Uncertainty analysis in near-surface soil moisture estimation on two typical land-use hillslopes

Purpose

Spatial prediction of near-surface soil moisture content (NSSMC) is necessary for both hydrologic modeling and land use planning. However, uncertainties associated with the prediction are always neglected and lack of quantitative analysis. The objective of this study was to investigate the influences of different sources of uncertainty on NSSMC estimation at two typical hillslopes (i.e., tea garden and forest).

Materials and methods

In this study, stepwise multiple regression models with terrain indices and soil texture were built to spatially estimate NSSMC on two typical land use hillslopes (tea garden and forest) at different dates. The uncertainties due to limited sample sizes used for developing regression models (uncertainty of model parameter), digital elevation model resolutions of 1, 2, 3, 4, and 5 m (uncertainty of terrain indices) and spatial interpolations of soil texture by kriging or cokriging with electromagnetic induction (uncertainty of soil texture), were investigated using bootstrap, resampling, and Latin hypercube sampling techniques, respectively.

Results and discussion

The accuracies of NSSMC predictions were acceptable for both tea garden (the Nash-Sutcliffe efficiency or NSE?=?0.34) and forest hillslopes (NSE?=?0.57). The model parameter uncertainty was more important on tea garden hillslope than on forest hillslope. A significant negative correlation (P?

Conclusions

Improving the regression model structure and the hillslope soil texture mapping are critical in the accurate spatial prediction of NSSMC on tea garden and forest hillslopes, respectively. This study presents techniques for analyzing three different uncertainties that can be used to identify the main sources of uncertainties in soil mapping.