Elsevier

Remote Sensing of Environment

Volume 162, 1 June 2015, Pages 186-195
Remote Sensing of Environment

Evaluation of multi-frequency bare soil microwave reflectivity models

https://doi.org/10.1016/j.rse.2015.02.015Get rights and content

Highlights

  • Multi-frequency, multi-angular rough bare soil reflectivity modelling

  • The GRMDM soil reflectivity model is more accurate than the SMDM model.

  • The GRMDM and SMDM models at higher frequencies (> 10 GHz) are not accurate.

  • The WM99 and QHN model gives similar RMSE.

  • QHN is better for soil parameterization because of less parameters to calibrate.

Abstract

Passive microwave remote sensing is commonly used to monitor the hydrological processes of the earth's surface (soil moisture) as well as key surface environmental processes (vegetation dynamics, snow cover, etc.). For many applications, it is important to model the effects of surface roughness. This study focuses on the unique PORTOS-93 measurement campaign that covered a wide range of bare soil conditions in terms of moisture, temperature and surface roughness. The PORTOS-93 campaign covers a frequency range from 1.41 to 90 GHz. In this study, based on the PORTOS-93 dataset, we compare the Wegmüller & Mätzler (1999) model (referred to as WM99) with the Wang & Choudhury (1981) model (referred to as QHN) and evaluate their abilities to simulate the soil surface brightness temperature (TB). We show that improved results were obtained by tuning the parameters of the two models to the entire PORTOS-93 dataset for each frequency separately compared to the model tuned independently of frequency. In addition we found that the Mironov et al. (2009) soil permittivity model is slightly more accurate at lower frequencies (< 11 GHz) for simulating soil permittivity than the Dobson et al. (1985) soil permittivity model for the PORTOS-93 site. We also found that both of these permittivity models need to be tuned at higher frequencies (> 20 GHz) and that using the Wang & Schmugge (1980) model tuned using parameters derived by Calvet et al. (1995a) at higher frequencies for the PORTOS-93 dataset gives best results. Using the proposed tuned reflectivity models for each frequency separately, we obtained an overall bias of − 0.20 K and 0.21 K for the WM99 and QHN models respectively between the measured and modelled brightness temperatures. This corresponds to a significant improvement in comparison to the use of the tuned WM99 model independently of the frequency (bias =  38.08 K).

Introduction

Passive microwave remote sensing is commonly used to monitor hydrological processes as in the European Space Agency's (ESA) Soil Moisture and Ocean Salinity (SMOS) mission (Kerr et al., 2001) and the National Aeronautics and Space Administration's (NASA) upcoming Soil Moisture Active and Passive (SMAP) mission (Brown et al., 2013). These missions mainly concentrate on L-band (1.41 GHz) microwave brightness temperatures (TB) to retrieve the soil moisture (SM). In order to extract SM, the contribution from the soil to the overall TB has to be isolated from other contributions such as the atmosphere (Kerr & Njoku, 1990), the vegetation (Wigneron et al., 2003) and the snow cover (Rautiainen et al., 2012). This signal also needs to take into account the soil surface properties such as surface roughness (Escorihuela et al., 2007, Wigneron et al., 2011) and soil texture (Njoku and Entekhabi, 1996, Wigneron et al., 2003).

The NASA National Snow and Ice Data Center (NSIDC) and the Japan Aerospace Exploration Agency (JAXA) provide standard soil moisture products from the Advanced Microwave Scanning Radiometer—Earth Observing System (AMSR-E) (Njoku et al., 2003, Shibata et al., 2003). The NSIDC AMSR-E Level 3 soil moisture product (Njoku et al., 2003) is based on an inversion algorithm from the 10.7 GHz and 18.7 GHz brightness temperature data using the empirical soil permittivity model of Wang and Schmugge (1980) and semi-empirical equations based on the Wang and Choudhury (1981) roughness model with three free parameters (hereinafter referred to as the QHN model): the roughness height (HR), a polarisation mixing parameter (QR) and a parameter accounting for the angular dependency of the reflectivity (NR). The JAXA product is based on the discrete ordinate method proposed by Tsang and Kong (1977) and the soil permittivity model of Dobson, Ulaby, Hallikainen, and El-Rayes (1985), hereafter referred to as the Soil Moisture Dielectric Mixing (SMDM) model. Validation studies show that the algorithms currently in use by JAXA and NASA still need improvements (see Jackson et al., 2010).

Studies for other environmental applications (land surface parameter retrieval such as vegetation and snow cover dynamics, or vegetation water content) based on higher frequencies (from 6 to 90 GHz) need accurate estimates of the soil signal under the canopy or the snow (e.g. Mätzler, 2006). In most of these studies, the soil contribution brings uncertainties that must be taken into account to improve the retrievals (see for example Calvet et al., 2011, Prigent et al., 2000).

