Computational Technology Resources - CCP

Keywords: soil hydrology, pedotransfer function, artificial neural network, support vector machine, data driven model, ensemble.

Summary

The water retention curve is one of the main soil hydraulic properties, which is used in simulating the water regime of soils. This curve is used to predict a soil's water storage, the water supply to plants, and for other tasks in soil water modelling. Various authors classified different types of pedotransfer function (PTF) evaluations such as point estimation methods, parameter estimation methods, and semi-physical methods. In this paper the authors focus on point estimation methods, which follow the direct approach by estimating the water content at predetermined pressure heads. The data-driven methods applied in this paper were trained to compute the water content for pressure head values of -2.5, -56, -209, -558, -976 and -3060 cm. An area of the Zahorska lowland in Slovak Republic was selected for testing the methods described. A total of 226 soil samples were taken from various localities in this area.

This task is basically regression task, which could be accomplished by various methods, among which most promising are various nonlinear data-driven techniques. Various data-driven techniques were analysed in this case study.

The first approach for modelling the PTFs used in this paper is the application of single data-driven models namely artificial neural networks (ANN) and support vector machines. This approach is mainly used for comparative purposes, e.g. for evaluating the possible gains obtained from the application of ensemble methodologies.

Two ensemble methods are compared in this study: bagging and additive regression. The training data in bagging were resampled using the bootstrap method to form five training sets of the same size as the original training data set (181 samples) from which the five ANN models were developed and combined to provide the predictions. A simple averaging of the five predictions from the ANN basic learners is used for the final prediction.

A second ensemble methodology applied was additive regression, which uses the Gaussian process as a base learner. In this work the Laplace kernel for this base learner was chosen by trial and error.

The accuracy of the predictions was evaluated by the correlation coefficient between the measured and predicted parameter values and by other statistics. From the results obtained it was shown that for this task ensemble data-driven methods work better than single data-driven methods. The best method was obtained by the additive regression ensemble methodology composed of the Gaussian process as base learner. This methodology produced approximately about 8% better results than the single ANN methodology in terms of the correlation coefficients.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £50 +P&P)

	Computational & Technology Resources an online resource for computational, engineering & technology publications
	not logged in - login
Front Page Browse CCP CSETS CTR IJRT Other Authors Search Purchase Guide FAQ Contact us	Civil-Comp Proceedings ISSN 1759-3433 CCP: 100 PROCEEDINGS OF THE EIGHTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY Edited by: B.H.V. Topping Paper 72 Pedotransfer Functions Development by means of the Ensemble Data-Driven Methodology M. Cisty, J. Bezak and J. Skalova Department of Land and Water Resources Management, Faculty of Civil Engineering, Slovak University of Technology Bratislava, Slovak Republic doi:10.4203/ccp.100.72 purchase the full-text of this paper Full Bibliographic Reference for this paper M. Cisty, J. Bezak, J. Skalova, "Pedotransfer Functions Development by means of the Ensemble Data-Driven Methodology", in B.H.V. Topping, (Editor), "Proceedings of the Eighth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 72, 2012. doi:10.4203/ccp.100.72 Keywords: soil hydrology, pedotransfer function, artificial neural network, support vector machine, data driven model, ensemble. Summary The water retention curve is one of the main soil hydraulic properties, which is used in simulating the water regime of soils. This curve is used to predict a soil's water storage, the water supply to plants, and for other tasks in soil water modelling. Various authors classified different types of pedotransfer function (PTF) evaluations such as point estimation methods, parameter estimation methods, and semi-physical methods. In this paper the authors focus on point estimation methods, which follow the direct approach by estimating the water content at predetermined pressure heads. The data-driven methods applied in this paper were trained to compute the water content for pressure head values of -2.5, -56, -209, -558, -976 and -3060 cm. An area of the Zahorska lowland in Slovak Republic was selected for testing the methods described. A total of 226 soil samples were taken from various localities in this area. This task is basically regression task, which could be accomplished by various methods, among which most promising are various nonlinear data-driven techniques. Various data-driven techniques were analysed in this case study. The first approach for modelling the PTFs used in this paper is the application of single data-driven models namely artificial neural networks (ANN) and support vector machines. This approach is mainly used for comparative purposes, e.g. for evaluating the possible gains obtained from the application of ensemble methodologies. Two ensemble methods are compared in this study: bagging and additive regression. The training data in bagging were resampled using the bootstrap method to form five training sets of the same size as the original training data set (181 samples) from which the five ANN models were developed and combined to provide the predictions. A simple averaging of the five predictions from the ANN basic learners is used for the final prediction. A second ensemble methodology applied was additive regression, which uses the Gaussian process as a base learner. In this work the Laplace kernel for this base learner was chosen by trial and error. The accuracy of the predictions was evaluated by the correlation coefficient between the measured and predicted parameter values and by other statistics. From the results obtained it was shown that for this task ensemble data-driven methods work better than single data-driven methods. The best method was obtained by the additive regression ensemble methodology composed of the Gaussian process as base learner. This methodology produced approximately about 8% better results than the single ANN methodology in terms of the correlation coefficients. purchase the full-text of this paper (price £20) go to the previous paper go to the next paper return to the table of contents return to the book description purchase this book (price £50 +P&P)
Back to top	©Civil-Comp Limited 2023 - terms & conditions