Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images

Transcription

1 Water Resour Manage (2011) 25: DOI /s Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images Xian Guan Jonathan Li William G. Booty Received: 31 May 2010 / Accepted: 7 February 2011 / Published online: 2 March 2011 Springer Science+Business Media B.V Abstract This study focuses on utilizing satellite remote sensing to monitor the water clarity of Lake Simcoe, Ontario, Canada, which has been suffering from the overload of phosphorus (TP) and therefore eutrophication for decades. The dataset includes 22 cloud-free Landsat-5 Thematic Mapper (TM) images, as well as the nearly simultaneous in-situ observations from 15 stations on the lake. Compared to the general model used to estimate the Secchi Disk Transparency (SDT), a parameter for water clarity measurements, an improved model is developed, from the TM images. The results of these estimations are validated using the in-situ data by linear regression, and the accuracies are measured by the coefficient of determination R 2. The results reveal an indication of high model fit between the majority of SDT predictions and the in-situ observations. Also, the improved SDT model provides higher prediction accuracies than the general one when applied to 68.2% (15 out of 22) of the images. The estimated clarity maps indicate that the turbid water is normally distributed at the nearshore areas and the northeastern region. Meanwhile, the southwestern lake has much clearer water than the other regions. In addition, the southern bay has always been suffering from a serious water quality problem even till now. The water clarity of Lake Simcoe shows strong seasonal patterns, that it is at its worst in August and September annually and much better in the other sampling seasons. On an overall scale, the lake water clarity kept relatively stable from 1987 until the fall of 1992, followed by a gradual increase until 2000, after a slight decrease, and lastly stayed consistent until the summer of Keywords Water clarity Landsat-5 TM imagery Lake Simcoe Monitoring X. Guan J. Li (B) Department of Geography and Environmental Management, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada junli@uwaterloo.ca W. G. Booty Aquatic Ecosystem Management Research Division, National Water Research Institute, Environment Canada, Burlington, Ontario L7R 4A6, Canada

2 2016 X. Guan et al. 1 Introduction Clean and adequate fresh water is essential to the welfare of human beings, as the water quality directly affects natural ecosystems, human health, and economic activities. Moreover, human activities directly affect water quality. The eutrophication problem caused by the overload of nutrients strongly threatens the lake ecosystem and therefore people s lives. Concentrated pollutants entering surface waters from specific locations and diffuse pollutants generated from local or small-scale activities add large quantities of nutrients, pathogens, and toxins to the lakes. A Canadian government report in 2001 stated that human activity resulted in more than 12 thousand tons of phosphorus (P) and 303 thousand tons of nitrogen (N) entering Canadian fresh, ground, and coastal waters in It also reported that municipal sewage, which adds approximately 5.6 thousand tons P, is the largest point source, while the discharge of industrial wastewater is also a considerable source (Chambers et al. 2001). These factors, together with the heightened concentration of suspended and dissolved matter caused by humans, lead to a rapid transition to a higher trophic level, which entails algal bloom events, anoxia, and eventually a dramatic deterioration of water quality (Henderson-Sellers and Markland 1987; Kondratyev et al. 1998). Decomposition of algal blooms can lead to foul odors and oxygen depletion, which can in turn, lead to fish deaths (Carpenter et al. 1998;Smith1998). These problems that inland lakes are suffering from make protecting and monitoring lake water quality a major concern for many local, provincial, and federal government agencies. Many lake-water monitoring programs have been set up to keep records of the change in water quality parameters. These programs usually include on-site sampling and laboratory analysis. Three water quality variables, including total phosphorus (TP), chlorophyll a (chl-a), and Secchi disk transparency (SDT) have been widely used by management agencies and organizations to indicate the trophic state of lake water. Generally, this ground-based monitoring achieves high accuracy of lake water quality measurements. However, the expense, time, and sampling frequency make ground-based monitoring impractical to be applied to large areas. The resolution is a compromise between spatial and temporal details (Kloiber et al. 2002a, b). Although it has not been officially admitted as an adopted policy tool, remote sensing techniques have been proven to be useful for monitoring water quality for decades. The use of this satellite-based technology is a cost-effective way to gather the needed information for regional water quality assessments in lake-rich countries such as Canada. Over the past years, various studies have been conducted to investigate the effectiveness of multispectral satellite images for monitoring water quality. For example, an attempt was made by Choubey (1994) to quantify the relationship between the variation in Indian Remote Sensing Satellite-1A Linear Imaging Self-Scanning System (IRS-1A LISS) radiance data and field measured change Secchi disk depth (SDD) for mapping water quality in reservoirs. Cox et al. (1998) investigates the potential of Landsat Thematic Mapper (TM) to assess water quality in the 11 reservoirs of the Catawba River basin in North Carolina, the United States. Giardino et al. (2001) used Landsat TM images to map some biophysical parameters (e.g., chl-a concentration, SDD and water surface temperature) in Lake Iseo, Italy. Mendoza et al. (2006) used NOAA-AVHRR, Landsat Multispectral Scanner (MSS) and TM, SPOT images, as well as black-and-white

