-a at Stations HB and C in 1996, all shifts being significant at P < 0.005. In all cases, except where no shifts were identified, detection was made with a higher level of confidence (lower P value) on model residuals (seasonality removed) than on original uncorrected series (Figure 4). In some cases this was very dramatic. At Station HB for the second (Dreissena) intervention, for example, the regime shift in TP beginning in 1995 was significant at P = 0.0075 on the residuals (seasonally “corrected” data), but on the original unadjusted data, a shift was detected only at P levels exceeding 0.18 (i.e. not statistically significant).

Figure 4.

May to October monthly mean total phosphorus data from Station B, 1972–1984 (per Figure 3a) showing a regime shift in 1978 (a) that was significant at P = 0.05, but not at P < 0.04. In contrast, there is a very well defined regime shift in the residual data series at P = 0.01 (b) after removal of seasonality modelled as the periodic function given in Figure 3a.

Figure 4.

May to October monthly mean total phosphorus data from Station B, 1972–1984 (per Figure 3a) showing a regime shift in 1978 (a) that was significant at P = 0.05, but not at P < 0.04. In contrast, there is a very well defined regime shift in the residual data series at P = 0.01 (b) after removal of seasonality modelled as the periodic function given in Figure 3a.

Rarely were any regime shifts detected that were not associated with either 1978 or 1995/1996—the dates representing the turning points for the two interventions explored here. Of those, only the positive shift identified in 2005 in Station B seasonally corrected Chl-a was statistically significant (P < 0.05).

Discussion

Over the past two decades, nonparametric statistical testing for long-term monotonic or abrupt step-trends has become the method of choice for detecting change in river and lake water quality data series (Cluis et al., 1989; Bodo 1989; Tsanis, 1993), including several applications to Great Lakes data series (Helsel, 1987; Tsanis and El-Shaarawi, 1992; Nicholls et al., 2001). The application of ARIMA modelling/forecasting/intervention an- alysis, however, has been applied only rarely to Great Lakes data sets (Nicholls et al., 1999a) or elsewhere (Lehmann and Rode, 2001; Kurunç et al., 2005). Other newer methods, like Rodionov's (2004) RSD test, originally intended for identifying “regime shifts” in climate-related time-series data should be used more widely in other environmental applications as they become better known (Andersen et al., 2009).

Although it is very clear that the major point-source TP reduction occurred in the winter of 1977–1978, a date for the establishment of Dreissena spp. in the upper bay is less clear. Bailey et al. (1999) reported the first arrivals in 1993, but a population large enough to elicit a detectable response in limnological variables may not have been established for a year or two after that. Indeed, the RSD tests indicated some step changes beginning in 1995 and others in 1996, depending on the variable and sampling location. Evidence for this delayed response was also supported by earlier preliminary multivariate analysis of Bay of Quinte phytoplankton community structure (Nicholls et al., 2002), which identified 1995 as a transition year between the pre- and post-Dreissena ecosystem states.

The suggestion of similarities in trophic state between upper, middle, and lower Bay of Quinte and western, central, and eastern Lake Erie was first made by Johnson and Owen (1971) and has been reinforced by the clear gradients in TP and chlorophyll reported here, still existing three decades after the 1970's point-source loading reductions. Undoubtedly, this gradient is not only the consequence of the downstream progressive dilution of the nutrient loads supplied to the upper bay but is also controlled by morphometric and hydrologic influences including the incursion of Lake Ontario water—a comparatively low-nutrient water mass that has been detected as far “upstream” in the Bay of Quinte as Station HB (Freeman and Prinsenberg, 1986).

