Search

• The more fitting points, the smaller the fractional-order derivative spectrum. • Fractional-order derivative can improve the correlation with total nitrogen content. • Uninformative variable elimination can reduce the data dimension. • The accuracy of the model constructed with sensitive bands was better. • The number of fitting points, order, and bands of the best model are 40, 0.67 and 272. Realizing the real-time nondestructive monitoring of soil total nitrogen (STN) content is an important task to promote precision agriculture development. In this study, a water regulation experiment was performed. Winter wheat and summer soybean soils were taken as the research objects. The effects of fractional-order derivatives (FOD) on soil spectral reflectance were studied, and the hyperspectral monitoring models of STN content were constructed based on the full spectrum band and the uninformative variable elimination (UVE) -extracted band. The main results were as follows: The FOD showed a gradual change in the integer-order derivative (IOD). When calculating the FOD, the more fitting points there were, the smaller the FOD spectrum, and the higher the correlation with the STN content. In the range of orders 0–1, the correlation increased with increasing order. In the range of orders 1–2, the correlation first decreased and then increased. When constructing the model, stepwise multiple linear regression (SMLR) was used, but the accuracies of the SMLR models were low. Afterward, the bands were extracted by UVE, and the models were constructed by partial least squares regression (PLSR). The number of bands extracted by UVE ranged from 11 to 534, mainly in the range of 1000–2450 nm. Under the same conditions, the accuracy of the PLSR model constructed with the UVE-extracted bands was higher than that of the model constructed with the full band spectrum. Among all PLSR models, when the number of fitting points, the order, and the number of bands were 40, 0.67, and 272, respectively, the model accuracy was the highest. The coefficient of determination of cross-validation (R2 cv), root mean square error of cross-validation (RMSE cv), coefficient of determination of validation (R2 v), root mean square error of validation (RMSE v), and relative analysis error of validation (RPD v) of the model were 0.7317, 0.2244 g·kg−1, 0.7937, 0.1976 g·kg−1, and 2.1904, respectively. This model could realize the hyperspectral monitoring of STN content in wheat and soybean fields. In this study, hyperspectral spectroscopy was used to monitor the STN content, which provided a theoretical basis and technical support for the observations. [ABSTRACT FROM AUTHOR]

For rapid estimation of soybean chlorophyll by hyperspectral technology, 76 soybean varieties were investigated by simulation of Landsat-8 satellite bands through spectral resampling technology and vegetation indices. Four different modeling methods viz. Support Vector Machine (SVM), Multiple Linear Regression (MLR), Partial Least Squares Regression (PLSR), and Back Propagation Neural Network (BPNN), were combined to analyze the response characteristics of resampled spectra to predict soybean chlorophyll and a prediction model based on vegetation indices was constructed. The results revealed that chlorophyll concentration and the corresponding spectral reflectance were inversely related. Also, Normalized Differential Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Transformed Vegetation Index (TVI) and Ratio Vegetation Index (RVI) were significantly correlated with chlorophyll and the GNDVI exhibited significant correlation with the whole growth period (r = 0.70). Furthermore, the BPNN model was the best in predicting the chlorophyll contents and vegetation index of the soybean at seed-filling stage (R v 2 = 0.846, RMSE v = 0.384, RPD v = 2.413). Based on the results, we propose that after spectral resampling, the soybean leaf chlorophyll content can be effectively predicted by BPNN with a combination of vegetation indices. [ABSTRACT FROM AUTHOR]

• RF model for FSP-OC outperformed than other models. • Best characteristic wavelength set achieved best prediction on SOC components. • Wavelet decomposition scales affected the band extraction and model performance. • A new method for optimal combination of characteristic wavelengths was proposed. Soil Organic Carbon (SOC) fixation plays a vital role in reducing greenhouse gas emissions and the formation of soil aggregates is recognized as a crucial process in SOC fixation. In the current study, we characterized spectral information on SOC and aggregate Organic carbon (OC) components by using a rapid and robust method. Soil samples were collected from different layers to obtain the SOC and aggregate OC components to record corresponding spectra. The Continuous Wavelet Transform (CWT) spectra were then combined with various characteristic wavelength selection methods to build Multivariate Linear Regression (MLR), BP neural network (BP), and Random Forest (RF) models. Finally, the optimal combination of characteristic wavelengths based on the best-performing model for each index was determined. The results demonstrated that the model's accuracy, which was built using the set of characteristic wavelengths of the best model for all metrics, was higher than that of the best model for a single metric. This was due to the correlation between SOC and aggregate OC components. The best set of characteristic wavelengths were those selected by the Sequential Backward Selection method with wavelet decomposition scales of 23, 26, and 27, and those selected by the Minimum Redundancy Maximum Relevance method with wavelet decomposition scale 27. The RF model for Free silt and clay particles organic carbon (FSP-OC) showed the highest accuracy among the four SOC prediction models. In addition, it was discovered that as the wavelet decomposition scales become more similar, the extracted characteristic wavelengths also become more similar, resulting in similar model predictions. In summary, the accuracy of the predictive model for SOC and its constituents can be demonstrably improved by reasonable combination of characteristic wavelengths selection method and wavelet decomposition scale. [ABSTRACT FROM AUTHOR]