3 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2017 aerial photographs to monitor changes in the water-surface area occupied by the Cuitzeo Lake, Mexico, during a period of Wu et al. (2009) explored the seasonal dynamics of water clarity and analyzed how water level, wind velocity and total precipitation influence this dynamics in Lake Dahuchi, China. Their study demonstrated a seasonal pattern of water clarity using the SDD recorded in the field and derived from Moderate Resolution Imaging Spectroradiometer (MODIS) images together. Kiage and Walker (2009) used four summer time-series ( ) of Normalized Difference Vegetation Index (NDVI) derived from MODIS imagery to investigate duckweed blooms and other floating vegetation in Lake Maracaibo, Venezuela. Giardino et al. (2007) investigated a procedure to map chl-a and tripton concentrations in the largest Italian lake, Lake Garda, from hyperspectral satellite data (with a 30 m resolution and more than 200 spectral channels) based on forward and inverse bio-optical modelling. Giardino et al. (2010) used the MEdium Resolution Imaging Spectrometer (MERIS) and MODIS images for monitoring water quality of Lake Trasimeno in central Italy, having compared the retrieved chl-a concentration, SDD and surface water temperature with the timeseries of the in-situ data. Among a variety of earth observation satellites, the relatively low cost, temporal coverage, spatial resolution, and data availability of the Landsat-5 TM imagery make it particularly useful for assessment of inland lakes (Kloiber et al. 2002a, b). Several studies have demonstrated a strong relationship between Landsat MSS or TM data and ground observations of water clarity and chl-a (e.g., Brown et al. 1977; Lillesand et al. 1983; Lathrop and Lillesand 1986; Lathrop 1992; Cox et al. 1998). In 1999, with the launch of Landsat-7, data distribution was returned to the public sector, and the cost of data acquisition dropped significantly. In particular, TM data is available free of charge. Along with today s powerful desktop computers and sophisticated software, processing and analysis of satellite imagery has become relatively inexpensive and easy to perform. It is also noticeable that several studies of water quality estimation by remote sensing focus on the algorithms for retrieving water quality themselves, while ignoring the lake water characteristics. After all, remote sensing serves as a tool or supplement for water quality monitoring in order to reduce human s burden, whether it will be treated as a policy or not. At the moment, few investigations are aiming at the long term trend in water quality monitoring of inland lakes, although many scholars choose to select one specific study area and focus on the water pollution problem itself. This study is motivated by the water quality deterioration, the weakness of conventional water quality programs, as well as the reduction of stations on Lake Simcoe. Four overall objectives are determined here: (1) to utilise and develop models for estimating SDT using the Landsat-5 TM images; (2) to establish empirical relationships between SDT and spectral information using linear regression analysis, (3) to identify both temporal and spatial water quality patterns of the study area and to map the excessive pollution regions in reference to the inland lake classification criteria, (4) to graph the overall water quality trends of the study area for a period of 22 years, from 1987 to The remainder of this paper is organized as follows. In Section 2 we describe the methodology designed for monitoring lake water clarity using multispectral satellite imagery. In Section 3 we demonstrate and analyze the results of the experiment and

4 2018 X. Guan et al. validate the satellite-derived information by in-situ observations. Finally, in Section 4 we present our conclusions and potential future work. 2 Methodology 2.1 Study Area Lake Simcoe, located in the southern part of Ontario, is the 12th largest lake in the province (see Fig. 1). It has a surface area of 722 km 2 in total, a maximum depth of 41 m, length of 30 km, and width of 25 km. It consists of the main basin and two large bays, which are Cook s Bay at the south and Kempenfelt Bay on the west side (Johnson and Nicholls 1989). Great significance has been empowered to this lake, where the fishery is an important industry generating more than 200 million dollars per year (Environment Canada2009). However, water-quality problems have existed since the 1970s, when the increasing phosphorus load to this lake contributed to the recruitment failure of cold-water fish, as well as the excessive growth of aquatic macrophytes and algae (Winter et al. 2007). Great attention has been paid to Lake Simcoe in order to improve its water quality. Examples of this include the Lake Simcoe Environmental Management Strategy (LSEMS), the Lake Simcoe Region Conservation Authority (LSRCA) and Fig. 1 Location of Lake Simcoe on a Canadian Map

