The binational Lake Ontario Lower Aquatic Foodweb Assessment program (LOLA) intensively sampled Lake Ontario in the Spring (April 28-May 3), Summer (August 10–11 and August 19–21) and Fall (September 21–25) of 2003. However, the timing of shipboard surveys often misses critical periods in biological productivity. We directly compared surface water temperature and chlorophyll a (chl a) measurements made during these cruises to Moderate Resolution Imaging Spectroradiometer (MODIS) satellite images for May 3, August 18, and September 21, 2003. Satellite measurements were strongly correlated to shipboard measurements for surface water temperature (r2 = 0.98) and chl a (r2 = 0.62, offshore sites only). The OC4 algorithm for chl a greatly overestimated nearshore sites because of the probable presence of color producing agents other than chl a. However, its relative reliability for offshore sites adds confidence in using the imagery to fill the gaps between sampling cruises.

Introduction

Concern about eutrophication of the Great Lakes led to the Great Lakes Water Quality Agreement of 1972 between Canada and the United States and the implementation of mandated phosphorus loading reductions. This agreement marked the start of a continuing legacy of Lake Ontario water quality monitoring programs that now includes nearly four decades (Kwiatkowski and El-Shaarawi, 1977, Millard et al., 2003; Holeck et al., 2008). These programs detected a decrease in Spring total phosphorus (TP) concentration from 20–25 μg l−1 in the early 1970's to just below the target concentration of 10 μg l−1 in 1986 (Millard et al., 2003). By 2003, TP had continued to decrease to < 8 μg l−1 (Holeck et al., 2008). Consequently, the implementation of management actions to reduce phosphorus was successful. However, an ever-present challenge for these monitoring programs has been to quantify the impact of the reduction in phosphorus on phytoplankton and zooplankton biomass. In turn, understanding the status of both phosphorus and the lower food web has been crucial for fishery managers in understanding the carrying capacity of the lake.

The Canadian Surveillance Program provided a thorough baseline for water quality in Lake Ontario in the 1970s (El-Shaarawi and Kwiatkowski, 1977). This program included 85 stations visited during 9–15 cruises between March and December. The high sampling frequency provided a thorough evaluation of the different scales of spatial and seasonal variability for several physical and biological parameters. From 1974 to 1979, Spring TP decreased 9.6 μg l−1 (35%) (Kwiatkowski, 1982). During this time period, Summer (June-August) chlorophyll a (chl a) concentration decreased by 0.45 μg l−1 (7.5%). However, the decrease of chl a was not consistent lake-wide—some nearshore zones displayed either no change or increased over the time period. Models (Thomann et al., 1977) had predicted that it would take 10–20 years for phytoplankton biomass to decrease, and a large decline in chl a (2 μg l−1 or 33%) occurred between 1982 and 1985, suggesting a ten-year delay in the response of phytoplankton to phosphorus reduction (Millard et al., 2003). Ecosystem change in Lake Ontario has complicated the relationship of phosphorus and phytoplankton biomass. For example, the decrease of chl a from an average of 4.5 μg l−1 in 1990 to 1.5 μg l−1 in 1996 was interpreted as an effect of dreissenid grazing (Figure 1, Millard et al., 2003). Shipboard monitoring programs have been successful in detecting large changes in phosphorus and chl a. It has been much more challenging to track subtle trends in phytoplankton biomass, particularly since seasonal variability can be large.

Ship time is expensive, and no current field-sampling program in Lake Ontario matches the seasonal or spatial coverage of the Canadian Surveillance program. The US EPA Great Lakes National Program Office (GLNPO) samples ten stations in Lake Ontario in April and August of each year. Lake-wide sampling for each Great Lake follows a five-year rotation. The binational Lake Ontario Lower Aquatic (LOLA) food-web program of 2003 included 28 sites sampled in three seasons (Spring (April), Summer (August), and Fall (September)) (Holeck et al., 2008) and was repeated in the Summer of 2008 at a smaller scale.

The objective of this paper is to compare chl a measurements from the three LOLA 2003 shipboard surveys to chl a estimates from satellite imagery. Satellite images provide an instantaneous measurement of surface water temperature and lake color for all of Lake Ontario and can place a single sampling site in the context of lake scale dynamics. Satellites collect images throughout the year and therefore can fill the long gaps between sampling programs.

