The Water Resources Management Division of the Department of Environment and Conservation performs routine water sampling to measure the physical and chemical parameters of select water bodies in Newfoundland and Labrador. Ionic concentration parameter measurement is performed during routine water sampling to complement some of the key indicator parameters measured in real time at these select water bodies. The collection, laboratory analysis and measurement of water samples is a time consuming process. Some of the common conducting ions measured during routine sampling are sodium, calcium, chloride and sulphate. These conducting ions can be estimated using continuously measured specific conductance after observing the effect of flow. The estimated measurement will help identify whether any local stressors are affecting the quality of water at a given point in time and hence save time and resources in performing routine sampling. This can also be applied in remote locations where routine sampling is not feasible. This paper compares four water bodies on the island part of Newfoundland and Labrador and estimates the ionic concentration using continuously measured specific conductance.

INTRODUCTION

The Water Resources Management Division (WRMD) of the Department of Environment and Conservation (ENVC) of the Province of Newfoundland and Labrador (NL) has established a near real time water quality (RTWQ) monitoring network throughout the province where key indicator water quality data are collected continuously. This water quality data can be used to monitor the health of aquatic ecosystems, establish trends and determine when specific water quality events occur. The information obtained from the network is needed by the WRMD to implement its mandate and allows managers and policy makers to make informed decisions on the early warning of adverse water quality events. The general public, policy makers, government agencies and private sectors greatly benefit from such timely data and information.

The water quality parameters measured through the real time monitoring system are water temperature, pH, dissolved oxygen (DO), specific conductance (SC) and turbidity. Percent saturation and total dissolved solids are two additional parameters calculated from DO and SC. These key indicator parameters provide significant information to better understand the water quality of a particular water body. Routine water quality grab sampling is also performed in these select water bodies. The grab sampling is part of the Quality Assurance/Quality Control (QA/QC) protocol for NL RTWQ program which is used to compare the accuracy of the real time parameters (pH, SC and turbidity) against those measured at an accredited laboratory. Some additional parameters analysed in the laboratory include sodium, calcium, chloride, and sulphate. The collection, shipping and analysis of grab samples in the laboratory require a significant amount of time to measure the ionic concentration of the sample contents and provide the results. This time lag can be greatly reduced if some of these ionic concentration parameters can be estimated in real time based on regression models using real time SC readings.

Conductivity reveals the presence of dissolved materials in water (Williams 1966; Thomas 1986) consisting of metallic ions, organic and inorganic materials. Studies performed by Lind (1970) had shown that it is possible to estimate the concentration of individual ionic constituents from continuous measurements of SC since the ionic composition remains constant even though the concentration may vary as a result of dilution. Miller et al. (1988) have shown that estimation of water quality constituents can be applied in locations where continuous water quality measurements have been discontinued and also as a part of the QA/QC program to verify chemical analyses of discrete water samples. Stevens et al. (1995) measured the relationship between electrical conductivity and nutrient content of animal slurries using correlation and linear regression analyses.

A detailed study by Granato & Smith (1999) in Northborough, Massachusetts applied regression analysis in their study to measure constituent calcium, sodium, and chloride on the basis of continuous records of SC of highway runoff. Christensen et al. (2002) and Ryberg (2006, 2007) also developed regression equations to estimate constituent concentration yields in water bodies in Kansas and North Dakota. El-Korashey (2009) has applied regression analysis to estimate sodium and chloride in Bahr El Baqar Drain in Egypt using electrical conductivity as an explanatory variable. These studies show that regression analysis is the standard method for estimating and modelling the relationship between SC and water quality constituents. Since the grab sample is collected at the same time as the RTWQ parameter measurement, it is possible to correlate some of the grab sample parameters with the RTWQ parameters. Among all real time parameters, SC is more likely to correlate with some of the ion dissolved parameters measured during grab sampling (Lind 1970). This was evident in the recent advancement of RTWQ program in NL discussed in Harvey et al. (2011).