Ground-based microwave experiments in combination with ground-truth measurements have been conducted to elaborate methods to relate SM to the L-band TB such as PORTOS (Wigneron, Laguerre, & Kerr, 2001), SMOSREX (de Rosnay et al., 2006) and also airborne campaigns such as SMEX03 (Jackson et al., 2005), CanEx-SM10 (Magagi et al., 2013) and SMAPEx (Panciera et al., 2014). These studies have mainly proposed soil surface roughness corrections to the L-band TB (Escorihuela et al., 2007, Schwank et al., 2010, Shi et al., 2002, Wang and Choudhury, 1981, Wigneron et al., 2011) but only a few studies have tried to apply these corrections to TB at higher frequencies (> 1.4 GHz, Calvet, Wigneron, Chanzy and Haboudane, 1995, Prigent et al., 2000, Wang et al., 1983, Wegmüller and Mätzler, 1999).

For example the PORTOS-93 (Wigneron et al., 2011) and SMOSREX (Escorihuela et al., 2007) campaigns were conducted to relate the bare soil geophysical parameters (soil moisture, temperature, surface roughness, textural composition, etc.) to the simulated soil reflectivity at L-band using the QHN roughness model. Analysis of the PORTOS-93 data showed that the last two parameters could be neglected (NR = QR = 0) with a low impact on the accuracy of the SM retrievals (Lawrence, Wigneron, Demontoux, Mialon, & Kerr, 2013). The SMOSREX study also showed that QR could be neglected but suggested that NR had to be specific for a given polarisation and found that NRV =  1 and NRH = 1 (Escorihuela et al., 2007). Nonetheless, the values of NRV and NRH at L-band are not yet well established (Lawrence et al., 2013, Wigneron et al., 2007). Prigent et al. (2000) have shown that for higher frequencies (23.8 GHz and higher), to use such a QHN approach, all three parameters (HR, QR and NR) had to be considered. They have also shown that all three parameters were dependent upon the surface properties and type (smooth soil, rough soil or covered with vegetation) meaning that these parameters needed to be tuned for each area and can vary in time depending on the surface type (texture, roughness, etc.). Goodberlet and Mead (2014) suggested that to model the effects of soil surface roughness at L-band, the QR parameter should be considered and that it is related empirically to the surface roughness. Also, they suggested that HR depended not only on the surface roughness but also on the Fresnel reflectivity (based on the soil surface permittivity which depends on soil moisture). This last result differs from previous studies. Finally, they showed that NRV =  2 and NRH = 1 which is similar to what Escorihuela et al. (2007) suggested. These different studies showed that there is a need to further investigate the effects of surface roughness on the soil reflectivity since there are still many questions concerning the calibration of the surface roughness parameters.

Other studies focusing mainly on vegetated areas used a similar approach to extract soil surface parameters from passive microwave TB at higher frequencies. Calvet, Wigneron, Chanzy, and Haboudane (1995) showed that it is possible to retrieve QR and HR at 23.8, 36.5 and 90 GHz for fields covered by sorghum and wheat canopies and estimate the soil contribution to the measured TB. The latter study also showed that the soil dielectric properties could be modelled for silt–loam soils using the model of Wang and Schmugge (1980) (hereafter referred to as WS80), provided that it has been calibrated to the specific conditions of the fields. In this study, the three frequencies had to be treated separately.

Pellarin, Kerr, and Wigneron (2006) simulated TB at C- (6.6 GHz) and X-bands (10.7 GHz) at global scale using the QHN model and have evaluated these simulations with the Scanning Multichannel Microwave Radiometer (SMMR) satellite TB. They showed that the QHN model was able to reproduce realistic values of TB at a global scale. Roy et al. (2012) showed that an optimization of the reflectivity model on the AMSR-E TB converged to specific values of HR and QR at 19 and 37 GHz over Canadian boreal forest sites. These last studies were conducted over vegetated areas where the soil contribution to the measured satellite TB is attenuated by the overlaying canopy making the retrieval of soil parameters more complex.

The most complete study at higher frequencies (> 10 GHz) was done by Wegmüller and Mätzler (1999) who developed a reflectivity model (hereinafter referred to as the WM99 model) that is based on the Mo and Schmugge (1987) parameterization for the frequency range of 1 to 100 GHz and incidence angle range of 20 to 70° for rough bare soils. They showed that the vertical (V-Pol) and horizontal polarisation (H-Pol) reflectivities were strongly correlated and that only one polarisation had to be modelled (either V or H) as a function of the soil variables, while the other could be derived from the former. Mo and Schmugge (1987) and Wegmüller and Mätzler (1999) found that it is preferable to model the H-Pol reflectivity as a function of the soil variables (soil roughness, moisture and temperature) and, in a second step, the V-Pol reflectivity can be computed from the modelled H-Pol reflectivity. Contrary to what was discussed by Calvet, Wigneron, Chanzy, Raju, and Laguerre (1995), they did not consider the reflectivities at different frequencies separately. This is mainly due to the fact that Wegmüller and Mätzler (1999) used the SMDM soil dielectric model (Dobson et al., 1985), whereas Calvet, Wigneron, Chanzy, Raju, and Laguerre (1995) used an empirical soil dielectric permittivity mixing model (Wang & Schmugge, 1980). More recently, a new soil permittivity model, referred to as the Generalized Refractive Mixing Dielectric Model (GRMDM), was developed (Mironov et al., 2013, Mironov et al., 2009). Recent studies (Goodberlet and Mead, 2014, Mialon et al., 2014, Wigneron et al., 2011, Wigneron et al., 2012) found that this new model provides accurate simulations of the soil dielectric constant in comparison to the SMDM model at L-band.

