Spectral Variables Research Articles

Novel Incremental Regularized Extreme Learning Machine Method Based on Improved Particle Swarm Optimization for Nonlinear Multivariate Calibration of Visible and Near-Infrared Spectroscopy

Novel incremental regularized extreme learning machine (IR-ELM) based upon improved particle swarm optimization (IPSO-IR-ELM) is reported for the nonlinear multivariate calibration of visible/near-infrared (Vis/NIR) spectroscopy. IR-ELM is employed to construct a nonlinear calibration model for samples. Combined with IPSO, three parts of the IR-ELM algorithm are intelligently optimized. First, the regularization coefficient of the initial network in IR-ELM is optimized by IPSO. Second, IPSO is used again to select the optimal input weights and hidden biases while adding new hidden nodes in IR-ELM. Third, the 2-norm of the output matrix in IR-ELM is introduced as the conditional constraint in IPSO for updating the particle position. The performance of the reported method was tested with two Vis/NIR spectra datasets: blood hemoglobin and water pH. Key spectral variables were selected by successive projections algorithm and employed to establish the calibration models. Compared with partial least squares, ELM, error minimized extreme learning machine, and IR-ELM, IPSO-IR-ELM achieved the highest accuracy and best generalization. For the blood hemoglobin dataset, the RMSEP (root mean square error of prediction) was 0.210 g·dL−1, and the R p 2 (coefficient of determination of prediction) was 0.973. For the water pH dataset, the RMSEP was 0.825, and the R p 2 was 0.899. The results demonstrate that IPSO-IR-ELM is an alternative nonlinear multivariate calibration approach for Vis/NIR spectroscopy.

Land use/land cover classification using hyperspectral soil reflectance features in the Eastern Himalayas, India

Information on periodic land use/land cover (LULC) changes are imperative for regional agricultural planning and policymaking. In this study, 332 soil samples were collected across five LULC types (coal mine degraded land, grassland, shifting cultivation or Jhum, plantation, and upland agriculture) of Eastern Indian Himalayas and subsequently characterized via traditional laboratory protocols and hyperspectral sensor. The soil physicochemical data and spectral data were separately used to classify six LULC types using three machine learning (ML) algorithms [support vector machine (SVM), random forest (RF), and K-nearest neighbors (KNN)]. Results indicated that the change in LULC directly impacted the dynamics of soil properties. The principal component analysis highlighted the interrelationships between the suite of soil physicochemical properties in classifying soil samples from different LULC types. All three ML algorithms exhibited that the physicochemical properties of the soil can perfectly classify LULC types. Spectral data from the hyperspectral sensor also demonstrated good classification accuracy. Overall, SVM performed better than RF and KNN, producing 93% and 78% classification accuracy using soil properties and spectral data, respectively. Moreover, the RF algorithm could select the influential soil and spectral variables for classifying LULC types. In the future, the approach tested herein can be used to classify several LULC types into the archived national soil spectral database. More research is needed to include a wide range of soil data and more LULC classes for a comprehensive classification using advanced deep learning tools.

Multivariate Calibration of the Composition of Low-Alloy Steels Using Preprocessed Low-Resolution Emission Spectra with Spectral Variables Selection

Multivariate Calibration of the Composition of Low-Alloy Steels Using Preprocessed Low-Resolution Emission Spectra with Spectral Variables Selection

Handheld and benchtop vis/NIR spectrometer combined with PLS regression for fast prediction of cocoa shell in cocoa powder

The fermented and dried cocoa beans are peeled, either before or after the roasting process, as peeled nibs are used for chocolate production, and shell content in cocoa powders may result from economically motivated adulteration (EMA), cross-contamination or misfits in equipment in the peeling process. The performance of this process is carefully evaluated, as values above 5% (w/w) of cocoa shell can directly affect the sensory quality of cocoa products. In this study chemometric methods were applied to near-infrared (NIR) spectra from a handheld (900–1700 nm) and a benchtop (400–1700 nm) spectrometers to predict cocoa shell content in cocoa powders. A total of 132 binary mixtures of cocoa powders with cocoa shell were prepared at several proportions (0 to 10% w/w). Partial least squares regression (PLSR) was used to develop the calibration models and different spectral preprocessing were investigated to improve the predictive performance of the models. The ensemble Monte Carlo variable selection (EMCVS) method was used to select the most informative spectral variables. Based on the results obtained with both benchtop (R2P = 0.939, RMSEP = 0.687% and RPDP = 4.14) and handheld (R2P = 0.876, RMSEP = 1.04% and RPDP = 2.82) spectrometers, NIR spectroscopy combined with the EMCVS method proved to be a highly accurate and reliable tool for predicting cocoa shell in cocoa powder. Even with a lower predictive performance than the benchtop spectrometer, the handheld spectrometer has potential to specify whether the amount of cocoa shell present in cocoa powders is in accordance with the Codex Alimentarius specifications.