Satellite measurement of surface water temperature is already well developed. NOAA's Great Lakes Coast Watch Program (http://coastwatch.glerl.noaa.gov/) has automated the process of producing daily maps of surface water temperature with a spatial resolution of 2.6 km in a program known as the Great Lakes Surface Environmental Analysis (GLSEA) (Schwab et al., 1999). These maps are based on AVHRR (Advanced Very High Resolution Radiometer) images and estimate temperatures in cloud covered areas by using previous imagery. These images have been ground-truthed with data from eight weather buoys with a mean difference of 0.5°C. One important application of this product has been the calculation of daily averages for lake-wide surface water temperature for each of the Great Lakes. Surface water temperature is an important variable for bulk hydrodynamic and ecological models and has now been available from 1992 to 2007. These measurements have been compared with a water temperature climatology model developed by Schneider et al. (1993) to detect change over time. Temperature images have also been used to mark surface water temperature fronts (Ullman et al., 1998) and to develop upwelling indices for Lake Michigan (Plattner et al., 2006). These temperature products have been useful, for example, for understanding the distribution of both Dreissena sp. in the Canadian nearshore of Lake Ontario (Wilson et al., 2006) and steelhead in Lake Michigan (Hook et al., 2004).

The use of satellite imagery for chl a concentration and hence phytoplankton biomass is much less developed for the Great Lakes. Lake color data is available from two satellites, SeaWIFS (Sea-viewing Wide Field-of-view Sensor) and MODIS (Moderate resolution Imaging Spectroradiometer). Algorithms (e.g. NASA's OC4) that convert ocean color to phytoplankton biomass (chl a) have been successful for open ocean (Case I, Morel and Prieur, 1977) conditions where phytoplankton is the primary color component (O'Reilly et al., 1998). These algorithms perform poorly in coastal and inland waters (Case II) where concentrations of colored dissolved organic matter (CDOM) and suspended matter (sm) can be as high as chl a concentrations. Atmospheric corrections and coastal boundary effects have also been problematic. Algorithms have been recently developed for Case II waters of the Great Lakes (Schuchmann et al., 2006; Pozdnyakov et al., 2005) that separate chl a, CDOM, and suspended minerals, and are now validated for Lake Michigan. A similar analysis is possible for Lake Ontario using the long history of in situ measurements of inherent optical properties (Bukata et al., 1991). While waiting for the development of a Lake Ontario specific algorithm, we tested the usefulness of the OC4 algorithm.

Methods

Shipboard sampling of chl a

Lake-wide field sampling in Lake Ontario in 2003 (hereafter referred to as LOLA) occurred aboard the EPA vessel Lake Guardian and the Canadian Coast Guard vessel Limnos (Figure 2). Three sampling periods included Spring (April 28-May3), Summer (August 10–11 and August 19–21) and Fall (Sept 21–25). Water quality parameters were measured from vertically integrated (to 20 m depth or 2 m above the bottom for shallow sites) epilimnion water samples. For chl a, up to 2 l of water was filtered on a Whatman GF/C filter at a pressure not exceeding 300 mm Hg, and the filter was frozen for later dissolving with 90% acetone (Holeck et al., 2008). Triplicate filters were collected at each site. We used the value for chl a (μg l−1) uncorrected for phaeophyton and other degradation pigments during the spectrophotometric measurement (i.e. before acid addition).

Satellite imagery

MODIS satellite imagery of the Great Lakes region (1 km resolution) was retrieved for several dates in 2003 from NASA's Ocean Color Web site (http://oceancolor.gsfc.nasa.gov). Three cloud-free images of Lake Ontario from May 3, August 18, and September 21, 2003 were compared to the three LOLA sampling periods. “Level 2” data has been processed by NASA from raw spectroradiometer data to estimate surface water temperature and chl a (μg l−1). The chl a product is derived using NASA's OC4 algorithm generally used for Case I open ocean waters. During the processing, pixels that are too close to the shore or covered in clouds are flagged as “−1.0”. Images spanning the period from April to October, 2003 were used to track seasonal changes of chl a in Lake Ontario.

Satellite images of surface water temperature and OC4-derived chl a were displayed using NASA's SeaDAS software (SeaDAS Version 5.3, http://oceancolor.gsfc.nasa.gov/seadas/). Parameter values for the 1 km pixel that contains each survey site were extracted using the “Ship Track” feature and a text file of GPS positions. In order to calculate lake-wide averages of surface water temperature and OC4-derived chl a, parameter values for all pixels within Lake Ontario were extracted.

