Models of L-Band Radar Backscattering Coefficients Over Global Terrain for Soil Moisture Retrieval

Physical models for radar backscattering coefficients are developed for the global land surface at L-band (1.26 GHz) and 40 ° incidence angle to apply to the soil moisture retrieval from the upcoming soil moisture active passive mission data. The simulation of land surface classes includes 12 vegetation types defined by the International Geosphere-Biosphere Programme scheme, and four major crops (wheat, corn, rice, and soybean). Backscattering coefficients for four polarizations (HH/VV/HV/350611873VH) are produced. In the physical models, three terms are considered within the framework of distorted Born approximation: surface scattering, double-bounce volume-surface interaction, and volume scattering. Numerical solutions of Maxwell equations as well as theoretical models are used for surface scattering, double-bounce reflectivity, and volume scattering of a single scatterer. To facilitate fast, real-time, and accurate inversion of soil moisture, the outputs of physical model are provided as lookup tables (with three axes; therefore called datacube). The three axes are the real part of the dielectric constant of soil, soil surface root mean square (RMS) height, and vegetation water content (VWC), each of, which covers the wide range of natural conditions. Datacubes for most of the classes are simulated using input parameters from in situ and airborne observations. This simulation results are found accurate to the co-pol RMS errors of to 3.4 dB (six woody vegetation types), 1.8 dB (grass), and 2.9 dB (corn) when compared with airborne data. Validated with independent spaceborne phased array type L-band synthetic aperture radars and field-based radar data, the datacube errors for the co-pols are within 3.4 dB (woody savanna and shrub) and 1.5 dB (bare surface). Assessed with spaceborne Aquarius scatterometer data, the mean differences range from ~ 1.5 to 2 dB. The datacubes allow direct inversion of sophisticated forward models without empirical parameters or formulae. This capability is evaluated using the time-series inversion algorithm over grass fields.