Multivariate Statistical Models for the Authentication of Traditional Balsamic Vinegar of Modena and Balsamic Vinegar of Modena on 1H-NMR Data: Comparison of Targeted and Untargeted Approaches.

This work aimed to compare targeted and untargeted approaches based on NMR data for the construction of classification models for Traditional Balsamic Vinegar of Modena (TBVM) and Balsamic Vinegar of Modena (BVM). Their complexity in terms of composition makes the authentication of these products difficult, which requires the employment of several time-consuming analytical methods. Here, 1H-NMR spectroscopy was selected as the analytical method for the analysis of TVBM and BVM due to its rapidity and efficacy in food authentication. 1H-NMR spectra of old (>12 years) and extra-old (>25 years) TVBM and BVM (>60 days) and aged (>3 years) BVM were acquired, and targeted and untargeted approaches were used for building unsupervised and supervised multivariate statistical modes. Targeted and untargeted approaches were based on quantitative results of peculiar compounds present in vinegar obtained through qNMR, and all spectral variables, respectively. Several classification models were employed, and linear discriminant analysis (LDA) demonstrated sensitivity and specificity percentages higher than 85% for both approaches. The most important discriminating variables were glucose, fructose, and 5-hydroxymethylfurfural. The untargeted approach proved to be the most promising strategy for the construction of LDA models of authentication for TVBM and BVM due to its easier applicability, rapidity, and slightly higher predictive performance. The proposed method for authenticating TBVM and BVM could be employed by Italian producers for safeguarding their valuable products.

Estimation and mapping of pasture biomass in Mongolia using machine learning methods

Mongolian pasture plays an essential role in the national economy. Reliable pasture biomass estimation is indispensable to support the agricultural sector and also sustainable livelihood in the country. The aim of this study is to determine an appropriate method to estimate and map pasture biomass in a forest-steppe area of Mongolia. For this purpose, machine learning methods such as random forest (RF), support vector machine (SVM), and partial least squares regression (PLSR) were compared. As data sources, spectral indices derived from Sentinel-2B image of 2019 and field-measured biomass sample datasets were used. To determine the optimal spectral predictor variables, initially, 20 spectral indices were evaluated using the PLSR. Of these, five indices (i.e. ATSAVI2, EVI, GRVI, IPVI and MSR) with the highest correlation coefficients (r ≥ 0.94) were considered for further analysis. These indices were also examined and validated by a variable importance analysis. Then, the RF, SVM, and PLSR models were applied to predict and map pasture biomass using the selected five indices. The PLSR method demonstrated the highest accuracy with coefficient of determination (R2) =0.899 and root mean square error (RMSE)=10.560 g/m2. The SVM technique showed the second highest accuracy with R2=0.837 and RMSE = 12.881 g/m2. The RF model gave the lowest accuracy with R2=0.823 and RMSE = 13.430g/m2. Our research showed that different machine learning models might be applied (because in all cases R2>0.82) for a pasture biomass estimation and mapping in the selected test site, but the best result could be achieved by the use of the PLSR.

Aboveground biomass estimation in dry forest in northeastern Brazil using metrics extracted from sentinel-2 data: Comparing parametric and non-parametric estimation methods

