Elsevier

Remote Sensing of Environment

Volume 173, February 2016, Pages 1-14
Remote Sensing of Environment

Soil moisture retrieval from AMSR-E and ASCAT microwave observation synergy. Part 1: Satellite data analysis

https://doi.org/10.1016/j.rse.2015.11.011Get rights and content

Highlights

  • Neural network active/passive microwave soil moisture retrieval approach

  • Preprocessing methods to highlight the satellite signal soil moisture information

  • The synergy of active/passive data improves the temporal correlations by 19%.

  • Data fusion is found to be the optimal synergy strategy.

Abstract

A study is performed to analyze the daily retrieval of soil moisture from the synergy of active and passive microwave data using observations from the ASCAT scatterometer and AMSR-E radiometer. The objective is to identify the information provided by each sensor and to analyze preprocessing methods – such as the day/night average, diurnal difference and microwave polarization difference index for AMSR-E and the incidence angle normalization and backscatter temporal index for ASCAT – to maximize the amount of soil moisture information extracted. Additionally, the data fusion and a posteriori synergy methodologies are compared to determine how to optimally exploit this combined information. This study is performed using a neural network (NN) to estimate soil moisture from a set of AMSR-E and ASCAT observations. ERA-interim/Land surface soil moisture fields are used to train the NN as well as to evaluate the performance of the different retrieval input datasets. It is shown that using the AMSR-E 7 GHz, 11 GHz, 19 GHz and 37 GHz channels with the three preprocessing methods highlights various surface contributions in the signal, in particular the soil moisture and surface temperature information, and greatly helps the retrieval in disentangling them. For ASCAT, the synergy effect is less significant and data preprocessed with the two methods analyzed yields very similar information. The information provided by the active and passive microwave sensors is found to be very complementary, such that a soil moisture retrieval using the combined active and passive information shows a significant improvement between 5% and 19% in the temporal correlation and a reduction of the retrieval uncertainty by 7%. The improvement in the spatial structure is smaller with a correlation increase of 2%. It is demonstrated that the choice of synergy method strongly impacts the retrieval improvement that can be achieved. Data fusion methods are shown to be better suited than a posteriori combination methods, due to their ability to exploit information complementarity. These results could help improve future active/passive soil moisture retrievals, such as from the Soil Moisture Active/Passive (SMAP) mission, through the application of similar preprocessing and synergy methods in order to better extract the soil moisture information provided.

Introduction

Soil moisture controls the water partitioning between runoff and infiltration (Assouline, 2013, Corradini et al., 1998, Philip, 1957) as well the surface energy partitioning, especially the ratio of sensible heat flux to latent heat flux (Bowen ratio) and hence it is critical to accurately provide the flux boundary condition to the atmosphere (Bateni and Entekhabi, 2012, Gentine et al., 2007, Gentine, 2009, Gentine et al., 2011b). As such soil moisture is a key variable coupling the land and the atmosphere as well as the energy and water cycles. Additionally, it is an essential component of the carbon cycle (McDowell, 2011, Sevanto et al., 2014). Despite its importance, numerical models still have difficulties accurately representing soil moisture (Koster et al., 2006, Guo et al., 2006, Koster et al., 2011, Orth and Seneviratne, 2013), in large part due to a lack of suitable observations for model evaluation. Consequently, in the last decades, a large amount of research has been directed at retrieving global soil moisture estimates from satellite observations. This involved sensors not optimized for soil moisture observations (Wagner et al., 1999, Owe et al., 2001, Owe et al., 2008, Aires et al., 2005) as well as dedicated sensors such as the Soil Moisture and Ocean Salinity (SMOS) mission (Kerr et al., 2010) and the recently launched Soil Moisture Active/Passive (SMAP) mission (Entekhabi et al., 2010). Most soil moisture retrieval products are derived from microwave observations, since these are more sensitive to soil moisture variations as a result of the link between the soil water content and its dielectric properties (Dobson and Ulaby, 1986, Schmugge et al., 1986). Auxiliary data from other sensors are often used to account for other contributions to the microwave signal or to serve as soil moisture proxies when microwave observations are not available (e.g., infrared data for surface temperature estimation, visible and near-infrared data for the vegetation state, multichannel atmospheric sounder data for the atmospheric state (Prigent, Aires, Rossow, & Robock, 2005)).

