Detection of nitrogen-overfertilized rice plants with leaf positional difference in hyperspectral vegetation index.

The main objective of this work was to compare the applicability of the single leaf (the uppermost leaf L1 and the third uppermost leaf L3) modified simple ratio (mSR(705) index) and the leaf positional difference in the vegetation index between L1 and L3 (mSR(705L1)-mSR(705L3)) in detecting nitrogen (N)-overfertilized rice plants. A field experiment consisting of three rice genotypes and five N fertilization levels (0, 75, 180, 285, and 390 kg N/ha) was conducted at Xiaoshan, Hangzhou, Zhejiang Province, China in 2008. The hyperspectral reflectance (350-2500 nm) and the chlorophyll concentration (ChlC) of L1 and L3 were measured at different stages. The mSR(705L1) and mSR(705L3) indices appeared not to be highly sensitive to the N rates, especially when the N rate was high (above 180 kg N/ha). The mean mSR(705L1)-mSR(705L3) across the genotypes increased significantly (P<0.05) or considerably from 180 to 285 kg N/ha treatment and from 285 to 390 kg N/ha treatment at all the stages. Also, use of the difference (mSR(705L1)-mSR(705L3)) greatly reduced the influence of the stages and genotypes in assessing the N status with reflectance data. The results of this study show that the N-overfertilized rice plants can be effectively detected with the leaf positional difference in the mSR(705) index.

[Discrimination and spectral response characteristic of stress leaves infected by rice Aphelenchoides besseyi Christie].

An ASD Field Spec Pro Full Range spectrometer was used to acquire the spectral reflectance of healthy and diseased leaves infected by rice Aphelenchoides besseyi Christie, which were cut from rice individuals in the paddy field. Firstly, foliar pigment content was investigated. As compared with healthy leaves, the total chlorophyll and carotene contents (mg x g(-1)) of diseased leaves decreased 18% and 22%, respectively. The diseased foliar content ratio of total chlorophyll to carotene was nearly 82% of the healthy ones. Secondly, the response characteristics of hyperspectral reflectance of diseased leaves were analyzed. The spectral reflectance in the blue (450-520 nm), green (520-590 nm) and red (630-690 nm) regions were 2.5, 2 and 3.3 times the healthy ones respectively due to the decrease in foliar pigment content, whereas in the near infrared (NIR, 770-890 nm) region was 71.7 of the healthy ones because of leaf twist, and 73.7% for shortwave infrared (SWIR, 1 500-2 400 nm) region, owing to water loss. Moreover, the hyperspectral feature parameters derived from the raw spectra and the first derivative spectra were analyzed. The red edge position (REP) and blue edge position (BEP) shifted about 8 and 10 nm toward the short wavelengths respectively. The green peak position (GPP) and red trough position (RTP) shifted about 8.5 and 6 nm respectively toward the longer wavelengths. Finally, the area of the red edge peak (the sum of derivative spectra from 680 to 740 nm) and red edge position (REP) as the input vectors entered into C-SVC, which was an soft nonlinear margin classification method of support vector machine, to recognize the healthy and diseased leaves. The kernel function was radial basis function (RBF) and the value of punishment coefficient (C) was obtained from the classification model of training data sets (n = 138). The performance of C-SVC was examined with the testing sample (n = 126), and healthy and diseased leaves could be successfully differentiated without errors. This research demonstrated that the response feature of spectral reflectance was obvious to disease stress in rice leaves, and it was feasible to discriminate diseased leaves from healthy ones based on C-SVC model and hyperspectral reflectance.

Spatial and seasonal characterization of net primary productivity and climate variables in southeastern China using MODIS data.

We developed a sophisticated method to depict the spatial and seasonal characterization of net primary productivity (NPP) and climate variables. The role of climate variability in the seasonal variation of NPP exerts delayed and continuous effects. This study expands on this by mapping the seasonal characterization of NPP and climate variables from space using geographic information system (GIS) technology at the pixel level. Our approach was developed in southeastern China using moderate-resolution imaging spectroradiometer (MODIS) data. The results showed that air temperature, precipitation and sunshine percentage contributed significantly to seasonal variation of NPP. In the northern portion of the study area, a significant positive 32-d lagged correlation was observed between seasonal variation of NPP and climate (P<0.01), and the influences of changing climate on NPP lasted for 48 d or 64 d. In central southeastern China, NPP showed 16-d, 48-d, and 96-d lagged correlation with air temperature, precipitation, and sunshine percentage, respectively (P<0.01); the influences of air temperature and precipitation on NPP lasted for 48 d or 64 d, while sunshine influence on NPP only persisted for 16 d. Due to complex topography and vegetation distribution in the southern part of the study region, the spatial patterns of vegetation-climate relationship became complicated and diversiform, especially for precipitation influences on NPP. In the northern part of the study area, all vegetation NPP had an almost similar response to seasonal variation of air temperature except for broad crops. The impacts of seasonal variation of precipitation and sunshine on broad and cereal crop NPP were slightly different from other vegetation NPP.

