Artificial Intelligence Scores High in Accuracy to Predict Water Contamination
|Momcilo Markus - (217) 333-0237, firstname.lastname@example.org|
Lisa Sheppard - (217) 244-7270, email@example.com
CHAMPAIGN, Ill. - New, effective solutions are revealed when scientists use computer programs that simulate human intelligence to forecast drinking water contamination in agricultural areas, according to Momcilo Markus, hydrologist at the Illinois State Water Survey (ISWS), a division of the Prairie Research Institute at the University of Illinois.
Nitrate loss from cropland and other sources can be flushed into local rivers and streams, and may exceed safe drinking water standards in public water supplies. When this happens, water suppliers must consider using costly treatment or other measures to correct the problem. Predictions on when high nitrate levels might occur in the future are needed for suppliers to make timely management decisions.
In a recent study, Markus and colleagues from the U.S., Italy, and the United Kingdom applied artificial intelligence to analyze six years of data from the Upper Sangamon River watershed, which discharges into Lake Decatur. The study objective was to investigate the tools to improve forecasting of high nitrate levels one week in advance of an event, such as a heavy rain. They tested the accuracy of data mining, or searching for patterns in large quantities of data to find exceptional solutions.
“With data mining, we can discover hidden relationships and their changes among variables that are not easily acquired by the usual methods of computer modeling,” Markus said. “Artificial neural networks and other technologies actually mimic human thinking and reasoning.”
Artificial neural network systems simulate intelligence by attempting to reproduce the types of physical connections that occur in human brains. Another technology that ISWS scientists have used is evolutionary algorithms, which are inspired by the theory of evolution in which the “fittest” or best solution has evolved. In this case, these technologies show which parameters and relationships provide the best prediction of nitrate in drinking water.
In these and similar studies, scientists feed the program with data, and the computer reveals which factors are most important in this particular challenge now, later, and in various cycles.
“In the traditional approach, we start with a model and assume it works,” Markus said. “In our studies, we didn’t have to assume anything; the computer sorts out the data and finds relationships between variables that we may not have realized.”
Markus and his colleagues have found that artificial neural networks are highly accurate in predicting drinking water contamination and detecting changing complexity of the relationships for different seasons. The evolutionary algorithms method was very useful in determining the most relevant predictors and types of predictive relationships. A multi-tool approach may be best for nitrate forecasting.
The engineers are facing numerous challenges in modeling water systems due to climate change, climate variability, land-use land-cover changes, urbanization, hydraulic alterations, and other factors. There is a great potential for data mining tools in modeling the increasingly complex water system and predicting its future.