Comparing geostatistics techniques and nonparametric k-nearest neighbor technique for predicting soil saturated hydraulic conductivity

Knowledge of the soil saturated hydraulic conductivity (Ks) is essential for irrigation management purposes and hydrological modeling، but it cannot often be measured because of practical and/or cost-related reasons. In this research، common geostatistical approaches with one type of the nonparametric lazy learning algorithms، a k-nearest neighbor (k-NN) algorithm، was compared and tested to estimate saturated hydraulic conductivity (Ks) from other easily available soil properties. In this research 151 soil samples were collected from arable land around Bojnourd and saturated hydraulic conductivity (Ks) was estimated from other soil properties including soil textural fractions، EC، pH، SP، OC، TNV، ρs and ρb. The nonparametric k-NN technique performed mostly equally well، in terms of Pearson correlation coefficient (r)، modeling efficiency (EF)، root-mean-squared errors (RMSE)، maximum error (ME) and coefficient of residual mass (CRM) statistics (r=0. 76، EF=0. 655، RMSE=42. 87، ME=26. 89 and CRM= -0. 11) and after that، Co-Kriging and simple kriging methods، performed better than others. It can be concluded that the k-NN technique is a competitive alternative to other techniques such as pedotransfer functions (PTFs) to estimate saturated hydraulic conductivity especially when for new data set redriving these functions is essential.