5 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2019 the Lake Simcoe Conservation Foundation (LSCF). As phosphorus is the main problem of Lake Simcoe, related studies have been conducted in order to lower the phosphorus concentration in this lake (e.g., Winter et al. 2005; Lu et al. 2006; Nicholls 1997). These studies contributed greatly to the management of Lake Simcoe, as they reveal that the TP loads are currently below the target of 75 metric tons/year (Winter et al. 2007). However, algae and macrophytes continue to be problems in the nearshore zone. In addition, the dissolved oxygen concentrations remain below the level (7 mg/l), which is considered to be protective of lake trout. Therefore, the lake water quality monitoring, as well as the efforts to reduce TP loads to this lake still need to be continued (Winter et al. 2007). The previous studies imply that the water-quality monitoring of Lake Simcoe still relies on conventional approaches. However, there has been a reduction of ground stations on this lake since Now only 11 stations are left for the SDT sampling (seefig. 2). This, coupled with limitations of the traditional water-quality monitoring programs, makes a supplementary or an alternative method necessary for the waterquality monitoring of Lake Simcoe. In this study, a remote sensing technique plays the role of a supplement to the old methodologies to estimate the water clarity from the TM images. Among the three water quality variables mentioned by Kloiber et al. (2002a, b), phosphorus is the one with least optical characteristics. Fig. 2 Location of the sampling stations in the Lake Simcoe area

6 2020 X. Guan et al. Moreover, the chl-a in Lake Simcoe was also estimated in the experimental stage. However, as the chl-a sampling points were not as many as the sampling points for SDT, the estimated chl-a did not show a reliable correlation with the in-situ observations. Therefore, SDT was finally determined as the indicator of eutrophication effects in Lake Simcoe. This study aims at demonstrating the spatial distribution and the trends of water clarity change in Lake Simcoe using Landsat-5 TM images. In 2008, the Canadian government opened a 30-million Canadian dollar fund for the Lake Simcoe clean-up, to reduce the environmental stresses (Environment News Service 2008). Therefore, the results of this study are of great value and significance to the whole environmental management of Lake Simcoe. 2.2 Satellite Images and Lake Reference Data This project has had data support from Environment Canada and the Ontario Ministry of Environment (MOE). With regard to the spatial coverage of Lake Simcoe, Landsat-5 TM imagery is considered well suited to provide broad spatial coverage and multi-year trends (Kloiber et al. 2002a, b). A total of 5 cloud-free and well-calibrated Landsat-5 TM images were provided by the Canada Centre for Inland Waters (CCIW), and 17 original Landsat-5 TM images were obtained from the United States Geological Survey (USGS) website free of charge. These images are from April 1987 to September The selection of data is based on the principle that the cloud coverage of any TM scene is less than 10%. These satellite images are uniformly projected to Zone 17N in the Universal Transverse Mercator (UTM) coordinate system, with the datum of the World Geodetic System 1984 (WGS84). It has been demonstrated that the satellite data and the ground measurement data should be taken almost simultaneously, for the reason that water clarity usually does not exhibit large and rapid fluctuations in a given lake (Brezonik et al. 2007). Very fortunately, 22 of the in-situ measurements are found very close to the dates of the available Landsat-5 TM images of Lake Simcoe. The in-situ SDT data were provided by the MOE. The MOE has been measuring a large number of water quality parameters of Lake Simcoe since There are a total of 15 ground sampling stations distributed on the lake (see Fig. 2). Sampling was conducted routinely every two weeks, during the ice-free seasons through each of those years. The water was analyzed in the laboratory to extract concentrations of different water quality parameters. Among the 22 images selected for this study, 16 of them are within a 5-day overlap, and 3 of them are within a 5 10 day overlap. Only three pairs of data have gaps longer than 10 days. The reason for doing this is that these images still have very positive correlations with the in-situ data. Considering that a trend analysis is one of the objectives of this study, they are kept in the data list on the condition that they do not affect the accuracy of the results (see Table 1). 2.3 Data Processing Both the satellite data and the in-situ data need to be pre-processed. Firstly, all the image data were re-sampled to the 30 m pixel size. After the geo-registration, water-only images were created by clipping the original images using the watershed boundary. The ATCOR-2 module was utilised for the atmospheric effect removal in ENVI software. A point layer was created and then converted to areas-of-interest

7 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2021 Table 1 Available stations for each pair of data Image date In-situ date No. of stations Data gap (day) (AOI) from the coordinates of in-situ data. The Digital Number (DN) values of the stations on the images were extracted through these AOIs and output into ASCII files, and then these values were compared with the in-situ SDT data. Different models were used for the SDT estimations by linear regression. Once the correlation coefficient R 2 was considered acceptable (more than 0.5 in this study specifically), a regression equation would be calculated and this equation can be used for the whole image and generates a clarity map. Finally the result maps needed to be smoothed, to remove the random or irregular pixels that seem to be discordant with the surrounding pixels. Therefore, a 5 5 median filter was used to make the maps more coherent and consistent. 2.4 Model Development and Validation SDT Estimation Using the General Equation After the data pre-processing and the creation of AOIs, the DN values of the station points were extracted from the images. Then these DN values of different bands were combined and calculated based on Eq. 1, which has been shown in other studies to be a general predictive equation for water clarity of inland lakes. This equation was regarded as a dependable predictor of SDT by Kloiber et al. (2002a, b) and it has been used in many studies (Kloiber et al. 2002a, b; Nelsonetal.2003; Olmanson et al. 2008). Then, the estimated SDT are compared with the ground measurements. ln (SDT) = a(tm1/tm3) + b(tm1) + c (1) TM1 and TM3 are the brightness values of blue and red bands, respectively. ln(sdt) stands for the natural logarithm of the SDT. a, b,andc are the coefficients fitting the