While the earlier studies were able to identify that regression analysis models can be applied to estimate water quality constituents in real time (Ryberg 2006, 2007), there was a lack in identifying how the model can be used as an integrated tool as a part of the NL RTWQ program. Although studies performed by USGS (Christensen et al. 2002) showed the application of regression models in US water bodies, the method was never applied as an integrated component in Canadian water bodies specifically in Newfoundland and Labrador. Moreover few studies (Granato & Smith 1999) have observed how local stressors such as urbanization, mining and industrial development in general can play a role in the variation of ionic parameter concentration which helps in the estimation of constituent parameters using real time water quality parameters. The effect of flow was observed in order to identify whether it plays any role in constituent parameter estimation. Integrating the regression model component with NL RTWQ program will help estimate the concentration of select ions at select locations in real time.

This paper shows that it is possible to optimize the resources and sampling time resulting in overall cost and time savings under the RTWQ Program using real time ionic concentration estimation. This approach will help the WRMD estimate the water quality in real time without waiting for laboratory analysis results. It will be very useful where tight timelines, budget constraints and human resources limitations are matters of concern. It can also be used to estimate water quality variables at remote sampling locations that are expensive and difficult to access. The results of this report will help to better understand how local stressors can lead to increased ionic concentration due to resulting elevated SC at impacted sites.

At first the methodology applied in this analysis is described followed by the description of the site locations. A brief discussion of the data collection, QA/QC and statistical analysis is then performed. The effect of flow on each of the parameters concentration is then shown. This is followed by description of regression models and whether there was any urban or developmental influence in the model results. Finally, the model verification and validation is performed along with concluding statements.

METHODOLOGY

Figure 1 shows the main components of developing the ionic concentration model and applying the model to estimate ionic concentration in real time. The model is developed by using regression analysis on grab sample ionic concentration and real time SC data.
Figure 1

Methodology to estimate real time ionic concentration data.

Figure 1

Methodology to estimate real time ionic concentration data.

The model is then applied using real time SC data on select stations to estimate the ionic concentration values for those stations. The model provides a measure of strength and variation of the relationship between real time and grab sample data. Site specific models for sodium, calcium, chloride and sulphate are developed for the four sampling locations across the island part of Newfoundland using SC data.

DESCRIPTION OF SAMPLING LOCATIONS

Water quality sampling sites across the island of Newfoundland were used to analyse site specific relationships between real time SC and sodium, calcium, chloride and sulphate ions measured during grab sampling. These sites have been sampled extensively for the last 4–5 years. The sites were selected based on the degree of anthropogenic activity taking place and the amount of dissolved solid material received by these water bodies in order that comparative analysis can be drawn from the data obtained.

Figure 2 shows the location of the four sites on the island part of Newfoundland from which the data are collected. The sites are: Leary's Brook, Waterford River, Humber River and Rattling Brook below bridge. These sites are chosen based on the degree of anthropogenic activity and availability of water quality data. Leary's Brook and Waterford River are located in an urban setting with a high level of anthropogenic impact from the surrounding areas. Rattling Brook is non-urban but in the middle of a construction site, while Humber River is non-urban with little impact from surrounding areas.
Figure 2

Geographic location of the four sites chosen for regression modelling.

Figure 2

Geographic location of the four sites chosen for regression modelling.

Leary's Brook at Clinch Crescent was the first RTWQ station established in 2001. The sampling site is located in a developed section of the City of St. John's close to Memorial University. One of the main shopping centres in the city is located immediately upstream of the sampling site where a portion of the river is channelled through a culvert. The total drainage area is 18.6 km square and is densely surrounded by houses, buildings, business facilities and major roads. Road salts are applied during the winter months which affect the water quality within the river. Significant accumulation of debris (tree branches, garbage, plastic materials, etc.) can be observed in the culvert area as a result of surrounding anthropogenic activities.

Waterford River at Kilbride station was established in 2005. The sampling site is situated near the downtown area of the City of St. John's. The total drainage area is 52.7 km square and is densely surrounded by houses, buildings, roads and highways. Major industrial areas are also located within the drainage basin. Road salts are applied during the winter months which affect the water quality within the river. The river is highly impacted as a result of surrounding anthropogenic influence which affects the quality of water at the sampling site.

The Humber River is the second largest river on the island of Newfoundland. The sampling station was established in December 2003. It is classified as a non-urban station. The total drainage area is 7,854.1 km square. There are a number of small communities located within the watershed but the overall population density is sparse. There are some transportation routes throughout the basin which are salted during the winter months. However, due to the large volume of water within the system, the ionic concentration is diluted.

