Search

China is the largest coal consumer in the world. The massive exploitation and utilization of coal resources have resulted in serious problems of heavy metal pollution and environmental contamination, such as soil degradation, water pollution, crop damage, and even threatening human lives. Therefore, monitoring soil heavy metal pollution quickly and in real time is an urgent task at present. This research not only formulated a new preprocessing method enlightened by few-shot learning for soil hyperspectral data but also combined it with other soil-related auxiliary information to extract effective information from the soil hyperspectrum, at the end of which different regression methods were adopted to predict soil heavy metal contamination. This test used 168 actual soil samples from the Eastern Junggar coalfield in Xinjiang for verification. Since copper in the soil is a trace element and the corresponding spectral characteristics are affected by other impurities, improper use of hyperspectral preprocessing methods may introduce interference information or may delete useful information, which makes the model effect unsatisfied. To effectively address the above-mentioned problems, the preprocessing method of this experiment combined the second-order differential derivation, and the data enhancement (DA) method together with the addition of auxiliary information to allow more effective features to be entered into the model. Next, the attentive interpretable tabular learning (TabNet) model was improved in three different ways using the original TabNet model and three improved TabNet models to create regression models. One of the improved TabNet models had the best effect, with a list of the top 30 features according to the degree of importance. Meanwhile, the regression prediction of Cu content using four different convolutional neural networks (CNNs) revealed that the model with the residual block was the strongest and slightly outperformed the improved TabNet model, but lacked interpretation of the input data. Besides, this experiment also employed different preprocessing methods for regression prediction on various models and found that the traditional preprocessing methods performed best in traditional regression models [e.g., partial least square regression (PLSR)] and underperformed in deep learning models. The selected optimal model was compared with PLSR and CNN models. The results indicated that both the improved TabNet model and the improved CNN model had better performance using the new preprocessing approach proposed in this article, with improved TabNet yielding a coefficient of determination ( $\text{R}^{2}$ ), root-mean-square error (RMSE), and the ratio of performance to interquartile range (RPIQ) of 0.94, 1.341, and 4.474, respectively. The improved CNN model had a coefficient of determination of 0.942, an RMSE of 1.324, and an interquartile range of 4.531 in the test dataset.

Drought is a climatic event that considerably impacts plant growth, reproduction and productivity. Toona sinensis is a tree species with high economic, edible and medicinal value, and has drought resistance. Thus, the objective of this study was to dynamically monitor the physiological indicators of T. sinensis in real time to ensure the selection of drought-resistant varieties of T. sinensis. In this study, we used near-infrared spectroscopy as a high-throughput method along with five preprocessing methods combined with four variable selection approaches to establish a cross-validated partial least squares regression model to establish the relationship between the near infrared reflectance spectroscopy (NIRS) spectrum and physiological characteristics (i.e., chlorophyll content and nitrogen content) of T. sinensis leaves. We also tested optimal model prediction for the dynamic changes in T. sinensis chlorophyll and nitrogen content under five separate watering regimes to mimic non-destructive and dynamic detection of plant leaf physiological changes. Among them, the accuracy of the chlorophyll content prediction model was as high as 72%, with root mean square error (RMSE) of 0.25, and the RPD index above 2.26. Ideal nitrogen content prediction model should have R2 of 0.63, with RMSE of 0.87, and the RPD index of 1.12. The results showed that the PLSR model has a good prediction effect. Overall, under diverse drought stress treatments, the chlorophyll content of T. sinensis leaves showed a decreasing trend over time. Furthermore, the chlorophyll content was the most stable under the 75% field capacity treatment. However, the nitrogen content of the plant leaves was found to have a different and variable trend, with the greatest drop in content under the 10% field capacity treatment. This study showed that NIRS has great potential for analyzing chlorophyll nitrogen and other elements in plant leaf tissues in non-destructive dynamic monitoring.

Drought is a climatic event that considerably impacts plant growth, reproduction and productivity. Toona sinensis is a tree species with high economic, edible and medicinal value, and has drought resistance. Thus, the objective of this study was to dynamically monitor the physiological indicators of T. sinensis in real time to ensure the selection of drought-resistant varieties of T. sinensis. In this study, we used near-infrared spectroscopy as a high-throughput method along with five preprocessing methods combined with four variable selection approaches to establish a cross-validated partial least squares regression model to establish the relationship between the near infrared reflectance spectroscopy (NIRS) spectrum and physiological characteristics (i.e., chlorophyll content and nitrogen content) of T. sinensis leaves. We also tested optimal model prediction for the dynamic changes in T. sinensis chlorophyll and nitrogen content under five separate watering regimes to mimic non-destructive and dynamic detection of plant leaf physiological changes. Among them, the accuracy of the chlorophyll content prediction model was as high as 72%, with root mean square error (RMSE) of 0.25, and the RPD index above 2.26. Ideal nitrogen content prediction model should have R2 of 0.63, with RMSE of 0.87, and the RPD index of 1.12. The results showed that the PLSR model has a good prediction effect. Overall, under diverse drought stress treatments, the chlorophyll content of T. sinensis leaves showed a decreasing trend over time. Furthermore, the chlorophyll content was the most stable under the 75% field capacity treatment. However, the nitrogen content of the plant leaves was found to have a different and variable trend, with the greatest drop in content under the 10% field capacity treatment. This study showed that NIRS has great potential for analyzing chlorophyll nitrogen and other elements in plant leaf tissues in non-destructive dynamic monitoring.

