Aarhus University Seal / Aarhus Universitets segl

Mathias Neumann Andersen

Modeling Root Length Density of Field Grown Potatoes under Different Irrigation Strategies and Soil Textures Using Artificial Neural Networks

Research output: Contribution to journal/Conference contribution in journal/Contribution to newspaperJournal articleResearchpeer-review

Standard

Modeling Root Length Density of Field Grown Potatoes under Different Irrigation Strategies and Soil Textures Using Artificial Neural Networks. / Ahmadi, Seyed Hamid; Sepaskhah, A R; Andersen, Mathias Neumann; Plauborg, Finn; Jensen, Christian Richardt; Hansen, Søren.

In: Field Crops Research, Vol. 162, 2014, p. 99-107.

Research output: Contribution to journal/Conference contribution in journal/Contribution to newspaperJournal articleResearchpeer-review

Harvard

APA

CBE

MLA

Vancouver

Author

Ahmadi, Seyed Hamid ; Sepaskhah, A R ; Andersen, Mathias Neumann ; Plauborg, Finn ; Jensen, Christian Richardt ; Hansen, Søren. / Modeling Root Length Density of Field Grown Potatoes under Different Irrigation Strategies and Soil Textures Using Artificial Neural Networks. In: Field Crops Research. 2014 ; Vol. 162. pp. 99-107.

Bibtex

@article{24ecc60917714944a47608d2b8e5ccb2,
title = "Modeling Root Length Density of Field Grown Potatoes under Different Irrigation Strategies and Soil Textures Using Artificial Neural Networks",
abstract = "Root length density (RLD) is a highly wanted parameter for use in crop growth modeling but difficult to measure under field conditions. Therefore, artificial neural networks (ANNs) were implemented to predict the RLD of field grown potatoes that were subject to three irrigation strategies and three soil textures with different soil water status and soil densities. The objectives of the study were to test whether soil textural information, soil water status, and soil density might be used by ANN to simulate RLD at harvest. In the study 63 data pairs were divided into data sets of training (80% of the data) and testing (20% of the data). A feed forward three-layer perceptron network and the sigmoid, hyperbolic tangent, and linear transfer functions were used for the ANN modeling. The RLDs (target variable) in different soil layers were predicted by nine ANNs representing combinations (models) of the eight input variables: soil layer intervals (D), percentages of sand (Sa), silt (Si), and clay (Cl), bulk density of soil layers (Bd), weighted soil moisture deficit during the irrigation strategies period (SMD), geometric mean particle size diameter (dg), and geometric standard deviation (σg). The results of the study showed that all the nine ANN models predicted the target RLD values satisfactorily with a correlation coefficient R2 > 0.98. The simplest and most complex ANN architectures were 3:2:1 and 5:5:1 consisting of D, SMD, dg, and D, Bd, SMD, σg, dg as the input variables, respectively. Low values of normalized root mean square error (NRMSE) (min = 0.101, max = 0.227) and mean absolute error (MAE) (min = 0.345 cm cm−3, max = 0.79 cm cm−3) proved the high capability of the ANN to predict RLD. The RLD prediction was more accurate in the top soil layers than in the deeper layers; this discrepancy could be possibly attributed to the non-homogenous root distribution in the three soil textures, soil pore structure, and nutrient availability. Results also implied that ANN is a strong modeling tool for simulating the RLD in small data sets. Conclusively, ANN is a powerful tool to predict RLD under a range of soil physical conditions with a high degree of accuracy and may be used in crop growth modeling.",
keywords = "Root length density, artificial neural network, modeling performance, potato, soil physical characteristics, irrigation strategies",
author = "Ahmadi, {Seyed Hamid} and Sepaskhah, {A R} and Andersen, {Mathias Neumann} and Finn Plauborg and Jensen, {Christian Richardt} and S{\o}ren Hansen",
year = "2014",
doi = "10.1016/j.fcr.2013.12.008",
language = "English",
volume = "162",
pages = "99--107",
journal = "Field Crops Research",
issn = "0378-4290",
publisher = "Elsevier BV",

}

RIS

TY - JOUR

T1 - Modeling Root Length Density of Field Grown Potatoes under Different Irrigation Strategies and Soil Textures Using Artificial Neural Networks

