Early Detection of Wireworm (Coleoptera: Elateridae) Infestation and Drought Stress in Maize Using Hyperspectral Imaging

Abstract

The objective of our research was to test hyperspectral imaging as a method for early detection and discrimination of biotic and abiotic stress in maize. We investigated the individual and combined effects of wireworm feeding and drought stress on leaf spectral responses and on various morphological and physiological traits of maize plants, selecting two hybrids with different tolerance to drought. Physiological parameters were determined at three time points (14, 21 and 28 days after adding wireworms and changing watering regime), along with hyperspectral imaging. Most of the differences in physiological characteristics between treatments were detected on day 21, when drought was the main cause of the negative physiological outcome, while the presence of wireworms only caused lower relative chlorophyll content, resulting in lower combined stress damage in some treatments. The morphological data showed greater wireworm damage to hybrid ZP341 and a greater negative effect of combined stress on hybrid FuturiXX. Hyperspectral imaging detected pest infestation and drought stress before they were detected by classical methods, with the highest overall accuracy on day 14 (84.7%) and the lowest on day 28 (67%). It can therefore be used as a method for early detection of wireworm infestation and/or drought in maize.

1. Introduction

The impact of climate change on European crop production has been described in numerous studies [1,2,3]. In the short term, climate change is expected to be felt in the form of extreme weather events, while in the long term, a gradual increase in average temperatures and changes in precipitation patterns are expected [4]. Maize growers will face short-term summer droughts, especially at the time of maize flowering, when water stress significantly affects plant and silk growth, kernel number and kernel weight, which can lead to considerable yield losses [5,6]. Another important aspect of the increase in average temperature is the change in the geographic distribution of pests, as well as changes in insect life history traits, i.e., warmer temperatures can result in multiple generations per season, longer activity, faster metabolism and higher insect survival rates due to milder winters [7,8].

Wireworms (Coleoptera: Elateridae) are among the most important pests of maize [8]. They are polyphagous subterranean larvae feeding mostly on plant roots, seedlings and stems [9,10,11,12]. Damage by wireworms occurs in maize mainly during germination, as they burrow into germinated seeds or the lower part of the young stem [13]. A changing climate could potentially intensify wireworm pressure, as it could affect the emergence time of overwintering adults, the moulting frequency of larvae, species composition in a given area and their feeding behaviour [14,15,16]. Larvae of Agriotes obscurus (Linnaeus, 1767) and Agriotes lineatus (Linnaeus, 1767) are reported to feed more intensively on roots in dry years [17,18], although these reports come from books, not peer-reviewed articles. Wireworms lose water rapidly in a dry environment due to the high permeability of their cuticle [19]. Therefore, to compensate for moisture loss, wireworms must either migrate to greater depths or increase feeding pressure to withstand desiccation.

Pests and pathogens trigger various defence mechanisms in plants, such as the production of specific secondary metabolites and proteins, changes in plant tissue structure and hypersensitivity reactions [20]. These changes, whether biochemical or structural, can be specific to certain biotic and abiotic stressors [21], which can be utilized in non-destructive phenotyping using optical sensors [22]. When electromagnetic radiation reach plants, the energy is absorbed, reflected or transmitted. Remote sensing can be used to detect changes in these effects, which allows us to determine biophysical characteristics of plants, such as colour, shape, texture and chemical composition [23,24]. In general, the spectral reflectance of healthy plants is low under visible light (VIS) due to light absorption by leaf pigments and high under near-infrared (NIR) light due to internal reflection by air pockets present in healthy plants. The opposite applies to plants under stress, as stressed plants tend to have lower chlorophyll content, resulting in lower VIS absorption and consequently higher VIS reflectance [25].

In agronomic studies, hyperspectral imaging can be used to determine plant phenology, plant nutrient requirements, water requirements, yield prediction, weed control and early detection of pests and diseases [24]. For the latter, research has focused primarily on early detection of fungi, bacteria and viruses (reviewed in [26]) and less on nematodes [27,28] and insects [29,30,31]. Hyperspectral imaging has been used in maize for a number of purposes, such as to discriminate maize varieties [32], classify kernel hardness [33], predict yield [34], determine leaf nitrogen concentration and chlorophyll content [35], discriminate genetically modified kernels [36] and for early detection of infections, such as Phaeosphaeria leaf spot [37], Aspergillus spp. [38,39], Fusarium verticillioides [40], maize dwarf mosaic virus, Helminthosporium maydis [41] and maize streak geminivirus [37], among others. However, studies on remote sensing of maize pest infestation are scarce [42]. To the best of our knowledge, hyperspectral imaging has never been used to detect early wireworm damage in maize.

Few studies have investigated the combined effect of multiple stressors on maize (see [43]). In this study, a series of abiotic (drought) and biotic (pest) treatments were applied to two maize hybrids in a greenhouse and periodically monitored for morphological, physiological or leaf spectral changes using various methods, such as photosynthetic gas exchange, PSII photochemistry and hyperspectral imaging. The objective of this research was to investigate the responses of maize to individual and combined stresses and whether hyperspectral imaging could be an effective method for early detection of wireworm infestations in maize in order to use precision agriculture techniques for wireworm control in the face of future climate change.