Xylella fastidiosa is a bacterial pathogen affecting many plant species worldwide. Recently, the subspecies pauca (Xfp) has been reported as the causal agent of a devastating disease on olive trees in the Salento area (Apulia region, southeastern Italy), where centenarian and millenarian plants constitute a great agronomic, economic, and landscape trait, as well as an important cultural heritage. It is, therefore, important to develop diagnostic tools able to detect the disease early, even when infected plants are still asymptomatic, to reduce the infection risk for the surrounding plants. The reference analysis is the quantitative real time-Polymerase-Chain-Reaction (qPCR) of the bacterial DNA. The aim of this work was to assess whether the analysis of hyperspectral data, using different statistical methods, was able to select with sufficient accuracy, which plants to analyze with PCR, to save time and economic resources. The study area was selected in the Municipality of Oria (Brindisi). Partial Least Square Regression (PLSR) and Canonical Discriminant Analysis (CDA) indicated that the most important bands were those related to the chlorophyll function, water, lignin content, as can also be seen from the wilting symptoms in Xfp-infected plants. The confusion matrix of CDA showed an overall accuracy of 0.67, but with a better capability to discriminate the infected plants. Finally, an unsupervised classification, using only spectral data, was able to discriminate the infected plants at a very early stage of infection. Then, in phase of testing qPCR should be performed only on the plants predicted as infected from hyperspectral data, thus, saving time and financial resources.

For the first time, a prototype HS-GC-MS-IMS dual-detection system is presented for the analysis of volatile organic compounds (VOCs) in fields of quality control of brewing hop. With a soft ionization and drift time-based ion separation in IMS and a hard ionization and m/z-based separation in MS, substance identification in the case of co-elution was improved, substantially. Machine learning tools were used for a non-targeted screening of the complex VOC profiles of 65 different hop samples for similarity search by principal component analysis (PCA) followed by hierarchical cluster analysis (HCA). Partial least square regression (PLSR) was applied to investigate the observed correlation between the volatile profile and the α-acid content of hops and resulted in a standard error of prediction of only 1.04% α-acid. This promising volatilomic approach shows clearly the potential of HS-GC-MS-IMS in combination with machine learning for the enhancement of future quality assurance of hops. Graphical abstract Electronic supplementary material The online version of this article (10.1007/s00216-020-02842-y) contains supplementary material, which is available to authorized users.

Hyperspectral remote sensing can be a powerful tool for detecting invasive species and their impact across large spatial scales. However, remote sensing studies of invasives rarely occur across multiple seasons, although the properties of invasives often change seasonally. This may limit the detection of invasives using remote sensing through time. We evaluated the ability of hyperspectral measurements to quantify the coverage of a plant invader and its impact on senesced plant coverage and canopy equivalent water thickness (EWT) across seasons. A portable spectroradiometer was used to collect data in a field experiment where uninvaded plant communities were experimentally invaded by cogongrass, a non-native perennial grass, or maintained as an uninvaded reference. Vegetation canopy characteristics, including senesced plant material, the ratio of live to senesced plants, and canopy EWT varied across the seasons and showed different temporal patterns between the invaded and reference plots. Partial least square regression (PLSR) models based on a single season had a limited predictive ability for data from a different season. Models trained with data from multiple seasons successfully predicted invasive plant coverage and vegetation characteristics across multiple seasons and years. Our results suggest that if seasonal variation is accounted for, the hyperspectral measurement of invaders and their effects on uninvaded vegetation may be scaled up to quantify effects at landscape scales using airborne imaging spectrometers.