In recent years, synergistic approaches have been introduced benefiting from the respective advantages of both passive and active sensors. These include both statistical methods (Aires et al., 2005, Liu et al., 2011, Das et al., 2011, Kolassa et al., 2013) as well as more physical approaches (Chauhan et al., 1994, Colliander and Xu, 2013, Dente et al., 2014). However, such a combination can be complex to implement. A traditional Radiative Transfer Model (RTM) inversion approach typically used for soil moisture retrievals would require a consistent coupled passive and active microwave RTM to merge data from different sensors. This is difficult to realize considering the availability and quality of all the necessary inputs at a global scale. As a result, the combination is often implemented at the retrieval product level rather than at the observation level or, in the case of SMAP, the active sensor data is used to downscale the passive microwave observations (Entekhabi et al., 2010, Das et al., 2011). The disadvantage of these approaches is that the complementarity and interdependence of information from different sensors is not optimally exploited. Here, a Neural Network (NN) based approach is presented, to facilitate data synergy and optimally exploit information complementarity in order to develop a daily soil moisture product (Aires et al., 2012, Kolassa et al., 2013).

When combining observations from different sensors the information content of each dataset needs to be considered. The signal received at the satellite is a combination of contributions from various soil parameters, the vegetation, and – to some extent – the atmosphere. In traditional RTM retrievals, the additional signal contributions are accounted for through the use of auxiliary data (e.g. Wigneron et al., 2007), whose availability and uncertainty impose strong constraints on the retrieval. However, since this information is inherently contained in the satellite observations, the potential of different preprocessing methods, such as polarization or diurnal indices, to highlight separate contributions is investigated here. This should help a soil moisture retrieval to maximize the amount of information and better account for the effects of additional signal contributions. A similar sensor information content analysis has been performed on a monthly scale for a combination of microwave, visible and infrared data by Aires et al. (2005) and Kolassa et al. (2013). On a daily timescale, the availability of visible and infrared surface observations becomes a limiting factor and the signal variability is much larger, which can limit the retrieval capacity. As a consequence, the study presented here aims to analyze the information content and potential of a microwave-only retrieval. In particular with regard to the new SMAP mission, it is of interest to develop strategies for the optimization of active/passive daily soil moisture retrievals without introducing the need for ancillary data, that are often not reliable, model dependent and difficult to obtain. The preprocessing methods investigated here are applied to individual channels, so that the results presented are directly applicable to SMAP, which will operate at L-band and should thus be more sensitive to soil moisture than the AMSR-E channels investigated here. Even more relevant for SMAP and other active/passive retrievals will be the results of the synergy method analysis conducted here, which determine how to optimally combine the information provided by active and passive sensors.

In summary, this paper addresses three questions:

  • 1.

    What is the information contained in the active and passive microwave observations? That is, which soil moisture variations and scales – both spatially and temporally – does each sensor capture and what other surface parameters are they sensitive to?

  • 2.

    How can the information provided by the observations be maximized? More specifically, are there preprocessing techniques that permit to highlight the soil moisture contribution to the satellite signal and help the retrieval disentangle additional contributions without relying on (often uncertain) auxiliary data?

  • 3.

    What is the retrieval benefit and optimal method of an active and passive microwave sensor synergy? Do the two sensor types provide complementary information that improves a retrieval when used in combination? If so, what is the best method to optimally exploit this information complementarity?

The focus of this manuscript is on the preparation of the microwave satellite data for a soil moisture retrieval. After identifying the best preprocessing methods and optimal synergy method, using ERA-Land soil moisture fields as a reference, and using these to compute a daily soil moisture product, this product has been evaluated against in situ observations, satellite retrievals and LSM soil moisture fields. The results of this study will be presented in a separate paper focussing on the presentation and evaluation of the NN soil moisture product.

Section 2 presents the datasets used in this study. Section 3 describes the methodology employed to answer the questions introduced above and Section 4 presents and discusses the findings of the analyses. Finally, Section 5 summarizes the conclusions of the study and provides perspectives for future analyses.

Section snippets

Datasets

The analysis presented here has been performed using data from two satellite microwave instruments, one active (ASCAT) and one passive (AMSR-E), as well as modeled soil moisture fields from ERA-interim/Land (see detail in Section 2.3). The study focusses on the overlap period of the two sensors, from August 2007 to October 2011. All data have been projected onto an equal area grid with a 0.25° resolution at the equator. All observations available for a given location and day (as a result of the

Methodology

The main objective of this study is to assess the soil moisture information content of different satellite observations and determine how it can be increased through various preprocessing methods as well as the exploitation of their complementarity. The general approach employed is to first calibrate a NN retrieval on a subset of the satellite observations and corresponding modeled soil moisture and subsequently estimate daily soil moisture from the complete set of satellite observations. The

