Study On Tobacco Hyperspectral Characteristic And Estimating Models Based On The Interaction Of Water And Nitrogen

Abstract: The levels of water and nitrogen are the most important limiting factors of crop growth. The shortage of water and nitrogen can cause a series of changes, such as leaf color, thickness, moisture content and morphological change, consequently causing the changes in spectral reflectance characteristic. The real-time monitoring and rapid diagnosis of the crop water and nitrogen is possible by remote sensing base on spectral reflectance characteristics to recognize the object. Compared with the traditional means, the more information can be obtained quickly by remote sensing. The remote sensing is indispensable basic technology for variable rate of water and fertilizer in Precision agriculture. The relationships between the physiological and biochemical of flue-cured tobacco canopy, fresh leaf, cured tobacco leaf and their hyperspectral reflectance characteristics were analyzed by remote sensing platform in different levels of water and nitrogen. The predictive models were established by stepwise regression, and the prediction results were good by comparing the measured values and model estimation values. The main results are as follows:(1) The spectral reflectance of flue-cured tobacco canopy and leaf was almost consistent in the same levels of nitrogen with different levels of the soil moisture content. In visible light region, the spectral reflectance of flue-cured tobacco was tended to decrease with the soil moisture content increasing, because of the decreasing of chlorophyll content in tobacco due to drought. In the near-infrared light region, the spectral reflectance was tended to increase with the increase of water, the main reason is that the aboveground tobacco growth was effected by the sufficient water, and resulting in the increase of leaf area, enhanced the ability of metabolism and increased the biomass of plant groups. In different growth stage, the spectral reflectance have exhibit the variation, which the lowest in root extension period, increasing in vigorous period, and then decreasing in maturity period in visible light region. The spectral reflectance have exhibit the variation In the near-infrared light region, which the rapid increasing phase from root extension period to vigorous period, and then decreasing in maturity period. This is because the rapid growth of flue-cured tobacco from root extension period to vigorous period and resulting in increase of leaves, increased thickness, and the cell structure of leaf become complicated. Those change will cause the increasing of reflectance rate after several reflections and scattering within the leaf. From the vigorous period to mature period, the spectral reflectance have begun to weak, because of the leaf senescence, less chlorophyll, the changes of leaf internal structure.in the same levels of soil moisture content with different levels of nitrogen content, the canopy and leaf of tobacco spectral reflectance were decreased with the amount of nitrogen increases in visible light region, the mainly reason due to the increase content of leaf pigment by increasing nitrogen. In the near-infrared light region, the canopy and leaf of tobacco spectral reflectance were increased with the increasing of nitrogen, which mainly relates to the differences in leaf structure. The reason is that the high content of nitrogen will result in enlarging of the gap of leaf cells, increasing the hydration degree of cell wall. In the visible range, the canopy spectral reflectance have exhibit the variation, which the lowest in root extension period, increasing in vigorous period, and then decreasing in maturity period. However, the trends of change were inconspicuous. The reflectance rate of Yunyan87 was higher than K326 at different stage, which caused by the variety maybe.The canopy reflectance has the same trends of change with soil moisture content in different water and nitrogen treatment. In the same soil moisture treatment, the tobacco canopy spectral reflectance of high nitrogen treatment was lower than the low nitrogen treatment’s in visible light region. However, the tobacco canopy spectral reflectance of high nitrogen treatment was higher than the low nitrogen treatment’s in near-infrared light region.(2) In this study, the ASD FieldSpec HandHeld spectroradiometer was used to measure the hyperspectral reflectance of canopy and leaf of tobacco. The relationship between several indexes of biophysical and biochemical and hyperspectral parameters of canopy and leaf was analyzed by correlation analysis, the closest relates of hyperspectral parameters with biophysical and biochemical indexes were filtrate and the predictive models of those indexes were established by stepwise regression.The first independent variables of regression equations of LAI, AFW, ADW were Rg by analyzing the spectral parameters of canopy without exception, and the correlation coefficient reached a significant level, the regression equations of first independent variable as Rg also were significant. It can be used as characteristic variable for LAI, AFW, ADW. the most relevant characteristic variables of Chlorophyll a, Chlorophyll b, carotenoid, total Chlorophyll content are SAVI; the most relevant characteristic variables of total nitrogen and nicotine are the Rg; the most relevant characteristic variables of Leaf moisture content and starch are the Db; the most relevant characteristic variables of total sugar and reducing sugar are the SDb; There are the direct and effective method for finding the characteristic variable of hyperspectral parameters with indexes of biophysical and biochemical of tobacco by the predictive models. useing 25 spectral parameters as independent variables, multiple stepwise regression analysis was performed. all the estimating models of biophysical and biochemical indexes were obtained the significance level. By randomly selected test samples to test, the relationship between the measured values and estimation values were significant in the indexes of biophysical and biochemical of tobacco canopy, indicating the more accurate and the better effect of estimation.By analyzing the spectral parameters of tobacco leaf shows that the most relevant characteristic variables of Chlorophyll a, Chlorophyll b, carotenoid, total Chlorophyll content areλr, the most relevant characteristic variables of total sugar and starch are Db, the most relevant characteristic variables of reducing sugar and total nitrogen are Rg, and the most relevant characteristic variables of leaf moisture content are (SDr-SDb)/(SDr+SDb). the most relevant characteristic variables of nicotine are SDr, The coefficient of determination of estimating models of tobacco biophysical and biochemical indexes were significant. The test show that the good predictive effect can be obtained by the estimating models by the sample randomly selected.(3) The relationships between Mineral elements、neutral aroma components、quality indexes of flue-cured tobacco and the hyperspectral parameters of canopy、leaf、cured tobacco leaf were analyzed, and the estimating models of those indexes were established by stepwise regression. The test indicating that the correlation relationship was exist between the hyperspectral parameters of tobacco and the mineral elements、neutral aroma components、quality indexes in different levels of water and nitrogen, and those parameters can be predicted accurately by the stepwise regression model. Consequently, it is feasible to predict the indexes of physiological and biochemical by the hyperspectral estimating models and the results can be used as the reference for monitoring the growth of tobacco and tobacco field management.(4) Although, most techniques of spectrum analysis have been studied and applied for precision agriculture. Currently, the models were established through extracting vegetation indexes by the hyperspectral reflectance. Estimation of vegetation index esmethod can achieve good results in predicting a variety of physiological and biochemical indexes of tobacco, but because of its general information using only a few wavelengths, the model predictive ability and stability cannot be guaranteed. The PCA method can effectively reduce dimension, and keeping the original important spectrum information. The information of each wave band can be excavated to achieve complementarities of information between each band, and to reduce the random interference from the little band. Therefore, the PCA method is dependable and universal fit. The neural network method has a strong nonlinear mapping ability, and it does not require normal distribution of data. The neural network model has strong ability of linear and nonlinear fitting, Has the incomparable superiority in the data fitting and simulation,but it is difficult to explain the decision by neural networks and the process of producing output.