Vegetation Reflectance Simulation And Bio-optical Property Inversion Based On Three-dimensional Radiative Transfer Model

As one of the terrestrial ecosystem components,vegetation can maintain ecological balance and improve the environment.Vegetation is inseparable from humans,thus playing an essential role in human survival and development.Remote sensing data are increasingly used for vegetation monitoring and inversion due to its improved measurement accuracy and spatial/spectral/temporal resolution,as well as the advances in remote sensing data interpretation methods.Traditional one-dimensional radiative transfer models affect the accuracy of vegetation reflectance simulations and bio-optical parameter inversions due to their inaccurate parameter estimates.Three-dimensional radiative transfer models based on real structures can well consider sensor observations and solar incidence directions,direct and diffuse scattering of surface irradiance,and the complex three-dimensional structure of the scene.However,the application of 3D radiative transfer simulation and inversion encounters problems in the following aspects:(1)Existing simulation models do not have continuous-time phase simulation capability due to the lack of knowledge of the spatial and temporal variation of key ground parameters.However,remote sensing images are mostly time series data,making it difficult to use together with remote sensing data.(2)Existing leaf spectral inversion methods are mainly applicable to densely vegetated areas.However,in scenes with complex components and many mixed pixels,such as cities,the inversion accuracy of leaf optical properties is seriously degraded.(3)The existing vegetation indices are easy to saturate in high vegetation cover areas,severely limiting their inversion capability.To address the problems mentioned above,based on the 3D radiative transfer model-based vegetation reflectance simulation and its bio-optical parameter inversion,we realize the 3D radiative transfer continuous-time phase simulation capability by coupling the growth model.Also,the accurately inverting of the spectral signatures of leaves in urban areas is achieved by introducing the Discrete Anisotropic Radiative Transfer(DART)calibration.Finally,we analyze the vegetation isolines behaviours and propose the intersection point right shift phenomenon based on the DART simulation data to mitigate the soil-adjusted vegetation index(SAVI)saturation effect in high vegetation cover areas.In order to achieve a 3D radiative transfer simulation capability in the continuous-time phase,a static 3D maize modelling model constructed from an extended L-system(ELSYS)was coupled with a dynamic maize growth equation using degree days as the growth factor and development rules describing canopy structure throughout the plant growing season from seedling emergence to the male flowering stage.Maize canopy reflectance for the corresponding scenarios was simulated using various 3D radiative transfer models for cross-validation,and a good agreement among them was achieved.The simulated canopy reflectances from the 3D radiative transfer model were compared with the 1D radiative transfer model under the same leaf area index(LAI)conditions as true values.Results show that the homogeneity assumptions of the 1D radiative transfer model under the same LAI conditions resulted in an approximately 1.5 times overestimation of foliage cover,leading to a significant overestimation of the reflectance in the near-infrared(NIR)in the nadir direction.This overestimation may be because the homogeneity assumption fails to consider that the vegetation seen at the nadir is the smallest part of the scene and that multiple scattering is mainly from vegetation.Considering that the nadir direction is a vital observation direction for the sensor and that the NIR is an essential band for vegetation monitoring,we believe that the overestimation due to the homogeneity assumption is not negligible and result in an underestimation of the amount of vegetation(LAI,etc.)in the remote sensing inversions based on the homogeneity assumption.In order to invert the optical properties of leaves in urban areas,the DART calibration was introduced.DART calibration first separates the single scattered reflectance of the canopy using a linear spectral mixture model and iteratively corrects the optical properties of the input leaves to simulate the canopy reflectance close to the mixed pixel separation.The inversion results were evaluated for accuracy assessment and sensitivity analysis.The mean relative errors of the ideal noise-free simulation experiment were 0.015,0.004,0.017,0.320,and 0.303 for ground,roof,water,tree,and shrub in all bands.Under experimental conditions with added noise(pixel shifting,geometric accuracy of the 3D scene and modulation transfer function with some deviation from the true value),significant errors in the inversion were observed: for ground,roof,water,trees and shrubs,the mean relative errors are 0.289,0.448,1.103,1.164 and 1.242 respectively.In descending order of importance,the parameters that have the most significant influence on the inversion accuracy of the optical properties of urban matter are the solar zenith angle,the spatial resolution of the satellite images,the pixel shifting,the inaccuracy of the 3D urban scene modelling and the modulation transfer function.In order to alleviate the saturation effect of SAVI in areas with high vegetation cover,the vegetation isolines were analyzed,and a right-shift phenomenon was proposed for the vegetation isolines intersection point.The right-shift phenomenon shows that as the vegetation cover increases in the homogeneous canopy(defined in this paper as a canopy with an clumping index equal to 1),the intersection points of the vegetation isolines and soil line gradually moves towards the positive red band axis.When the intercept of the vegetation isolines is smaller than the intercept of the soil line,the final intersection point can reach the positive red band area.The right-shift phenomenon successfully resolves the two major debates in the current academic community and achieves a dialectical unification from a new perspective.Considering that the optimal soil adjustment factor is the negative value of the horizontal coordinate of the intersection of the vegetation isolines and the soil line,based on the right-shift phenomenon,the hypothesis that the optimal soil adjusted factor should be negative in high vegetation cover areas is put forward.Results show that the optimal soil adjustment factor is approximately equal to-0.148 when the average LAI equal 5.35;the optimal soil adjustment factor is approximately equal to-0.183 when the average LAI is 6.72.The hypothesis can significantly mitigate the saturation effect of SAVI and improve the accuracy of LAI estimation.In this paper,vegetation reflectance simulation and bio-optical property inversion based on three-dimensional radiative transfer model parameters is the research object in terms of both simulation and inversion.The first half focuses on the modelling and reflectance simulation of a 3D vegetation scene with a coupled growth model.The second half focuses on the inversion of plant biochemical parameters using the 3D radiative transfer model or data from its simulations.Potential applications include providing high-quality analytical validation data for sensor design and adequate data support for quantitative remote sensing inversion modelling,spatial and temporal scale conversion,and multi-source data assimilation to achieve the objective of "quantitative simulation,quantitative evaluation".