Results and discussion

The following sections discuss the results of the preprocessing and synergy method analysis of AMSR-E and ASCAT data for a soil moisture retrieval. The purpose of the study presented here is to determine the information content and type that can be extracted from active and passive microwave observations, in particular whether they can be used to maximize the soil moisture information provided to a retrieval. Additionally, two methods for combining multi-instrument data were analyzed to

Conclusions

An investigation of the different AMSR-E channels and various preprocessing methods showed that each captures a different aspect of the overall surface signal and thus used in combination they greatly improve the retrieval quality. It was shown that the AMSR-E day/night average best captures soil moisture spatial variations, whereas the day/night difference is better suited to capture temporal variations. Additionally, the day/night difference helps to distinguish between very different regimes

Acknowledgments

The authors would like to acknowledge funding from NASA-ROSES grant NNX15AB30G entitled Development of a Neural Network Scheme for SMAP Retrieval of Soil Moisture at the Global Scale and Assimilation into NWP Centers. Furthermore, some of the developments have been supported by SMOS + Neural Networks ESA ESRIN project under contract 4000105455/12.

References (63)

  • C. Albergel et al.

    A bare ground evaporation revision in the ECMWF land-surface scheme: Evaluation of its impact using ground soil moisture and satellite microwave data

    Hydrology and Earth System Sciences

    (2012)
  • R.L. Armstrong et al.

    Northern Hemisphere EASEGrid weekly snow cover and sea ice extent version 3

    (2005)
  • S. Assouline

    Infiltration into soils: Conceptual approaches and solutions

    Water Resources Research

    (2013)
  • G. Balsamo et al.

    ERA-Interim/Land: A global land-surface reanalysis based on ERA-Interim meteorological forcing

  • G. Balsamo et al.

    A revised hydrology for the ECMWF model: Verification from field site to terrestrial water storage and impact in the Integrated Forecast System

    Journal of Hydrometeorology

    (2009)
  • G. Balsamo et al.

    Evolution of land surface processes in the IFS

    ECMWF Newsletter

    (2011)
  • S.M. Bateni et al.

    Relative efficiency of land surface energy balance components

    Water Resources Research

    (2012)
  • S.M. Bateni et al.

    Mapping evaporation and estimation of surface control of evaporation using remotely sensed land surface temperature from a constellation of satellites

    Water Resources Research

    (2013)
  • S. Boussetta et al.

    Impact of a satellite-derived Leaf Area Index monthly climatology in a global Numerical Weather Prediction model

    International Journal of Remote Sensing

    (2013)
  • N. Chauhan et al.

    Use of discrete scatter model to predict active and passive microwave sensor response to corn: Comparison of theory and data

    IEEE Transactions on Geoscience and Remote Sensing

    (1994)
  • A. Colliander et al.

    Normalized residual scattering index applied to aquarius L-band measurements

    IEEE Geoscience and Remote Sensing Letters

    (2013)
  • G. Cybenko

    Approximation by superpositions of a sigmoidal function

    Mathematics of Control, Signals, and Systems

    (1989)
  • N.N. Das et al.

    An algorithm for merging SMAP radiometer and radar data for high-resolution soil–moisture retrieval

    IEEE Transactions on Geoscience and Remote Sensing

    (2011)
  • M.C. Dobson et al.

    Active microwave soil moisture research

    IEEE Transactions on Geoscience and Remote Sensing

    (1986)
  • W.A. Dorigo et al.

    Error characterisation of global active and passive microwave soil moisture datasets

    Hydrology and Earth System Sciences

    (2010)
  • E. Dutra et al.

    An improved snow scheme for the ECMWF land surface model: Description and offline validation

    Journal of Hydrometeorology

    (2010)
  • D. Entekhabi et al.

    The soil moisture active passive (SMAP) mission

    Proceedings of the IEEE

    (2010)
  • J. Figa-Saldaña et al.

    The advanced scatterometer (ASCAT) on the meteorological operational (MetOp) platform: A follow on for European wind scatterometers

    Canadian Journal of Remote Sensing

    (2002)
  • B.A. Forman et al.

    Estimating passive microwave brightness temperature over snow-covered land in North America using a land surface model and an artificial neural network

    IEEE Transactions on Geoscience and Remote Sensing

    (2014)
  • P. Gentine

    Spectral behavior of the coupled land-atmosphere system

    (2009)
  • P. Gentine et al.

    The diurnal behavior of evaporative fraction in the soil–vegetation–atmospheric boundary layer continuum

    Journal of Hydrometeorology

    (2011)
  • Cited by (0)

    View full text