SM Estimation Under Tropical Forests Using UHF

In this paper, we report on the performance of a semiempirical algorithmfor the retrieval of soil moisture (SM) under dense tropical forests using ultrahigh frequency (UHF) polarimetric synthetic aperture radar (SAR) data. The algorithm is a simplification of a 3-D coherent model of forest canopy based on the distorted Born approximation (DBA). The simplified model reduces the number of parameters and preserves the three dominant scattering mechanisms of volume, volume-surface, and surface for three polarized backscattering coefficients, i.e., σ_HH, σ_HV, and σ_VV, at UHF frequencies. The inversion process uses the Levenberg-Marquardt nonlinear least squares method to estimate the three model parameters: vegetation aboveground biomass, integrated SM up to a certain depth, and surface roughness. The performance of the inversion process is examined by first using simulation data where the initial values of the inversion process vary randomly and then using airborne UHF SAR data acquired in Costa Rica over La Selva Biological Station. The results with simulated data show that the inversion process is not significantly sensitive to initial values considering they are in the range of ±50% of the true value. A root-mean-square error (RMSE) of less than 4% can be achieved in retrieving the SM.

The use of an alternate inversion approach without initial conditions using a genetic algorithm is less efficient (> 120 times longer time) and produces larger error with simulated data (RMSE = 11%) than the Levenberg-Marquardt estimation method. The inversion model simultaneously produces a biomass and SM distribution at 100-m spatial resolution. The RMSE of biomass estimation is 38 Mg/ha (15% relative error) when compared with 28 field plots. Over the plots where SM ground measurements are available, but not at the exact same day as the radar flight occurred, the total volumetric RMSE is 13.6%. However, only two gr- und measurements were very close to the flight day (three days apart), and for those, the SM estimate has about 3% absolute volumetric error. At the P-band, the SM sensing depth is inversely correlated with the SM allowing to map the spatial variations of SM close to the average root zone or hydrological active horizon of soils in tropical ecosystems.