HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Figures Only Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Cockx, L. Articles by Van Coillie, F. M. B. PubMed Articles by Cockx, L. Articles by Van Coillie, F. M. B. Agricola Articles by Cockx, L. Articles by Van Coillie, F. M. B. Related Collections Electromagnetic Induction, EMI Spatial Variability Statistics

PEDOLOGY

Extracting Topsoil Information from EM38DD Sensor Data using a Neural Network Approach

^a Research Group Soil Spatial Inventory Techniques, Dep. of Soil Management, Ghent Univ., Coupure 653, 9000 Gent Belgium
^b Dep. of Soil Science, Faculty of Agriculture, Univ. of Peradeniya, Sri Lanka
^c Geo Solutions, Veldkant 37, 2550 Kontich
^d Lab. of Forest Management and Spatial Information Techniques, Ghent Univ., Coupure 653, 9000 Gent, Belgium

^* Corresponding author (liesbet.cockx{at}ugent.be)

Electromagnetic induction soil sensors are an increasingly important source of secondary information to predict soil texture. In a 10.5-ha polder field, an EM38DD survey was performed with a resolution of 2 by 2 m and 78 soil samples were analyzed for sub- and topsoil texture. Due to the presence of former water channels in the subsoil, the coefficient of variation of the subsoil clay content (45%) was much larger compared with the topsoil (13%). The horizontal (EC_a–H) and vertical (EC_a–V) electrical conductivity measurements displayed a similar pattern, indicating a dominant influence of the subsoil features on both signals. To extract topsoil textural information from the depth-weighted EM38DD signals we turned to artificial neural networks (ANNs). We evaluated the effect of different input layers on the ability to predict the topsoil clay content. To identify the response of the topsoil, both EM38DD orientations were used. To examine the influence of the local neighborhood, contextual EC_a information by means of a window around each soil sample was added to the input. The best ANN model used both EC_a–H and EC_a–V data but no contextual information: a mean squared estimation error of 2.83%² was achieved, explaining 65.5% of the topsoil clay variability with a variance of 0.052%². So, with the help of ANNs, the prediction of the topsoil clay content was optimized through an integrated use of the two EM38DD signals.

Abbreviations: ANN, artificial neural network • CV, coefficient of variation • EC_a, apparent electrical conductivity • EC_a–H, EC_a measurement in the horizontal orientation • EC_a–V, EC_a measurement in the vertical orientation • EMI, electromagnetic induction • MSEE, mean squared estimation error • R_n, normalized ranks score • R_o, overall rank score • RNE, relative nugget effect • VAR, variance • WS, window size