Monitoring Growth Parameters With Hyperspectrum Under Different Vegetation Coverage Conditions In Wheat

Non-destructive and quick monitoring of leaf nitrogen (N) status and growth characters are important for precision management and yield prediction in crop production. Hyperspectral remote sensing technique, with advantages of more wavebands, high resolution and rich data, is becoming a key technique for the non-destructive monitoring of crop growth parameters. However, the accuracy of remote-sensing monitoring is significantly affected by the soil background at the beginning or middle periods of crop growth, which restricts the popularization and application of the remote sensing technique. Therefore, in order to reduce and elimate the influences from soil background, it is essential to develop monitoring models with strong universality and mechanism. The primary objective of this study is to systematically analysis the quantitative relationships between canopy spectrum and nitrogen status or growth characters. And then this paper will discuss the effective methods to reduce the influences of soil background. Finally, the optimum spectral parameters and quantitative models for estimating leaf N status and growth characters will be explored through the extraction of the change patterns of canopy spectrum under varied nitrogen levels and planting densities. On the basis of a series of experiments with varied cultivars, N rates and planting densities, the results would provide technical support for non-destructive monitoring and precision diagnosis of wheat growth.Based on the features of canopy spectrum under different nitrogen levels and planting densities, a comprehensive quantitative analysis was carried out on the relationships between five different spectral parameters (Ratio vegetation index(RVI), Normalized difference vegetation index (NDVI), Soil-adjusted vegetation index(SAVI), Optimal soil-adjusted index (OSAVI), Perpendicular vegetation index(PVI)) and leaf nitrogen content(LNC) in wheat.These spectral parameters are designed based on the combinations by the original spectrum and the derivative spectrum from all possible two-band within 350-2500nm. Then these spectral parameters were modified by Fractional Vegetation Cover (FVcover), and further adopting the linear mixture model to explore the methods for eliminating and reducing the influences of soil background. The results indicated that the selected spectral parameters were significantly affected by the soil background. In addition, the coefficient of determination (R2) of these spectral parameters was about0.55, and the sensitive regions with high correlations between spectral parameters and LNC were similar (500nm, green region). This confirms that the different spectral parameters do not really provide different independent information, and the highest accuracy of spectral parameters was OSAVI(R514, R469) L=0.04.Moreover, the R2value of the first derivative spectral parameters increased to0.59. However, the testing performances for the models of the first derivative parameters were poor and the RRMSE was higher. Therefore, these implied that the models were unstable. Furthermore, the FVcover modified soil adjusted spectral parameter, had the R2value of0.62, the lower RRMSE of0.13, and less sensitive to the LAI、LDW、FVcover and LNA, which was newly established. Although the linear mixture model did not provide a good accuracy because of the complicated field circumstances, it has great potentialities. Besides, three band spectral parameters and red edge parameters were less sensitive to LNC than two band spectral parameters. Therefore, the newly developed spectral parameter NDVI/(l+FVcover) could be used to monitor LNC under varied conditions of FVcover in wheat.Canopy spectrum and leaf nitrogen accumulation (LNA) under series different FVcover were obtained, a systematic analysis was conducted on the quantitative relationships between LNA and those five spectral parameters. The results showed that the R2between LNA and spectral parameters of original spectrum was0.7, and the optimal waveband combinations were located in the near infrared region (670-762nm). The spectral parameters of derivative spectrum had higher R2than the original spectrum, the testing accuracy of the models was better as well. The best performing spectral parameter was SAVI(FD856, FD740)L=-0.1, with the R2value of0.803, RRMSE of0.235and slope of1.009. Moreover, newly established three-band spectral parameter had higher accuracy for estimating LNA but poor performance of model testing. Furthermore, the red edge parameters had low correlation with LNA, and the area difference between left and right peaks based on IGUS had a relative higher accuracy. Finally, NDVI(Rλ1, Rλ2)/(1+FVcoverr) was introduced from the Chapter three, which is sensitive to LNC. However, this model of the sensitive spectral parameter did not perform well, due to the differences of composition between LNA and LNC. Therefore, SAVI(FD856, FD740) L=-o.1is a good indicator of LNA under varied FVcover in wheat.Based on the acquisitions of Leaf Area Index (LAI), Leaf Dry Weight (LDW) and FVcover and the canopy spectrum under different FVcover by the ASD spectroradiometer, a series of quantitative analyses were carried out on the relationships between spectral parameters and LAI, LDW and FVcover, respectively. The results revealed that SAVI(R762, R724) L=1, a soil adjusted spectral parameter which is based on the combinations of original spectrum, had good accuracy on the estimations of LAI, LDW and FVcover with the R2value of0.8. In addition, both the derivative spectral parameter of OSAVI(FD1146, FD758) and the area parameter of right peak computed for the FD (first derivative) had reliable performances and the similar R2with SAVI(R762, R724).The three band spectral parameter (R762-R724-R856)/(R762+R724+R856) was less sensitive to LAI, LDW and FVcover than other spectral parameters.Among these earlier published spectral parameters, WDV、MSAVI2, Gm-1, DVI(810,560) were also highly correlated with LAI, LDW and FVcvoer, the R2values of them were all above0.75which were little lower than the spectral parameters explored in this article. Testing performance of the monitoring models with the independent experiment dataset indicated that SAVI(R762, R724) had the highest accuracy while OSAVI(FD1146, FD758) and right area were not stable. Besides, the performance of published GM-1was relatively better, but was generally worse than SAVI(R762, R724) Therefore, SAVI(R762, R724) L=1could be reliably used for the estimation of LAI, LDW and in wheat.