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

Hyperspectral remote sensing technique, with the characteristics of high resolution, consecutive wavebands and rich data, can significantly enhance the ability of detecting growth parameters of crop. In this study, experiments with rice including different varieties, nitrogen and planting density practices were conducted in four years at different eco-sites. Based on analysis of canopy spectral information and assay of physic-chemical index in rice plant, canopy spectral reflectance was measured with hyper-spectral radiometer (FieldSpec Pro FR2500) over the whole growth periods. The characteristics of hyper-spectral reflectance under different experimental conditions and their correlations with nitrogen status and growth characters in rice were investigated. In addition, the sensitive spectrum parameters and quantitative regression models were established for nitrogen status and growth characters, the prospective results would provide technical basis for non-destructive monitoring and precision diagnosis of rice growth.Based on the field experiments with different varieties, nitrogen rates and planting densities in rice, a comprehensive analysis was made on the quantitative relationships between hyperspectral vegetation indices of canopy and its leaf area index (LAI), leaf dry weight (LDW) and vegetation coverage (VC), respectively. The results showed that a good performance of the simple ratio index SR(826,746) on estimating LAI, LDW and VC. Among all earlier published indices, GVI、SAVI and SARVI(MSS) were most highly correlated with LAI, LDW and VC, while the performance level of these vegetation indices was lower than SR(826,746). Testing result of the monitoring models with an independent experiment dataset indicated that the spectral index of SR(826,746) gave accurate growth estimation under the experimental conditions. Based on the datasets of experiment1-3before and after closure between two rows, we tested the newly proposed and earlier published indices. It was proposed that LAI and LDW (and VC) could be monitored by spectral index SR(826,746), and LDW could be also monitored by spectral index DI(854,760) as well. The ratio index SR(826,746) could be reliably used for the estimation of LAI, LDW and VC in rice.Based on comprehensive consideration of LAI and VC, and the influence of water and soil background on canopy reflectance spectra. A comprehensive analysis was made on the quantitative relationships between hyperspectral vegetation indices of rice canopy and its leaf nitrogen concentration (LNC). The aim purpose was to develop a spectral index highly correlated to canopy LNC, but less influenced by canopy leaf area index (LAI) and vegetation coverage (VC). The results showed that the simple ratio index SR(553,537) which using two green bands and the three bands index (R605-R521-R682)/(R605+R521+R682) could be used to estimate canopy LNC. Furthermore, it had a good performance on eliminating the effects of LAI and VC. The best spectral index based on two first derivative bands for estimating canopy LNC was the difference index DI(D875, D645). In general, the hyperspectral indices based on spectral reflectance performed better than the spectral indices based on the first derivative spectra. The tests with independent dataset also indicated that the monitoring models based on SR(553,537) and (R605-R521-R682)/(R605+R521+R682) had R2of0.69and0.72, RRMSE of14%and21%, however,(R605-R521-R682)/(R605+R521+R682) appeared instability among four years. Although NDVIg-b and ND(503,483) had a good performance overall, the most of earlier published spectral indices showed poor performance on LNC estimation, however they were highly correlated with LAI and VC. Therefore, the newly developed spectral index SR(553,537) could be reliably used for estimation of LNC in rice.Further analyzing the quantitative relationships between hyperspectral vegetation indices of rice canopy and its canopy leaf nitrogen accumulation (LNA), the analysis results indicated that the simple ratio index SR(770,752) could be used to estimate LNA (R2=0.9). The modified SR index mSR(770,752) which is modified based on SR(770,752) by introducing the reflectance at445nm was also a good predictor of LNA (R2=0.88). The tests with independent dataset also indicated that the monitoring models based on SR(770,752) and mSR(770,752) had good estimation accuracy for LNA in rice. In addition, SR(770,752) was slightly better than mSR(770,752) in estimation accuracy and error; The SLOPE value of mSR(770,752)(SLOPE=0.98) is significantly higher than that of SR(770,752), which shows a better fitness between the measured and estimated LNA. The best spectral index based on two first derivative bands for estimating LNA was the normalized difference index ND(D754, D7oo). Moreover, the hyperspectral indices based on spectral reflectance performed better than the spectral indices based on the first derivative spectra. Based on the datasets of experiment1-3before and after closure between two rows, we tested the newly proposed and earlier published spectral indices, the result showed that SR(770,752) and mSR(770,752) had better performance than other indices. Therefore, the ratio index SR(770,752) and the modified index mSR(770,752) could be reliably used for the estimation of LNA in rice.