برآورد رطوبت در نقطه پژمردگي دائم و ظرفيت زراعي خاك¬هاي شمال و شمال شرق ايران با استفاده از روش هاي نزديك ترين K همسايه و شبكه هاي عصبي مصنوعي

Estimating of Water Content in FC and PWP in North and North East of Iranʹs Soil Sampels Using k-Nearest Neighbor and Artificial Neural Networks

Field capacity and permanent wilting point are the most important potential points in modeling and management of agicultrual products water requirement. Direct methods of identifying water content are time consuming and expensive. So use of pedo-transfer functions for conveting easily mesurablae characterictiscs to hydraulic charachterisitcs is a suitable solution to solve this problem. In this research efficency of NeuroMultistep outflow and K nearest neighborhood Models in deriving PTFs for identifying water contetnt in FC and PWP for 122 soil samples of north and north east of Iran was investigated. In addition effect of different input parameters and data for deriving in both methods was specifyied. Results showed that in generall KNN method (RMSE= 0.027) had better results than NNs models (RMSE= 0.037). Also we can say that sensitivity of NNs to quality and kind of deriving data is very high and heterogeneous data can decrease the efficency of these models and increase the RMSE around 100%. Also the results showed that considering one or more hydraulic parameters as an input can improve the modeling results. Key words: k-Nearest Neighbor, Artificial Neural Networks, Pedotransfer function, Field capacity, Permanent wilting point