Spatial distribution and possible sources of SMOS errors at the global scale

SMOS (Soil Moisture and Ocean Salinity) data have now been available for over two years and, as part of the validation process, comparing this new dataset to already existing global datasets of soil moisture is possible. In this study, SMOS soil moisture product was evaluated globally by using the triple collocation method. This statistical method is based on the comparison of three datasets and produces global error maps by statistically inter-comparing their variations. Only the variable part of the errors are considered here, the bias errors are not treated by triple collocation. This method was applied to the following datasets: SMOS Level 2 product, two soil moisture products derived from AMSR-E (Advanced Microwave Scanning Radiometer)–LPRM (Land Parameter Retrieval Model) and NSIDC (National Snow and Ice Data Center), ASCAT (Advanced Scatterometer) and ECMWF (European Center for Medium range Weather Forecasting). The resulting errors are not absolute since they depend on the choice of the datasets. However this study showed that the spatial structure of the SMOS was independent of the combination and pointed out the same areas where SMOS performed well and where it did not. This global SMOS error map was then linked to other global parameters such as soil texture, RFI (Radio Frequency Interference) occurrence probabilities and land cover in order to identify their influences in the SMOS error. Globally the presence of forest in the field of view of the radiometer seemed to have the greatest influence on SMOS error (56.8%) whereas RFI represented 1.7% according to the analysis of variance from a multiple linear regression model. These percentages were not identical for all the continents and some discrepancies in the proportion of the influence were highlighted: soil texture was the main influence over Europe whereas RFI had the largest influence over Asia.

• The spatial distribution of SMOS errors is obtained from triple collocation.
• SMOS data have been compared to AMSR-E (LPRM and NSIDC), ASCAT and ECMWF.
• SMOS gives good results over North America, Africa, central Asia and Australia.
• Possible sources of SMOS errors are identified by performing an analysis of variance.
• SMOS error is mainly due to the soil texture, the forest and the interferences (RFI).