The physical influence of Lake Ontario water on the lower bay was apparent in the phosphorus and chlorophyll concentrations at Station C. Nicholls et al. (2001) reported monotonic declines in phosphorus concentrations in the raw-water intakes for the municipalities of Cobourg and Kingston (west and east, respectively, of the lower Bay of Quinte) of 0.7–0.9 μg L-1 yr-1 during the 1980s and 1990s. This, together with the apparent lack of a well-defined negative step-trend in TP at Station C in 1978 (when such declines were strongly manifested in the upper bay), suggests that the longer term monotonic declines in the lower bay are more strongly influenced by Lake Ontario than by the upper Bay of Quinte. There were dramatic declines in chlorophyll in north-eastern Lake Ontario (municipal water intake samples; Nicholls, 2001), but because these were mostly associated with a later time period (the 1990s), they were associated with the Zebra Mussel (Dreissena) establishment in Lake Ontario. Declines in chlorophyll at Bay of Quinte Station C during the 1990's were likely more related to Dreissena effects in northeastern Lake Ontario and in the lower bay than to lingering impacts of phosphorus loading reductions to the upper bay. In general, it can be concluded that the TP loading intervention had a greater relative impact on the upper bay than in the lower bay while the reverse is true for the Dreissena intervention (greater relative impact in the lower bay than in the upper bay).

Results from the three independent analytical approaches were in very good agreement with respect to the levels of significance of the identified steps or shifts. Those stations and variables showing no statistically significant step-trends in the nonparametric tests (i.e. global means pre- and post- interventions identical in Table 1) were also those stations/variables that had measured data values clustered around the ARIMA prediction line for the post-intervention periods, and likewise did not reveal statistically significant RSD shifts.

In both the nonparametric step-trend and ARIMA approaches, the analyst must provide an a priori input on the date of the first sample date for the step (in the first approach) or the date of the intervention in the ARIMA modelling. Although some simple tools are available to assist in this largely subjective determination (e.g. the shape of CUSUM or double-mass curves (Nicholls, 2001), a more objective way to identify the end-date of a pre-intervention period in a data set is now available in Rodionov's (2004) RSD test. It was applied here after removal of strong seasonal components in the data sets because the RSD procedure is most efficient for residual data sets that have been thus “pre-whitened” (Rodionov, 2006).

Regime shifts identified by RSD that coincide with those used in nonparametric tests for step-trends and for use in ARIMA/intervention modelling clearly help to build consensus about intervention effects. Conversely, regime shifts identified by the RSD approach that do not coincide with the timing of the known interventions help to provide perspective on the relative significance of factors other than those directly related to the interventions that may conspire to affect the structure of the time-series.

The declines in TP and chlorophyll reported here for the Bay of Quinte support findings from other Great Lakes locations (Nicholls, 2001; Nicholls et al., 1999a, 2001). Clearly, dreissenids are effective in removing TP from the water column during filter feeding of suspended particulate matter (containing phosphorus and chlorophyll) and delivering TP to the lake sediment layer in feces, pseudofeces and dead dreissenids. The normal transference of TP through the plankton and necton communities is thus disrupted; instead, TP is directed to the benthos (Hecky et al., 2004).

Declines in the chlorophyll-to-TP ratio associated with the establishment of Dreissena in the Lower Great Lakes are well documented. May to October Chl-a/TP ratios in northern, nearshore waters of Lake Erie were 2–6 times higher prior to Dreissena establishment, than after (Nicholls et al., 1999b). The likely explanation for the decline in the chlorophyll-to-TP ratio associated with Dreissena establishment is as follows: Essentially all chlorophyll-bearing particles contribute to the total P pool, but a portion of the total P pool is not associated with chlorophyll because it includes non-chlorophyllous particulate P and dissolved P (but, in a P-limited system, dissolved P is likely to be a minor component of the total P pool during the growing season). It follows then, that the filter-feeding habit of Zebra Mussels will remove P-bearing particles (both chlorophyllous and non-chlorophyllous particles) from the water column. Being organic, the chlorophyllous particles will largely be digested and assimilated by the mussels, but the inorganic P-bearing particles will be ejected directly or “packaged” into pseudofeces and expelled back into the environment from which they came. The net result is a reduction in both chlorophyll and total P concentrations in the water column, but because of the feedback of rejected inorganic particulate P, the rate of water-column chlorophyll decline exceeds the P decline, resulting in a negative step trend in the chlorophyll-to-phosphorus ratio.

Conclusions

The three methods of time-series analysis applied here to the 37-year record of phosphorus and chlorophyll-a data from the Bay of Quinte have revealed the persistence of strong gradients between the upper bay (Belleville region) to the lower bay (between Glenora and Lake Ontario). The Remedial Action Plan objective of 0.030 mg l-1 (May–October average) for the upper Bay of Quinte was not achieved consistently in recent years, even while under the influence of the water column clearing effects of filter-feeding Dreissenid Mussel populations established about the mid-1990s. Reductions in chlorophyll-a throughout the Bay of Quinte were significant after the establishment of Dreissena spp. Reductions in total phosphorus, however, were much more significant in the upper bay following point-source control of phosphorus loading initiated in the winter of 1977.

