Extending RAPID model to simulate forest microwave backscattering

摘要

The retrieval of vegetation parameters benefits significantly from the data fusion of optical and microwave signals. The integration of accurate forward models in both regions can play an important role in supporting these fusion approaches. Because of the different imaging mechanisms used in optical and microwave wavelength domains, the forward models in the two domains have been generally developed separately based on the different specifications of the scene. The inconsistencies between optical and microwave models, such as confusing input/output parameter definitions, different scattering theories and discrepant model complexity, make the data fusion difficult to conduct and lead to different results in terms of accuracy and computer time. Therefore, it is of great interest to develop a unified three-dimensional (3D) model using one scattering theory, identical input and similar complexity. To our knowledge, there are very few 3D models that can accomplish this task. By extending the Radiosity Applicable to Porous IndiviDual Objects (RAPID) model for the optical region, a general radiosity model (RAPID2) was proposed in this paper for the microwave region. This is the first time radiosity theory has been applied in microwave remote sensing, which invents a new way to solve the radar multiple scattering more efficiently. RAPID2 has four new functions: projecting translucent objects, tracking specular scattering, separating polarization components and imaging radar signals. The relationship between the radar cross section (RCS) and the bi-directional reflectance factor (BRF) is bridged. The modified Stokes vector and Mueller matrix are integrated into radiosity formulas to unify the scattering process between the optical and microwave regions. RAPID2 can simulate double-bouncing and multiple scattering effects over vegetated 3D scenes containing topography. The simulated radar images can well reflect the distinct radar geometric features, including layover, foreshortening and shadows. Validation over two forest sites shows good agreement with AIRSAR backscattering data (errors 3.4 dB). The demonstrated results show the importance of the incident azimuth angle (variation up to 2 dB), slope (variation up to 5 dB), and multiple scattering effects (contribution up to 2 dB), which should be considered in forest parameter inversion.