Surface soil moisture estimation using IEM model with calibrated roughness

Abstract: The ENVISAT/ASAR image is an important remote sensing data source for estimating soil moisture, and the integral equation model (IEM) is the most widely used, physically based radar backscatter model for bare soil and sparsely vegetated landscapes. However, the soil moisture retrieval from ASAR images using the IEM is not fully operational at present, mainly due to the difficulties in the parameterization of soil surface roughness and the elimination of spatial and temporal variation of soil roughness. The IEM simulated backscattering coefficients are often in poor agreement with satellite radar measurements because of un-accurate description of the surface roughness, especially the correlation length l parameter. Baghdadi proposed to replace correlation length l with a fitted parameter lopt for the IEM, which can be expressed as the function of root mean square height hRMS and incidence angle. So far, there is still lack of application of this method in semi-arid areas. This paper applied this approach in the Walnut Gulch Experimental Watershed of southeast Arizona, and showed that the IEM performed better in simulating radar backscattering coefficient when lopt was used as the input. Based on the improvement in radar backscattering coefficient simulation, lopt and hRMS are replaced by the combined roughness Zs (hRMS2/ lopt), and the relationship between surface roughness Zs, soil moisture and the simulated backscatter coefficients is analyzed. The results showed that the simulated backscattering coefficient was logarithmically correlated with both Zs and soil moisture. Then, maps of Zs in two dates are estimated with a logistic regression equation using the difference between backscattering coefficients at incidence angles of IS6 and IS2. Using Zs estimates and IEM simulated backscattering coefficients, the empirical formula of soil moisture inversion under two incidence angles was established with the nonlinear least squares method for VV (vertical vertical) polarization mode. On analyzing the parametric formula of simulated IEM data, a semi-empirical method was further applied based on Taylor series expansion. Therefore, two surface roughness and two soil moisture maps are obtained using ASAR images in two dates, i.e., August 18 and August 24, 2004. Comparison between the surface roughness maps in two dates shows that the surface roughness has similar spatial distribution characteristics, but the surface roughness on August 18 was less than that on August 24. Dynamic changes of the surface roughness in two dates are consistent with the occurrence of rainfall events. Comparison between the estimated soil moisture with observations of 19 stations in the Walnut Gulch watershed shows that the correlation coefficients were 0.785 and 0.837 between the observed and the empirically estimated soil moisture, and 0.900 and 0.863 between the observed and the semi-empirically estimated soil moisture, for August 18 and August 24 respectively. It means that both the empirical method and the semi-empirical method are effective, but the semi-empirical method performs better. The method quantifies the impact of surface roughness on IEM model simulations and the influence of roughness change on surface roughness estimation, which is effective for retrieving soil moisture at the watershed scale.