Statistical analysis

Shipboard and satellite measurements of surface water temperature and chl a for LOLA sites were compared using linear regression in Microsoft Excel. Sites were categorized as nearshore (< 30 m depth) and offshore (> 30 m depth) in the analysis.

Averages for Lake Ontario surface water temperature and OC4-derived chl a were calculated at several dates in 2003 using all cloud-free pixels of Lake Ontario. In order to describe the spatial variability of the parameters at different times of the year we developed empirical variograms in S-PLUS (Version 8.0,Tibco Software). Variograms have been used to evaluate the spatial variation of satellite detected chl a in two lakes in Scotland (Hedger et al., 2001).

We extracted 1 km pixel data for the two parameters for four Lake Ontario images collected on June 15, July 14, August 18, and September 26, 2003. Variograms are graphical representations of how the variance of a parameter changes with distance and are used in the “kriging” interpolation method. Generally sites closer to each other are more similar and therefore show less variance than sites further apart. Variance (gamma) increases within a range of distance until it levels off at a “sill” value. An appropriate continuous mathematical model was selected using S-PLUS.

Results

Comparison of shipboard and satellite measurements during the LOLA surveys

The ranges in shipboard surface water temperature did not overlap for April (2.0 to 6.4°C), August (21.0 to 25.0°C), and September, 2003 (8.9 to 19.9°C). The large range in September was due to wind driven coastal upwelling. Overall, shipboard and satellite measurements of Lake Ontario surface water temperature were strongly correlated with a slope of 1 and a positive intercept of 1°C (r2 = 0.98, Figure 3).

The ranges in shipboard measured surface chl a during April and August, 2003 overlapped (0.7 to 1.9 μg l−1) but chl a was significantly higher in September 2003 (1.7 to 4.1 μg l−1). There were several data points from the three surveys where satellite OC4-based chl a measurements were very high relative to shipboard measured chl a (Figure 4). Most of these sites were nearshore (< 30 m). Two additional cases in August were just offshore of the nearshore/offshore boundary. The OC4 algorithm is expected to be least reliable in nearshore conditions where other color producing components such as suspended sediments and dissolved organic matter are prevalent. When nearshore values were omitted, the correlation of shipboard and satellite measured chl a for the three seasons improved considerably (r2 = 0.62, Figure 5).

Spatial variability of lake surface water temperature and chl a within satellite data

Variograms for monthly images from June 15 to September 26, 2003 demonstrate that pixels within a distance of 0 to 40 km are spatially autocorrelated (Figure 6A and Figure 6B). Lake Ontario surface water temperature and OC4 chl a were most variable (high gamma value for sill) on June 15. OC4-derived chl a was relatively variable on August 18 even though surface water temperature was least variable at that time.

Discussion

Compatibility of shipboard and satellite measurements

Overall, shipboard and satellite measurements were strongly correlated for surface water temperature (r2 = 0.98) and surface chl a (r2 = 0.62, offshore sites only) for the 2003 Lake Ontario LOLA sampling. These findings increase confidence in satellite data as an effective way to fill the gaps between shipboard surveys.

The correlations of shipboard and satellite measurements of both surface water temperature and chl a were weaker for individual LOLA cruises. There were several potential reasons for this discrepancy. The ranges of the two parameters during individual cruises were narrow. The two measurements were displaced in time, with the shipboard samples taken over the time span of a week, while the satellite image is collected at one time. Features such as upwelling can change dramatically day to day and could likely limit the correlation of the two measurements. Due to the difference in timing, the shipboard surveys and satellite images cannot be directly compared to confirm spatial patterns within the satellite data.

Shipboard and satellite measurements of chl a also differ in that the shipboard measurement integrated the upper 20 m of the water column, while the satellite measurement used surface color. As noted previously, the OC4 algorithm was developed for cases where chl a is the sole color producing agent, a condition that often does not apply in lakes. This analysis confirms that the OC4 algorithm is not reliable in nearshore regions where other color producing agents including sediment and dissolved organic matter are present, leading to anomolously high satellite estimates. Reliable measurements of nearshore chl a depends on the separation of the components of lake color for Lake Ontario as has been done for Lake Michigan. However, the OC4 algorithm's relative effectiveness in offshore regions indicates great potential for this tool.