2. Materials and Methods

2.1. Study Species

Maize hybrids were selected based on a set of functional DNA markers associated with drought (in)tolerance identified by Dr. Barbara Pipan (personal communication); therefore, ZP341 (Institut za kukuruz “Zemun polje”, Belgrade, Serbia) was selected as a drought-intolerant variety, and FuturiXX (RAGT Semences SAS, Rodez, France) was selected as a drought-tolerant variety. Because the seeds were obtained from commercial seed suppliers, we had to remove the pesticide coating, which could affect the herbivory results. For this purpose, the kernels were vortexed in 1% sodium hypochlorite for 1 min, rinsed with water, vortexed in 0.1% Tween 80 for 1 min, rinsed with water, vortexed three times in sterile dH₂O and dried. Viability and germination were not affected by this procedure.

Larvae of Agriotes lineatus were obtained from a laboratory colony established from adult elaterid beetles collected in the early summer of 2019 and reared at the Agricultural Institute of Slovenia according to [44]. In February 2020, the size of the larvae was sufficient for experimental use (1.6 ± 0.23 cm) [9] therefore they were moved to a smaller pot filled with soil and fed with carrot and placed in the refrigerator (5 °C) to slow down development until the start of the experiment. To ensure that only healthy and active larvae were included in the experiment, each larva was placed in the centre of a moist filter paper 10 cm in diameter before the experiment. If the larva reached the edge of the filter paper within 30 s, it was considered active and suitable for further experimental use [45].

2.2. Experimental Design

A factorial pot experiment was conducted in the greenhouse of the Agricultural Institute of Slovenia from April to June 2020 under natural and artificial illumination and a 12 h photoperiod at a 20 ± 3.9 °C/17 ± 2.2 °C day/night temperature and 60 ± 17.2/73 ± 7.2% day/night relative humidity. We tested the physiological, morphological and leaf spectral responses of two maize hybrids (FuturiXX vs. ZP341) to two watering regimes (water-deficient vs. well-watered) and to belowground herbivory (no wireworms vs. wireworm-infested). Three maize kernels of the same hybrid were planted in 3 L plastic pots with a diameter of 20 cm, each filled with 3.5 kg of autoclaved substrate mixture of 1.5 parts fine-grain (MP1/G), 1.5 parts coarse-grain (MP4) quartz sand (Termit, Moravče, Slovenia) and 1 part Potgrond P fine peat substrate (KlasmannDeilmann, Geeste, Germany). After 8 days, the tallest plant was left in the pot for further experimental use, and the other two were discarded. The 56 maize plants were evenly divided among eight treatment combinations (n = 7 per treatment; Table 1). Each pot received 300 mL of the Valentin NPK 7-3-6 fertiliser (Semenarna Ljubljana, Ljubljana, Slovenia) in a concentration of 13.5 mL/l on a weekly basis for the next four weeks. In the fourth week of maize growth, four larvae of A. lineatus were added to half of the pots near the base of the stems. At the same time, the watering regime was changed to induce drought stress (Table 2). Under the well-watered regime, we watered the plants to the point of soil saturation, discarded excess water and weighed each pot. In the water-deficient treatment, we stopped watering all plants until they reached 88–91% of the weight of the saturated pot (after about one week). The 88–91% threshold came from a previously conducted pilot experiment in which we monitored leaf water potential in maize plants on a daily basis. We started with plants watered at field capacity and terminated the experiment when the rate of water loss in the pot began to level off, water potential fell below a set threshold and plants showed signs of drought stress (e.g., wilting or loss of turgor). We adjusted the watering regime by weighing the pots before watering so that the plants were maintained at 88–91% of the weight of the saturated pot under water-deficient treatment and at >95% of the weight of the saturated pot under well-watered treatment. Pots were also randomised and rearranged weekly in order to average out the positional effect on plants [46].

2.3. Drought Stress and Wireworm Herbivory Damage Evaluation

Several methods were used to evaluate drought stress and herbivory damage by wireworms on maize plants.

2.3.1. Physiological Parameters

Data were acquired on days 14, 21 and 28 after adding A. lineatus larvae and changing the watering regime. Chlorophyll a fluorescence parameters related to PSII photochemistry were measured with a MiniPAM pulse-amplitude-modulated fluorometer (Heinz Walz GmbH, Pfullingen, Germany) on all plants on the same leaves as the relative chlorophyll content was measured (see below). The maximum quantum efficiency of PSII photochemistry (F_v/F_m) was measured on 10 min dark-adapted leaves when all PSII reaction centres were open (State 1). It was calculated using Equation (1):

(1)${\mathrm{F}}_{\mathrm{v}}/{\mathrm{F}}_{\mathrm{m}}=\frac{{\mathrm{F}}_{\mathrm{m}}–{\mathrm{F}}_0}{{\mathrm{F}}_{\mathrm{m}}},$

by first measuring the minimal fluorescence (F₀) using a low-intensity measuring light (0.15 µmol m⁻²s⁻¹ PAR), after which a saturating pulse (7000 µmol m⁻² s⁻¹ PAR for 0.8 s) was applied close to all PS2 reaction centres (State 2) to measure maximal fluorescence (F_m) [47].

