The Added Utility Of Nonlinear Techniques In Rescaling Soil Moisture Time Series

Abstract:

Soil moisture is one of the key parameters in geophysical processes. There are different techniques available for the retrieval of this variable (i.e., hydrological models, in-situ observations, or remote sensing). Although the soil moisture values obtained from these different platforms have temporal and spatial consistency, they often have differences in their signal (i.e. variance) components and/or in their scales. Therefore, it is necessary to rescale these datasets to a reference dataset in order to avoid any added bias during preprocessing phase. There are different linear and non-linear rescaling methods developed for rescaling of time series. However the added utility nonlinear rescaling methods compared to linear methods remained not well explored. In this study accuracy of 12 different linear and non-linear methods in rescaling of soil moisture time series are evaluated over 4 USDA ARS watersheds using LPRM-, NOAH-, API-, and station-based datasets. Results showed that generally non-linear methods (especially artificial neural networks) result in improved accuracy than linear methods.