PubAg

Main content area

Integrating real-time and manual monitored data to predict hillslope soil moisture dynamics with high spatio-temporal resolution using linear and non-linear models

Author:
Zhu, Qing, Zhou, Zhiwen, Duncan, Emily W., Lv, Ligang, Liao, Kaihua, Feng, Huihui
Source:
Journal of hydrology 2017 v.545 pp. 1-11
ISSN:
0022-1694
Subject:
bedrock, monitoring, nonlinear models, soil heterogeneity, soil water, subsurface flow, topography, water distribution
Abstract:
Spatio-temporal variability of soil moisture (θ) is a challenge that remains to be better understood. A trade-off exists between spatial coverage and temporal resolution when using the manual and real-time θ monitoring methods. This restricted the comprehensive and intensive examination of θ dynamics. In this study, we integrated the manual and real-time monitored data to depict the hillslope θ dynamics with good spatial coverage and temporal resolution. Linear (stepwise multiple linear regression-SMLR) and non-linear (support vector machines-SVM) models were used to predict θ at 39 manual sites (collected 1–2 times per month) with θ collected at three real-time monitoring sites (collected every 5mins). By comparing the accuracies of SMLR and SVM for each depth and manual site, an optimal prediction model was then determined at this depth of this site. Results showed that θ at the 39 manual sites can be reliably predicted (root mean square errors <0.028m3m−3) using both SMLR and SVM. The linear or non-linear relationship between θ at each manual site and at the three real-time monitoring sites was the main reason for choosing SMLR or SVM as the optimal prediction model. The subsurface flow dynamics was an important factor that determined whether the relationship was linear or non-linear. Depth to bedrock, elevation, topographic wetness index, profile curvature, and θ temporal stability influenced the selection of prediction model since they were related to the subsurface soil water distribution and movement. Using this approach, hillslope θ spatial distributions at un-sampled times and dates can be predicted. Missing information of hillslope θ dynamics can be acquired successfully.
Agid:
5633745
Handle:
10113/5633745