Four plants were randomly selected from each treatment for photosynthetic gas exchange and relative chlorophyll content measurements. One leaf per plant was analysed using an LI-6400 XT portable photosynthesis system (LI-COR Biosciences, Lincoln, NE, USA) at ambient air temperature, air humidity and reference CO₂ concentration (400 μmol mol⁻¹), with a stable light intensity of 1000 μmol photons m⁻² s⁻¹ from an internal LED light source. The following parameters were obtained: net photosynthesis (µmol/m²/s), leaf stomatal conductivity (mmol m⁻² s⁻¹), effective quantum yield of PSII (Fv’/Fm’) and transpiration rate (mmol H₂O m⁻² s⁻¹).

Relative chlorophyll content measurements were performed using a SPAD-502 Plus portable chlorophyll (Chl) meter (Konica Minolta Optics, Tokyo, Japan). Chlorophyll content was calculated as the average value from five measurements taken randomly along the entire length of the 5th leaf (counted from the bottom–up).

Leaf samples were taken immediately after measurements to determine relative water content (Rwc). Leaf relative water content was measured using the standard technique [48] based on Equation (2):

(2)$\mathrm{Rwc}=\frac{\mathrm{FW}-\mathrm{DW}}{\mathrm{TW}-\mathrm{DW}},$

where FW, DW and TW are leaf fresh weight, dry weight and turgid weight, respectively.

2.3.2. Hyperspectral Imaging

Data were acquired on days 14, 21 and 28 after the addition of A. lineatus larvae and changing the watering regime. Plants were imaged from above in a darkened room on a black background using two pushbroom hyperspectral sensors, a Hyspex VNIR-1600 and a SWIR-384 (Norsk Elektro Optikk AS, Oslo, Norway), and illuminated with two calibrated halogen lamps with homogenous light intensity in the range of 400–2500 nm. The lamps were turned on 15 min before imaging, to stabilize temperature drift and ensure spatial uniformity of illumination [49]. Data were recorded in a combined range of 400–2500 nm and in a total of 448 spectral bands (bandwidth/spatial resolution: 3.6 nm/1 mm and 5.4 nm/2.5 mm, respectively). A calibrated diffuse 20% grey reference plate (SphereOptics, Herrsching am Ammersee, Germany) was included in each image and used to calculate reflectance. Each line was acquired twice to increase the signal-to-nose ratio. The acquired images were radiometrically calibrated to radiance units (W sr⁻¹ m⁻²).

2.3.3. Plant Morphology and Herbivory Damage

Plant morphology data (shoot length, shoot dry weight, root dry weight and stem diameter) and herbivory damage (number of tunnels in roots and stem due to wireworm feeding) were acquired at the end of the experiment, that is, on day 29 after adding A. lineatus larvae and changing the watering regime. The soil from an individual pot was transferred to a plastic tub and inspected for remaining A. lineatus larvae. Maize stem diameter was calculated as the average of two perpendicular measurements taken from the stem base. Shoot length was measured from the stem base to the tip of the apical shoot. Plant material was then cut into smaller pieces and placed in a paper bag; the shoot and root of each plant were separated, dried at 60 °C for 72 h and weighed.

2.4. Statistical Analysis

Because data on maximum quantum efficiency of PSII (F_v/F_m) were collected for all plants at all three time points, we were able to perform a two-way repeated measures ANOVA to evaluate the effect of different treatments over time on the maximum quantum efficiency of PSII, where time was defined as a within-subject factor. In the case of significant two-way interaction, analysis of simple main effects of treatment at each time point and simple main effects of the time variable were investigated, followed by pairwise comparison with Bonferroni’s adjusted p-value [50].

The effect of pests, watering regime and maize hybrid (fixed effects) on plant physiological and morphological parameters and herbivory damage (number of tunnels in plants) was analysed using a one-way ANOVA, followed by a post hoc Tukey HSD test. For data with non-normal distribution, a non-parametric Kruskal–Wallis test was used, followed by Dunn’s post hoc test, where the p-value was adjusted using Benjamini–Hochberg (BH) correction.

A three-way ANOVA was used to simultaneously evaluate the effect of pests, watering regime and maize hybrid (fixed effects) on plant physiological and morphological parameters at individual time points. In the case of a significant two-way interaction, simple main effects of the first variable at each level of the second variable were investigated, followed by pairwise comparison with Bonferroni’s adjusted p-value. In the case of a significant three-way interaction, simple two-way interactions at each level of the third variable were investigated, followed by an analysis of simple main effects at each level of a second variable. Multiple pairwise comparisons were used to determine which group means were significantly different. Statistical significance of a simple two-way interaction and simple main effect were accepted at a Bonferroni-adjusted alpha level. Analyses were performed with “rstatix” [51], “tidyverse” [52] and “emmeans” [53] packages in R version 3.6.1 [54]. We used principal component analysis (PCA) to visualise multiple correlations among the different physiological and morphological maize parameters depending on the treatment used. “FactoMineR” [55] and “factoextra” [56] packages were used for PCA.

Due to the small sample size, a non-parametric permutational multivariate analysis of variance (PERMANOVA) with 999 permutations [57] was performed to determine the effect of treatments (fixed effect) on all physiological and morphological maize parameters at a single time point. Euclidean distance was used to construct a similarity matrix. To this end, the “EcolUtils” [58] and “vegan” [59] packages were used.

