Evaluation of empirical and semi-empirical backscattering models for surface soil moisture estimation J. Álvarez-Mozos, M. González-Audícana, and J. Casalí Abstract. Several empirical and semi-empirical backscattering models have been proposed to offer alternative expressions for the inversion of surface parameters from radar data, but the applicability and adequacy of the models for different surface conditions and sensor configurations have not been clearly assessed. A number of empirical and semi-empirical models are studied in this paper to assess the applicability of the models for different conditions that are often found over agricultural areas. The performance of the models is evaluated first analytically by comparing their simulations with those obtained using the theoretical integral equation model (IEM) and geometrical optics model (GOM). The model estimations are then compared with RADARSAT-1 observations acquired over an experimental catchment. The results show very different model behaviour depending on the surfaces roughness conditions and incidence angle. This study highlights the importance of carefully selecting the backscattering model to be used in radar applications. Résumé. Plusieurs modèles empiriques et semi-empiriques de rétrodiffusion ont été proposés pour offrir des expressions alternatives à l’inversion des paramètres de surface à partir des données radar. Cependant, l’applicabilité et la pertinence des modèles aux différentes conditions de surface et configurations de capteurs n’ont pas été évaluées de façon satisfaisante. Dans cet article, nous étudions un nombre de modèles empiriques et semi-empiriques en vue d’évaluer leur applicabilité aux diverses conditions souvent observées en milieu agricole. La performance des modèles est évaluée tout d’abord de façon analytique, c’est-à-dire en comparant leurs simulations avec celles obtenues avec les modèles théoriques IEM (« integral equation model ») et GOM (« geometrical optics model »). Ensuite, les estimations des modèles sont comparées avec des observations RADARSAT-1 acquises au-dessus d’un bassin versant expérimental. Les résultats montrent un comportement très différent des modèles selon les conditions de rugosité de surface et l’angle d’incidence. Cette étude souligne l’importance de bien sélectionner le modèle de rétrodiffusion à utiliser dans les applications radar. [Traduit par la Rédaction] Álvarez-Mozos et al. 188 Introduction Surface soil moisture (SM) is a variable that plays a crucial role in many processes occurring at the soil–atmosphere interface. The knowledge of the moisture content of the soil over a field or a catchment can be very helpful for hydrological, agronomic, and meteorological applications (Schmugge et al., 2002; Moran et al., 2004). However, soil moisture characterization is a complicated task due to its high spatial and temporal variability (Wilson et al., 2004). In addition, most soil moisture measuring devices developed so far consist of point-based probes. Therefore, estimating the moisture content of fields or larger areas by means of remote sensing observations is at present an attractive challenge. Soil moisture sensing can best be approached using either passive or active microwave sensors (Du et al., 2000; Schmugge et al., 2002). However, passive sensors have a very coarse spatial resolution and are thus limited to small-scale applications. Consequently, active microwave (radar) sensors represent the best alternative for hydrological and agronomic applications. The backscattering coefficient, σ 0 , obtained from radar sensors is directly related to the dielectric properties of the soil surface being observed, which in turn are mainly dependent on its moisture content (Ulaby et al., 1986). Radar-based SM retrieval has been intensively studied in the last decades. Three main approaches have been generally followed (Moran et al., 2004): (i) empirical linear regression models relating the backscattering coefficient to SM which are valid for invariant roughness, vegetation, and scene-acquisition conditions (Glenn and Carr, 2004; Álvarez-Mozos et al., 2005); (ii) change detection techniques for monitoring SM dynamics, assuming that surface roughness and vegetation cover change more slowly than does SM (Wickel et al., 2001); and (iii) electromagnetic scattering models that simulate the surface backscattering process and can be inverted to retrieve SM (Ulaby et al., 1986; Fung, 1994). The first two approaches have limited validity because they require the surface characteristics apart from SM to remain unchanged and the sensor parameters to be exactly the same. If identical sensor parameters were needed, the revisit time of most sensors would be on the order of several weeks, which is generally insufficient for most 176 © 2007 CASI Can. J. Remote Sensing, Vol. 33, No. 3, pp. 176–188, 2007 Received 22 May 2006. Accepted 22 March 2007. Published on the Canadian Journal of Remote Sensing Web site at http://pubs.nrc-cnrc.gc.ca/cjrs on 18 July 2007. J. Álvarez-Mozos, 1 M. González-Audícana, and J. Casalí. Department of Projects and Rural Engineering, Public University of Navarre, Arrosadia s/n, 31006 Pamplona, Spain. 1 Corresponding author (e-mail: jesus.alvarez@unavarra.es).