Extraction And Application Of Rice Growth Parameters Based On Multi-source And Multi-temporal Remote Sensing Data

Rice is one of the main grain crops in China and south-east Asia.Monitoring crop growth is important for field management,yield prediction,disaster prevention and control.The biophysical and biochemical parameters such as leaf area index(LAI)and pigment content were widely used as indicators for rice growth monitoring.Hyperspectral technology has strong band continuity,high spectral resolution and abundant spectral information,so it can be real-time,rapid,efficient,non-destructive to acquire rice growthment,nutritional status and their changes.It provides technical support and theoretical basis for effective information management of precision agriculture.In this paper rice experiments including different nitrogen levels and growth stages were conducted.Based on rice canopy reflectance,biophysical and biochemical parameters,the aim of this study is to establish models between hyperspectral reflectance and pigment content,LAI.Obtaining the spatial distribution of rice LAI at key growth period based on Environmental Satellite(HJ-1A)image data,in order to achieve large scale observation of rice growth.According to the above research,the main results are as follows:1.Based on the rice canopy hyperspectral data,15 vegetation indices were calculated and got.Using the leave-one cross validation we carried out 5 kinds of traditional regression analysis(linear,exponential,logarithm,power,parabola),so as to establish rice LAI hyperspectral estimation model and choose optimal vegetation index at different growth stages.Sensitivity analysis of vegetation index calculating LAI was carried out by noise equivalent(NE).The results indicated that normalized difference vegetation index(NDVI)and new vegetation index(NVI)were sensitive to the change of LAI at the tillering stage;The green normalized difference vegetation index(GNDVI),ratio vegetation index(RVI-3),modified simple ratio index(MSR)had highly sensitivity and estimated accuracy at the jointing stage;At later stage of rice growth,GNDVI,modified normalized difference vegetation index(mNDVI),and MSR were more suitable to LAI estimation than other indices.The traditional regression model of rice LAI in the whole growth period which was constructed by vegetation index is of low accuracy,and it's difficult to estimate rice LAI in the whole growth stage with uniform vegetation index.Using partial least squares regression modeling method,the coefficient of determination of the modeling set and the validation set can respectively reach 0.87 and 0.81,RMSEC is 0.612,RMSEP is 0.856,PRD is more than 2,which means rice LAI can be accurately estimated during the whole period.2.According to the rice canopy hyperspectral data(400~750nm),four band indexes: band depth(BD),band depth ratio(BDR),normalized band depth index(NBDI)and band depth normalized to band area(BNA),were calculated via continuum removal processing.Principal component analysis(PCA)was used to reduce dimensions of hyperspectral data and determine 10 principle components,which were introduced into Back Propagation(BP)neutral network as input variables.As a result,the estimation model of BD and BP(R2=0.61,RMSEP=0.128 mg·g-1)has the highest estimation accuracy on the carotenoid content in rice leaves,while the estimation model of BNA and BP(R2=0.73,RMSEP=0.343 mg·g-1)has the highest estimation accuracy on chlorophyll content in rice leaves.A further comparison between BDA & BP Model and the Vegetation Index Best Regression Model was made to obtain the BP neutral network model based on the band depth analysis,which indicates a better solution to the saturation problem and a higher estimation precision of rice leaf pigment content.3.According to spectral response function of HJ-1A satellite images,the blue,green,red and near-infrared bands of HJ-1A satellite were simulated based on the ground survey point and canopy reflectance of rice during grain filling stage in experimental field.The correlation between 12 vegetation indices and rice LAI was analyzed,and the rice LAI estimation model was established by selecting the vegetation index with the largest correlation coefficient.The results showed that the accuracy of GRVI two polynomial regression rice LAI estimation model in the filling stage was highest,and the model was LAI=-0.027*GRVI2+1.125*GRVI+0.028.The modeling and validation sets of R2 reached 0.89 and 0.80 respectively,RMSEC and RMSEP were both lower,PRD is greater than 2,the model is excellent.Using GRVI-LAI estimation model,the spatial distribution of rice LAI in the filling stage was obtained.Due to the limitation of the atmospheric condition and the spatial resolution of HJ-1A image,the LAI predicated value in the spatial distribution map is generally lower than the LAI measured value of the corresponding ground point.