干旱区生态环境知识资源中心

Soil moisture is a key geophysical variable in hydrological and meteorological processes. Accurate and current observations of soil moisture over meso to global scales used as inputs to hydrological, weather and climate modelling will benefit the predictability and understanding of these processes. At present, satellite platforms are active in mapping global surface soil moisture jointly at sub-daily intervals and mesoscale resolutions. However to correctly interpret observed variations and assimilate them in hydrological and weather models, the error structures of the retrieved soil moisture data need to be better understood and characterised.

In this paper we investigate the utility of a recently proposed method to quantify the variance of stochastic noise in passive and active satellite soil moisture products. While it is typical to analyse the difference between satellite retrievals and ground truth in the time domain, this method is based on quantifying the differences between retrieved soil moisture and a standard water-balance equation in the conjugate Fourier domain. The method, which referred to as Spectral Fitting (SF), is applied to estimate the errors in passive and active retrievals over Australia (10-44 degrees South, 112-154 degrees East). In particular we consider the AMSR-E (Advanced Microwave Scanning Radiometer - Earth Observing System) LPRM (Land Parameter Retrieval Method), CATDS (Centre Aval de Traitement des Donnees SMOS) SMOS (Soil Salinity and Ocean Salinity), and TU-WIEN (Vienna University of Technology) ASCAT (Advanced Scatterometer) soil moisture products. The results are compared against the errors estimated using the standard method of triple collocation (TC) with AMSR-E, SMOS and ASCAT as the data triplet.

Our analyses show that the SF method is able to recover similar and reasonable error maps that reflect sensitivity of retrieval errors to land surface and climate characteristics over Australia. As expected, more vegetated and wetter areas are usually associated with higher errors. Additionally for SMOS and ASCAT, the dry cooler desert areas of southern Australia also show higher errors, in contrast to lower errors over the hotter dry desert of central Australia. The reverse is the case for AMSR-E. These patterns are also reflected in the spatial error maps of TC analysis and the direct comparisons of SF and TC estimates show moderate-to-good correlations: 0.64 for AMSR-E, 0.68 for SMOS, and 0.68 for ASCAT. However the SF yields lower estimates than TC at the high end of the range. On one hand, this is perhaps expected given rationale of the SF method to estimate only the stochastic/high-frequency components of the total errors. On the other hand, the simple error model and implementation of TC with non-coincident overpass times can also over-estimate the errors.

This work therefore presents an additional perspective on satellite soil moisture observation errors (in the Fourier domain) that may complement other error estimation approaches (in the time domain), thereby improving our understanding of the sources and types of errors.