Acknowledgements

I thank the many regular staff and students of the Laboratory Services Branch, Ontario Ministry of the Environment (Kingston and Toronto laboratories) for several thousand chlorophyll and phosphorus analyses of Bay of Quinte samples over the past 37 years. Students and staff of Fisheries and Oceans, Canada collected most of the water samples from the Bay of Quinte stations.

References

Andersen, T., Carstensen, J., Hernández-García, and Durte, C. M.
2009
.
Ecological thresholds and regime shifts: approaches to identification
.
Trends in Ecology and Evolution
,
24
:
49
57
.
Anon
.
1969
.
Report to the International Joint Commission on the pollution of Lake Erie, Lake Ontario and the international section of the St. Lawrence River
, Vol.
I
,
Ottawa
:
Queen's Printer
.
International Lake Erie and Lake Ontario - St. Lawrence River Water Pollution Boards
Bailey, R. C., Grapentine, L., Stewart, T. J., Schaner, T., Chase, M. E., Mitchell, J. S. and Coulas, R. A.
1999
.
Dreissenidae in Lake Ontario; impact assessment at the whole lake and Bay of Quinte spatial scales
.
J. Great Lakes Res.
,
25
:
482
491
.
Bodo, B. A.
1989
.
Robust graphical methods for diagnosing trend in irregularly spaced water quality time series
.
Environ. Monitoring and Assessment
,
12
:
407
428
.
Box, G. E.P. and Jenkins, G. M.
1970
.
Time Series Analysis Forecasting and Control. 2nd Ed
,
Oakland, CA
:
Holden-Day
.
Box, G. E.P. and Tiao, G. C.
1975
.
Intervention analysis with applications to economic and environmental problems
.
J. Amer. Statistical Assoc.
,
70
(
349
):
70
79
.
Cluis, D., Langlois, C., Van Coillie, R. and Laberge, C.
1989
.
Development of a software package for trend detection in temporal series: application to water and industrial effluent quality data for the St. Lawrence River. Environ
.
Monitoring and Assessment
,
12
:
429
441
.
CoHort Software
.
1995
.
CoStat
Minneapolis, Minnesota, , U.S.A.
55419
Freeman, NG. and Prinsenberg, S. J.
1986
. “
Exchange flows in the Adolphus Reach/North Channel
”. In
Project Quinte:point source phosphorus control and ecosystem response in the Bay of Quinte, Lake Ontario
Edited by: Minns, C. K., Hurley, D. A. and Nicholls, K. H.
27
39
.
Can. Spec. Publ. Fish. Aquat. Sci. 86
Hasler, A. D.
1969
.
Cultural eutrophication is reversible
.
BioScience
,
19
:
425
431
.
Hecky, R. E., Smith, E. H., Barton, D. R., Taylor, W. D., Charlton, M. N. and Howell, T.
2004
.
The nearshore phosphorus shunt: a consequence of ecosystem engineering by dreissenids in the Laurention Great Lakes
.
Can. J. Fish. Aquat. Sci.
,
61
:
1285
1293
.
Helsel, D. R.
1987
.
Advantages of non-parametric procedures for analysis of water quality data
.
J. Hydrol. Sci.
,
32
:
179
190
.
Human Edge Software
.
1992
. “
The Forecasting Edge
”.
San Kateo, CA
:
Human Edge Software Corporation
.
Johnson, M. G. and Owen, G. E.
1971
.
Nutrients and nutrient budgets in the Bay of Quinte, Lake Ontario
.
J. Wat. Pollut. Control Fed.
,
43
:
836
853
.
Kurunç, A., Yürekli, K. and Çevik, O.
2005
.
Performance of two stochastic approaches for forecasting water quality and streamflow data from Yeşilırmak River, Turkey
.
Environmental Modelling & Software
,
20
:
1195
1200
.
Lehmann, A. and Rode, M.
2001
.