Caatinga is a dry forest, which occupies almost one million square kilometers in the Brazilian territory, representing more than 11% of the territory. However because of the extensive areas to be inventoried and the difficulties in collecting data from this natural vegetation using traditional techniques, the operation becomes very costly, often resulting in low sampling intensity owing to cost and time constraints. Therefore, remote sensing tools have become very useful for minimizing the difficulties caused by traditional methods. Given this context, our study aims to evaluate the performance of spectral and texture variables extracted from an MSI/Sentinel-2 image in the estimation and mapping of AGB in a fragment (natural vegetation area) of Caatinga vegetation in northeastern Brazil. The original reflectance values, the ratio between spectral bands, vegetation indices, texture metrics (extracted from the 3 × 3, 5 × 5, 7 × 7, and 9 × 9-pixel windows), and fraction images (proportion of the vegetation, soil, and shadow/water components) were associated with AGB. These variables were derived from a forest inventory with systematic sampling composed of 40 plots of 900 m2 each. We evaluated the estimates using a parametric model of multiple linear regression (MLR) and a non-parametric model of an artificial neural network (ANN). The texture metrics (angular second moment, correlation, entropy, and contrast), extracted in a 7 × 7-pixel window, were the best predictors. The parametric model was the most accurate for AGB estimation. The MLR and ANN techniques showed bias validation of 2.58% and 2.79%, and relative root mean square errors of 12.02% and 16.60%, respectively. AGB estimates ranged from 1.49 Mg ha−1 to approximately 73 Mg ha−1 on the MLR and ANN maps. The textural metrics showed the potential to accurately predict AGB, providing valuable information for the use of open access data. Such information is very useful in the management of natural resources in large areas, such as the Caatinga vegetation. The use of optical data for AGB estimation in the region is relatively unexplored and this study contributes to filling that gap or better understand their application for deriving AGB in a dry forest in the semiarid region of Sergipe.

Prediction of essential oil content in spearmint (Mentha spicata) via near-infrared hyperspectral imaging and chemometrics

Spearmint (Mentha spicata L.) is grown for its essential oil (EO), which find use in food, beverage, fragrance and other industries. The current study explores the ability of near infrared hyperspectral imaging (HSI) (935 to 1720 nm) to predict, in a rapid, nondestructive manner, the essential oil content of dried spearmint (0.2 to 2.6% EO). Spectral values of spearmint samples varied considerably with spatial coordinates, and so the use of averaging the spectral values of a surface scan was warranted. Data preprocessing was done with Multiplicative Scatter Correction (MSC) or Standard Normal Variate (SNV). Selection of spectral input variables was done with Least Absolute Shrinkage and Selection Operator (LASSO), Principal Component Analysis (PCA) or Partial Least Squares (PLS). Regression was executed with linear regression (LASSO, PLS regression, PCA regression), Support Vector Machine (SVM) regression, and Multilayer Perceptron (MLP). The best prediction of EO concentration was achieved with the combination of MSC or SNV preprocessing, PLS dimension reduction, and MLP regression (1 hidden layer with 6 nodes), achieving a good prediction with a ratio of performance to deviation (RPD) of 2.84 ± 0.07, an R2 of prediction of 0.863 ± 0.008, and a RMSE of prediction of 0.219 ± 0.005% EO. These results show that NIR-HSI is a viable method for rapid, nondestructive analysis of EO concentration. Future work should explore the use of NIR in the visible spectrum, the use of HSI for determining EO in other plant materials and the potential of HSI to determine individual compounds in these solid plant/food matrices.

Predicting coffee water potential from spectral reflectance indices with neural networks

Leaf water potential is one of the main parameters used to assess water relations in plants by revealing levels of tissue hydration. It is commonly measured with the Scholander pressure chamber; which demands hard work and a time-consuming process. On the other hand, there is a diversified literature demonstrating the assessments of several plant variables via indices of leaf reflectance, that also present direct and indirect relationships with water potential. The aim of this work is to exploit spectral variables to estimate the water potential of coffee plants by using computational intelligence approaches. Data was collected in the cities of Santo Antônio do Amparo and Diamantina, Brazil, from 2014 to 2018. Two neural networks (Multi-Layer Perceptron) were designed to estimate and classify leaf water potential based on spectral variables. Moreover, a classifier and an estimator based on decision tree were also developed. The results showed that the artificial neural network model was superior as an estimator when compared with the decision tree model, with an average confidence index of 0.8550. On the other hand, decision trees showed a slightly higher performance as a classifier, with an overall accuracy of 88.8% and a Kappa index of 70.07%. We concluded that the leaf reflectance indices may be properly used to build accurate models for estimating coffee water potential. The indices PRI, NDVI, CRI1 and SIPI were the most relevant ones for estimating and classifying the coffee water potential.

