FAYETTEVILLE, Ark. - Researchers use water quality models to estimate the effectiveness of water management practices. Accurate models are essential to avoiding expensive implementation of ineffective practices, but acquiring the data to build them is expensive. University of Arkansas researchers have determined the appropriate amount of data necessary to produce accurate results in water quality models.

"Ninety percent of water quality assessments use a model," said Indrajeet Chaubey, assistant professor of agricultural and biological engineering. "In most instances, they are the only way to establish the total maximum daily load (TMDL) accurately. They are also used to find hot spots within a watershed where best management practices (BMP) should be implemented and to evaluate the impact of various BMPs on water quality."

Chaubey recently conducted a study to determine how the model results are affected by land use and soil data. He worked with Thomas Soerens, associate professor of civil engineering; Mark Nelson, director of the water quality laboratory at the Arkansas Water Resource Center, and graduate student Amy Cotter and associate professor Tom Costello of agricultural and biological engineering. Chaubey will present their findings this week at the American Society of Agricultural Engineers annual meeting in Chicago.

The researchers looked at prediction uncertainty in the Soil and Water Assessment Tool (SWAT) model, which has become widely accepted and is used by the U.S. Environmental Protection Agency. They evaluated the effects of data quantity from geographic information systems (GIS) to find the lowest data resolution necessary to accurately characterize a watershed.

Resolution is a measure of the proportion of the smallest individually accessible portion of an area to the overall size of the area. A higher resolution means that more details can be discerned. But as resolution increases, the amount of data that the model must process also increases dramatically and the modeling process becomes increasingly difficult. Chaubey wanted to find the type and quantity of data that had the greatest effect on the model output accuracy.

"While it is always desirable to have as much detailed GIS data as possbile, often the cost of the data collection determines the resolution of the GIS data used in model applications," Chaubey explained

Although GIS use has enhanced the ability to simulate watershed-scale water quality processes, Chaubey says that the extent to which soil survey, elevation and land use data should be used is not known. Spatial resolution - the distance between data points - has a direct effect on output errors in the model. However, some studies had found wider spacing to be more inaccurate, while others found that closer spacing introduced errors.

Researchers used the Lincoln Lake watershed in northwest Arkansas to evaluate the model use of data. First, they ran seven scenarios at input resolutions ranging from 30 x 30 meters to 1000 x 1000 meters to predict discharge, nitrate and phosphorus loads. Then they compared the model predictions with actual data take from the watershed between 1997 and 2000.

"The watershed, a sub-basin of the Illinois River, is primarily pasture and mixed forest with numerous poultry and beef operations. Lincoln Lake is a secondary water supply for the city of Lincoln," explained Chaubey. "It has been monitored since 1992. Degradation of its water quality is believed to have been caused by runoff of nutrients from surface-applied animal manure in the watershed."

Chaubey’s team found that watershed and stream characterization became increasingly less accurate as the resolution became lower. Although it varied among the parameters, overall the GIS resolution had a significant impact on model output uncertainty, with topography resolution having the strongest effect. For example, the researchers found that stream flow and nitrates were underestimated when topographic resolution decreased. However, this did not always result in decreased export of total phosphorus.

"Watershed topography was the key factor in determining the accuracy of SWAT model predictions at different resolutions," said Chaubey. "However, our research found that the data resolution required to achieve 90 percent accuracy depends on which output is of interest - discharge, nitrate or phosphorus loads - and may range between 30 and 1000 meters."