A key success in the comparison was that both shipboard and satellite methods were effective at picking up an increase in offshore chl a from August to September. This finding suggests that satellite data can provide at least a qualitative tracking of chl a between sampling cruises.

Filling the gaps between the LOLA 2003 cruises

The comparability of the shipboard and satellite measurements encourages a closer use of satellite imagery to interpolate between the sample cruises. Satellite images collected between the LOLA cruises illustrate the development of common lake-scale seasonal features (Figure 7). During the May LOLA sampling, surface water temperature was cold, ranging from 2.0 to 6.4°C. By June 14, the thermal bar was evident as shallow coastal water warmed earlier than offshore regions. The coastal water was also more strongly colored than water offshore. By July 8, an upwelling event in the northwest region was evident as cool surface water temperatures. By the time of LOLA sampling in August, surface water temperatures were warm. At this time, a narrow coastal band of colored water was evident. By the time of the September LOLA cruise, the lake had cooled considerably and offshore chl a had increased.

A more quantitative assessment of the seasonal trends in surface water temperature and chl a tracks overall averages calculated from all cloud-free pixels for Lake Ontario. The GLSEA analysis of surface water temperature for Lake Ontario includes a peak at 23.64°C on August 18, 2003, close in time to the August shipboard surveys (Figure 8). A similar approach can be used for seasonal trends in chl a. This analysis suggests that offshore chl a in Lake Ontario had a bimodal pattern in 2003 with peaks on June 15 and September 10, 2003 (Figure 9). The Fall peak was close in time to the September shipboard survey. Lowest chl a values occurred on July 30.

The variograms confirm that surface water temperature and chl a exhibit the highest spatial variability in Lake Ontario in mid-June. This is the time in which the thermal bar develops, isolating relatively warm coastal water from cold offshore water. It also coincides with the first peak in chl a and suggests that phytoplankton respond to periods of high variability.

Conclusions

The OC4 algorithm developed for use in oceans was able to pick up seasonal changes in chl a concentration in the offshore region of Lake Ontario. This success bodes well for using satellite imagery for tracking the seasonality of physical and biological changes in offshore waters in the months from April until October. However, as expected the algorithm was not reliable in the nearshore waters of depth < 30 m. Several factors contribute to this failure included turbidity from rivers and the resuspension of bottom sediments. Accurate chl a measurements from satellite images of Lake Ontario require the future development of a Lake Ontario specific algorithm that considers the contribution of all color producing agents including sediment, organic matter, and calcium carbonate. Several investigators are making significant progress towards this goal.

These findings do encourage increased coordination of lake wide sampling programs with the extensive long-term collection of available satellite data for Lake Ontario. Resources for shipboard field studies are increasingly limited. The information from these studies can greatly benefit from being placed in the context of satellite imagery. The seasonal changes in phytoplankton concentrations are dynamic, rapidly respond to the physical structure of the water column, and are easily missed. Annual phytoplankton production in the offshore region of Lake Ontario is very low and needs to be closely watched towards understanding the current carrying capacity of the lake in support of the ecosystem and its important fisheries.

Acknowledgements

I thank the crews of the USEPA R/V Lake Guardian and CCGS Limnos for collecting the shipboard data. I also acknowledge the SeaDAS Development Group at NASA GSFC for the use of SeaDAS software and the Ocean Color Group for providing access to MODIS data. I acknowledge the leadership of Fred Luckey and Jack Kelly of the USEPA, and Vi Richardson of Environment Canada in the LOLA program and Mohi Munawar's work towards publishing and presenting our findings. Ed Mills, Lars Rudstam, and Kristen Holeck encouraged me to pursue analysis of the satellite imagery. This work was funded by EPA Grant CR-83209001 to Cornell University. This is contribution #266 of the Cornell Biological Field Station.

