A hybrid inversion scheme for estimating surface variables of vegetation is proposed under Bayesian Network（BNet） theory

and then is used to estimate chlorophyll content of winter wheat leaves（Cab） and Leaf Area Index（LAI） of canopy.A coupled physical model named PROSPECT+SAIL was chosen to generate simulation data set

which means that the SAIL model uses the leaf reflectance and transmittance derived from PROSPECT model to simulate canopy directional reflectance.Results derived from simulation data and SHUNYI Experiment in 2001 data show that both LAI and Cab can be estimated with an appreciated accuracy under the proposed scheme

except that there are about 10% of total points falling into failure inversion.Then an uncertain data handling method

which considers the measured data as the random variables obeying Gaussian distribution

is employed to solve the failure problem.As a result the failure points are removed successfully though the RMSE of estimated the two variables is larger slightly.The presented hybrid inversion scheme is a knowledge-based inferring mechanism in principle

so the updated information content in the inversion process is quantitatively calculated thanks to the concept of entropy introduced from thermodynamics.Contrasting to the conditional entropy

the posteriori entropy calculated according to our proposed probability revision algorithm is not a descending parameter.This property can give some indications in estimating the information content parameters and the currently used data

that is to say

if the data are consistent with the previously derived information of estimated parameters

then there is descending entropy

otherwise

it is ascending.In the last section of this paper

some discussions are presented about the problem on how to estimate and control the information stream

especially when the inversed physical model is nonlinear.