P-band SAR Tomography And Temporal Decorrelation Analysis On Tropical Forests

Forest biomass is a central parameter to better understand the global change and carbon cycle.However,there is no current global gridded biomass data due to the great difficulty in the investigation of forest resources.Recently the use of Synthetic Aperture Radar(SAR)to estimate forest biomass has attracted more and more attention because of its potential for the assessment of forest biomass in global scale.Nowadays many countries are devoted to the assessment of global forest biomass via long wavelengths spaceborne SAR missions due to their advantage of the under-foliage penetration capabilities,in 2017 a research on civil use P-band spaceborne SAR mission is also launched by Shanghai Academy of Spaceflight Technology and Wuhan University.Among all the proposed spaceborne missions the most mature one is P-band BIOMASS mission proposed by European Space Agency(ESA).The main driver of the BIOMASS mission is to provide new measurements that will allow to estimate forest Above Ground Biomass(AGB)worldwide with unprecedented resolution and accuracy.The BIOMASS mission will last five years and comprise a tomographic phase and a interferometric phase.The tomographic phase will gather 7 repeat-pass acquisitions with a revisit time 3 days during the first 14 months,whereas the interferometric phase will use only three images in the rest of mission lifetime.BIOMASS will provide three basic observables: Polarimetric SAR(PolSAR)intensities,forest height by Polarimetric SAR Interferometry(PolInSAR),and forest vertical structure at different polarizations by SAR Tomography(TomoSAR).Propelled by BIOMASS preparatory activities,tomographic imaging of forested area has been addressed in a number works in the last years,producing both theoretical developments and experimental results.Concerning AGB retrieval,the use of P-band TomoSAR was found to provide a clear improvement over using polarimetric intensities alone.It was shown that the backscattered power at30 m above the ground is best correlated to the AGB.This can be ascribed to removal of the ground contributions in the backscatter signal,minimizing the perturbing effects of local topography and soil moisture and thus improving the relationship correlation to AGB.Based on this explanation a pure interferometric method: ground notching is recently proposed to cancel out the ground echo by using just two coherent radar images in order to support the interferometric phase of BIOMASS.It has been demonstrated that the ground notched intensity is highly correlated to the AGB.Both the TomoSAR and ground notching results were found by analyzing stacks of airborne SAR images acquired in a single day,which allowed to neglect the impact of temporal decorrelation associated with random motions and varying dielectric conditions within the vegetation layer.This condition,however,cannot be retained in the case of BIOMASS,for which the revisit time will be of 3 days.Besides,spatial baseline has a great impact on ground notching,a single baseline allows accurate estimation only when a proper baseline is chosen.To address the issues above,the researches in this thesis were carried out over two P-band multi-baseline and multi-temporal tropical forests dataset: TropiScat and AfriSAR in the framework of BIOMASS mission.The main works are summarized as follows:(1)A BIOMASS tomography simulation method using P-band data from the TropiScat archive was proposed according to the acquisition time and revisit time of BIOMASS.The effect of temporal decorrelation on the quality of tomography,the stability of backscattered power in both canopy and ground layers was studied under changing weather conditions.The AGB retrieval error caused by temporal decorrelation was also investigated by using the linear model between backscattered power at 30 m above the ground and forest AGB.The experimental results prove the validity of the use of P-band TomoSAR for forest biomass estimation when considering the effect of temporal decorrelation.(2)The temporal decorrelation caused by two single weather element: rain and wind in full polarimetric channels was studied,and a temporal decorrelation model based on wind speed was proposed.The vertical distribution of temporal coherence before and after rain was studied,the time series of temporal coherence with different wind speed was also explored,giving an insight into the effect of temporal decorrelation on forest scattering mechanisms of P-band TomoSAR.The proposed temporal decorrelation model was demonstrated to be more suitable than Brownian motion model in the case of short temporal baseline,contributing to the investigation of temporal decorrelation compensation method in order to improve the stability and accuracy of forest vertical structure information extraction and biomass estimation.(3)The temporal decorrelation within the volume scattering components at a much larger scale was studied to complement the results from TropiScat.A volume temporal decorrelation analysis model was propose based on ground notching,which is used to cancel out the ground echo and single out volume scattering based on single-baseline acquisitions.Based on the proposed volume temporal decorrelation analysis model,the temporal decorrelation within volume scattering and its effect on the intensity of ground notched images was studied over P-band AfriSAR Lopé dataset.(4)The relationship between interferometric ground notching and TomoSAR was studied from a perspective of signal processing.The effect of the length spatial baseline and the incident angle variation on ground notching was also studied.A multi-baseline ground notching method was proposed,reducing the dependence on spatial baseline.Moreover,the use of the three methods of TomoSAR,ground notching and multibaseline ground notching on forest biomass estimation was studied,experimental results indicates a great improvement using multi-baseline ground notching compared to the single-baseline one.