Study On Simulation Model Of Cotoon Grow Information Based On Hyperspectral Data

Hyperspectral data obtained by using of hyperspectral remote sensing have strong continuity and large amount of information, which plays an important role in improving the ability of monitoring crop growth information. This study set up cotton plot experiment under different nutrient levels in order to form the different growth conditions of cotton. The canopy spectral of cotton had measured by using the portable spectrometer in different growth period, and some agronomic parameters which can reflect the growth information of cotton were measured at the same time. In addition, spectral reflectance of cotton single leaf was determined in different growth stages, corresponding chlorophyll content and anthocyanin content of cotton single leaf were measured also. According to the characteristics of cotton spectral information, this study extracted 27 typical spectral variables, sensitive band variables based on the analysis results of agronomic parameters and raw spectral reflectance and the first derivative reflectance were selected also. Analysis of the correlation between these variables and parameters of agriculture, and then the linear and nonlinear regression equation for different agronomic parameters were constructed based on the higher correlation variables from the results of correlation analysis, finally, the accuracy were analyzed for some models which determine coefficient higher in order to choose the best estimation models for cotton growth information. The main results are as follow:1. The single leaf spectral information influenced by external environmental factors was relatively small because of its collection indoor. Therefore, the predictive precisions of the chlorophyll content models and the leaf anthocyanin content estimation model was higher than that of chlorophyll content model and canopy leaf anthocyanin content estimation models. It is best to estimate the canopy leaf anthocyanin content in the blooming period, other cotton growth parameters estimated should based on the whole growth period data according to the analysis results.2. In the estimated models of chlorophyll content of leaf and canopy, the optimum variable models were established based on the first derivative spectral sensitive bands. The single variable linear estimation model of single leaf chlorophyll content which established using of DR757 as variables determine coefficient was 0.8656**, and the estimated accuracy was 97.06%. The model based on the first derivative spectral reflectance DR753 by using nonlinear regression method obtained the most satisfied results for the estimation of canopy chlorophyll content, the determine coefficient was 0.4488**, but in the precision inspection process we found that the accuracy of the model is lower than the accuracy of the model using of R709 as variable, the relative error of two models were 11.6% and 6.55%, and the correlation coefficient between the value obtained in the prediction equation using of DR753 as variable and the measured value was relatively lower, therefore, whether the estimation model was the optimal model needs further research.3. The optimal prediction model of individual leaf anthocyanin content was unitary quadratic regression equation with DR755 as the independent variable, RMSE=0.28, estimation accuracy was 97.44%, the correlation coefficient between measured and estimated value was 0.84**. Flower and boll stage was the best period to estimate anthocyanin content in canopy leaves, the model based on the typical variable of(Rg-Rr) /(Rg+Rr) obtained the most satisfied results, the estimation accuracy can reach 88.45%, RMSE=1.43.4. The results of analysis in the different growth period show that the correlation between leaf area index, aboveground biomass, leaf water content, plant water content and 27 typical spectral variables was not strong, even there is no correlation between the spectra extraction variables and these agronomic parameters in some growth period. The results of analysis based on the data of whole period show that higher correlation with leaf area index, aboveground fresh biomass, aboveground fresh biomass, leaf water content, plant water content were SDr/SDb, SDr/SDy, SDr/SDy,(SDr-SDb)/(SDr+SDb) and the red edge position λr respectively, the correlation coefficients were 0.601, 0.637, 0.648,-0.685,-0.577, reached a significant level. Moreover, these parameters with the sensitive band variable have good correlation, and the correlation with the first derivative spectral sensitive bands was higher relatively. The optimal estimation models were screened out based on the precision test. The results show that using DR742 as variable to simulate cotton leaf area index was best, the precision of model was 73.73; the optimal prediction models of aboveground fresh biomass and aboveground fresh biomass were power function regression equation with SDr/SDy as the independent variable, the precision of models were 94.81% and 95.3%; the optimal prediction models of leaf water content and plant water content were unitary quadratic regression equation with DR537 and DR1408 as the independent variable respectively, the precision of models were 97.65% and 97.37%.