Long-term behaviour and cross-correlation water quality analysis of the river Elbe, Germany
.
Water Research
,
35
:
2153
2160
.
Minns, C. K., Hurley, D. A. and Nicholls, K. H.
1986
. “
Project Quinte: point source phosphorus control and ecosystem response in the Bay of Quinte, Lake Ontario
”.
86
Can. Spec. Publ. Fish. Aquat. Sci.
.
Eds.
Nicholls, K. H.
2001
.
CUSUM phytoplankton and chlorophyll functions illustrate the apparent onset of Dreissenid Mussel impacts in Lake Ontario
.
J. Great Lakes Res.
,
27
:
393
401
.
Nicholls, K. H. and Hopkins, G. J.
1993
.
Recent changes in Lake Erie (north shore) phytoplankton: cumulative impacts of phosphorus loading reductions and the Zebra Mussel introduction
.
J. Great Lakes Res.
,
19
:
637
647
.
Nicholls, K. H., Standke, S. J. and Hopkins, G. J.
1999a
. “
Effects of Dreissenid Mussels on nitrogen and phosphorus in north shore waters of Lake Erie
”. In
State of Lake Erie (SOLE) - Past, Present and Future
, Edited by: Munawar, M., Edsall, T. and Munawar, I.
323
336
.
Leiden, , The Netherlands
:
Ecovision World Monograph Series. Backhuys Publ.
.
Nicholls, K. H., Hopkins, G. J. and Standke, S. J.
1999b
.
Reduced chlorophyll-to-phosphorus ratios in nearshore Great Lakes waters coincide with the establishment of Dreissenid Mussels
.
Can. J. Fish. Aquat. Sci.
,
56
:
153
161
.
Nicholls, K. H., Hopkins, G. J., Standke, S. J. and Nakamoto, L.
2001
.
Trends in total phosphorus in Canadian near-shore waters of the Laurentian Great Lakes: 1976–1999
.
J. Great Lakes Res.
,
27
:
402
422
.
Nicholls, K. H., Heintsch, L. and Carney, E. C.
2002
.
Univariate step-trend and multivariate assessments of the apparent relative effects of P loading reductions and Zebra Mussels on the phytoplankton of the Bay of Quinte, Lake Ontario
.
J. Great Lakes Res.
,
28
:
15
31
.
Prince, A. T. and Bruce, J. P.
1972
. “
Development of nutrient control policies in Canada
”.
Ottawa
:
Inland Waters Branch, Dept. of the Environment
.
Tech. Bull. No. 51
Robinson, G. W.
1986
. “
Water quality of the Bay of Quinte, Lake Ontario, before and after reductions in phosphorus loading
”. In
Project Quinte: point source phosphorus control and ecosystem response in the Bay of Quinte, Lake Ontario
, Edited by: Minns, C. K., Hurley, D. A. and Nicholls, K. H.
50
58
.
Can. Spec. Publ. Fish. Aquat. Sci
.
86
Rodionov, S. N.
2004
.
A sequential algorithm for testing climate regime shifts
.
Geophys. Res. Lett.
,
31
:
L09204
doi:
Rodionov, S. N.
2006
.
The use of pre-whitening in climate regime shift detection
.
Geophys. Res. Lett.
,
31
:
L12707
doi:
Slater, R. W. and Bangay, G. E.
1980
. “
Action taken to control phosphorus in the Great Lakes
”. Edited by: Loehr, R. C., Martin, C. S. and Rast, W.
Ann Arbor, Michigan
:
Ann Arbor Science
.
Proceedings of the 1979 Conference on Phosphorus Management Strategies for Lakes
Tsanis, I. K.
1993
.
WQTA—a water quality trend analysis program
.
Water Poll. Res. J. Canada
,
28
:
311
335
.
Tsanis, I. K. and El-Shaarawi, A. H.
1992
.
Trend evaluation of water quality parameters in the Niagara and St. Lawrence Rivers
.
Environ. Monit. Assessment
,
23
:
205
218
.
Warwick, W. F.
1980
.
Paleolimnology of the Bay of Quinte, Lake Ontario: 2800 years of cultural influence
.
Can. Bull. Fish. Aquat. Sci.
, :
206