Inter-Seasonal Estimation of Grass Water Content Indicators Using Multisource Remotely Sensed Data Metrics and the Cloud-Computing Google Earth Engine Platform

Indicators of grass water content (GWC) have a significant impact on eco-hydrological processes such as evapotranspiration and rainfall interception. Several site-specific factors such as seasonal precipitation, temperature, and topographic variations cause soil and ground moisture content variations, which have significant impacts on GWC. Estimating GWC using multisource data may provide robust and accurate predictions, making it a useful tool for plant water quantification and management at various landscape scales. In this study, Sentinel-2 MSI bands, spectral derivatives combined with topographic and climatic variables, were used to estimate leaf area index (LAI), canopy storage capacity (CSC), canopy water content (CWC) and equivalent water thickness (EWT) as indicators of GWC within the communal grasslands in Vulindlela across wet and dry seasons based on single-year data. The results illustrate that the use of combined spectral and topo-climatic variables, coupled with random forest (RF) in the Google Earth Engine (GEE), improved the prediction accuracies of GWC variables across wet and dry seasons. LAI was optimally estimated in the wet season with an RMSE of 0.03 m−2 and R2 of 0.83, comparable to the dry season results, which exhibited an RMSE of 0.04 m−2 and R2 of 0.90. Similarly, CSC was estimated with high accuracy in the wet season (RMSE = 0.01 mm and R2 = 0.86) when compared to the RMSE of 0.03 mm and R2 of 0.93 obtained in the dry season. Meanwhile, for CWC, the wet season results show an RMSE of 19.42 g/m−2 and R2 of 0.76, which were lower than the accuracy of RMSE = 1.35 g/m−2 and R2 = 0.87 obtained in the dry season. Finally, EWT was best estimated in the dry season, yielding a model accuracy of RMSE = 2.01 g/m−2 and R2 = 0.91 as compared to the wet season (RMSE = 10.75 g/m−2 and R2 = 0.65). CSC was best optimally predicted amongst all GWC variables in both seasons. The optimal variables for estimating these GWC variables included the red-edge, near-infrared region (NIR) and short-wave infrared region (SWIR) bands and spectral derivatives, as well as environmental variables such as rainfall and temperature across both seasons. The use of multisource data improved the prediction accuracies for GWC indicators across both seasons. Such information is crucial for rangeland managers in understanding GWC variations across different seasons as well as different ecological gradients.

Alpha and beta diversity mapping in Indian tropical deciduous forests using high-fidelity imaging spectroscopy

The present study evaluates the potential of airborne hyperspectral and space-borne multispectral satellite datasets to determine alpha (α) and beta (β) diversity combining in-situ floristic composition in tropical deciduous forests (TDFs) of Mudumalai Tiger Reserve (MTR), India. 43 forest plots (each of 0.04 ha) were sampled to record the species diversity in MTR. The very high-resolution AVIRIS-NG hyperspectral images and temporal Sentinel-2 multispectral images were used to extract spectral variables by associating each field plot with corresponding pixels of the satellite images. A high positive correlation of field-based α-diversity (Shannon-Wiener diversity index (H′)) with AVIRIS-NG based α-diversity (R2: 0.83 (2016); R2: 0.82 (2018)) was observed compared to Sentinel-2 based α-diversity (R2: 0.50 to 0.63) with high correlation during post-monsoon (R2: 0.63). Principal coordinate analysis of AVIRIS-NG based β-diversity signifies a low diversity (H′: <1.00) in the eastern parts (0.13% of MTR) in contrast to moderate (H′: 1.00 to 2.00; 13.39% of MTR) to high diversity (H′ >2.00; 86.49% of MTR) in the remaining parts of MTR. The study highlights the application of high spectral-spatial AVIRIS-NG images for precise tree diversity mapping and the high temporal-spatial Sentinel-2 satellite images for assessing the phenological variations of the forest species. The integration of high spatial-hyperspectral and temporal multispectral satellite datasets with field-based observations for fine-scale tree species diversity mapping at geographical and temporal scales is the novelty of the study that may effectively contribute in devising effective forest management plans for biodiversity conservation.