Here, we propose to evaluate the WM99 and QHN models using the unique PORTOS-93 multi-angular, bi-polarisation and multi-frequency dataset. To do so, 1) an evaluation of the permittivity modelling based on the SMDM, GRMDM and WS80 models was made in the 1–90 GHz range of frequencies, 2) then, we evaluated and compared the two semi-empirical soil reflectivity models (WM99 and QHN) and 3) finally a comparison between different tuned approaches of the WM99 and QHN models was made.

Section snippets

Experimental dataset

The PORTOS-93 dataset is thoroughly described in Wigneron et al. (2001) and only a short description of the measurements will be given here. The measurements were taken over seven bare fields at the Institut National de la Recherche Agronomique (INRA) Avignon Remote Sensing test site during the period of April 20th to July 10th, 1993 (Table 1). The sites are silty clay loam fields with a textural composition of 62% silt, 11% sand and 27% clay. The surface roughness parameters are given in

Evaluation of the permittivity models

To evaluate the performance of the SMDM, GRMDM and WS80 models, the soils with a smooth surface were chosen (σ < 5 mm and l > 60 mm, e.g. sites 9 and 17 in Table 1). The modelled reflectivities using the different permittivity models and the Fresnel equations were compared to the measured reflectivities at all the five frequencies and for both H and V polarisations (Fig. 1).

Table 2 presents the mean biases and the associated standard deviations between the modelled and measured reflectivities shown

Discussion

In this section, the results highlighted in the previous section are discussed in more details.

Conclusion

The objective of this study was to evaluate two soil reflectivity models using the unique multi-angular and multi-frequency PORTOS-93 bare soil measurements and to obtain a simple and accurate surface reflectivity model to simulate the brightness temperature (TB). Considering the TB measurements at H-Pol at 1.41, 10.65, 23.8, 36.5 and 90 GHz over smooth soil surfaces, it was found that the GRMDM model (Mironov et al., 2009) is more accurate to simulate the soil permittivity than the SMDM model (

Acknowledgements

The authors would like to thank Dr. Alexandre Roy for his helpful discussion which contributed to this paper. We also acknowledge the contribution of the National Sciences and Engineering Research Council of Canada for their funding, and INRA for the POSTOS-93 database.

References (43)

  • J.-C. Calvet et al.

    Sensitivity of passive microwave observations to soil moisture and vegetation water content: L-band to W-band

    IEEE Transactions on Geoscience and Remote Sensing

    (2011)
  • D.P. Dee et al.

    The ERA-Interim reanalysis: configuration and performance of the data assimilation system

    Quarterly Journal of the Royal Meteorological Society

    (2011)
  • M.C. Dobson et al.

    Microwave dielectric behavior of wet soil — Part II: Dielectric mixing models

    IEEE Transactions on Geoscience and Remote Sensing

    (1985)
  • M. Escorihuela et al.

    A simple model of the bare soil microwave emission at L-band

    IEEE Transactions on Geoscience and Remote Sensing

    (2007)
  • M. Goodberlet et al.

    Improved models of soil emission for use in remote sensing of soil moisture

    IEEE Transactions on Geoscience and Remote Sensing

    (2012)
  • M.A. Goodberlet et al.

    A model of surface roughness for use in passive remote sensing of bare soil moisture

    IEEE Transactions on Geoscience and Remote Sensing

    (2014)
  • T. Jackson et al.

    Polarimetric scanning radiometer C- and X-band microwave observations during SMEX03

    IEEE Transactions on Geoscience and Remote Sensing

    (2005)
  • T.J. Jackson et al.

    Validation of advanced microwave scanning radiometer soil moisture products

    IEEE Transactions on Geoscience and Remote Sensing

    (2010)
  • Y.H. Kerr et al.

    Semiempirical model for interpreting microwave emission from semiarid land surfaces as seen from space

    IEEE Transactions on Geoscience and Remote Sensing

    (1990)
  • Y. Kerr et al.

    Soil moisture retrieval from space: The soil moisture and ocean salinity (SMOS) mission

    IEEE Transactions on Geoscience and Remote Sensing

    (2001)
  • H. Lawrence et al.

    Evaluating the semiempirical H–Q model used to calculate the L-band emissivity of a rough bare soil

    IEEE Transactions on Geoscience and Remote Sensing

    (2013)
  • Cited by (41)

    View all citing articles on Scopus
    View full text