Retrieval Of Leaf Area Index Based On Vegetation Canopy Spectra BRDF Model

Leaf area index is generally defined as the total green leaf area per unit horizontal ground surface area,LAI represents a boundary for the exchange of energy,gas and momentum between the biosphere and the atmosphere and an important indicator representing green vegetation growing trends and monitoring global climate change.The traditionally direct method has certain damage to the vegetation,which could save the manpower and material resource even the measurements can meet the needs of practical application in high accuracy,so it's hard to acquire large area data.However,RS(Remote sensing)is an indirect and non-contact measurement method,which is widely used because remote sensing could acquire large area,high efficiency and all-weather images without damaging to vegetation.During the processes of remote sensing image data acquisition,we found the reflectance is often used as an important parameter reflecting the physical and chemical properties of biological objects.Nevertheless,high precise surface reflectance products were usually related to the sensor geometric attitude and the atmospheric environment.In this paper,a LAI estimation method considering the BRDF effect of vegetation canopy was proposed.The method used PROSAIL model to obtain simulation data,and the data as the training samples which used to construct BP neural networkThe content of this paper may contains as follows:Firstly,integrated with the measured data,reference data and priori knowledge,the research proposed to construct a multi-angle parameter database based on vegetation radiation transmission model.Through combining parameter database,a large total of vegetation canopy spectra data were simulated.And then those simulated data could use to calculate the parameters of volume scattering and geometric scattering of Ross-Li BRDF model.Secondly,Empirical statistical models are built based on RVI,DVI,NDVI,MSR,EVI1,EVI2,RDVI,ARVI,SAVI,OSAVI,MSAVI,NLI calculated with the vegetation canopy spectra to estimate LAI,and evaluated with simulation samples and wheat experiments.The validation show that the model based on NDVI has the highest accuracy among the 12 indexes in estimating LAI,and the test result of experiments is R2=0.78,MAE =0.380505,RMSE=0.486441.thirdly,based on the super fault tolerance and nonlinear mapping ability of neural network,a method to estimate LAI based on BP neural network is proposed.The vegetation canopy multispectral reflecting the characteristics of vegetation are used as the inputs and LAI as output of neural network to build LAI estimation model.In this paper,two four-layer BP neural network are used to model the situation whether take the directional characteristics of vegetation canopy into consideration.The neural network 1 neglects the geometrical optical properties of vegetation canopy,and the neural network 2 counts the vegetation canopy BRDF effect,which means Ross-Li kernel parameters regarded as the geometric optical characteristics are applied to the training of neural networks.The simulation vegetation canopy spectra and wheat experiments are used to test the networks,the accuracy tested with experiments show that the neural network 1 ignoring the optical characteristics of vegetation canopy is R2=0.80859,MAE=0.45998,RMSE=0.34936,and network 2 counts BRDF effect is R2=0.82973,MAE=0.44297,RMSE=0.33886.The results show that: 1)NDVI has higher estimating accuracy,better robustness and more inversion results among the 12 vegetation indices based on empirical statistical models.2)Because of the unique nonlinear fitting ability,the neural network estimate model based on the vegetation canopy reflectance spectrum is superior to the NDVI empirical statistical model.3)The neural network considering BRDF effect of vegetation canopy spectra is better than the one does not.Therefore,a LAI retrieving method with neural network based on BRDF model of vegetation canopy is proposed,which can improve the estimating accuracy of LAI to a certain extent.