Discrimination of rice panicles by hyperspectral reflectance data based on principal component analysis and support vector classification.

Detection of crop health conditions plays an important role in making control strategies of crop disease and insect damage and gaining high-quality production at late growth stages. In this study, hyperspectral reflectance of rice panicles was measured at the visible and near-infrared regions. The panicles were divided into three groups according to health conditions: healthy panicles, empty panicles caused by Nilaparvata lugens Stål, and panicles infected with Ustilaginoidea virens. Low order derivative spectra, namely, the first and second orders, were obtained using different techniques. Principal component analysis (PCA) was performed to obtain the principal component spectra (PCS) of the foregoing derivative and raw spectra to reduce the reflectance spectral dimension. Support vector classification (SVC) was employed to discriminate the healthy, empty, and infected panicles, with the front three PCS as the independent variables. The overall accuracy and kappa coefficient were used to assess the classification accuracy of SVC. The overall accuracies of SVC with PCS derived from the raw, first, and second reflectance spectra for the testing dataset were 96.55%, 99.14%, and 96.55%, and the kappa coefficients were 94.81%, 98.71%, and 94.82%, respectively. Our results demonstrated that it is feasible to use visible and near-infrared spectroscopy to discriminate health conditions of rice panicles.

[Estimating the severity of rice brown spot disease based on principal component analysis and radial basis function neural network].

An ASD Field Spec Pro Full Range spectrometer was used here to acquire the spectral reflectance of healthy and disease leaves cut from rice plants in the field. The leaf disease severity of rice brown spot was determined by estimating the percentage of infected surface area of rice leaves in the laboratory through phytopathologist's observation. Three steps were taken to estimate leaf disease severity of rice brown spot. The first step was that different spectra transforming methods, namely, resampling spectrum (10 nm interval), the first- and second-order derivative spectrum based on raw hyperspectral reflectance, were conducted. The second step was that the principal component analysis (PCA) was examined to obtain the principal components (PCs) from the above transformed spectra to reduce the spectra dimensions of hyperspectral reflectance and simplify the data structure of hyperspectra. The last step was that the resampling and PCs spectra entered the Radial Basis Function neural network (RBFN) as the input vectors, and the disease severity of rice brown spot entered RBFN as the target vectors. RBFN is an effective feed forward propagation neural network, which is based on the linear combinations of corresponding radial basis functions. In general RBFN can be used to solve the problems such as regression or classification with high operation rate and efficient extrapolation capability, and quickly designed with zero error to approximate functions. The total dataset (n = 262) was divided into two subsets, in which three quarters (n = 210) was the training subset to train the neural network, and the remaining quarter (n = 52) was the testing dataset to conduct the performance analysis of neural network. The spread constants of RBFN and various data processing methods were investigated in detail. The best prediction result was obtained by PCs spectra based on the first-order derivative using RBFN model, the root mean square of prediction error (RMSE) was small (7.73%) in the testing dataset, and the next was the resampling spectra with RMSE of 8.75%. This research demonstrated that it was feasible and reliable to estimate the disease severity of rice brown spot based on PCA-RBFN and hyperspectral reflectance at the leaf level.

[Study on hyperspectral estimation model of crop vegetation cover percentage].

In order to boost the study and application of hyperspectral remote sensing for the estimation of crop vegetation coverage percentage, an ASD FieldSpec Pro FRTM spectroradiometer was used for canopy spectral measurements of rape, corn and rice at different vegetation cover levels and photos of individual plants were taken simultaneously in order to calculate the vegetation cover percentage in computer. Firstly, data of three crops respectively and the mixed data of them were used to make correlation analysis between vegetation coverage percentage and reflectance spectra There was a high correlation between them and no obvious difference in correlation coefficient among different types of crop in the region of blue, red and near-infrared band. This indicated that it was feasible to make correlation analysis and build estimation model using mixed data Secondly, mixed data were used as unique analytical data to calculate red edge variables and pair combination of bands in the region of blue, red and near-infrared band was used to calculate normal difference vegetation index (NDVI). Hyperspectral estimation models with NDVI and red edge variable as independent variable were built individually. The correlation coefficient of the former was larger than the latter, which indicated that NDVI was most effective for the estimation of vegetation coverage percentage. Effective wavelength combinations of NDVI for vegetation cover percentage estimation were determined based on the principle of higher correlation coefficient. NDVI combined with bands in the regions from 350 to 590 nm and from 710 to 1150 nm or bands in the regions from 590 to 710 nm and from 710 to 1300 nm are most effective for vegetation coverage percentage estimation. The best estimation model is simple quadratic equation using NDVI(696-921) as independent variable. The correlation coefficient matrix shows that most of the correlation coefficients of vegetation coverage percentage and NDVI combined with bands in the regions from 630 to 690 nm and from 760 to 900 nm are larger than 0.8. These two band regions correspond to TM3 and TM4 of landsat 4,5,7. It proves that NDVI(TM3-TM4) can be used to and has been used to simulate vegetation coverage percentage. In order to further the study, TM3 and TM4 of Landsat5 was modeled according to spectral response function to calculate NDVI. Correlation analysis was made with NDVI and corresponding vegetation coverage percentage. The correlation coefficient of them was 0.80 and the regression equation was verified by experimental data. This is exploratory research for the calculation of vegetation coverage percentage using TM data in large area.