2.5. Hyperspectral Data Preprocessing and Analysis

The radiometrically corrected images were segmented into five classes (background, soil, pots, reference panel and plants) using spectral information divergence (SID), which uses the Kullback–Leibler information measure to match image pixels to references [60]. Regions of interest (ROI) for segmentation were selected by hand in the first image in Envi 5.2 (Harris Geospatial, Broomfield, CO, USA); the same ROIs were used for segmentation for all images. SID classifications were used to create segmentation masks, and only leaf-area and reference-plate pixels were extracted as separate classes for further analysis. Hyperspectral images were segmented using SiaPy (Spectral imaging analysis for Python) software [61] (Figure 1).

Reference panel pixels were filtered using the median absolute deviation (MAD) method to remove outliers and increase the accuracy of reflectance calculations. Values outside ± 2 MAD were removed from further calculations, and the remaining pixels were used to calculate the mean spectra of the reference plate.

The reflectance was calculated for each leaf-area pixel according to Equation (3):

(3)${\mathrm{R}}_{\mathrm{i}}=\frac{{\mathrm{I}}_{\mathrm{i}}-{\mathrm{D}}_{\mathrm{i}}}{{\mathrm{W}}_{\mathrm{i}}-{\mathrm{D}}_{\mathrm{i}}}\ast 0.2,$

where I_i represents the radiance of the i-th band of leaf-area pixels, W_i is the radiance of the i-th band of the reference panel and D_i is the dark current of the sensors of the i-th band. The leaf-area pixels of each plant were then filtered using MAD, and values outside ± MAD were removed. The mean spectra of each plant were then calculated.

The reflectance spectra were smoothed and transformed using second-order Savitzky–Golay derivatives with second-order polynomials and a smoothing window size of 11 points. This procedure reduces noise while preserving shape and important information and highlights small spectral variations in the data. The dimensionality of the data was reduced using sparse partial least squares discriminant analysis (sPLS-DA; [62]). PLS-DA uses annotated data labels to maximize the variance between classes and has been used extensively as a method for dimensionality reduction. The sparse version of PLS-DA performs both variable selection and classification in a single step [62]. sPLS-DA was validated with 10-fold repeated 5-fold cross validation, and Mahalanobis distances were used for class prediction and latent variable selection (LV). The relevance of spectral bands was assessed using variable in projection (VIP) analysis. Dimensionality reduction was repeated three times for each set of analyses: (1) differentiation of all classes (2) drought stress detection and (3) pest detection. Spectral data were visualised using principal component analysis (PCA). Both PCA and PLS are linear methods. Unlike partial least squares methods, PCA does not consider data labels (i.e., a response variable); therefore, it is an unconstrained ordination method. PCA explains the variance–covariance structure of a dataset and aims to increase the variance of the features themselves. Both PLS and PCA can be used as dimensionality reduction methods, with PLS generally performing better in this regard since it is a supervised method (i.e., a response variable is present). If one increases the number of features without increasing the number of training samples, the dimensionality of the feature space becomes sparser, i.e., the sample density decreases exponentially. This can lead to overfitting (the generalizability of the model is low), as it becomes easier to find a good, albeit random, solution to a classification problem. By using dimensionality reduction methods, we reduce the feature space while minimizing information loss and improving relevant features.

The features extracted from dimensionality reduction were then used in a support vector machine (SVM) classification with a radial basis function kernel. This kernel allows for the computation of nonlinear boundaries between groups and thus captures more complex relationships between data points. This approach allows for more accurate classifications with less overfit compared to simpler methods such as random forests or gradient boosting. The data were split into training and test datasets in a 70/30 ratio. Hyperparameters (cost function (c) and sigma) were tuned using a grid search. Models were trained with a radial basis function kernel and validated with 10-fold repeated 5-fold cross validation. Final validations were performed using the test set. All hyperspectral analyses and data visualisations were performed in R [49] using the “mixOmics” [63], “kernlab” [64], “caret” [65], “prospectr” [66], “ggplot2” [67] and “doParallel” [68] packages.

3. Results

3.1. Physiological Parameters

There was a statistically significant interaction between treatment and time on the maximum quantum efficiency of PSII (F_{(11.56, 79.27)} = 3.071; p = 0.002); therefore, the simple main effect of treatment was analysed at each time point and was found to be significant only on day 14. Considering the Bonferroni-adjusted p-value, wireworm-infested and well-watered FuturiXX plants (treatment Y/W) had significantly lower maximum quantum efficiency of PSII than ZP341 plants (treatments N/D, N/W and Y/W) and lower maximum quantum efficiency of PSII than water-deficient FuturiXX plants without wireworms (treatment N/D). In addition, FuturiXX plants with wireworm infestation and water deficiency (treatment Y/D) had significantly lower maximum quantum efficiency of PSII than water-deficient ZP341 plants without wireworms (treatment N/D) on day 14. There was a significant effect of time on wireworm-infested and well-watered FuturiXX plants (treatment Y/W) (F_{2, 12} = 11.8; p = 0.008); the maximum quantum efficiency of PSII increased with time (Table 3).