References

Bukata, R. P., Jerome, J. H., Kondratyev, K. Y. and Pozdnyakov, D. V.
1991
.
Estimation of organic and inorganic matter in inland waters: optical cross sections of Lakes Ontario and Ladoga
.
J. Great Lakes Res.
,
17
(
4
):
461
469
.
El-Shaarawi, A. H. and Kwiatkowski, R. E.
1977
.
A model to describe the inherent spatial and temporal variability of parameters in 1974
.
J. Great Lakes Res.
,
3
:
177
183
.
Hedger, R. D., Atkinson, P. M. and Malthus, T. J.
2001
.
Optimizing sampling strategies for estimating mean water quality in lakes using geostatistical techniques with remote sensing
.
Lakes and Reservoirs: Research and Management
,
6
:
279
288
.
Holeck, K. T., Watkins, J. M., Mills, E. L., Johannsson, O. E., Millard, E. S., Richardson, V. and Bowen, K.
2008
.
Spatial and long-term temporal assessment of Lake Ontario nutrients, chlorophyll a, and zooplankton
.
Aquatic Ecosystem Health and Management
,
11
(
4
):
377
391
.
Hook, T. O., Rutherford, E. S., Brines, S. J., Schwab, D. J. and McCormick, M. J.
2004
.
Relationship between surface temperature and steelhead distributions in Lake Michigan
.
North Am. J. Fisheries Management
,
24
(
1
):
211
221
.
Kwiatkowski, R. E.
1982
.
Trends in Lake Ontario Surveillance parameters, 1974–1980
.
J. Great Lakes Res.
,
8
(
4
):
648
659
.
Kwiatkowski, R. E. and El-Shaarawi, A. H.
1977
.
Physico-chemical surveillance data obtained for Lake Ontario, 1974 and their relationship to chlorophyll a
.
J. Great Lakes Res.
,
3
(
1–2
):
132
143
.
Millard, E. S., Johannsson, O. E., Neilson, M. A. and El-Shaarawi, A. H.
2003
. “
Long-term, seasonal, and spatial trends in nutrients, chlorophyll a and light attenuation in Lake Ontario
”. In
State of Lake Ontario (SOLO)-Past, Present, and Future, pp. 97–132.
, Edited by: Munawar, M.
Burlington, ON
:
Ecovision World Monograph Series, Aquatic Ecosystem Health and Management Society
.
Morel, A. and Prieur, L.
1977
.
Analysis of variations in ocean colour
.
Limnology and Oceanography
,
22
:
709
722
.
O'Reilly, J. E., Maritorena, S., Mitchell, B. G., Siegel, D. A., Carder, K. L. and Garver, S. A.
1998
.
Ocean color chlorophyll algorithms for SeaWiFS
.
J. of Geophys. Res.
,
103
(
C11
):
2493
2495
.
Plattner, S., Mason, D. M., Leshkevich, G. A., Schwab, D. J. and Rutherford, E. S.
2006
.
Classifying and forecasting coastal upwellings in Lake Michigan using satellite derived temperature images and buoy data
.
J. Great Lakes Res.
,
32
(
1
):
63
76
.
Pozdnyakov, D., Schuchman, R., Korosov, A. and Hatt, C.
2005
.
Operational algorithm for the retrieval of water quality in the Great Lakes
.
Remote Sensing of the Environment.
,
97
(
3
):
352
370
.
Schneider, K., Assel, R. A. and Croley, T. E.
1993
.
Normal water temperature and ice cover of the Laurentian Great Lakes: A computer animation, data base, and analysis tool
,
Ann Arbor, MI
:
NOAA Great Lakes Env. Res. Lab.
.
NOAA Tech. Memo. ERL GLERL-81
Schuchman, R., Korosov, A., Hatt, C., Pozdnyakov, D., Means, J. and Meadows, G.
2006
.
Verification and application of a bio-optical algorithm for Lake Michigan using SEAWIFS: a 7-year inter-annual analysis
.
J. Great Lakes Res.
,
32
(
2
):
258
279
.
Schwab, D. J., Leshkevich, G. A. and Muhr, G. C.
1999
.
Automated mapping of surface water temperature in the Great Lakes
.
J. Great Lakes Res.
,
25
(
3
):
468
481
.
Thomann, R. V., Winfield, R. P. and Szumski, D. S.
1977
.
Estimated responses of Lake Ontario phytoplankton biomass to varying nutrient levels
.
J. Great Lakes Res.
,
31
(
1–2
):
123
131
.
Ullman, D., Brown, J., Cornillon, P. and Mavor, T.
1998
.
Surface temperature fronts in the Great Lakes
.
J. Great Lakes Res.
,
24
(
4
):
753
775
.
Wilson, K. A., Howell, E. T. and Jackson, D. A.
2006
.
Replacement of zebra mussels by quagga mussels in the Canadian nearshore of Lake Ontario: the importance of substrate, round goby abundance, and upwelling frequency
.
J. Great Lakes Res.
,
32
(
1
):
11
28
.