Abstract: [Objective] This paper developed a method of forest above-ground biomass(AGB) estimation based on the technology of multi-baseline InSAR tomography, aiming to solve the problem of saturation effect and support mapping global forest biomass.[Method] The experiments were carried out over the site of Paracou, French Guiana. The tropiSAR 2009 P-band multi-baseline airborne InSAR data and 85 forest plot investigation data were used as the key data sources. Firstly, three-dimension distribution information of the tomographic relative reflectivity for HH polarization was obtained. Accordingly, the ground elevation was retrieved, and the terrain topography was removed from HV polarization data. Secondly, three-dimension distribution information of the tomographic relative reflectivity for HV polarization was obtained and converted to ground geometry by geocoding. Finally, correlation analysis between in situ AGB measurements and the extracted tomographic relative reflectivity at different heights (5 m interval) were implemented. The forest AGB estimation model was built and assessed by leave-one-out cross-validation.[Result] Negative correlations were found for the layers of tomographic relative reflectivity at the height below 20 m, with the best correlation of -0.58 for the 5 m layer. Positive correlations were found for the layers of tomographic relative reflectivity at the height above 20 m, with the best correlation of 0.63 for the 25 m layer. The 5 m layer made the accuracy of the forest AGB estimation model to be on the order of 88.44% with RMSE of 49.85 t·hm^-2 (RRMSE of 13.56%). The 25 m layer made the accuracy of the forest AGB estimation model to be on the order of 88.82% with RMSE of 47.30 t·hm^-2 (RRMSE of 12.87%). The forest AGB estimation model could be refined by combining the 5 m layer and the 25 m layer, with the accuracy of 89.17% and RMSE of 46.45 t·hm^-2 (RRMSE of 12.63%).[Conclusion] The saturation effect in tropical forest would be solved based on the technology of multi-baseline InSAR tomography. High-precision forest AGB could be retrieved either with the 5 m layer or with the 25 m layer, which demonstrated that tomographic relative reflectivity within forest was an effective indicator to forest AGB. The forest AGB estimation model could be refined by combining different layers, which demonstrated that making the best of the forest vertical structure information may further improve the accuracy of forest AGB estimation.

Key words: multi-baseline InSAR, tomography technology, forest vertical structure, forest above-ground biomass, tropical rainforest