8 2022 X. Guan et al. in-situ data by the regression analysis. These coefficients can be used to estimate the SDT values for the entire lake in the same Landsat image, once they fit the equation well (Brezonik et al. 2002). In this study specifically, however, the SDT value itself, rather than its natural logarithm, has better correlation with the satellite brightness values. Therefore, the in-situ SDT itself is compared with the estimated SDT from satellite images instead, and this relation can be described in Eq. 2, wherethesdt E1 refers to the estimated SDT from TM images. SDT E1 = a(tm1/tm3) + b(tm1) + c (2) SDT Estimation Using the Improved Equation A preliminary comparison is made between the reflectance values extracted from TM data and the in-situ measured SDT values. It shows that sampling points with low SDT are frequently accompanied with a high reflectance of TM3. Moreover, the reflectance values of TM3 alter following the change of SDT regularly, whereas the brightness values of TM1 do not show such a related trend. Therefore, TM3 is regarded as a better predictor than TM1 for the SDT estimation in this study. In Eq. 3, SDT E2 refers to the estimated SDT from TM images. TM3 reflectance values are subsequently employed to replace the TM1 reflectance in Eq. 2, in order to improve the prediction accuracies of SDT. SDT E2 = a(tm1/tm3) + b(tm3) + c (3) Regression Analysis After utilizing different models to estimate SDT, the best-fit equations, describing the relationship between each of the estimated data and the corresponding in-situ data, are determined by regression analysis. Linear regression analysis, modeled by a least-squares function, is used to give information on the relationship between a dependent variable and one or more independent variables to the extent that information is contained in the data (Kaw and Kalu 2008). The correlation between two variables reflects the degree to which the variables are related and the Pearson s correlation is a commonly used measure of correlation. These estimations are measured by the correlation coefficient R 2. The maximum positive correlation is If the two standardized variables covary positively and perfectly, or vary oppositely and perfectly, then the correlations will both equal to 1.00 (Rummel 1976). When computed in a sample, it is designated by the letter r and is sometimes called Pearson r. Each of these 22 images was compared with the corresponding in-situ measured data. Two couples of variables were compared by correlating to the calibration data, one couple is the reflectance value of TM1 and TM1/TM3, while the other is the reflectance value of TM3 and TM1/TM3. The couple with higher coefficients R 2 were utilized as the independent variables for linear regression analysis, and the in-situ SDT was employed as the dependent variable. The resulting regression equations were applied to the entire lake area of those images in order to map and simulate the SDT distribution of Lake Simcoe.

9 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images Results and Discussion 3.1 Evaluation of Lake Clarity Estimation SDT Evaluation for Individual Datasets In this study, the SDT calculated from TM images were compared to the in-situ observations in three ways, which include: (1) the individual comparison, (2) the comparison for year 2008, and (3) the comparison of the entire dataset. First, we compared the satellite-derived SDT values to the in-situ observations individually. Regression equations relating each TM image to the corresponding insitu SDT are presented in Table 2, which includes the R 2, and standard deviation of estimation (SEE) as well. The results indicate that the estimations using both models have fair agreements with the in-situ SDT data. The generally used SDT model is related to the ground measurements with the R 2 ranging from to 0.836, the average R 2 of and the average SEE of On the other hand, the improved model in this study has the R 2 ranging from to 0.861, the average R 2 of 0.622, and the average SEE of In other words, the composed TM3 and TM1/TM3 is better correlated to the calibration data than the combination of TM1 and TM1/TM3 in general. Additionally, 68.2% (15 out of 22 images) of these comparisons have higher R 2 by using the combination of TM3 and TM1/TM3, while only seven images have higher Table 2 Regression equations for prediction of SDT from TM images Date R 2 SEE R 2 SEE Regression Equations Variables (TM1,TM1/TM3) (TM3, TM1/TM3) B1=TM1; B3=TM3 1987/04/ (B1/B3)+0.241B /05/ (B1/B3)+0.053B /05/ (B1/B3) 0.652B /06/ (B1/B3) 0.140B /08/ (B1/B3)+0.096B /07/ (B1/B3)+0.025B /09/ (B1/B3) 0.199B /07/ (B1/B3) 0.554B /08/ (B1/B3)+0.088B /05/ (B1/B3) 0.587B /07/ (B1/B3) 0.634B /08/ (B1/B3) 0.254B /10/ (B1/B3)+0.103B /07/ (B1/B3) 0.621B /08/ (B1/B3)+0.036B /05/ (B1/B3) 0.192B /06/ (B1/B3)+0.231B /07/ (B1/B3) 0.416B /07/ (B1/B3) 0.919B /08/ (B1/B3) 0.477B /09/ (B1/B3) 0.328B /09/ (B1/B3)+0.218B

