BOISE — Since 2007, Mel Kunkel has used his unique computer model to offer extraordinarily accurate annual streamflow predictions for the Boise Basin on Oct. 1, the first day of each water year.
The Idaho Power hydrometeorologist said plus or minus 40 percent is a common margin of error for forecasters who would dare make their predictions that early. The Oct. 1 projections rendered by Kunkel’s model have proven accurate within plus or minus 6.5 percent — and they are months earlier than most others.
Idaho Power has asked Kunkel to expand his model to include the entire Snake River system through Idaho.
Once he has modeled the additional regions, he’ll share his results with other state water managers and experts. He hopes to have predictions available for a few new basins for the next water year.
He devised his model as a Boise State University Ph.D. project.
“For the first three or four years, folks were fairly skeptical that I was going to give a forecast that far ahead of time,” Kunkel said. “Since about 2010, there’s been quite a bit of interest in it.”
For the current water year, Kunkel predicted streamflows that are 108 percent of normal for the Boise River.
“Up until the start of February, I thought I’d blown it very badly,” Kunkel said, noting that the wet February has put his estimates well within the margin of error. “Currently, the Upper Boise is sitting at 105 percent of snow-water equivalent.”
Kunkel’s model is based on weather data from 1981 to 2002.
Other models select one of several weather indexes by identifying the index that most strongly correlates with moisture outcomes. His model takes the process a step further, utilizing a probability simulation known as a Monte Carlo series, a tool often used to identify biases inherent in models.
His computers run a million combinations of 17 of the 21 years in his data set, identifying those weather indexes correlated with actual weather conditions about 85-90 percent of the time. For each chosen index — such as El Nino/La Nina or the Pacific-North American Index — he plugs values from the months preceding the water year into a regression analysis, which is a statistical process for estimating relationships among variables, to generate a flow estimate.
Kunkel is also seeking to update his model to create season-specific flow estimates, which managers of irrigation districts and canal companies say would be especially useful.
“People have spent a lot of time developing a model that works good for a year or two, and then it doesn’t,” Kunkel said. “This technique allows for adaptability — rebuild the models every year or two and use strongest predictors each time.”
The Natural Resources Conservation Service doesn’t start making its water forecasts until Jan. 1 — three months after Kinkel makes his forecast, said Ron Abramovich, NRCS water supply specialist.
Abramovich said Kunkel has consistently “nailed it” in the Boise Basin, and he believes his forecasts for other basins will be viewed by irrigators as crucial information in making early decisions.
“He’s been on the cutting edge of using current information and the capabilities of computers to process (moisture predictions) in a timely manner,” Abramovich said.
Abramovich also appreciates that Kunkel predicts streamflow volumes, which is more useful to irrigators than simple snowpack levels.