The scientists used data mining and classification trees to break down traditional soil maps into more detailed digital representations. By analyzing soil descriptions and landscape data, they created models that matched soil components in the original maps about 56% to 65% of the time. When comparing their models to independent soil samples, they found that the models correctly predicted soil classes around 22% to 24% of the time, increasing to 39% to 44% within a 60-meter radius. They also developed a method to measure uncertainty in their predictions, identifying four levels of uncertainty ranging from 7% to 43%.