Forest CHM Inversion Based On Vegetation Coherence Scattering Model Using Airborne InSAR/PolInSAR Data

Forest canopy height model(CHM),as a crucial parameter of the forest vertical structure,is of major importance in the detection of forest harvesting,forest degradation,and forest fire.Moreover,above-ground forest biomass can be estimated by the forest CHM via allometric equations,which would assist in studying the regional and global carbon cycles,carbon trade,and climatic variation.Interferometric synthetic aperture radar(In SAR)and Polarimetric In SAR(Pol In SAR)techniques are sensitive to the vertical distribution of scatterers,and have gradually developed into the efficient techniques for the estimation of forest vertical structure.The CHM of large-scale trees,in particular,can be quickly and effectively inverted by In SAR/Pol In SAR in different wavelengths(X-,L-,and P-bands)and with different forest types(boreal,temperate,and tropical forests).However,there are still three major problems when extracting forest CHM by In SAR/Pol In SAR.We will introduce the three problems as following:(1)Vertical structure backscatter of forest is influenced by a combination of factors,including the frequencies of the radar waves and the forest biophysical parameters(i.e.,density,species).The deeper penetration of P-band SAR and the heterogeneity of vegetation structure lead to the diversity of forest vertical backscatter,and the resulting parameter estimation error is thus a critical element in limiting the accuracy of P-band Pol In SAR forest CHM inversion.(2)Sub-look In SAR has higher observation efficiency in comparison with Pol In SAR,due to the wider swath and higher spatial resolution.However,the accuracy of CHM inversion based on existing sub-look In SAR method is attenuated by the inaccurate coherence scattering modeling and uncertain parameter calculation.(3)In the complex unit circle,a two-dimensional(2-D)ambiguous error of the pure volume coherence is caused by the joint influence of the residual ground scattering contribution and the temporal decorrelation,which needs to be removed for the high-precision CHM.Moreover,the heterogeneity of the forest vertical structure must be considered in the temporal decorrelation compensation.Therefore,considering the above three problems,this paper focuses on the following works:(1)Based on the study of the Gaussian mean and standard deviation effect in the volume-only coherence,a 2-D Gaussian vertical backscatter(GVB)model with constrained mean is proposed to alleviate the backscatter-induced error,which avoids underdetermined 3-D parameter inversion.Then,the ground scattering contribution of the measured pure volume coherence can be removed by searching for the optimal pure volume coherence in the dual-baseline inversion.As a result,a robust dual-baseline inversion method based on the 2-D GVB model is proposed to remove the major errors from both the backscatter profile and ground scattering contribution.The results show that the root-mean-square error(RMSE)of the proposed approach was 3.26 m,which represents an average improvement of 33.7% over random volume over ground based dual-baseline inversion.(2)Impacts of backscatter in forest height estimation based on the models of random volume over ground(RVo G)(σ＞0),RVo G(σ(27)0),and Gaussian vertical backscatter(GVB)were investigated in the complex plane,and then,with the fusion of a small amount of Pol In SAR data and light detection and ranging(Li DAR)data,a random forest(RF)classification method based on multi-dimensional backscatter observations is proposed to obtain the optimal backscatter function and forest CHM from the results based on the different models in each pixel.The proposed method was tested with single-and multi baseline Pol In SAR data in the P-band,and the RMSEs of the proposed approach were 2.85 m and 2.69 m,respectively,which represented average improvements of 20.6% and 17.7% over the optimal single-model inversion.(3)Path difference of the sublook In SAR signals,which is the major factor to influence the backscattering variations,is considered in the scattering matrix.Then,the TF-RVo G model,which is used to describe the relationship between the sublook coherences and the forest biophysical parameters,is proposed from the derivation of the coherence scattering modeling based on the scattering matrix.Moreover,based on the average phase difference of the sub-aperture coherences,an adaptive judgment criterion of the pure volume coherence is constructed,and a three-stage inversion method suitable for the TF-RVo G model is proposed for CHM retrieval.The result shows that the proposed method using In SAR data provides a RMSE of 5.61 m and14.3% improvement with respect to the classical three-stage inversion results from Pol In SAR data.(4)Based on the empirical relationship between D.I and extinction coefficient(where D.I means the distance ratio index to describe the position of pure volume coherence in the coherence line),the RVo G+MTD model,which takes into account the spatial heterogeneity of forest,is proposed to alleviate the 2-D ambiguous error of the pure volume coherence caused by the joint influence of the residual ground scattering contribution and the temporal decorrelation.The inversion accuracy of RMSE=2.54 m is obtained when both the overall error of temporal decorrelation and the residual error of temporal decorrelation caused by the spatial heterogeneity of forest is considered in the proposed method,demonstrating a noticeable improvement of 32.8% compared with results from the RVo G+VTD method which introduces the fixed temporal decorrelation factor.