In this study

vegetation indices（VI） were firstly calculated from the simulated band reflectance spectra of NOAA11-AVHRR

AQUA-MODIS and LANDSAT5-TM using rice canopy hyperspectral reflectance spectra

which were acquired in 2002 and 2004.Regression models for estimating LAI were fitted by that of 2002 and validated by the VIs data sets of 2004.Estimating accuracies between models were compared and analysis of VIs’ sensitivity to changes in LAI was performed.Due to differences in spectral response function

the simulated red band reflectance of AVHRR was higher

whereas near infrared reflectance was lower than that of TM or MODIS for identical rice canopy.Compared with the models of VIs calculated from the red band reflectances

LAI showed closer linear relationship in models of ratio VIs from the red-edge or green band reflectances and closer exponential relationship in models of normalized VIs from the green band reflectances.Among others

the model using Red-edge Ratio Vegetation Index（Red-edge RVI） showed the highest estimating accuracy not only for calibration but also the highest one for external validation

furthermore the decrease in estimating accuracy between calibration and external validation was lowest.The models of Green Normalized Difference Vegetation Index（GNDVI） and Green Ratio Vegetation Index（GRVI） held the second highest position of estimating accuracies for external validation

the models of normalized difference vegetation index（NDVI） and enhanced vegetation index（EVI） were the third highest and those of RVI were the worst.Although AVHRR bands have broader ranges of spectral response than TM and MODIS

their NDVI and RVI models exhibited higher estimating accuracies for rice LAI in this study.The sensitivities of all indices decreased with the increase of LAI values

but difference of sensitivities between normalized VIs and ratio VIs was significant.The normalized VIs exhibited much higher sensitivities with rapidly decreased rate than the ratio VIs at LA1 values of less than 1.2

while the ratio VIs did higher sensitivities than the normalized VIs at LA1 values of more than 1.2.Red-edge RVI and GRVI exhibited not only higher linearity relationship with LAI but also higher sensitivities at LAI values of more than 2.8.When comparison of model fitness between linear and nonlinear regression models is made

it should be done not by directly comparing the coefficients of determination of regression models but the coefficients of determination of linearity relationship between the predicted and the observed.