USE OF SURROGATE MODELS AS AN EARLY WARNING SYSTEM FOR CHANGING WATER-QUALITY CONDITIONS IN THE KANSAS RIVER, KS

The Kansas River and its alluvial aquifer are primary sources of drinking water for about 800,000 people in northeastern Kansas. Water-treatment facilities that use the Kansas River as a source water supply employ various chemical and physical processes to remove contaminants before distribution. In-situ water-quality parameters that are measured and transmitted in near real-time, such as turbidity, specific conductance, and pH, can be used as indicators for the necessity to alter treatment processes to ensure effective removal of contaminants. Early warning of changing water-quality conditions, including cyanobacteria and associated toxin and taste-and-odor compounds, can provide drinking-water treatment facilities time to proactively implement appropriate treatment strategies that help ensure safe drinking water and public-health protection. To estimate changes in water-quality conditions in near-real time, continuous and discrete water-quality data have been collected at two U.S. Geological Survey (USGS) streamflow sites on the lower Kansas River since 2012: 1) Kansas River at Wamego, KS and 2) Kansas River at De Soto, KS. These data were used to develop surrogate linear regression models that continuously estimate constituent concentrations for major ions, dissolved solids, alkalinity, nutrients, suspended sediment, and indicator bacteria. Additionally, surrogate logistic regression models were developed to continuously estimate the probability of occurrence above selected thresholds of interest for cyanobacteria, microcystin, and geosmin. These estimates are available in near-real time on the USGS National Real-Time Water Quality (NRTWQ) website at https://nrtwq.usgs.gov/ks/. Linear models generally have performed well since initial development in 2016. However, the logistic models did not perform well in subsequent years, likely because of flow-driven environmental conditions in the system. Continued monitoring of continuous and discrete water-quality data is essential for further model refinement and evaluation. This will be useful for characterizing changes in water-quality conditions through time, indicating changes in water-quality conditions that may affect drinking-water treatment processes, and characterizing potentially harmful cyanobacterial events.