10 2024 X. Guan et al. R 2 using the general predictive variables for SDT estimation. Furthermore, a total of 5 comparisons of the commonly used combination highlighted in Table 2, show poor correlations with the calibration data, while the improved one in this study has a significantly better correlation. These comparisons show that the improved model has a better agreement with the calibration data, which indicates that the TM3 and TM1/TM3 combination can greatly improve the accuracies of SDT estimations from Landsat-5 TM images in this study. In Table 2, the estimations of three datasets (1987/04/19, 1989/06/11 and 2008/06/15) are not accurate enough that the R 2 is between 0.4 and 0.5. There are even two comparisons (1995/07/30 and 2008/07/01) with the R 2 lower than Since all the Landsat images have been corrected to remove the atmospheric effects, the possible reason to explain these poor agreements is the seasonal influence. Olmanson et al. (2008) and Stadelmann et al. (2001) pointed out in their studies of Minnesota Lakes that the late-summer period is the best index period for the remote sensing of water clarity. Two reasons are concluded for this perspective: one is that most lakes have their lowest level of water clarity in this period; the other is that the lake water clarity has the minimum short-term variability in the late-summer season (Olmanson et al. 2008). The lake area has a minimum value of precipitation in the summer, therefore, the status of lake water is relatively stable, and the constituents of lake water are possibly less changed during this time. The employed satellite data have a seasonal variability, which include the early spring, late spring, early summer, late summer and early fall. In this study, the late-summer is defined as August and September. Table 2 shows that the vast majority of comparisons in the late-summer period have very close relationships, whereas those which are poorly related data are mostly collected in the spring or early summer. For Lake Simcoe, as the early spring is the period right after the snow melting, it is understandable that there may be an internal mixing in the lake. Also, locations in a lake can vary due to wind mixing fairly rapidly. All of these factors can contribute to the instability of lake water and the low agreements between the satellite-estimated SDT and the in-situ SDT SDT Evaluation for 2008 The available data in this study are more sufficient in 2008 than the other years. A total of 6 images and the corresponding in-situ data were collected in 2008, from the very early summer to fall. This section shows the comparison between these two datasets. The scatter plot in Fig. 3a shows that the satellite estimated SDT and the insitu SDT have a fair agreement with the R 2 of Figure 3b shows a similar comparison, which only uses the late-summer data in The linear relationship is much closer than the previous one and the R 2 increases to The late-summer comparison has a more stable correlation in that most of the samples are close to the reference line. Therefore, it can support the suggestion that the late-summer can be regarded as a better period for the lake water clarity prediction in However, compared with Fig. 4, which is the scatter plot of in-situ SDT against the estimated SDT for late-summer period in the entire database, the data in 2008 still have a better agreement. The gap between these two results indicates that the close dates can lead to similar water clarity conditions, as the lake water quality does not show big differences during a short period.

11 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2025 Fig. 3 (a) Comparison between in-situ SDT and estimated SDT in 2008, (b) comparison between the in-situ and the estimated SDT in the late-summer 2008 (June October) SDT Evaluation for Entire Database This section shows the comparison results for the estimated SDT and in-situ SDT of the entire database. Firstly, all of the data of the late-summer period are employed in Fig. 4a, and then the dataset is extended to the entire database in Fig. 4b. Both pairs of the datasets have good agreements, with the R 2 of and 0.712, respectively. Therefore, it can be concluded that the Landsat-5 TM images are reliable enough to predict the water clarity in Lake Simcoe. Also, the advantage of sampling in the latesummer is demonstrated again, as the comparison in that period has a higher R 2 than the others. In the late-summer period, the lake usually has its lowest level of water clarity (Olmanson et al. 2008). As a result, there is more light and signals reflected to the remote sensor. Therefore, the relationships between the reflectance values and in-situ data become stronger. 3.2 Spatial and Temporal Analysis The clarity map estimated by the regression equations can indicate the spatial distribution of the turbid areas. Figure 5 is the blue-and-white map displaying the Fig. 4 (a) Comparison between the in-situ SDT and the estimated SDT in the late-summer period (June October), (b) Comparison between in-situ and estimated SDT for entire database