The effects of increasing ambient ozone (O3) concentrations on food security has become a major concern as the demand for agricultural productivity is projected to increase significantly over the next several decades. In this contribution, the responses of common soybean genotypes (AK-HARROW, PI88788, DWIGHT, PANA, and WILLIAMS82) to ambient O3 are characterized using hyperspectral data and foliar biophysical, mineral nutrient concentrations and soybean yield. Specifically, leaf reflectance spectra measured at different growth stages and canopy layers were used to examine the spectral indices that were most strongly correlated with leaf physiological status. The effects of elevated O3 on six important nutrients (K, Ca, Mg, Fe, Mn and Cu) were evaluated by analyzing the variations in nutrient concentrations at two critical growth stages with increasing ambient O3 concentration using Partial Least Square Regression (PLSR). Lastly, the identified best spectral indices and the robust nutrient prediction models were extrapolated to the entire growth period to explore their ability to track the effects of ambient O3 concentrations on soybean physiology and nutrient uptake. The results showed that fluorescence yield (ΔF/Fm’) and photochemical quenching (qP) appear to be good indicators of soybean physiological responses to O3 stress that are echoed by the harvest index (HI). Newly identified normalized difference spectral index (NDSI) [R416, R2371] always had the highest correlation (R2 > 0.6) with ΔF/Fm’, qP and electron transport rate (ETR, μmol m−2 s−1) compared to the published indices. Additionally, there were significant and broad spectral regions in visible and near infrared region that were well-correlated with ΔF/Fm’ and selected NDSIs that were applicable to satellite observations. The results of nutrient modeling using PLSR explained 54–87% of the variance in nutrient concentrations, and the predicted mineral nutrient accumulation throughout the growing season reflected the responses of ozone tolerant and sensitive genotypes well. NDSI [R416, R2371] demonstrated great potential in regard to its sensitivity in tracking plant physiological responses to changing ambient O3 concentrations. The outcome of this research has potential implications for development of space-based observation of large-scale crop responses to O3 damage, as well as for biotechnological breeding efforts to improve ozone tolerance under future climate scenarios.

Sewage sludge application to agricultural soils is a common practice in several countries in the European Union. Nevertheless, the application dose constitutes an essential aspect that must be taken into account in order to minimize environmental impacts. In this study, near infrared reflectance spectroscopy (NIRS) was used to estimate in sewage sludge samples several parameters related to agronomic and environmental issues, such as the contents in organic matter, nitrogen and other nutrients, metals and carbon fractions, among others. In our study (using 380 biosolid samples), two regression models were fitted: the common partial least square regression (PLSR) and the penalized signal regression (PSR). Using PLSR, NIRS became a feasible tool to estimate several parameters with good goodness of fit, such as total organic matter, total organic carbon, total nitrogen, water-soluble carbon, extractable organic carbon, fulvic acid-like carbon, electrical conductivity, Mg, Fe and Cr, among other parameters, in sewage sludge samples. For parameters such as C/N ratio, humic acid-like carbon, humification index, the percentage of humic acid-like carbon, the polymerization ratio, P, K, Cu, Pb, Zn, Ni and Hg, the performance of NIRS calibrations developed with PLSR was not sufficiently good. Nevertheless, the use of PSR provided successful calibrations for all parameters. This work has been financed by the Ministerio de Ciencia e Innovación (currently Economía y Competitividad) of Spain and European Regional Development Funds (FEDER, “Una manera de hacer Europa”) (Project AGL2009-12371-C02-01) and also by the Generalitat Valenciana (ACOMP/2010/177).

Olive fruits of three different cultivars (Moraiolo, Dolce di Andria, and Nocellara Etnea) were monitored during ripening up to harvest, and specific and total phenols were measured by HPLC (High Pressure Liquid Chromatography). On the same olive samples (n = 450), spectral detections were performed using a portable NIR (Near Infrared)-AOTF (Acousto Optically Tunable Filter) device in diffuse reflectance mode (1100-2300 nm). Prediction models were developed for the main phenolic compounds (e.g., oleuropein, verbascoside, and 3,4-DHPEA-EDA) and total phenols using Partial Least Squares (PLS). Internal cross-validation (leave-one-out method) was applied for calibration and prediction models developed on the data sets relative to each single cultivar. Validation of the models obtained as the sum of the three sample sets (total phenols, n = 162; verbascoside, n = 162; oleuropein, n = 148; 3,4-DHPEA-EDA, n = 162) were performed by external sets of data. Obtained results in term of R(2) (in calibration, prediction and cross-validation) ranged between 0.930 and 0.998, 0.874-0.942, and 0.837-0.992, respectively. Standard errors in calibration (RMSEC), cross-validation (RMSECV), and prediction (RMSEP) were calculated obtaining minimum error in prediction of 0.68 and maximum of 6.33 mg/g. RPD ratios (SD/SECV) were also calculated as references of the model effectiveness. This work shows how NIR-AOTF can be considered a feasible tool for the on-field and nondestructive measurement of specific and total phenols in olives for oil production.

The sensory properties of fat in milk were examined by sensory descriptive analysis. To date, no single food additive has been completely successful in mimicking the sensory properties of fat in milk. This experiment investigated the effects of various factors and combinations thereof on sensory properties and perceived fattiness of milk, and compared them to the actual fat content (0.1; 1.3 and 3.5% fat milk was used). The other factors studied were the addition of thickener, whitener, cream aroma and homogenisation. Multivariate data analytical methods (Partial Least Squares Regression) were applied for analysis of the data. The three former additional factors contributed significantly to perceived fattiness of the milk, and homogenisation had a small but not significant effect. It was shown that a combination of thickener, whitener and cream aroma in 0.1% fat milk was approximately successful in mimicking sensory properties of 1.3% fat milk.