Skip to main content
ARS Home » Research » Publications at this Location » Publication #113678
Title: A BOOTSTRAP-NEURAL NETWORK APPROACH TO PREDICT SOIL HYDRAULIC PARAMETERS

Author
item SCHAAP, MARCEL - UC RIVERSIDE, CA
item LEIJ, FEIKE - UC RIVERSIDE, CA
item Van Genuchten, Martinus

Submitted to: Book Chapter
Publication Type: Book / Chapter
Publication Acceptance Date: 10/1/1999
Publication Date: N/A
Citation: N/A

Interpretive Summary: Not Required

Technical Abstract: Indirect estimation using pedotransfer PTFs of soil hydraulic properties has long received considerable attention. PTFs use regressions to predict soil hydraulic parameters from surrogate data such as soil texture and bulk density. Ideally, PTFs should provide both accurate predictions as well a measure of the reliability of those predictions. Most reviews of PTFs have focused on the accuracy, in terms of how well a particular PTF predicts hydraulic parameters of an independent data set. The reliability of PTF predictions can be quantified using the probability distribution of a prediction. Such information, normally not available, is strongly dependent upon the distribution of data in the original calibration data set. In this paper we present information about the use of a combined bootstrap-neural network procedure to predict water retention parameters, the saturated and unsaturated hydraulic conductivity, and their associated probability distributions. We also present user-friendly software that implements the developed neural-network pedotransfer functions.