[Predicting nitrogen concentrations from hyperspectral reflectance at hyperspectral reflectance at leaf and canopy for rape].

An experiment was designed to determine whether nitrogen concentrations could be predicted from reflectance (R) spectra of rape leaves in laboratory, and, if so, whether the predictive spectral features could be correlated with nitrogen concentration of simple canopies of rape. The best predictors for nitrogen in leaves appeared with first-difference transformations of R, and the bands selected were similar to those found in other studies. Shortwave infrared bands were best predictors for nitrogen. In the shortwave infrared region, however, the absolute differences in reflectance at critical bands were extremely small, and the bands of high correlation were narrow. High spectral and radiance resolution are required to resolve these differences accurately. Variability in canopy reflectance in shortwave infrared region was at least an order of magnitude beyond that necessary to detect signals from chemicals. The variability in first-difference R and log 1/R on canopy scales were related to the arrangement of trees with respect to direct solar radiation, instrument noise, leaf fluttering, and small change in atmospheric moisture. The first-difference of reflectance R based regressions prediction of nitrogen concentration at canopy level gets a good fitness.

Monitoring rice nitrogen status using hyperspectral reflectance and artificial neural network.

Over use of nitrogen fertilization can result in groundwater pollution. Tools that can rapidly quantify the nitrogen status are needed for efficient fertilizer management and would be very helpful in reducing the environmental pollution caused by excessive nitrogen application. Remote sensing has a proven ability to provide spatial and temporal measurements of surface properties. In this study, the MLR (multiple linear regression) and ANN (artificial neural network) modeling methods were applied to the monitoring of rice N (nitrogen concentration, mg nitrogen g(-1) leaf dry weight) status using leaf level hyperspectral reflectance with two different input variables, and as a result four estimation models were proposed. RMSE (root-mean-square error), REP (relative error of prediction), R2 (coefficient of determination), as well as the intercept and slope between the observed and predicted N were used to test the performance of models. Very good agreements between the observed and the predicted N were obtained with all proposed models, which was especially true for the R-ANN (artificial neural network based on reflectance selected using MLR) model. Compared to the other three models, the R-ANN model improved the results by lowering the RMSE by 14.2%, 32.1%, and 31.5% for the R-LR (linear regression based on reflectance) model, PC-LR (linear regression based on principal components scores) model, and PC-ANN (artificial neural network based on principal components scores) model, respectively. It was concluded that the ANN algorithm may provide a useful exploratory and predictive tool when applied on hyperspectral reflectance data for nitrogen status monitoring. Besides, although the performance of MLR was superior to PCA used for ANN inputs selection, the encouraging results of PC-based models indicated the promising potential of ANN combined with PCA application on hyperspectral reflectance analysis.

Characterizing and estimating rice brown spot disease severity using stepwise regression, principal component regression and partial least-square regression.

Detecting plant health conditions plays a key role in farm pest management and crop protection. In this study, measurement of hyperspectral leaf reflectance in rice crop (Oryzasativa L.) was conducted on groups of healthy and infected leaves by the fungus Bipolaris oryzae (Helminthosporium oryzae Breda. de Hann) through the wavelength range from 350 to 2,500 nm. The percentage of leaf surface lesions was estimated and defined as the disease severity. Statistical methods like multiple stepwise regression, principal component analysis and partial least-square regression were utilized to calculate and estimate the disease severity of rice brown spot at the leaf level. Our results revealed that multiple stepwise linear regressions could efficiently estimate disease severity with three wavebands in seven steps. The root mean square errors (RMSEs) for training (n=210) and testing (n=53) dataset were 6.5% and 5.8%, respectively. Principal component analysis showed that the first principal component could explain approximately 80% of the variance of the original hyperspectral reflectance. The regression model with the first two principal components predicted a disease severity with RMSEs of 16.3% and 13.9% for the training and testing dataset, respectively. Partial least-square regression with seven extracted factors could most effectively predict disease severity compared with other statistical methods with RMSEs of 4.1% and 2.0% for the training and testing dataset, respectively. Our research demonstrates that it is feasible to estimate the disease severity of rice brown spot using hyperspectral reflectance data at the leaf level.