• Applying spectral resampling method to simulate the multi-sources satellite bands. • Constructed three-band indexes were high related with SAOC. • Evaluating the three calibrated methods on predicting SAOC. Soil aggregate-associated organic carbon (SAOC) is critical for maintaining soil quality and environmental health. To realize non-destructive and potentially applicable assessments on SAOC by using the hyperspectral technology, the resampling method was adopted to simulate the different satellites bands and the three multivariate methods of support vector machine (SVM), back propagation neural network (BPNN) and random forest (RF) were also implemented to calibrate the predictive models on SAOC. The results showed that the computed Three-band index 2 (TBI2) was proved to be highly related to both soil micro- and macroaggregate-associated organic carbon (OC) among all spectral indexes, and the highest correlation coefficient (R = −0.443) was obtained between the microaggregate-associated OC and TBI2 derived from the simulated bands 7, 4, 10 of Sentinel-2 satellite. For the macroaggregate-associated OC, the good correlation (R = −0.422) was also acquired with TBI2 which was constructed by using the simulated bands 7, 6, and 1 from the Sentinel-2 satellite. The calibrated models show that accepted accuracy for microaggregate-associated OC was achieved with R2 = 0.773, RMSE = 0.853, RPD = 1.808, and it outperformed than macroaggregates model (R2 = 0.709; RMSE = 0.969, RPD = 1.255) which was limited in the practical application with lower predictive accuracy. The results of correlation analysis and multivariate models performance indicated that the microaggregate-associated OC is highly related to responding to the spectral band simulated with different satellites than macroaggregates. This study provides a possible way to link the hyperspectral and remote sensing satellites, and open up opportunity for large-scale estimation on SAOC. [ABSTRACT FROM AUTHOR]

• This study explored the monitoring performance of vegetation index from the perspective of algebraic operation. • The performance of three-band index is better than two-band index. • Optimal band distribution and monitoring performance of vegetation index is greatly influenced by calculation formula. The vegetation indices (VIs) derived from the different band combinations can be used for monitoring crop quality traits. We conducted field experiments over two years time to investigate critical growth stages across four varieties and by using different nitrogen (N) application rates. In order to explore and evaluate the performance of different VIs on estimation of the canopy chlorophyll content (CCC) of winter wheat, the published and modified indices were optimized by using the random band combination through original spectrum (OS) and first-order differential (FD) treatment. The results showed that the first derivative processing improved the correlation between the red edge band and winter wheat CCC. The three-band VI can break the restriction of the number of bands on the extraction of target information, relieved the saturation problem of the dual-band VI, and improved the monitoring accuracy of winter wheat CCC. The index 2 × R1-R2-R3 was found to be the best VI for assessing the CCC of winter wheat based on the original and first-order differential spectrum (calibration R2 > 0.733), R2 and RMSE of validation set were 0.688, 0.755 and 1.515, 1.336, respectively. In addition, the index expression formula R1/(R2 × R3) was recommended as a favorable choice for monitoring the agronomic traits of crop. Moreover, the VI is suggested to use at the red edge position band to monitor crop growth indicators. In conclusion, the use of VI can better monitor winter wheat CCC which could provide a theoretical basis for precision agriculture. [ABSTRACT FROM AUTHOR]

• FOSGD was apply to the pretreatment of hyperspectral data of corn leaves. • The best PLS model was obtained when the fractional order equaled to 1 and 2. • The PLS model with CARS and random frog wavelength selection got the best results. • The method in this paper may provide an effective way to predict the LWC of corn. Water status is critical since it affects photosynthetic efficiency and limits crop yield. Thus it is essential to calculate the water content of crop fast and nondestructively. In this work, hyperspectral reflectance in 900–1700 nm was used to estimate water content of corn leaves. The fractional order Savitzky-Golay derivation (FOSGD) was used to pretreat the hyperspectral data. The result indicated that the best performance of partial least square (PLS) model was obtained when the fractional order equaled to 1 and 2. In order to make a PLS model simpler, three wavelength selection approaches, variable importance in the projection (VIP), competitive adaptive reweighted sampling (CARS), and random frog were performed to extract the sensitive wavelengths. The result demonstrated that the PLS model with CARS and random frog wavelength selection did exhibit the best performance and extracted 23 feature wavelengths, with the coefficient of determination (R2) up to 0.91 for calibration and 0.92 for validation. The root mean squared error (RMSE) decreased to 0.049 for calibration and 0.044 for validation. Meanwhile, 80 percent of the feature wavelengths extracted by these two methods were similar. The findings of the current study may provide an effective way to predict the water content of corn leaves using a combination of FOSGD-CARS-PLS or FOSGD-random frog-PLS. [ABSTRACT FROM AUTHOR]