Improving water content estimations using penetration resistance and principal component analysis

详细信息    查看全文

• 作者：Hossein Bayat^a ; ^{h.bayat@basu.ac.ir} ; Mohammad Reza Neyshaburi^b ; ¹ ; ^{neyshmr@hotmail.com} ; Kourosh Mohammadi^c ; ² ; ^{kourosh.mohammadi@gmail.com} ; Nader Nariman-Zadeh^d ; ³ ; ^{nnzadeh@yahoo.com} ; Mahdi Irannejad^e ; ⁴ ; ^{iranajad@aut.ac.ir}
• 关键词：Artificial neural networks ; Penetration resistance ; Principal components ; Water retention ; Estimation
• 刊名：Soil & Tillage Research
• 出版年：2013
• 出版时间：May, 2013
• 年：2013
• 卷：129
• 期：Complete
• 页码：83-92
• 全文大小：394 K

文摘

Several predictors, such as soil penetration resistance (PR), have been used to estimate the soil water retention curve (SWRC), but this has been limited because most soil databases lack PR data. It would be very useful if the predicted PR could improve SWRC estimation. Also, using principal components (PCs) as predictors may help to improve SWRC estimations. One hundred and forty-eight soil samples were collected from two provinces in Iran. Soil physical properties and water content at 1, 5, 25, 50, and 1500 kPa matric suction were determined. Penetration resistance was measured (for 24 core samples) and predicted (for the rest 124 samples). Principal component analysis (PCA) was applied to all 33 original variables and eight PCs were selected. Pedotransfer functions were developed using artificial neural networks (ANNs) to estimate water content at the measured matric suctions. Using PR significantly improved SWRC estimates. Organic matter, mean weight diameter, and saturated hydraulic conductivity improved SWRC estimation around field capacity. Using macroporosity (Mp) and microporosity (Mip), improved the estimation of SWRC. This result may highlight the importance of water content at 4 kPa in determining the overall SWRC shape. A significant improvement in the pedotransfer functions accuracy and reliability occurred when the PCs were included in the list of inputs. As a preliminary analysis, PCA could simplify the pattern recognition process for ANNs. Therefore SWRC predictions could have been improved without additional measurements, by using predicted PR and PCs as predictors.