Fourier-Transform Infrared Spectral Inversion of Soil Available Potassium Content Based on Different Dimensionality Reduction Algorithms

Estimating the available potassium (AK) in soil can help improve field management and crop production. Fourier-transform infrared (FTIR) spectroscopy is one of the most promising techniques for the fast and real-time analysis of soil AK content. However, the successful estimation of soil AK content by FTIR depends on the proper selection of appropriate spectral dimensionality reduction techniques. To magnify the subtle spectral signals concerning AK content and improve the understanding of the characteristic FTIR wavelengths of AK content, a total of 145 soil samples were collected in an agricultural site located in the southwest part of Sichuan, China, and three typical spectral dimensionality reduction methods—the successive projections algorithm (SPA), simulated annealing algorithm (SA) and competitive adaptive reweighted sampling (CARS)—were adopted to select the appropriate spectral variable. Then, partial least squares regression (PLSR) was utilized to establish AK inversion models by incorporating the optimal set of spectral variables extracted by different dimensionality reduction algorithms. The accuracy of each inversion model was tested based on the coefficient of determination (R2), root mean square error (RMSE) and mean absolute value error (MAE), and the contribution of the inversion model variables was explored. The results show that: (1) The application of spectral dimensionality reduction is a useful technique for isolating specific components of multicomponent spectra, and as such is a powerful tool to improve and expand the predicted potential of the spectroscopy of soil AK content. Compared with the SA and CARS algorithms, the SPA was more suitable for soil AK content inversion. (2) The inversion model results showed that the characteristic wavelengths were mainly around 777 nm, 1315 nm, 1375 nm, 1635 nm, 1730 nm and 3568–3990 nm. (3) Comparing the performances of different inversion models, the SPA–PLSR model (R2= 0.49, RMSE = 22.80, MAE = 16.82) was superior to the SA–PLSR and CARS–PLSR models, which has certain guiding significance for the rapid detection of soil AK content.

Rapid determination of essential oils functional groups using compositional methods and VisNIR spectroscopy

Essential oils (EOs) are natural products formed by plant volatile compounds. EOs are frequently used in the cosmetic and food industries as well as for domestic purposes, because of their physiochemical, biological and sensory properties. The functional groups (FG), corresponding to various chemical structures present in EOs, are responsible for their biological activities. Therefore, simple, rapid, and economical techniques suitable to characterize the EOs features by measuring their contents, are of great interest. Near-infrared spectroscopy (NIRS) highlights because of its rapidity, and being no-contaminant, as a potential solution. Multivariate correlation methods are commonly used to build NIRS calibrations. These methods were designed for the real space, that is for values comprised between - ∞ and + ∞. However, EOs components are co-dependent data restricted to a simplex space. These are the so-called compositional data (CoDa), needing specific methods to be correlated with a set of spectral explicative variables. In this study, compositional visible and near-infrared (VISNIRS) models have been assessed to quantify the FG of the analyzed EOs. For this purpose, the FG were organized according to their greater frequency in 1) alcohol; 2) ether; 3) ester; 4) aldehydes; 5) ketones, and the hydrocarbon fraction representing the remainder EOs mass, to characterize them. The approach of this study, based on compositional models from VISNIRS spectra, has provided a satisfactory predictive performance for the quantitative estimation of the main FG of the EOs. The proposed approach can be an alternative to traditional chemical methods to characterize EOs.

Chromium in soil detection using adaptive weighted normalization and linear weighted network framework for LIBS matrix effect reduction

Rapid and accurate detection of agricultural soil chromium (Cr) is of great significance for soil pollution assessment. Laser-induced breakdown spectroscopy (LIBS) could serve as a rapid and chemical-free method for hazardous metal analysis compared with conventional chemical methods. However, the detection of LIBS is interfered by uncertainty and matrix effect. In this study, an average strategy combined with linear weighted network (LWNet) was proposed to reduce the uncertainty. Adaptive weighted normalization-LWNet (AWN-LWNet) framework was proposed to reduce the matrix effect in two soil types. The results indicated that LWNet outperformed traditional machine learning and achieved the average relative error (ARE) of 2.08 % and 3.03 % for yellow brown soil and lateritic red soil, respectively. Moreover, LWNet could effectively mine Cr feature peaks even under the low spectral resolution. AWN-LWNet was the optimal model compared with commonly used models to reduce matrix effect (ARE=4.12 %). Besides, AWN-LWNet greatly reduced the number (from 22016 to 72) of spectral variables for model input. By extracting Cr peaks from models, the difference of Cr peaks intensity could be intuitively observed, which served as spectral interpretation for matrix effect reduction. The two methods have the potential to realize the detection of hazardous metals in soil by LIBS.