On day 14, hybrid ZP341 had significantly higher net photosynthesis (F_{1, 30} = 4.71; p = 0.038) and relative chlorophyll content (F_{1, 30} = 7.03; p = 0.0127) compared to FuturiXX, and water-deficient plants had a borderline significantly lower transpiration rate compared to well-watered plants (F_{1, 30} = 4.39; p = 0.0447). Most differences between treatments were found on day 21 (Figure 2), when wireworm-infested plants had significantly higher relative water content (χ2(1) = 4.30, p = 0.0382) and higher effective quantum yield of PSII (χ2(1) = 5.82, p = 0.0159), but lower relative chlorophyll content (F_{1, 29} = 10.62; p = 0.0029). Water-deficient plants had significantly lower relative water content (χ2(1) = 4.62, p = 0.0317), net photosynthesis (F_{1, 30} = 6.91; p = 0.0134), leaf stomatal conductivity (F_{1, 30} = 12.01; p = 0.0016), effective quantum yield of PSII (χ2(1) = 6.19, p = 0.0129) and transpiration rate (χ2(1) = 7.99, p = 0.0047). Regarding hybrids, FuturiXX had significantly higher relative water content (χ2(1) = 6.38, p = 0.0116) and borderline significantly higher effective quantum yield of PSII (χ2(1) = 3.99, p = 0.0458) compared to ZP341. There was also a statistically significant three-way interaction between pests, watering regime and hybrid for the effective quantum yield of PSII. Consequently, a simple main effects analysis was conducted in which well-watered FuturiXX plants without wireworms (treatment N/W) showed higher values of effective quantum yield of PSII (F_{1, 24} = 7.72; p = 0.01) compared to well-watered ZP341 plants without wireworms (treatment N/W).

On day 28, results were similar to day 14, i.e., hybrid ZP341 had statistically higher relative chlorophyll content than FuturiXX (F_{1, 30} = 7.91; p = 0.0086) and well-watered plants had higher transpiration rates than water-deficient plants (F_{1, 30} = 5.09; p = 0.0314) (not shown).

3.2. Hyperspectral Imaging

Comparatively accurate identification of infested plants was achieved by combining sparse partial least squares discriminant analysis and support vector machine classification of hyperspectral data. With 19 latent variables explaining 95.03% of the variance, a 100% success rate was achieved in identifying the imaging session (i.e., day 14, 21 or 28 after addition of wireworms and changing the watering regime), although principal component analysis (PCA) showed no groups or patterns in the data (Figure 3). When all three imaging sessions were pooled into one dataset, 29 latent variables (LV) (explaining 97.9% of the variance) were selected for SVM classification of eight classes. The overall accuracy was 0.63; the accuracy per class ranged from 0.58 to 0.975. Identifying infested plants achieved a 0.98 success (19 LVs, 95.3%), and water stress was identified with 0.959 success (13 LVs, 90.5%) (Table 4). Classification accuracy increased when data were analysed separately according to imaging session. The overall accuracy of individual treatments increased to a minimum of 0.67 and a maximum of 0.847, with the lowest accuracy achieved on day 28 (0.67, 15 LVs, 91.4%). Detection of infestation and water stress achieved a 100% success rate in all three imaging sessions (Table 5).

For each classification, the relevant wavelengths were also determined using Variable importance in projection (VIP) analysis (Figure 4). Classification using pooled samples identified multiple wavelengths in the visible part of the spectrum, although the mean spectra had similar reflectance patterns (Figure 5). Wavelengths between 750 and 890 nm were identified for treatment and pest detection but not for drought stress. On the other hand, drought stress detection exhibited relevant wavelengths around 2200–2300 nm, whereas pest infestation and treatment detection did not. There were only a handful of identified wavelengths in the near-infrared (NIR) and shortwave infrared (SWIR) parts of the spectrum (800 to 2500 nm) that had similar patterns in all three cases, for example, between 1600 and 1700 nm. Nevertheless, the mean spectra of the infested plants had lower reflectance than those of the healthy plants. In contrast, water-deficient plants exhibited higher mean reflectance in the same spectral range than well-watered plants (Figure 5). Several additional relevant wavelengths were observed when the data were split between imaging sessions (IS). In all cases, there were fewer relevant wavelengths in the visible part of the spectrum in the imaging session on day 14 than on days 21 and 28, and there were several more in the NIR and SWIR regions. For pest and drought stress detection, the imaging sessions on day 14 and 28 showed more overlap than with the imaging session on day 21.

3.3. Plant Morphology

Regarding morphological parameters (Figure 6), wireworm-infested plants had a significantly shorter shoot length (χ2(1) = 7.81, p = 0.0052) and lower shoot dry mass (F_{1, 54} = 19.03; p = 0.0001); FuturiXX plants had a significantly lower mean stem diameter (χ2(1) = 9.02, p = 0.0027), shoot dry mass (χ2(1) = 4.72, p = 0.0298) and root dry mass (χ2(1) = 10.89, p = 0.001) compared to ZP341; and water-deficient plants had a significantly shorter shoot length (F_{1, 54} = 64.48; p < 0.0001) and lower average stem diameter (F_{1, 54} = 11.82; p = 0.0011) and shoot dry mass (F_{1, 54} = 20.72; p = 0.0001) compared to well-watered plants. There was a statistically significant two-way interaction between pests and hybrid for average stem diameter and between pests and watering regime for shoot dry mass. Consequently, a simple main effects analysis was conducted in which wireworm-infested FuturiXX plants were found to have a smaller average stem diameter compared to wireworm-infested ZP341 plants (F_1,26 = 12.7; p = 0.0014), and a lower dry shoot mass was observed in water-deficient plants that were either wireworm-infested (F_{1, 26} = 8.87; p = 0.0062) or uninfested (F_{1, 26} = 25.57; p < 0.0001) compared to well-watered plants. There was also a statistically significant three-way interaction between pests, watering regime and hybrid for shoot and root dry mass. Consequently, a simple main effects analysis was conducted in which well-watered FuturiXX plants without wireworms (treatment N/W) were found to have lower dry shoot (F_{1, 48} = 14.2; p = 0.0004) and root masses (F_{1, 48} = 24.7; p < 0.0001) compared to well-watered ZP341 plants without wireworms (treatment N/W).