12 2026 X. Guan et al. Fig. 5 Map of the estimated SDT of August 17, 1991 estimated SDT of In this map, the lighter the colour is, the clearer the water is. This means that the dark blue area has the lowest SDT and therefore the most turbid water. It is obvious that the central part of Lake Simcoe water is clearer than the nearshore area, and the turbid area is concentrated at the northeast shore and the southern nearshore area. In the scale of the whole Lake Simcoe, the northeast part of this lake has lower water clarity than the southwest part. This indicates that the northeast lake suffers from a more severe water clarity problem than the southwest part. Also, the water quality gradually decreases following the direction from northeast towards the southwest. Figure 6 presents a line chart showing a trend of monthly average SDT, monthly maximum and minimum SDT dating from 1987 to 2008, based on the fix-point in-situ measurement data. During this period, Lake Simcoe had an overall monthly average SDT of 4.05 m, a maximum monthly average SDT of 8.79 m, in May 2005 and a minimum monthly average of 0.4 m, in April According to the trend of the monthly averaged SDT, on an overall scale, there are three obvious periods of change: (1) the lake water clarity kept relatively stable from 1987 until fall 1992, (2) followed by a gradual rising up until 2000, (3) after a little decreasing, lastly stayed consistent until the summer of As three of the ground stations (H3, H4 and H6) were not in use during 1993 to 2000, it is possibly one explanation of the decreasing in the third stage. These three areas were always turbid, and that would dramatically lower the average SDT of the entire lake, considering there were only 15 fixed stations. On the other side, the maximum average SDT

13 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2027 Fig. 6 Monthly SDT trend from 1987 to 2008 shows a continuously increasing trend, which indicates the strategies executed by the agencies also contributed to this clarity upgrading. On the other hand, from an annual scale, Lake Simcoe appears to have a very similar water clarity trend every year. SDT is minimized in the late summer period,

14 2028 X. Guan et al. normally in August or September. Also, the SDT is relatively higher before and after these two months, which means it decreased to the lowest point and increased after that, during each of these 22 years. More specifically, Cook s Bay, located at the southernmost part of Lake Simcoe and about 60 km north of Toronto, is suffering from the most severe environmental problems. According to the in-situ measured data, stations H3, H4 and H6, which are the only three observation stations located within the Cook s Bay area, always have a very low SDT. Based on the in-situ data, it is normal for this bay (station H3, H4 and H6) to have the SDT less than 1 m, corresponding with a chl-a concentration higher than 80 μg/l or even 140 μg/l. According to Carlson s lake water quality criteria (Carlson and Simpson 1996), Cook s Bay should be classified as eutrophic, or hypereutrophic. The monthly minimum value of the green line in Fig. 6 gives an explicit explanation of the monthly SDT of the Cook s Bay area, as it is the most turbid part in the Lake Simcoe area. It can also be concluded that although the water quality of Lake Simcoe has improved a lot since the 1990s, the water quality in Cook s Bay has been continuously poor. This SDT trend of Lake Simcoe can also be examined using the concentration map. In Fig. 7, 18 SDT maps, calculated by the regression equations from the TM images are compared to exhibit the change of water clarity over these years. The results of calculation are classified into five colors using the density slice colour table. Dark red, light blue, cyan, dark blue and black mean the SDT between 0 2, 2 4, 4 8, 8 12 and m, respectively. The SDT trend presented in this way is coincident with the line chart. For instance, the averaged SDT in April 1987 was quite low on the line chart, so most of the water on Fig. 7 is classified as dark red. Figure 8 shows that there was a continuous increasing of water clarity from August 1990 to July There was a slight water clarity decrease since August 2001, while it recovered soon in Influence of Other Parameters It is noticeable that some of the estimated SDT are not well-correlated to the insitu data. Besides the errors caused by the sensor and the atmosphere, there are two reasons that may contribute to those results. The first reason is that most of the observation stations are located very close to the shore area. The eroded soil makes the nearshore water turbid as well. On the other hand, there is usually a confusion of the pixels in satellite imagery between the water and soil if the spatial resolution of the satellite images is not high enough. Sometimes even if the stations are located perfectly on the imagery, it still may turn out that the DN values of soil rather than water are extracted. Therefore, the location of the station may affect the correlations. Another reason is the chl-a and DOC in the lake water. These parameters influence not only the water quality itself, but also the absorption of light. According to the in-situ data provided by MOE, the DOC concentration stays quite stable around 4 mg/l until 1995, and it starts to increase gradually since 1996 and becomes unstable. In peak times, the DOC concentration can even achieve 10 mg/l. The average concentrations were about 2 mg/l higher during the period after 1996 than before Also, the chl-a concentration in Lake Simcoe was high during these

15 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2029 (a) (b) (c) (d) (e) (f) (g) (h) (i) Fig. 7 Estimated water clarity maps of Lake Simcoe ( ) years. Although the chl-a concentration in the main basin has been lowered a lot since 1992, it stays in a high stage around the Cook s Bay area. It is believed that the DOC and green pigment can absorb the sunlight and therefore cause the parameters to be underestimated.