Nondestructive Determination of Leaf Nitrogen Content in Corn by Hyperspectral Imaging Using Spectral and Texture Fusion

The nitrogen content is an important indicator affecting corn plants’ growth status. Most of the standard hyperspectral imaging-based techniques for nondestructive detection of crop nitrogen content use a single feature as the input variable of the model, which reduces the generalization ability of the prediction model. To this end, a prediction model for the nitrogen content of corn leaves based on the fusion of image and spectral features is proposed. In this study, corn leaves at the modulation stage were studied, samples with different nitrogen levels were numbered, and their hyperspectral data in the wavelength range of 400~1100 nm were collected. The average spectrum of the models was used as valid spectral information. First-order derivatives, standard normal variables transformation (SNV), Savitzky-Golay (S-G) smoothing, and normalization were selected to preprocess the spectral features. The CARS-SPA algorithm was used to screen sensitive spectral variables. The gray level co-currency matrix (GLCM) was chosen to extract the texture image features of the test samples. Corn leaf spectral and texture image features were fused and modeled as target features. Partial least squares regression (PLSR) and support vector machine regression (SVR) were used to predict corn leaves’ nitrogen content. The results showed that the image and spectral-based fusion models improved the prediction performance to some extent compared to the univariate models. The PLSR model based on feature fusion predicted the best results, in which the RP2 and RMSEP were 0.987 and 0.047. This method provides a reliable theoretical basis and technical support for developing nondestructive and accurate detection of nitrogen content in corn leaves.

Mapping and dynamic monitoring of military training-induced vegetation cover loss using Sentinel-2 images and method comparison.

Sustainable management of the US Army installations is critical for military training and readiness of forces. However, monitoring military training-induced vegetation cover disturbances using remote sensing data is challenging due to the lack of methodology for optimizing the selection of spectral variables or predictors and spatial modeling methods. This study aimed to propose and demonstrate a methodological solution for this purpose. The study was conducted in the Fort Riley installation in which three training areas were selected to map and monitor the training-induced vegetation cover loss. Sentinel-2 images and field observations of percentage vegetation cover (PVC) were combined at a spatial resolution of 10m by 10m to map PVC and its dynamics by comparison of two predictor selection methods and five spatial modeling algorithms based on a total of 304 spectral variables from the images before and after the training. Results showed that overall, the correlation-based predictor selection method reduced the relative root mean square error (RRMSE) of PVC predictions by 4.44% than the random forest (RF)-based predictor selection. Machine learning methods including RF, neural network, and support vector machine overall reduced the RRMSE of PVC predictions by 42.83% compared with multiple linear regression and k-nearest neighbors. Combining correlation-based predictor selection and RF modeling, coupled with leave one out cross validation, provided the greatest potential of increasing the accuracy of monitoring the vegetation cover loss. The findings provided useful implications to develop a near real-time system of monitoring military training-induced vegetation cover loss.

Digital Mapping of Root-Zone Soil Moisture Using UAV-Based Multispectral Data in a Kiwifruit Orchard of Northwest China

Accurate estimation of root-zone soil moisture (SM) is of great significance for accurate irrigation management. This study was purposed to identify planted-by-planted mapping of root-zone SM on three critical fruit growth periods based on UAV multispectral images using three machine learning (ML) algorithms in a kiwifruit orchard in Shaanxi, China. Several spectral variables were selected based on variable importance (VIP) rankings, including reflectance Ri at wavelengths 560, 668, 740, and 842 nm. Results indicated that the VIP method effectively reduced 42 vegetation indexes (VIs) to less than 7 with an evaluation accuracy of root-zone SM models. Compared with deep root-zone SM models (SM40 and SM60), shallow root-zone SM models (SM10, SM20, and SM30) have better performance (R2 from 0.65 to 0.82, RRMSE from 0.02 to 0.03, MAE from 0.20 to 0.54) in the three fruit growth stages. Among three ML algorithms, random forest models were recommended for simulating kiwi root-zone SM during the critical fruit growth period. Overall, the proposed planted-by-planted root-zone SM estimation approach can be considered a great tool to upgrade the toolbox of the growers in site-specific field management for the high spatiotemporal resolution of SM maps.