3.4. PCA of Physiological and Morphological Maize Parameters

Four separate PCAs were performed: (1) physiological traits of maize on day 14, (2) physiological traits of maize on day 21, (3) physiological traits of maize on day 28 (Figure 8) and (4) morphological traits of maize (Figure 7). On day 14, the first two principal components had eigenvalues greater than 1, explaining 70.8% of the total variance, with leaf stomatal conductivity, transpiration rate, net photosynthesis and effective quantum yield of PSII contributing most to explanation of the variability of the first axis (Dim1) and relative chlorophyll content, maximum quantum efficiency of PSII and relative water content contributing most to explaining the variability of the second axis (Dim2) (see variance of each principal component in Table S1). On day 21, the first two principal components had eigenvalues greater than 1, explaining 78.3% of the total variance, with transpiration rate, leaf stomatal conductivity, net photosynthesis and effective quantum yield of PSII contributing most to explaining the variability of the first axis (Dim1) and maximum quantum efficiency of PSII and relative chlorophyll content contributing most to explaining the variability of the second axis (Dim2). On day 28, the first two principal components had eigenvalues greater than 1, explaining 73.3% of the total variance, with the effective quantum yield of PSII, leaf stomatal conductivity, transpiration rate and net photosynthesis contributing most to explaining the variability of the first axis (Dim1) and relative chlorophyll content and relative water content contributing most to explaining the variability of the second axis (Dim2). The first two principal components of the morphological data explain 82.1% of the total variance, with only the first principal component having an eigenvalue greater than 1. Shoot mass and root mass are the variables that contribute most to explaining the variability of the first axis (Dim1), and the average stem diameter and shoot length contribute most to explaining the variability of the second axis (Dim2).

The PCA shows a clear clustering of both morphological and physiological data based on watering regime, hybrids and the presence of wireworms. This was also confirmed by the PERMANOVA test, which showed significant differences in morphology among all variables (Table S2a) and significant differences in physiology among hybrids on day 14 and among all variables on day 21 (Table S2b).

3.5. Herbivory Damage

The number of tunnels caused by wireworms in maize was higher in water-deficient plants but not significantly (χ²(1) = 3.123, p = 0. 0.077). The number of tunnels was significantly higher in hybrid ZP341 compared to FuturiXX (χ²(1) = 5.367, p = 0. 0.0205); specifically, wireworm-infested and water-deficient ZP341 plants (treatment Y/D) had a higher number of tunnels than wireworm-infested and well-watered FuturiXX plants (treatment Y/W) (χ²(3) = 9.189, p = 0. 0.0269) (Figure 6).

4. Discussion

Visual detection of maize responses to wireworm infestations often lacks accuracy and speed to prevent significant yield losses. The results of our study show that hyperspectral remote sensing can provide accurate early detection of wireworm infestations in maize. Although PCA did not show any data groupings, classifications using SVM were accurate. Principal component analysis transforms data using an eigenvalue decomposition into N components, with each new component accounting for a decreasing proportion of the variance. This algorithm does not attempt to group the data in any way. However, it is a method for data visualisation and can be used to look at an entire set of samples. In our case, the components with the greatest variance were not relevant to our classification problem. In this regard, sPLS-DA performs better as a dimensionality reduction method. However, PLS-DA score plots of whole datasets can be misleading, especially when the number of variables exceeds the number of samples (as is often the case in metabolomics studies). In such cases, correlations may occur by chance due to many variables [69]. This effect can be overcome by using n-repeated k-fold cross validation and a train/test split to check for overfitting.

