On the ability of AirSAR to measure patterns of dielectric constant at the hillslope scale

This paper assesses the potential for measuring spatial patterns of dielectric constant using AirSAR images at high spatial resolutions (up to 10 m pixel sizes). Dielectric constant is strongly related to soil moisture. Data from the 10 ha Tarrawarra experimental catchment in south-eastern Australia are used. Ground data collected include detailed spatial patterns of dielectric constant measured with time domain reflectometry on a 10 m grid over the 10 ha catchment (∼1000 measurements), as well as surface roughness and vegetation characteristics. The average dielectric constant was 16.9, soil moisture was 29.6% volumetric, rms surface roughness height was 1.7 cm and average pasture biomass was 0.18 kg/m2. At the 30 m pixel scale, empirical comparisons between the ground-measured dielectric constant and various band and polarization combinations result in R2 values of 0.22–0.43 for L- and P-band, but were 0.00 for C-band. Using the Dubois model [IEEE Trans. Geosci. Remote Sensing 33 (1995a) 915; IEEE Trans. Geosci. Remote Sensing 33 (1995b) 1340)] in forward mode and comparing backscatter coefficients improved the R2 for L-band to 0.31 and 0.37 for hh and vv polarizations, respectively. Some systematic underestimation occurred. Using the L-band images, the Dubois model operated in inverse mode produced an acceptable estimate of the spatial mean dielectric constant (and soil moisture) but was unable to extract any useful information on the spatial pattern at the 30 m pixel scale. An analysis of the change in error with pixel scales between 10 and ∼100 m suggests that aggregation up to a pixel size of about 50 m is required to remove most of the random error in the inverted dielectric constant (soil moisture) estimate.