Soil salinity estimation: Effects of microwave dielectric spectroscopy and important frequencies

AbstractThe microwave dielectric constant is a key bridge in establishing the relationship between microwave remote sensing and soil salinity (electrical conductivity, EC). However, the response between microwave dielectric spectrum type, frequency, and soil salinity is still unclear. The purpose of this study is to reveal the dielectric spectrum and frequency range closely related to soil salinity. In this study, 129 surface soil samples were collected, and soil EC and microwave dielectric spectrum (frequency range of 0.3–20 GHz) were measured. Four variable selection methods were used to select important spectral variables, and the relationship between these spectral variables and soil salt was constructed using a random forest model. The results showed that the combination of (real part) and (imaginary part) revealed the most implicit information related to soil salinity. Considering the important spectral variables, the model accuracy was generally improved to the full‐dielectric spectrum model (the ratios of the performance to the interquartile distance (RPIQ) improved by 0.11–0.40); furthermore, the variable iterative space shrinkage (VISSA) method could most effectively select important spectral variables (RPIQ = 2.32). The selected spectral‐RF models could provide important frequencies for estimating soil salinity; moreover, the low‐frequency range (0.3–4 GHz) and new important frequency ranges (e.g., 7 and 15 GHz) able to respond to soil salinity were identified. This study provides a new strategy for monitoring the soil salinization risk using multiple frequencies microwave remote sensing techniques.

Hyperspectral Inversion Model of Relative Heavy Metal Content in Pennisetum sinese Roxb via EEMD-db3 Algorithm

Detection rapidity and model accuracy are the keys to hyperspectral nondestructive testing technology, especially for Pennisetum sinese Roxb (PsR) due to its extremely high adsorptive heavy metal content. The study of the resolution of PsR is conducive to the analysis of the accumulated heavy metal content in its different parts. In this paper, the contents of Cd, Cu and Zn accumulated in the old leaves, young leaves, upper stem, middle stem and lower stem, as well as the hyperspectral data of the corresponding parts, were measured simultaneously in both fresh and dry states. To begin, the spectral data of PsR were preprocessed by using Ensemble Empirical Mode Decomposition-Daubechies3 (EEMD-db3), Savitzky–Golay (SG), Symlet3 (sym3), Symlet5 (sym5), and multiplicative scatter correction (MSC). The 40 samples were divided into 32 training sets and 8 validation sets. The preprocessed spectral data were transformed by the first derivative (FD) and reciprocal logarithm (log(1/R)) to highlight the singularities using binary wavelet decomposition. After screening the significant bands from the correlation curve, the competitive adaptive reweighted sampling (CARS) and successive projection algorithm (SPA) were applied to extract the spectral characteristic variables, which were used to establish the partial least-squares (PLS) regression and multiple stepwise linear regression (MSLR) inversion models of Cd, Cu, and Zn contents. Based on EEMD-db3 pretreatment, the inversion model of Zn in the dry (fresh) state had R2 values of 0.884 (0.880), NRMSE values of 0.179 (0.253) and RPD values of 3.191 (3.221), indicating excellent stability and predictive performance. The findings of this study can not only aid in the rapid nondestructive detection of heavy metal adsorption in various parts of PsR, but can also be applied to guide the development and use of animal feed.

Correlation analysis of modern analytical data - a chemometric dissection of spectral and chromatographic variables.

The Standard Practices for Infrared Multivariate Quantitative Analysis (ASTM E1655) provide a guide for determining physicochemical properties of materials using multivariate calibration techniques applied to chemical sources that have high multicollinearity and correlated information. Partial least squares (PLS) is the most widely used multivariate regression method due to its excellent prediction capabilities and easy optimization. Initially applied to chromatographic data, PLS has also shown great results in near-infrared (NIR) and mid-infrared (MIR) spectroscopies. However, complex chemical matrices with low correlation may not be efficiently modeled using PLS or other multivariate analyses limited by grouping similar information (such as latent variables or principal components). Therefore, this study aims to evaluate the multicollinearity of different analytical techniques, such as high-temperature gas chromatography (HTGC), NIR, MIR, hydrogen nuclear magnetic resonance (1H NMR), carbon-13 nuclear magnetic resonance (13C NMR), and Fourier transform ion cyclotron resonance mass spectrometry coupled to the electrospray source in positive and negative ionization modes (ESI(±)FT-ICR). Descriptive statistics (coefficient of determination, R2) and principal component analysis (PCA) were used to identify the distribution of correlated information. Results showed that NIR and MIR spectroscopies exhibited a higher percentage of correlated variables, while 13C NMR and ESI(±)FT-ICR MS had more discrete profiles. Therefore, PLS development may be more effectively applied to NIR, MIR, and 1H NMR data, while 13C NMR and mass spectra may require other algorithms or variable selection methods in combination with PLS.