Inversion Of Leaf Area Index Of Spring Wheat Based On Radiative Transfer Model And Chris Data

Leaf area index (LAI) is an important structure parameter of vegetation ecosystems, which can represent the living conditions of vegetation. At the same time, it is a key ecological parameter in crop growth monitoring, yield prediction and estimation.So it is very important to inversion LAI accurately.The CHRIS (Compact High Resolution Imaging Spectrometer) data is the main data resources which is acquired in June 4,2008 in this paper, besides ASTER,SPOT5 and BJ-1 remote sensing data were also selected. The goal is to inverse LAI of spring wheat combing with the radiative transfer model (PROSAIL), and at last verified the results according to the in-situ data.In this paper, study has been made as follows:(1) Acquiring and processing of data sets. It mainly includes hyperspectral and multi-angle remote sensing data -CHRIS, multi-spectral remote sensing data and ground data (spectrum, structural parameters, physicochemical parameters, etc).(2) Based on the in-situ data and some literatures values, the range of inputing parameters of model was determined, besides the sensitivity of model parameters was analized using uncertainty and sensitivity matrix. From the results of sensitivity analysis, four sensitive parameters were chose which contained LAI,Average Leaf Angle-ALA,chlorophyll Cab and observation zenith angle. And then by comparing the modeling and field spectral data, the forward model was validated.(3) The optimal band was selected using segmented principal component analysis and the results were band 4, band 9 and band 15,which center wavelength was 551.1nm,696.9nm and 871.5nm separately. At last using this method realized data reduction.(4) Based on the model analysis and band selection, using typical ranges of those four sensitive parameters and values of other parameters, some canopy spectral reflectance data set of training samples and validation samples were simulated as training data and veritied data. The inversion models were built using quick artificial neural network (ANN), and then verified the inversion models using correlation coefficient and root mean square error. The results showed that the method that quick ANN had high accuracy, multi-angle models were better than single-angle model, correlation coefficient of two angles increased by10.465% and RMSE reduced 4.3% which compared to single angle inversion,with the number of angle sets increases, the inversion accuracy increases with the improvement decreases.(5) Inversion and validation of multi-angle hyperspectral data.Combined with verificated accuracy of inversion modeling, the optimal model (0degree,0 and 36 degree,three and four angles)were used in the remote sensing imaging inversion, and then verified the inversion result according to ground measurement data. It suggested that RMSE reduced with angle increases apart from five angles,and the maximum correlation coefficient is three angles combination,the followed was the combination of three angles,two angles combination, four angles combination,negative degree of 36, combination of five angles. the minimum correlation coefficient is zero angle. The inversion precision of combination of three angles is better than single angle.