Title: Estimating root zone soil moisture using multilayer feedforward neural networks based on Levenberg-Maquardt and scaled conjugate gradient algorithms

Authors: John Ogony Odiyo; Rachel Makungo

Addresses: Department of Hydrology and Water Resources, School of Environmental Sciences, University of Venda, Private Bag X 5050, Thohoyandou 0950, South Africa ' Department of Hydrology and Water Resources, School of Environmental Sciences, University of Venda, Private Bag X 5050, Thohoyandou 0950, South Africa

Abstract: This paper presents the use of multilayer feedforward neural networks for root zone soil moisture estimation based on Levenberg-Maquardt (LM) and scaled conjugate gradient (SCG) algorithms. Data driven approaches such as artificial neural networks overcome some limitations of remote sensing, conceptual and semi-analytical models in root zone soil moisture estimation. This creates the need to compare neural networks training algorithms to determine those with best estimation accuracy. Observed volumetric soil moisture at 80 cm depth, rainfall and evaporation data were used to estimate soil moisture at 120 and 180 cm depths, within MATLAB. SCG trained network underestimated and/or overestimated soil moisture as compared to LM trained network. LM trained network has better performance and estimation accuracy. Estimated soil moisture is useful for irrigation scheduling, hydrological modelling, and groundwater recharge estimation in the study area.

Keywords: algorithm; estimation accuracy; Levenberg-Maquardt; multilayer feedforward; neural networks; root zone; scaled conjugate gradient; SCG; soil moisture.

DOI: 10.1504/IJHST.2019.102421

International Journal of Hydrology Science and Technology, 2019 Vol.9 No.4, pp.399 - 411

Received: 08 Feb 2017
Accepted: 30 Dec 2017

Published online: 24 Sep 2019 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article