Rattling Brook below bridge station was established in December 2006 on the south eastern Avalon Peninsula. The total drainage area is 32.57 km square. It is within the construction zone of a commercial processing facility. Major work resulting from the construction of the processing facility occurs along the river and access to the sampling sites is controlled due to security and safety concerns. The river is moderately impacted with ionic concentration due to sparse population and the presence of the processing plant and facilities.

REAL TIME WATER QUALITY MONITORING NETWORK

Monitoring instruments shown in Figure 3 are submerged in a representative section of the water body which continuously measures RTWQ parameter data for each sampling station.
Figure 3

Water quality parameter sensors.

Figure 3

Water quality parameter sensors.

Data collected from these instruments are then transmitted and collected as shown in Figure 4.
Figure 4

Real time data retrieval and management.

Figure 4

Real time data retrieval and management.

All RTWQ parameter data are retrieved through the Automatic Data Retrieval System (ADRS), a series of microcomputer based programs which automatically collects and distributes the near RTWQ data. Real time stations are continuously recording large amounts of data with intervals ranging from every 15 minutes to an hour. All parameter data are logged by a data logger at the RTWQ monitoring stations and are transmitted through Geostationary Operational Environmental Satellite (GOES) or dial-up. In the GOES system, when the field instrument records a measurement to the data logger, it is then transmitted via a GOES transmitter to the National Environmental Satellite Data Information System (NESDIS) that is operated by National Oceanic and Atmospheric Administration (NOAA) in Maryland, USA. The ADRS system obtains the data through connection via an internet IP address and populates an Oracle database. In the dial-up system, the data are transmitted from the data logger via a modem directly to the ADRS processing application which then populates an Oracle Database Server.

The ADRS system extracts the raw data from the Oracle database and sorts them according to pre-specified parameter order (pH, DO, etc.). It collects the data, creates web products and transfers these products to the WRMD website. The product includes graphs of each parameter recorded at each RTWQ station, allowing a visual representation of the parameters. These graphs aid in identifying trends over specific time periods and provide a method for tracking any immediate disturbances or changes in water characteristics. The graphs are available for public viewing and are updated approximately every 2 hours. The obtained data can be exported and downloaded by the user.

DATA ANALYSIS AND SUMMARY

Data collection

Conductivity data are obtained using the RTWQ monitoring network. Grab sample data are obtained using the monthly grab samples collected by WRMD measured from an accredited laboratory. Daily average flow data were collected using Water Survey Canada's (WSC) centrally managed web based database HYDAT. The grab samples were collected from January 2006–September 2010 for Leary's Brook, Waterford River and Humber River and January 2007–September 2010 for Rattling Brook below bridge. The data from these periods were used for model calibration while an additional year and a half of data consisting of 10 grab samples were used for model validation.

QA/QC

The collected real time data undergo QA/QC check. This involves deployment, calibration and removal of the instrument on a 30-day cycle (Figure 5). At the beginning of the cycle, comparison of real time data (pH, DO, temperature, conductivity, turbidity) is made between the instrument being deployed on site (field instrument) and another freshly calibrated instrument (QA/QC instrument) in order to measure the accuracy of the field instrument.
Figure 5

Thirty-day calibration and deployment cycle.

Figure 5

Thirty-day calibration and deployment cycle.

The grab sample is collected when the field instrument is freshly calibrated and deployed back to site. This is done to ensure data accuracy of the field instrument when comparison is made to the result of the grab sample data. The collected grab sample is analysed and measured at an accredited laboratory. Real time data are again collected and compared between the field and QA/QC instrument at the end of the 30-day deployment period. Any significant shift in field instrument measurement is identified at this time.

To further measure the accuracy of the field instrument measurement, all the collected data are entered into a spreadsheet which is transformed into automated graphs. These graphs are used to identify visual drift, anomalous values and whether field instrument readings are within acceptable ranges for the given sensors.

Statistical analysis

Descriptive statistical analyses of the data for all four stations are performed for flow, SC and grab sample parameters (sodium, calcium, chloride and sulphate).