Spectral signatures in the visible region of the spectrum (400–750 nm) differed relatively little between treatment groups, although some relevant wavelengths were identified. These are generally associated with plant pigments, which are themselves affected by drought stress and plant immune responses to pests and diseases. Nevertheless, changes in absorbance around 470 nm have been linked to peroxidase and catalase activity, which, in turn, are related to plants coping with oxidative stress. More significant differences were observed in the NIR (750–1000 nm) and SWIR (900–2500 nm) regions. These wavelengths were identified as relevant for drought stress detection and are related to the second and third overtones (vibrational spectral bands with higher energy than the fundamental spectral band of a molecule after that molecule has absorbed a unit of energy and entered its excited state) of aliphatic (e.g., at 828, 919, 1167, 1395, 1759 and 2455 nm) and aromatic carbohydrates (e.g., at 802, 1015, 2150 and 2477 nm), as well as lipids (at 2140 nm), cellulose (at 2270 and 2335 nm) and water (at 1205 and 2096) [70]. Interestingly, the water absorption band at 980 nm was only relevant for pest identification but not for drought stress detection. It is possible that this water absorption band was overshadowed by other spectral bands that have a better explanatory ability for drought stress detection in maize. While several wavelengths overlap in drought stress and pest detection, there are a few notable differences. For example, in pest detection, wavelengths associated with N-H and O-H stretching in proteins showed better discrimination ability (e.g., at 2064, 2167 and 2178 nm). Furthermore, in addition to aliphatic and aromatic carbohydrates, simpler hydrocarbons (such as glucose) also helped in the detection of infested plants. Plants respond to pest infestations by producing secondary metabolites and altering the concentrations and/or activity of defence-related enzymes; for example, plants responded to nematode infestations by increasing the production and activity of chitinase, peroxidase and catalase. Peroxidases are associated with defence mechanisms against pathogens, and catalases catalyse the decomposition of antimicrobial agents (such as hydrogen peroxide) to water and oxygen [71]. Our results were partially confirmed in [42]; to the best of our knowledge, this is the only publication in which the damage of two maize insect pests was predicted using hyperspectral imaging, namely the fall armyworm (Spodoptera frugiperda Smith & Abbot, 1797) and the green-belly stink bug (Dichelops melacanthus Dallas, 1851). Using the random forest model, researchers achieved an overall accuracy of up to 96.7% in distinguishing damaged from undamaged maize plants, especially in the near-infrared range. Although random forest can achieve good classification accuracy, it is prone to overfitting, and great care must be taken in model training and validation to reduce this effect. In our study, we combined a dimensionality reduction method with support vector machines and used repeated cross validation to minimize overfitting. Final validations were then performed on a separate test dataset that the algorithm had not yet “seen”. While this approach significantly increases computation time, it can provide more accurate classification models that can still produce acceptable results on new data.

Maize is sensitive to drought at all stages, but the greatest yield losses are caused by stress in the late vegetative and early reproductive stages, especially during pollination and flowering [5,72,73]. However, the measurements in this particular experiment were conducted prior to these stages because greenhouse space was limited, and we could not conduct experiments with larger plants/pots. We also wanted to avoid damage that might occur to larger plants during their manipulation for measurements. If the experiments were continued at later growth stages, even more obvious signs of drought stress would be expected, especially signs of a reduction in photosynthesis, which could be observed in addition to growth reduction [74,75]. Overall, our results show that the average stem diameter and dry shoot mass, as well as shoot length, of water-stressed plants were significantly lower compared to well-watered plants. One of the effects of drought was also evident on day 14 in the form of a lower transpiration rate, followed by stomatal closure on day 21 to prevent water loss, which, in turn, resulted in lower effective quantum yield of PSII and lower net photosynthesis. In contrast to drought, maize is most susceptible to wireworm attack in the early stages of development [14,76]. In this experiment, wireworms were added to the maize pots after the most critical period (i.e., when maize is between the seedling and four (six)-leaf stages). At the end of the experiment, we evaluated the damage caused by the wireworms and found that most tunnels and feeding wounds occurred at the stem base and adventitious roots. During this stage, the maize plants were already large enough that such damage was not lethal, and they responded by forming new adventitious roots. Nevertheless, such herbivory can weaken the plant, resulting in stunting or wilting of the plant [17]. Consistent with this theory, wireworm-infested maize plants in our experiment were generally shorter, had lower shoot dry mass, lower maximum quantum efficiency of PSII and lower relative chlorophyll levels. A rather unexpected result was that the wireworm-infested plants had higher relative water content and higher effective quantum yield of PSII, which will be discussed later.

The effects of climate change on plants are complex and anything but simple. Nevertheless, the belowground environment is more stable than the aboveground environment, and the effects of rising global temperatures on increased larval feeding and movement are expected only in the upper soil layers [11]. In contrast, soil moisture or lack thereof may trigger more significant behavioural responses, such as intense foraging for moisture or feeding at greater depths [17,18]. Wireworms lose water rapidly in a dry environment [19]. Therefore, a combination of dry and hot weather is likely to cause larvae to move deeper into the soil, where temperatures are lower, food is scarce and development rates are consequently slower or, on the other hand, to intensify root foraging in search of adequate moisture [11,77]. Low humidity may also adversely affect wireworms during their metamorphosis from larva to pupa [19] and during egg laying and development between April and June [78,79].

The plant stress hypothesis states that plants are more favourable to insect feeding in stressful situations due to increased nutrient availability, especially nitrogen and carbohydrates, and reduced defence capability [80]. Therefore, it is expected that changes in plant morphology and physiology would be more severe in the case of multiple stressors. However, plant–insect interactions may change under drought conditions. In [81,82], the combined effects of herbivory and drought were studied, and plants were found to be more resistant to herbivory under mild drought conditions. In this case, plant defence mechanisms could be triggered by abiotic stress, resulting in increased resistance to herbivory. In addition to the presence of stressors, their intensity should also be considered, as a change in the intensity of one stressor can have either a synergistic or antagonistic effect on another stressor [83]. Additionally, the combined effects in our experiment were by no means one-sided and, in some cases, caused less physiological damage than expected based on the results of either stressor alone. Looking at the results on day 21, when the greatest differences between treatments were observed in the plants, drought was the dominant factor in negative physiological outcomes in both maize hybrids, as both hybrids had low stomatal conductivity, effective quantum yield of PSII, transpiration rate and relative water content. However, in the water deficit treatments with the FuturiXX hybrid, in which wireworms were also added (treatment Y/D), most of the values of the mentioned parameters were higher, and the plants seemed to be less stressed; for example, the relative water content under treatment Y/D was comparable to the control (treatment N/W) and significantly higher than under the treatment of drought exposure with no pests (treatment N/D). Another surprising result is the fact that the presence of wireworms in well-watered plants did not have negative effects on maize physiological traits in any of the hybrids; in fact, some values were even higher. For example, the effective quantum yield of PSII in ZP341 (treatment Y/W) increased by 36% when wireworms were present compared with the control (treatment N/W). The presence of root herbivory may have mitigated the negative effects of drought while slightly improving—or at least not significantly affecting—the physiological characteristics of the well-watered plants (Figure 2). However, it should be noted that the plants were already four weeks old when the wireworms were added to the pots and had therefore passed the time when wireworms pose the greatest risk to their survival. In this case, the wireworms could represent a mild stress to the maize, making the plant’s defence mechanisms more capable of effectively responding to drought stress. Overall, there were no significant differences between treatments in physiological parameters measured by classical methods on day 28, suggesting that maize adapts to both biotic and abiotic stresses.

