Study On Rice Condition Monitoring With Remote Sensing In Northwest Region

Rice is one of the world’s most important food crops. The yield and quality of rice play an important role in world food security. Biophysical and biochemical parameters such as chlloryphy(SPAD), leaf area index(LAI) and leaf nitrogen content(LNC) is good indicators for assessing rice growth. Obtaining these parameters quickly and accurately is very important for scientific fertilizer application and effective field management. In this study rice experiments including different years, nitrogen and growth stages were conducted at different sites. Based on ground hypersptral reflectance, UAV hyperspectral imagery and satellite-borne multispectral images, by using hyperspectral remote sensing, information extraction, statistic analysis and Parameter mapping technology, the aim of this study was to investigate the correlations between hyperspectral reflectance and SPAD, LAI and LNC, in addition establish the quantitative regression models. The prospective results would provide reliable data technical support for precision agriculture. The main results are as follows.(1) Based on the field experiments with different nitrogen status and growth stages in rice, a systematic analysis was made on the hysperspectral and red edge characteristics of rice canopy, the results showed that the reflectance of rice showed some seasonal pattern as affected by growth stages. From seeding to heading stage, the spectral reflectance of rice canopy decreased gradually at visible range and increased at near-infrared region. With the increase of nitrogen levels the spectral reflectance increased at visible range and decreased at near-infrared region. The red edge position of rice canopy varied between 690~740 nm during the whole stages. The red edge parameters shifted to longer wavelength from seeding to heading stage, while after heading these parameters shifted to shorter wavelength. All of them were increase with the increasing nitrogen levels.(2) By correlation analysis, the sensitive bands and the best spectral index were found for rice SPAD. Rice canopy spectral reflectance at 698 nm showed a significant maximum negative correlation with SPAD. There were more highly relevant wavelength range between first derivative of spectral reflectance and SPAD and the band width was relatively narrow. The prediction ability of the model of rice SPAD based on the variable of spectral index BND was better than the regression model based on diagnostic bands. For the leaf level SPAD, it varied at different parts of the leaf, it gradually reduced from the leaf base to tip.(3) Based on the knowledge of rice LAI varied from different growth stages and nitrogen level we systematic analyzed the quantitative relationship between LAI and NDSI, DSI, RSI and MSACI2 using any two band combination within 400~2400 nm. The results showed that exponential model with RSI(848, 752) could be the best for predicting rice LAI. After optimized by least squares support vector machine(LS-SVM), the prediction accuracy of the model had improved slightly, but the setting of the SVM model parameters were rather cumbersome, and the model structure was more complex. Thus it was not conducive to the process of remote sensing.(4) Rice leaf nitrogen content decreased with the growth progress, and increased with the increase of nitrogen level. Correlation analysis showed that in the range of 450~746 nm the spectral reflectance and leaf total nitrogen content was negatively correlated and between 754~1000 nm they were positively correlated. By analyzing the correlation between leaf nitrogen content and NDSI,RSI 450 ~ 950 nm spectral reflectance of any two of the original band formed and RSI spectral index and leaf nitrogen content, and on this basis to build based on the spectral reflectance of the first derivative of any combination of two band spectral index of RSI. The results showed that the linear model based on RSI(D738, D522) was better than others, RMSE and RE were minimized. Although the LNC regression model based on partial least squares used the full spectrum band, but its prediction accuracy was lower than RSI(D738, D522) model.(5) With UAV hyperspectral imagery which had the advantage of image-spectrum merging, the inversion of the spatial distribution of rice SPAD, LAI and LNC were achieved at regional scale. The inversion results were more consistent with the actual situation of the ground, indicating that low-altitude UAV hyperspectral imagery has certain advantages for crop monitoring in small region. In order to monitor rice growth at a larger scale, multi-spectral models of rice physicochemical parameters were built by simulating the spectral reflectance of GF-1 satellite. The results showed that multivariate linear model of rice SPAD based on GNDVI, NPCI and NDVI had better predictability, and the best predictive model of LAI and LNC are exponential model and linear model constructed by RVI. GF-1 satellite was more suitable for monitoring crop condition at large scale due to its low ground resolution and the lack of expression of details...