Soil Moisture Retrieval from Active/Passive Microwave Observation Synergy Using a Neural Network Approach

Abstract:

In November 2014 NASA will launch the Soil Moisture Active/Passive (SMAP) mission carrying an L-band radiometer and radar sensor to observe surface soil moisture globally. This new type of instrument requires the development of innovative retrieval algorithms that are able to account for the different surface contributions to the satellite signal and at the same time can optimally exploit the synergy of active and passive microwave data. In this study, a neural network (NN) based retrieval algorithm has been developed using the example of active microwave observations from ASCAT and passive microwave observations from AMSR-E. In a first step, different preprocessing techniques, aiming to highlight the various contributions to the satellite signal, have been investigated. It was found that in particular for the passive microwave observations, the use of multiple frequencies and preprocessing steps could help the retrieval to disentangle the effects of soil moisture, vegetation and surface temperature. A spectral analysis investigated the temporal patterns in the satellite observations and thus assessed which soil moisture temporal variations could realistically be retrieved. The preprocessed data was then used in a NN based retrieval to estimate daily volumetric surface soil moisture at the global scale for the period 2002-2013. It could be shown that the synergy of data from the two sensors yielded a significant improvement of the retrieval performance demonstrating the benefit of multi-sensor approaches as proposed for SMAP. A comparison with a more traditional retrieval product merging approach furthermore showed that the NN technique is better able to exploit the complementarity of information provided by active and passive sensors. The soil moisture retrieval product was evaluated in the spatial, temporal and frequency domain against retrieved soil moisture from WACMOS and SMOS, modeled fields from ERA-interim/Land and in situ observations from the International Soil Moisture Network. In all cases, the NN soil moisture estimates showed a good ability to capture the spatial and temporal structures. Additionally, it could be shown that the NN technique yields model-compatible SM estimates and thus constitutes a useful tool for satellite data assimilation.