Modeling above Ground Biomass Using Multi-Frequency SAR Data - a Case Study Over Gabonese Tropical Forest

Forest plays an important role in regulating earth's carbon cycle. Forest ecosystem stores large amount of carbon in trees and soils, by taking CO2 from the atmosphere. Hence, burning of fossil fuels and deforestation may lead to elevated levels of CO2 in the atmosphere. In order to understand the impact of these disturbance in carbon cycle leading to perturbation in climatic system, it is essential to have a proper estimate of carbon stock and fluxes associated with it. Also, mapping forest parameters at finer spatial and temporal resolution assists us to delineate the degradation and deforestation of ecosystem, helping us to prioritize the criteria for sustainable forest management. Space-borne remote sensing techniques provides a feasible way for mapping global forest periodically with acceptable level of accuracy, helping improved modelling of carbon cycle and climate changes. In recent time, SAR remote sensing has shown the capability to map global forest biomass because of it weather and daylight independence along with penetrating capability, especially at lower frequency. With number of future SAR missions such as BIOMASS (P-band), Tandem-L (L-band), NISAR (L- and S-band), NovaSAR (S-band), SAOCOM (C-band) and RADARSAT constellation (C-band), the challenge would be the integration of results from datasets at different frequencies over global scale.

In this work, we examine the multi-frequency behavior of full polarimetric and interferometric observables at L- and P-band over Gabonese forest. Here, we estimate the biomass independently using the polarimetric and interferometric approaches. Estimation of forest biomass from backscatter and interferometric coherence, depends on its sensitiveness to structural component that varies with radar parameters. We develop a multitemporal, multifrequency backscatter model to estimate the biomass form the backscatter of multitemporal datasets. In case of interferometric approach, it is not possible to directly relate the biomass to the observables. Hence, we estimate the tree height from the interferometric coherence and relate it to biomass using height biomass allometric equations. Estimating forest parameter form interferometric coherence is a not straight forward, but requires inversion of models with high complexity. Here, we use random volume over ground (RVoG), one of the most widely used model and is validated over wide variety of forest environments, corrected for terrain distortion. The phase diversity coherence optimization algorithm was adopted to estimate the two coherences with largest separation. The inversion of RVoG model begins with estimation of ground topography by fitting a straight line between two optimized coherence and followed by retrieval of forest tree heights. In this case, we have considered the effect of temporal decorrelation as zero. Finally, two biomass estimates from the backscatter and interferometric approach are combined using Bayesian minimum mean square error model.