Radiometric calibration of high resolution UAVSAR data over hilly, forested terrain

SAR backscatter data contain both geometric and radiometric distortions due to underlying topography and the radar viewing geometry. Thus, applications using SAR backscatter data for deriving various scientific products (e.g. above ground biomass) require accurate absolute radiometric calibration. The calibration process involves estimation of the local radar scattering area through knowledge of the imaged terrain, which is often obtained through DEMs. High resolution UAVSAR data over a New Hampshire boreal forest test site was radiometrically calibrated using a low resolution SRTM DEM, and different calibration methods were tested and compared. Heteromorphic methods utilizing DEM integration are able to model scattering area better than homomorphic methods based on the local incidence or projection angle with a resultant backscatter calibration difference of less than 0.5 dB. Additionally, the impact of low DEM resolution on the calibration was investigated through a Fourier analysis of different topographic classes. Power spectra of high-resolution airborne lidar DEMs were used to characterize the topography of steep, moderate, and flat terrain. Thus, errors for a given low resolution DEM associated with a particular topographic class could be quantified through a comparison of its power spectrum with that from the lidar. These errors were validated by comparing DEM slope derived from SRTM and lidar DEMs.

The impact of radiometric calibration on the biomass retrieval capabilities of UAVSAR data was investigated by fitting second-order polynomials to backscatter vs. biomass plots for the HH, HV, and VV polarizations. LVIS RH50 values were used to calculate biomass, and the process was repeated for both uncalibrated and area calibrated UAVSAR images. The calibration improved the $R2$ values for the polynomial fits by 0.7-0.8 for all three polarizations but had little effect on the polynomial coefficients. The Fourier method for predicting DEM errors was used to predict biomass errors due to the calibration. It was revealed that the greatest errors occurred in the near range of the SAR image and on slopes facing towards the radar.