Premium
Performance of Soil Cation Exchange Capacity Pedotransfer Function as Affected by the Inputs and Database Size
Author(s) -
Khodaverdiloo Habib,
Momtaz Hamidreza,
Liao Kaihua
Publication year - 2018
Publication title -
clean – soil, air, water
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.444
H-Index - 66
eISSN - 1863-0669
pISSN - 1863-0650
DOI - 10.1002/clen.201700670
Subject(s) - pedotransfer function , cation exchange capacity , soil science , soil water , silt , soil test , linear regression , environmental science , mathematics , statistics , hydraulic conductivity , geology , paleontology
Cation exchange capacity (CEC) have diverse applications from soil classification and management to agricultural/environmental simulations. Direct measurement of CEC is difficult. The objective of this study is to investigate the effect of calibration dataset size on accuracy and validity of soil CEC pedotransfer functions (PTFs) derived for selected arid–semi arid soils. Furthermore, the effect of the type of inputs on performance of the PTFs is evaluated. The soil organic carbon content along with either soil sand, silt, and clay fractions or d g are used as input variables. A large dataset ( n = 1141) and another independent dataset ( n = 205) are applied to develop and evaluate the validity of PTFs, respectively. Multiple linear regression analysis is used to correlate soil CEC with the inputs. To investigate the database‐size effect, random subsets with different sizes ( n = 50, 100, 200, 300, 500, 700, 900) from the whole dataset are used for the development of PTFs. The accuracy of the PTFs decreased to some extent and then remained relatively constant, but their reliability increased with increasing the dataset size. Clay has the strongest correlation with CEC ( r = 0.41 to 0.59, depending on the dataset size), so that using clay content resulted in more accurate PTFs than d g . The critical value of CEC is found to be about 15–16 cmol c kg −1 , below which the PTFs overestimate CEC. The results support the hypothesis that the number of samples have to be large enough (≥500) to develop both accurate and reliable PTFs.