AU - Ahmadi, Seyed Hamid

AU - Sepaskhah, A R

AU - Andersen, Mathias Neumann

AU - Plauborg, Finn

AU - Jensen, Christian Richardt

AU - Hansen, Søren

PY - 2014

Y1 - 2014

N2 - Root length density (RLD) is a highly wanted parameter for use in crop growth modeling but difficult to measure under field conditions. Therefore, artificial neural networks (ANNs) were implemented to predict the RLD of field grown potatoes that were subject to three irrigation strategies and three soil textures with different soil water status and soil densities. The objectives of the study were to test whether soil textural information, soil water status, and soil density might be used by ANN to simulate RLD at harvest. In the study 63 data pairs were divided into data sets of training (80% of the data) and testing (20% of the data). A feed forward three-layer perceptron network and the sigmoid, hyperbolic tangent, and linear transfer functions were used for the ANN modeling. The RLDs (target variable) in different soil layers were predicted by nine ANNs representing combinations (models) of the eight input variables: soil layer intervals (D), percentages of sand (Sa), silt (Si), and clay (Cl), bulk density of soil layers (Bd), weighted soil moisture deficit during the irrigation strategies period (SMD), geometric mean particle size diameter (dg), and geometric standard deviation (σg). The results of the study showed that all the nine ANN models predicted the target RLD values satisfactorily with a correlation coefficient R2 > 0.98. The simplest and most complex ANN architectures were 3:2:1 and 5:5:1 consisting of D, SMD, dg, and D, Bd, SMD, σg, dg as the input variables, respectively. Low values of normalized root mean square error (NRMSE) (min = 0.101, max = 0.227) and mean absolute error (MAE) (min = 0.345 cm cm−3, max = 0.79 cm cm−3) proved the high capability of the ANN to predict RLD. The RLD prediction was more accurate in the top soil layers than in the deeper layers; this discrepancy could be possibly attributed to the non-homogenous root distribution in the three soil textures, soil pore structure, and nutrient availability. Results also implied that ANN is a strong modeling tool for simulating the RLD in small data sets. Conclusively, ANN is a powerful tool to predict RLD under a range of soil physical conditions with a high degree of accuracy and may be used in crop growth modeling.

AB - Root length density (RLD) is a highly wanted parameter for use in crop growth modeling but difficult to measure under field conditions. Therefore, artificial neural networks (ANNs) were implemented to predict the RLD of field grown potatoes that were subject to three irrigation strategies and three soil textures with different soil water status and soil densities. The objectives of the study were to test whether soil textural information, soil water status, and soil density might be used by ANN to simulate RLD at harvest. In the study 63 data pairs were divided into data sets of training (80% of the data) and testing (20% of the data). A feed forward three-layer perceptron network and the sigmoid, hyperbolic tangent, and linear transfer functions were used for the ANN modeling. The RLDs (target variable) in different soil layers were predicted by nine ANNs representing combinations (models) of the eight input variables: soil layer intervals (D), percentages of sand (Sa), silt (Si), and clay (Cl), bulk density of soil layers (Bd), weighted soil moisture deficit during the irrigation strategies period (SMD), geometric mean particle size diameter (dg), and geometric standard deviation (σg). The results of the study showed that all the nine ANN models predicted the target RLD values satisfactorily with a correlation coefficient R2 > 0.98. The simplest and most complex ANN architectures were 3:2:1 and 5:5:1 consisting of D, SMD, dg, and D, Bd, SMD, σg, dg as the input variables, respectively. Low values of normalized root mean square error (NRMSE) (min = 0.101, max = 0.227) and mean absolute error (MAE) (min = 0.345 cm cm−3, max = 0.79 cm cm−3) proved the high capability of the ANN to predict RLD. The RLD prediction was more accurate in the top soil layers than in the deeper layers; this discrepancy could be possibly attributed to the non-homogenous root distribution in the three soil textures, soil pore structure, and nutrient availability. Results also implied that ANN is a strong modeling tool for simulating the RLD in small data sets. Conclusively, ANN is a powerful tool to predict RLD under a range of soil physical conditions with a high degree of accuracy and may be used in crop growth modeling.

KW - Root length density

KW - artificial neural network

KW - modeling performance

KW - potato

KW - soil physical characteristics

KW - irrigation strategies

U2 - 10.1016/j.fcr.2013.12.008

DO - 10.1016/j.fcr.2013.12.008

M3 - Journal article

VL - 162

SP - 99

EP - 107

JO - Field Crops Research

JF - Field Crops Research

SN - 0378-4290

ER -