16 2030 X. Guan et al. (j) (k) (l) (m) (n) (o) (p) (q) (r) Fig. 8 Estimated water clarity maps of Lake Simcoe ( ) 4 Conclusions This study focused on estimating the water clarity change of Lake Simcoe from 1987 to 2008, via the approach of comparing Landsat-5 TM images and the in-situ data, including SDT and chl-a concentration. Specifically, this research was able to: (1)

17 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2031 build the empirical relationships between the TM images and the in-situ water clarity data, so as to estimate SDT of Lake Simcoe during a period of 22 years, from 1987 to (2) Identify the turbid regions based on the clarity maps calculated from the regression equations. (3) Analyze the seasonal variation of the SDT estimations. (4) Simulate the water clarity trends of the past 22 years by the ways of both the line chart and the concentration maps. The estimated results from TM images reveal that the predicted SDT has a close agreement with the in-situ data. A total of 77.3% comparisons have good correlation coefficients. Moreover, TM3 is found to be a better predictor than TM1 for the SDT estimation; 68.2% of the comparisons show higher predictive efficiency by utilising the improved model, composed of TM3 and the TM1/TM3 ratio. The possible reasons for the comparisons with low correlation are the chl-a, as well as the DOC in Lake Simcoe, which can strongly absorb the blue and green light and therefore weaken the signals back to the satellite. It is indicated that the SDT estimations have seasonal variations, i.e., the late-summer comparisons frequently lead to high agreements. The clarity maps show that the turbid area is normally centralized at the nearshore and the northeast part of Lake Simcoe. The overall trend of SDT begins with a stable status from 1987 to fall 1992, followed by a rising up till 2000, and lastly stays consistent to the summer of As the in-situ stations on Lake Simcoe have been continuously reduced since 1995, the satellite estimated water quality parameters can play a role to supplement the traditional in-situ measurements. Nevertheless, some limitations still exist in this study and they can possibly be improved in the future. The first limitation is the seasonal variability. As the late-summer period is the best season to estimate the lake water clarity, more accurate results would be given during this period. However, almost 50% of the data in this study were collected in the spring and fall seasons. As a consequence, some of these images have poor agreements with the in-situ data. The reasons for this unsatisfactory data collection include: (1) the 16-day temporal resolution of Landsat-5 TM, (2) the demand for cloud-free images during the ice-free period, (3) the coincidence with the in-situ data, and (4) the limited free data availability on the USGS website. Consequently, the odds of qualified data are actually low. Not surprisingly, only 22 TM images were acquired during the whole 22-year period. In order to stimulate the long-period trend, the unification of seasons has to be sacrificed. In spite of this, the seasonal variation can be overcome by combining MODIS and Landsat images as the datasets. The high temporal resolution of MODIS images can guarantee an adequate latesummer data selection, while the fair spatial resolution of Landsat images can exhibit the spatial patterns of water quality parameters. Moreover, the spatial resolution of TM images is not high enough for some sampling stations, such as the stations H3, H4, and H6, located at Cook s Bay, where the shape of the lake becomes quite narrow. Therefore, this region can easily get confused with the agricultural lands surrounding it on the images. As a result, when estimated by the regression equations, the pixels within this region cannot be calculated correctly. Since Cook s Bay is an important area that is worth monitoring, owing to its consistently poor water quality, images with higher spatial resolution (e.g., SPOT) are required in the future studies. Last but not least, the improved equation for SDT estimation needs to be applied to other study areas. Although TM3 and the TM1/TM3 ratio have impressive estimation efficiency in this study, the universality of this prediction model is still

