Estimation of Dissolved-Solids Concentrations Using Continuous Water-Quality Monitoring and Regression Models at Four Sites in the Yuma Area, Arizona and California
The Colorado River is controlled by an extensive system of dams and canals diverting much of the water in the United States for agricultural and municipal uses. The Water Treaty of 1944 guarantees that 1.5 million acre-feet of Colorado River water is delivered to Mexico from the United States annually. Additionally, the water delivered must meet water-quality criteria as defined in Minute 242. The study developed site-specific relations of specific conductance, water temperature, and dissolved-solids concentrations at four monitoring sites in the Yuma area between Imperial Dam and the United States–Mexico southerly international boundary to predict dissolved-solids concentrations using surrogate regression-modeling approaches. The surrogate models that were developed and real-time estimates of dissolved solids in water aide water managers in making timely decisions on water operations to better meet the water-quality obligations of Minute 242.
Instream specific conductance and water temperature data were recorded every 15 minutes at 4 sites: (1) the Colorado River above Imperial Dam, (2) the Colorado River below Cooper wasteway, (3) Yuma Main Drain above Arizona– Sonora boundary, and (4) the 242 lateral above Main Drain at Arizona-Sonora boundary (see location map to the right). Discrete water-quality samples were collected at each site either biweekly or monthly over the period January 2017 to March 2019. Dissolved-solids concentration, or salinity, was computed for each sample as the sum concentration of 11 major elements. Multiple linear regression models relating specific conductance (and other factors) to dissolved-solids concentration were developed independently for all four sites based on the data collected.
Instream specific conductance for all sites from January 2017 to March 2019 ranged from 966 to 3,030 microsiemens per centimeter at 25 degrees Celsius. Dissolved-solids concentrations from discrete samples ranged from 690 to 2,580 milligrams per liter (mg/L) during the same period. Streamflow ranged from 0 to 11,700 cubic feet per second for all sites. All sites were evaluated for representative stream conditions relative to the cross section and the placement of the multi-parameter sensor. All but one site, Colorado River below Cooper wasteway, had consistently well mixed instream conditions, and specific conductance was well represented by the multi-parameter probe. Inflows from the Pilot Knob Hydro-electric Plant to the Colorado River 1.5 miles upstream from Cooper wasteway resulted in a poorly mixed stream cross section at the station when the powerplant was operational and releasing water to the river. During periods of poor mixing the specific conductance values from the instream monitor were always greater than the mean channel specific conductance values, as determined from discrete samples. A correction factor based on individual discrete sample data was developed and applied over time in order for the in-place specific conductance values to better represent the mean channel conditions.
Regression surrogate models were developed using concurrent measurements of discrete and continuous water-quality data collected from January 2017 through March 2019. Regression models were developed following methods described in Rasmussen and others (2009) and Helsel and others (2020). Ordinary least-squares regression was used to evaluate and determine the optimal continuous water-quality parameters (explanatory variables) to be used as surrogates for predicting discrete measures of dissolved solids (response variable) (See figure below).
The multiple linear-regression models explained 87.6 to 96.9 percent of the variation in dissolved solids, and the root mean square error ranged between about 6 and 27 mg/L. Specific conductance was the primary factor at all four sites and explained 87.6 to 94 percent of variation in dissolved solids concentrations. Water temperature, as an indicator of seasonal variations, was determined to be a statistically significant secondary factor at both Colorado River above Imperial Dam and Colorado River below Cooper wasteway that explained an additional 6.9 and 2.1 percent of variation in dissolved solids, respectively. The multiple-linear regression models represented between 85 and 95 percent of all the continuous specific conductance values measured at the four sites during the study.
Ongoing analysis of the regression models are occurring with models being evaluated for consistency over time. Instream dissolved-solids water-quality samples continue to be collected monthly at all sites and are compared to the computed dissolved solids values to ensure model robustness and data validity.
The Colorado River is controlled by an extensive system of dams and canals diverting much of the water in the United States for agricultural and municipal uses. The Water Treaty of 1944 guarantees that 1.5 million acre-feet of Colorado River water is delivered to Mexico from the United States annually. Additionally, the water delivered must meet water-quality criteria as defined in Minute 242. The study developed site-specific relations of specific conductance, water temperature, and dissolved-solids concentrations at four monitoring sites in the Yuma area between Imperial Dam and the United States–Mexico southerly international boundary to predict dissolved-solids concentrations using surrogate regression-modeling approaches. The surrogate models that were developed and real-time estimates of dissolved solids in water aide water managers in making timely decisions on water operations to better meet the water-quality obligations of Minute 242.
Instream specific conductance and water temperature data were recorded every 15 minutes at 4 sites: (1) the Colorado River above Imperial Dam, (2) the Colorado River below Cooper wasteway, (3) Yuma Main Drain above Arizona– Sonora boundary, and (4) the 242 lateral above Main Drain at Arizona-Sonora boundary (see location map to the right). Discrete water-quality samples were collected at each site either biweekly or monthly over the period January 2017 to March 2019. Dissolved-solids concentration, or salinity, was computed for each sample as the sum concentration of 11 major elements. Multiple linear regression models relating specific conductance (and other factors) to dissolved-solids concentration were developed independently for all four sites based on the data collected.
Instream specific conductance for all sites from January 2017 to March 2019 ranged from 966 to 3,030 microsiemens per centimeter at 25 degrees Celsius. Dissolved-solids concentrations from discrete samples ranged from 690 to 2,580 milligrams per liter (mg/L) during the same period. Streamflow ranged from 0 to 11,700 cubic feet per second for all sites. All sites were evaluated for representative stream conditions relative to the cross section and the placement of the multi-parameter sensor. All but one site, Colorado River below Cooper wasteway, had consistently well mixed instream conditions, and specific conductance was well represented by the multi-parameter probe. Inflows from the Pilot Knob Hydro-electric Plant to the Colorado River 1.5 miles upstream from Cooper wasteway resulted in a poorly mixed stream cross section at the station when the powerplant was operational and releasing water to the river. During periods of poor mixing the specific conductance values from the instream monitor were always greater than the mean channel specific conductance values, as determined from discrete samples. A correction factor based on individual discrete sample data was developed and applied over time in order for the in-place specific conductance values to better represent the mean channel conditions.
Regression surrogate models were developed using concurrent measurements of discrete and continuous water-quality data collected from January 2017 through March 2019. Regression models were developed following methods described in Rasmussen and others (2009) and Helsel and others (2020). Ordinary least-squares regression was used to evaluate and determine the optimal continuous water-quality parameters (explanatory variables) to be used as surrogates for predicting discrete measures of dissolved solids (response variable) (See figure below).
The multiple linear-regression models explained 87.6 to 96.9 percent of the variation in dissolved solids, and the root mean square error ranged between about 6 and 27 mg/L. Specific conductance was the primary factor at all four sites and explained 87.6 to 94 percent of variation in dissolved solids concentrations. Water temperature, as an indicator of seasonal variations, was determined to be a statistically significant secondary factor at both Colorado River above Imperial Dam and Colorado River below Cooper wasteway that explained an additional 6.9 and 2.1 percent of variation in dissolved solids, respectively. The multiple-linear regression models represented between 85 and 95 percent of all the continuous specific conductance values measured at the four sites during the study.
Ongoing analysis of the regression models are occurring with models being evaluated for consistency over time. Instream dissolved-solids water-quality samples continue to be collected monthly at all sites and are compared to the computed dissolved solids values to ensure model robustness and data validity.