Table 1 displays the descriptive statistical measurements of grab sample parameters and the corresponding real time SC and daily average flow data for all stations. The statistical measurements show higher variations in sodium and chloride while a lower variation in calcium and sulphate for Leary's Brook and Waterford River. The high variations can be due to increased snowmelt or storm runoff that takes place during seasonal weather changes. Sulphate data were not included for Humber River and Rattling Brook due to the presence of less than detection (LTD) limit values and lack of enough variation in data values. For Humber River the high volume in water in addition to high flow dilutes the ionic concentration of most of the parameter values.

Table 1

Descriptive statistical analysis for flow, SC and grab sample data for all stations

StationsParametersSizeMinMaxMeanMedianQ1aQ3b
Leary's Brook Flow (m3/s) 31 0.06 1.68 0.49 0.4 0.2 0.64 
SC (μS/cm) 31 148.1 1,346 450.6 360.2 287 505 
Na (mg/L) 31 26 270 77.77 63 45 88 
Ca (mg/L) 31 4.2 16 8.25 10 
Cl (mg/L) 31 35 420 122.9 94 71 130 
SO4 (mg/L) 31 18 9.8 11 
Waterford River Flow (m3/s) 30 0.32 16.8 2.28 1.24 0.73 1.89 
SC (μS/cm) 30 235 1,417 529.4 438.5 369.3 534.3 
Na (mg/L) 30 33 280 92.6 68 60 94.3 
Ca (mg/L) 30 21 11,1 11 13 
Cl (mg/L) 30 51 550 146.7 110 90.3 142.5 
SO4 (mg/L) 30 22 12 11 10 12.3 
Humber River Flow (m3/s) 29 161 497 259.1 237 194 303 
SC (μS/cm) 29 25.5 43.4 34.94 35.6 31.75 38.65 
Na (mg/L) 29 3,6 2,53 2,6 
Ca (mg/L) 29 3.7 5.9 4.34 4.1 4.65 
Cl (mg/L) 29 4,03 
Rattling Brook Flow (m3/s) 29 0.17 5.68 1.2 0.99 0.56 1.54 
SC (μS/cm) 29 27.2 41.5 34.13 35.1 31.5 36.35 
Na (mg/L) 29 5.2 4.38 4.4 4.7 
Ca (mg/L) 29 1.2 1.76 1.7 1.55 
Cl (mg/L) 29 6.58 
StationsParametersSizeMinMaxMeanMedianQ1aQ3b
Leary's Brook Flow (m3/s) 31 0.06 1.68 0.49 0.4 0.2 0.64 
SC (μS/cm) 31 148.1 1,346 450.6 360.2 287 505 
Na (mg/L) 31 26 270 77.77 63 45 88 
Ca (mg/L) 31 4.2 16 8.25 10 
Cl (mg/L) 31 35 420 122.9 94 71 130 
SO4 (mg/L) 31 18 9.8 11 
Waterford River Flow (m3/s) 30 0.32 16.8 2.28 1.24 0.73 1.89 
SC (μS/cm) 30 235 1,417 529.4 438.5 369.3 534.3 
Na (mg/L) 30 33 280 92.6 68 60 94.3 
Ca (mg/L) 30 21 11,1 11 13 
Cl (mg/L) 30 51 550 146.7 110 90.3 142.5 
SO4 (mg/L) 30 22 12 11 10 12.3 
Humber River Flow (m3/s) 29 161 497 259.1 237 194 303 
SC (μS/cm) 29 25.5 43.4 34.94 35.6 31.75 38.65 
Na (mg/L) 29 3,6 2,53 2,6 
Ca (mg/L) 29 3.7 5.9 4.34 4.1 4.65 
Cl (mg/L) 29 4,03 
Rattling Brook Flow (m3/s) 29 0.17 5.68 1.2 0.99 0.56 1.54 
SC (μS/cm) 29 27.2 41.5 34.13 35.1 31.5 36.35 
Na (mg/L) 29 5.2 4.38 4.4 4.7 
Ca (mg/L) 29 1.2 1.76 1.7 1.55 
Cl (mg/L) 29 6.58 

aQ1 = first quartile.

bQ3 = third quartile.

Seasonal flow profile

Interval plots in Figure 6 show the seasonal variation for flow data. The monthly flow is plotted using daily average flow and the standard deviation for each month. Four to five years of daily average flow data obtained from HYDAT are used in this graph. The highest seasonal flow for most stations occurred during the month of April to May while the lowest seasonal flow occurs during July to October. The rise in flow corresponds to snow melt which immediately dips with a low flow season in the summer months.
Figure 6

Seasonal flow interval plot using 4–5 years of flow data for all stations.

Figure 6

Seasonal flow interval plot using 4–5 years of flow data for all stations.

Ionic concentration and real time parameter variability

Based on the statistical measurements, the variability of the grab sample parameters for all stations was tested using box plots and regression models developed. Figure 7 shows the application of box plots to compare the variability of conductivity, sodium, calcium and chloride values across locations. Sulphate values were not tested due to the presence of LTD values for two of the four stations. The median line was connected to all the stations to observe the variability and difference in parameter values. The urban stations showed more variability in parameter values in comparison to the rural stations. This can be noted by looking at the flatness of the box plots for the rural stations and the presence of outliers in the urban stations.
Figure 7

Comparisons of parameter values across locations.

Figure 7

Comparisons of parameter values across locations.

RESULTS

The assumptions for linear regression are tested using residual plots for the tested water quality parameters in Minitab™. The independence, homoscedasticity, and normality of the error distribution assumption for all parameters were fulfilled. The obtained grab samples were plotted against flow and SC to identify their effect on ionic concentration. This is further measured using correlation and regression models. Based on the strength of the model, those parameters which show a strong relationship were further validated using grab samples collected after the period of model development.

Effect of flow on parameter concentration

The effect of flow on parameter concentration is discussed in Clissie et al. (1996). Figure 8 shows scatter plots with Lowess lines to see if a relationships exists between the grab sample parameter concentration and average daily flow. Although in many water bodies flow plays a major role in controlling parameter concentration, the Lowess lines in this case displays a lack of clear patterns between parameter concentration and flow. As instantaneous flow data were not available at the time of parameter concentration measurement, the average daily flow may dampen some of the effects of flow on parameter concentration.
Figure 8

Effect of flow on parameter concentration.

Figure 8

Effect of flow on parameter concentration.

To further identify whether flow plays a role in ionic concentration parameter estimation, the flow for each station is normalized using the drainage area (Armstrong et al. 2004) and correlation was performed for each of the ionic concentration parameters with the normalized flow. Table 2 displays the result of the correlation between the ionic concentration parameters and normalized flow. The table displays lack of correlation between flow and ionic concentration parameters. Sulphate values were not tested for Humber River and Rattling Brook due to the presence of LTD values.

Table 2

Correlation of ionic concentration with normalised flow

StationSodium (mg/L)Calcium (mg/L)Chloride (mg/L)Sulphate (mg/L)
Leary's Brook −0.11 −0.28 −0.13 −0.13 
Waterford River 0.088 −0.17 0.12 −0.04 
Rattling Brook −0.17 −0.13 −0.11 NA 
Humber River 0.15 0.40 0.29 NA 
StationSodium (mg/L)Calcium (mg/L)Chloride (mg/L)Sulphate (mg/L)
Leary's Brook −0.11 −0.28 −0.13 −0.13 
Waterford River 0.088 −0.17 0.12 −0.04 
Rattling Brook −0.17 −0.13 −0.11 NA 
Humber River 0.15 0.40 0.29 NA 

Effect of conductivity on ionic parameter concentration

Figures 912 show the scatter plots with Lowess lines to check the linear patterns for sodium, calcium, chloride and sulphate with respect to SC for Leary's Brook and Waterford River. Due to the presence of LTD values, sulphate plot is not shown for Humber River and Rattling Brook Below Bridge. For Leary's Brook and Waterford River the Lowess lines displays linear patterns for each of the parameters against SC. The linearity is more apparent for sodium, chloride and sulphate. Only calcium shows linear pattern for Rattling Brook. The Lowess lines for Humber River and remainder of the Rattling Brook parameter do not display linear patterns against SC.
Figure 9

Leary's Brook parameter scatterplot with Lowess lines vs conductivity.

Figure 9

Leary's Brook parameter scatterplot with Lowess lines vs conductivity.

Figure 10

Waterford River parameter scatterplot with Lowess lines vs conductivity.

Figure 10

Waterford River parameter scatterplot with Lowess lines vs conductivity.

Figure 11

Humber River parameter scatterplot with Lowess lines vs conductivity.

Figure 11

Humber River parameter scatterplot with Lowess lines vs conductivity.

Figure 12

Rattling Brook below bridge parameter scatterplot with Lowess lines vs conductivity.

Figure 12

Rattling Brook below bridge parameter scatterplot with Lowess lines vs conductivity.

Regression models

Due to non-normality and the presence of outliers in most of the above parameter data values, log transformation was performed on the original data. Ordinary least square (OLS) was applied on the log transformed data using Minitab™. Bias correction (Duan 1983) was performed on the log transformed model for Leary's Brook and Waterford River. However due to lack of correlation between SC and ionic concentration data it was not performed on Humber River and Rattling Brook. The results are shown below in Table 3.

Table 3

OLS model for sodium, calcium, chloride and sulphate in the four stations

StationsComputed variableVariable range (mg/L)Regression modelR-squareaP-valuebBias corr.c
Leary's Brook Sodium Na: 26–270 log(Na) = −0.97 + 1.08 × log(SC) 98% 0.992 
Calcium Ca: 4.2–16 log(Ca) = −0.81 + 0.65 × log(SC) 80% 1.015 
Chloride Cl: 35–420 log(Cl) = −0.87 + 1.11 × log(SC) 96% 1.028 
Sulphate SO4: 6–18 log(SO4) = −0.22 + 0.46 × log(SC) 88% 1.005 
Waterford River Sodium Na: 33–280 log(Na) = −1.02 + 1.09 × log(SC) 96% 1.022 
Calcium Ca: 5–21 log(Ca) = −0.49 + 0.56 × log(SC) 77% 1.009 
Chloride Cl: 51–550 log(Cl) = −0.99 + 1.15 × log(SC) 91% 1.023 
Sulphate SO4: 7–22 log(SO4) = −0.18 + 0.46 × log(SC) 72% 1.011 
Humber River Sodium 2–3.6 log(Na) = 0.5 + 0.1 × log(SC) 0% 0.734 NA 
Calcium 3.7–5.9 log(Ca) = 1.1 + 0.09 × log(SC) 1% 0.549 NA 
Chloride 3–5.0 log(Cl) = 1.3 + 0.02 × log(SC) 0% 0.916 NA 
Rattling Brook Sodium Na: 3.0–5.2 log(Na) = 0.23 + 0.26 ×log(SC) 6% 0.192 NA 
Calcium Ca: 1.2–3.0 log(Ca) = −1.76 + 1.31 × log(SC) 58% NA 
Chloride Cl: 5.0–9.0 log(Cl) = 0.092 + 0.47 × log(SC) 12% 0.103 NA 
StationsComputed variableVariable range (mg/L)Regression modelR-squareaP-valuebBias corr.c
Leary's Brook Sodium Na: 26–270 log(Na) = −0.97 + 1.08 × log(SC) 98% 0.992 
Calcium Ca: 4.2–16 log(Ca) = −0.81 + 0.65 × log(SC) 80% 1.015 
Chloride Cl: 35–420 log(Cl) = −0.87 + 1.11 × log(SC) 96% 1.028 
Sulphate SO4: 6–18 log(SO4) = −0.22 + 0.46 × log(SC) 88% 1.005 
Waterford River Sodium Na: 33–280 log(Na) = −1.02 + 1.09 × log(SC) 96% 1.022 
Calcium Ca: 5–21 log(Ca) = −0.49 + 0.56 × log(SC) 77% 1.009 
Chloride Cl: 51–550 log(Cl) = −0.99 + 1.15 × log(SC) 91% 1.023 
Sulphate SO4: 7–22 log(SO4) = −0.18 + 0.46 × log(SC) 72% 1.011 
Humber River Sodium 2–3.6 log(Na) = 0.5 + 0.1 × log(SC) 0% 0.734 NA 
Calcium 3.7–5.9 log(Ca) = 1.1 + 0.09 × log(SC) 1% 0.549 NA 
Chloride 3–5.0 log(Cl) = 1.3 + 0.02 × log(SC) 0% 0.916 NA 
Rattling Brook Sodium Na: 3.0–5.2 log(Na) = 0.23 + 0.26 ×log(SC) 6% 0.192 NA 
Calcium Ca: 1.2–3.0 log(Ca) = −1.76 + 1.31 × log(SC) 58% NA 
Chloride Cl: 5.0–9.0 log(Cl) = 0.092 + 0.47 × log(SC) 12% 0.103 NA 

aR-square: the proportion of variation in the response data that is explained by the predictor.

bP-value: statistical significance between the association between the response and predictor.

cBias corr.: bias correction performed according to (Duan 1983).

The table below shows that SC can explain the variation of the ionic concentration parameters (indicated by high R-square values) for Leary's Brook and Waterford River. This is further explained by the statistically significant P-values less than 0.05. The association is stronger in sodium and chloride in comparison with calcium and sulphate. In the case of Rattling Brook below bridge and Humber River, SC cannot explain the variation of the ionic concentration parameters (low R-square values) with the exception of calcium for Rattling Brook below bridge. The weaker association is explained by P-value greater than 0.05.

The models obtained for Leary's Brook, Waterford River and Rattling Brook below bridge are used for validation. The Humber River is not used since there was no variation in any of its parameter values.

Model validation

The graphs in Figures 1315 show the ionic concentration estimation and validation of parameters used for Leary's Brook, Waterford River and Rattling Brook below bridge modeling. The OLS models from Table 3 were used to estimate ion concentration (sodium, calcium, chloride and sulphate) in real time. The model is represented by a line in the graph. The corresponding calibration grab sample values were placed as points within the graph to see how closely they fit to the model. As shown in the graphs, the calibration grab samples lie closely to the regression model line.
Figure 13

Leary’ Brook – model comparison (OLS) with actual grab samples.

Figure 13

Leary’ Brook – model comparison (OLS) with actual grab samples.

Figure 14

Waterford River – model comparison (OLS) with actual grab samples.

Figure 14

Waterford River – model comparison (OLS) with actual grab samples.

Figure 15

Rattling Brook below bridge – model comparison (OLS) with actual grab samples.

Figure 15

Rattling Brook below bridge – model comparison (OLS) with actual grab samples.

In order to validate the model, 10 additional grab samples were used after the model development. The validation grab samples are represented in a different shape in order to distinguish between calibration and validation samples. As shown in the graph, the validation grab sample values lies closely to the model line. The fitness of the values is closer for sodium and chloride as expected from the model line.

DISCUSSION

The obtained results demonstrate that ionic concentration of select parameters (sodium, calcium, chloride, and sulphate) can be accurately estimated by applying a regression analysis model to available SC real time data at select locations. An important finding was that the success of each site-specific model was dependent on the variability of the parameter values of the grab sample data. It was evident that the sites which were influenced by local stressors such as road salt, road runoff and generally increased anthropogenic influence, displayed greater variability in the measured ionic concentrations and subsequently stronger models. At the stations where the water quality was less impacted the variability of measured ionic concentrations was less and the models were unsuccessful.

These key findings will lead to an added dimension for the NL RTWQ program whereby models will be developed across the network at select locations to estimate ionic concentration, where applicable. This added functionality will provide the general public, policy makers, government agencies and other stakeholders with additional data for use. More importantly it will allow regulators to address water quality issues in a proactive manner as they arise.

With many agencies facing growing budget constraints and reduced availability of human resources, the methodology behind this tool ensures the data needed to better manage water resources are available in near real time at decreased costs leading to improved water resources management. More specifically, by estimating surrogate parameters on a continuous basis in near real-time, regulators will have a better characterization of the water body over time and be able to identify when water quality is being impacted by surrounding land uses. For example, the impact of road salting operations on Leary's Brook will be more thoroughly documented and understood from a temporal perspective by reporting the estimated ionic parameters in a continuous nature. This enhancement in available information will lead to improved policy and decision making.

It is evident that the prediction of select parameters provides the needed data while reducing the amount of financial and human resources needed to obtain the data, especially in relation to limited access to remote locations.

CONCLUSIONS

It is evident from the results presented that increased variation within grab sample measurements leads to a better regression model. This has been observed in the case of Leary's Brook and Waterford River as well as for calcium in Rattling Brook below bridge. The variation in the level of ionic concentration is largely due to the presence of anthropogenic influence within these rivers. In the case of Humber River with little anthropogenic influence, the ionic concentration of most parameter measurements were below the detection limit, and hence it was difficult to apply any statistical tests to identify if a relationship exists between real time parameters and grab samples. The volume of water in that river diluted most of the parameter concentrations which is represented in the low measurements of parameter values.

This study will aid in estimating ionic concentration in real time for the sites where a good fit for regression was obtained. It will also reduce the time delay required to measure water quality constituents at the laboratory by estimating ionic concentration instantaneously. Using the real time parameter SC, the model will help predict the surrogate parameters (sodium, chloride, calcium and sulphate) in real time which can be viewed graphically. In order to maintain the accuracy of the model, it must be calibrated every year when additional grab samples results are available. This will adjust the model accuracies based on the updated grab sample values.

Potential parameters of interest can be estimated in emerging real time sites using real time parameters as predictors by applying the methodological analysis applied in this study. One such parameter is total suspended solids (TSS) which can be estimated using real time turbidity. This would be beneficial to industries monitoring real time water quality parameters who would like to ensure that the TSS values are in compliance with the current regulations. Another area of application of this model methodology is to identify the impact of water quality due to the application of road salts. Operational decisions can be made in a proactive manner when estimated data are readily available to stakeholders.

ACKNOWLEDGEMENTS

Feedback on statistical analysis was provided by Dr Leonard Lye, Professor of Engineering and Applied Science from Memorial University. The paper was thoroughly reviewed by colleagues Kyla Brake and Robert Wight (Environmental Scientists at WRMD) with valuable insights. Technical advice relating to the ‘Data collection and management’ portion of the report was provided by Leona Hyde (Environmental Monitoring Specialist at WRMD). Finally, assistance was provided by colleague Keith Abbott (Environmental Scientist at WRMD) in the generation of the station location maps.

REFERENCES

REFERENCES
Armstrong
D. S.
Parker
G. W.
Richards
T. A.
2004
Evaluation of Streamflow Requirements for Habitat Protection by Comparison to Streamflow Characteristics at Index Streamflow-Gaging Stations in Southern New England. U.S
.
Geological Survey Water-Resources Investigations Report 03-4332
,
USA
.
Christensen
V. G.
Rasmussen
P. P.
Ziegler
A. C.
2002
Real time water quality monitoring and regression analysis to estimate nutrient and bacteria concentrations in Kansas streams
.
Water Sci. Technol
.
45
,
205
219
.
El-Korashey
R.
2009
Using regression analysis to estimate water quality constituents in Bahr El Baqar Drain
.
J. Appl. Sci. Res
.
5
,
1067
1076
.
Granato
G. E.
Smith
K. P.
1999
Estimating Concentrations of Road-Salt Constituents in Highway-Runoff from Measurements of Specific Conductance
.
U.S. Geological Survey
,
USA
.
Lind
C. J.
1970
Specific Conductance as a Means of Estimating Ionic Strength
.
U.S. Geological Survey Professional Paper
,
USA
, pp.
272
280
.
Miller
R. L.
Bradford
W. L.
Peters
N. E.
1988
Specific conductance: theoretical considerations and application to analytical quality control
.
U.S. Geological Survey Water-Supply Paper 2311
.
Ryberg
K. R.
2006
Continuous water-quality monitoring and regression analysis to estimate constituent concentrations and loads in the Red River of the North, Fargo, North Dakota, 2003–05: U.S. Geological Survey Scientific Investigations Report 2006-5241
.
Ryberg
K. R.
2007
Continuous water-quality monitoring and regression analysis to estimate constituent concentrations and loads in the Sheyenne River, North Dakota, 1980–2006: U.S. Geological Survey Scientific Investigations Report 2007-5153
.
Stevens
R. J.
O'Bric
C. J.
Carton
O. T.
1995
Estimating nutrient content of animal slurries using electrical conductivity
.
J Agric. Sci
.
125
(
2
),
233
238
.
Water Survey of Canada
2013
,
Environment Canada
,
Canada
.