18 2032 X. Guan et al. doubtable. One feasible approach is to employ this method to other research sites of the existing studies using the general SDT prediction equation, and to compare the efficiency with those studies afterwards. Acknowledgements We are grateful to the referees for their constructive comments on this article. References Brezonik PL, Kloiber SM, Olmanson LG, Bauer ME (2002) Satellite and GIS tools to assess lake quality. Technical Report 145, Water Resources Center, University of Minnesota Brezonik PL, Olmanson LG, Bauer ME, Kloiber SM (2007) Measuring water clarity and quality in Minnesota lakes and rivers: a census-based approach using remote-sensing techniques. Cura Reporter, Water Resources Center, University of Minnesota, St Paul Brown D, Warwick R, Skaggs R (1977) Reconnaissance analysis of lake condition in east-central Minnesota. Report No Minnesota Land Management Information System, Center for Urban and Regional Affairs, University of Minnesota, Minneapolis Carlson RE, Simpson J (1996) A coordinator s guide to Volunteer Lake monitoring methods. North American Lake Management Society, Madison, Wisconsin. eutrophication_sd.htm. Accessed 26 January 2011 Carpenter SR, Caraco NF, Correll DL, Howarth RW, Sharpley AN, Smith VH (1998) Nonpoint pollution of surface waters with phosphorus and nitrogen. Ecol Appl 8: Chambers PA, Guy M, Roberts ES, Charlton MN, Kent R, Gagnon C, Grove G, Foster N (2001) Nutrients and their impact on the Canadian environment, Agriculture and Agri-Food Canada, Environment Canada, Fisheries and Oceans Canada, Health Canada and Natural Resource Canada Choubey VK (1994) Monitoring water quality in reservoirs with IRS-1A-LISS-I. Water Resour Manage 8: Cox RM, Forsythe RD, Vaughan GE, Olmsted LL (1998) Assessing water quality in the Catawba River reservoirs using Landsat Thematic Mapper satellite data. Lake Reserv Manage 14: Environment Canada (2009) Cleaning up Lake Simcoe: Part of the Government of Canada s Action Plan for Clean Water. Accessed 26 January 2011 Environment News Service (2008) Canada opens CAN$30m fund to clean popular vacation lake. Accessed 26 January 2011 Giardino C, Pepe M, Brivio PA, Ghezzi P, Zilioli E (2001) Detecting chlorophyll, Secchi disk depth and surface temperature in a sub-alpine lake using Landsat imagery. Sci Total Environ 268:19 29 Giardino C, Brando VE, Dekker AG, Strmbeck N, Candiani G (2007) Assessment of water quality in Lake Garda (Italy) using Hyperion. Remote Sens Environ 109: Giardino C, Bresciani M, Villa P, Martinelli A (2010) Application of remote sensing in water resource management: the case study of Lake Trasimeno, Italy. Water Resour Manage 24: Henderson-Sellers B, Markland HR (1987) Decaying lakes. Wiley, Chichester Johnson MG, Nicholls KH (1989) Temporal and spatial variability in sediment and phosphorus loads to Lake Simcoe, Ontario. J Great Lakes Res 15: Kaw A, Kalu E (2008) Numerical Methods with Applications, lulu.com Kiage LM, Walker ND (2009) Using NDVI from MODIS to Monitor Duckweed Bloom in Lake Maracaibo, Venezuela. Water Resour Manage 23: Kloiber SM, Brezonik PL, Olmanson LG, Bauer ME (2002a) A procedure for regional lake water clarity assessment using Landsat multispectral data. Remote Sens Environ 82:38 47 Kloiber SM, Brezonik PL, Bauer ME (2002b) Application of Landsat imagery to regional-scale assessments of lake clarity. Water Res 36: Kondratyev KY, Pozdnyakov DV, Pettersson LH (1998) Water quality remote sensing in the visible spectrum. Int J Remote Sens 19: Lathrop RG (1992) Landsat Thematic Mapper monitoring of turbid inland water quality. Photogramm Eng Remote Sens 58: Lathrop RG, Lillesand TM (1986) Utility of Thematic Mapper data to assess water quality. Photogramm Eng Remote Sens 52:

19 Monitoring Lake Simcoe Water Clarity Using Landsat-5 TM Images 2033 Lillesand TM, Johnson WL, Deuell RL, Lindstrom OM, Meisner DE (1983) Use of Landsat data to predict the trophic state of Minnesota lakes. Photogramm Eng Remote Sens 49: Lu Q, Duckett F, Nairn R, Brunton A (2006) 3-D eutrophication modeling for Lake Simcoe, Canada. In: American Geophysical Union 2006 Fall Meeting, December San Francisco, CA Mendoza ME, Bocco G, Bravo M, Granados EL, Osterkamp WR (2006) Predicting water-surface fluctuation of continental lakes: a RS and GIS based approach in Central Mexico. Water Resour Manage 20: Nelson SAC, Soranno PA, Cheruvelil KS, Batzli SA, Skole DL (2003) Regional assessment of lake water clarity using satellite remote sensing. Residence time in lakes: science, management, education. J Limnol 62(1):27 32 Nicholls KH (1997) A limnogolical basis for a Lake Simcoe phosphorus loading objective. Lake Reserv Manage 13: Olmanson LG, Bauer ME, Brezonik PL (2008) A 20-year Landsat water clarity census of Minnesota s 10,000 lakes. Remote Sens Environ 112: RummelBrezonik PL, Kloiber S RJ (1976) Understanding correlation. powerkills/uc.htm. Accessed 26 January 2011 Smith VH (1998) Limitations, and frontiers in ecosystem ecology. Springer, New York Stadelmann TH, Brezonik PL, Kloiber S (2001) Seasonal patterns of chlorophyll a and Secchi Disk Transparency in lakes of East-Central Minnesota: implications for design of ground- and satellite-based monitoring programs. Journal of Lake and Reservoir Management 17: Winter JG, Walters M, Willox C (2005) Scientifically derived phosphorus loading objective and adaptive watershed management for Lake Simcoe, Canada. In: 53rd Joint Meeting of American Geophysical Union, May, Louisiana Winter JG, Eimers MC, Dillon PJ, Scott LD, Scheider WA, Willox CC (2007) Phosphorus inputs to Lake Simcoe from 1990 to 2003: declines in tributary loads and observations on lake water quality. J Great Lakes Res 33: Wu G, de Leeuw J, Liu Y (2009) Understanding seasonal water clarity dynamics of Lake Dahuchi from in situ and remote sensing data. Water Resour Manage 23: