Structural Biomass Estimation from L-band Interferometric SAR and Lidar

After a review of biomass estimation from interferometric SAR (InSAR) at all bands over the last 15 years, and a brief review of lidar biomass estimation, this paper discusses structure and biomass estimation from simultaneously acquired (not repeat-track) InSAR at L-band. We will briefly discuss the history of regression of biomass to InSAR raw observations (coherence and phase) and structural parameters (height, standard deviation, Fourier component). Lidar biomass estimation from functions of the waveform will be discussed. We review our structural and biomass estimation results for C-band InSAR at vertical polarization for 12-14 baselines in La Selva Biological Station, Costa Rica. C-band vertical scales were between 12 and 100 m for structure estimation, but only between 50 and 100 m for biomass estimation, due to phase calibration problems at the shorter vertical wavelengths (larger baselines). Most of the talk will be spent on L-band, simultaneously acquired multibaseline InSAR, also at La Selva, acquired at vertical polarization. Because the vertical interferometric scale is proportional to the radar altitude times the wavelength over the baseline length, the AirSAR aircraft had to be flown very low (1.2 km) to realize vertical scales at L-band of 60 m and higher. Our lidar biomass estimation suggests that vertical scales of 14 m-100 m are optimal for biomass estimation. We will try three different approaches to biomass estimation with the limited high vertical scales we have available: 1) We will regress biomass to Fourier transforms as in the C-band and lidar study, but with 60 m - 100+ m vertical scales we do not expect accuracies to be as high as for the lidar demonstration (58 Mg/ha RMS scatter of estimated about field biomass for biomasses up to 450 Mg/ha), which used Fourier vertical wavelengths of 15 m-20 m. In addition to using Fourier components, 2) we will report the use of the derivative of the InSAR complex coherence with respect to Fourier frequency or vertical scale for the very short baselines (high vertical scales). At very long scales (low vertical spatial frequency), this derivative is close to the radar-power-weighted average height, to which we will regress field-measured biomass. Since we have many short baselines, we will try to use the second derivative of InSAR coherence (Fourier transform) with respect to vertical frequency to estimate radar-power vertical standard deviation and will regress biomass to that as well. 3) We will estimate average vegetation heights (as opposed to radar-power heights) from the available short baselines, attempting to separate out ground effects present in the previous two approaches. Biomass regression will then be done to those estimated average heights. The accuracy of each approach and the differences which can be reasonably gleaned between L-band and C-band structure and biomass estimation will be explored.