Wireworms caused a higher number of tunnels in water-deficient maize plants than in well-watered plants. This could be due to the fact that wireworms were also affected by drought and were more active in searching for water sources [18]—in this case, maize roots and stalks. Based on the morphological data, hybrid ZP341 showed higher tolerance to wireworms than FuturiXX, as it had a significantly higher average stem diameter in the presence of wireworms. On the other hand, the number of tunnels per plant was also significantly higher in hybrid ZP341, especially in water-deficient plants (treatment Y/D). One possible explanation, as mentioned earlier, is that plants under stress are more favourable to herbivores. Therefore, as a drought-intolerant hybrid, ZP341 might begin to decrease the concentration of defence compounds or accumulate nitrogen and carbon in response to drought stress, making it more nutritious to herbivores [84,85]. However, we did not perform measurements to confirm this hypothesis. Overall, the negative effects of combined drought and herbivore stress on maize morphological traits were more pronounced in our case than under individual stress but not to the same extent in both hybrids (Figure 6).

Because wireworms spend several years in the soil developing to the adult stage [10,11,12], a reliable method for detecting wireworm infestation would be beneficial for a holistic wireworm IPM management strategy [14,86]. The results of this research can be used as an early warning system for wireworm infestations after successful field validation. While this would not allow growers to protect the crop that signals wireworm herbivory (i.e., this year’s crop) due to changed hyperspectral signatures, they could take preventative measures to protect next year’s crop, e.g., crop selection, crop rotation, use of soil insecticides, etc. [87]. The results of our research can also be used to identify wireworm-infested and uninfested fields in geographically large or diverse regions. Such area-wide monitoring of infestation pressure can be used in IPM programmes and various crop insurance schemes to reduce pesticide use [14,88].

5. Conclusions

The objective of this study was to investigate the physiological, morphological and spectral response of leaves of two maize hybrids (one drought-tolerant and one drought-sensitive) to two stressors—drought and belowground herbivory—and to test the effects of combined stress. In this research, we also evaluated the ability of hyperspectral imaging to rapidly detect biotic and abiotic stress in maize. To the best of our knowledge, this is the first study to use hyperspectral imaging for early detection of wireworm infestation in a crop. We identified the most important spectral signatures for distinguishing between treatment groups in the near-infrared (NIR) and shortwave infrared (SWIR). Wavelengths associated with aliphatic and aromatic carbohydrates, lipids, cellulose and water appear to be relevant for drought stress detection, while wavelengths associated with N-H and O-H stretching in proteins, aliphatic and aromatic carbohydrates and simpler hydrocarbons appear to be relevant for wireworm infestation detection. Our results show that the greatest differences in physiological traits measured by classical methods occurred on day 21 after the addition of wireworms and the change in watering regime. However, hyperspectral remote sensing was able to discriminate between abiotic and biotic stresses as early as day 14 with 84.7% accuracy, making it a reliable method for rapid and accurate prediction of wireworm infestation in maize challenged by drought. Our measurements of maize physiology also indicate that the combined effects were not necessarily synergistic and that maize tolerated drought stress better in the presence of mild herbivory. On the other hand, the combined effect of drought and herbivory on maize morphology was greater than the effect of the individual stressors, although a larger sample size would be beneficial to draw firmer conclusions.

This research has highlighted several future challenges that need to be addressed, namely validation of the presented laboratory study in the field with differing wireworm infestation rates, acquisition of hyperspectral images after shorter wireworm infestation periods (e.g., 1–10 days after wireworm infestation) and hyperspectral visualisation of wireworm infestation on young maize plantlets that are more susceptible to wireworm herbivory. Despite these necessary validation steps, this research provides evidence that hyperspectral imaging could be used in various IPM wireworm management to facilitate a reduction in pesticide use.

Author Contributions

Conceptualization, J.R., E.P., U.Ž., S.Š., M.K. and D.V.; methodology, E.P., J.R., U.Ž., S.Š., M.K. and D.V.; software, U.Ž., J.L. and A.V.; formal analysis, E.P., and U.Ž.; investigation, E.P., A.V., P.Ž., M.K., N.S., S.Š., D.V., D.L., J.L., U.Ž. and J.R.; writing—original draft preparation, E.P. and U.Ž.; writing—review and editing, E.P., A.V., P.Ž., M.K., N.S., S.Š., D.V., D.L., J.L., U.Ž. and J.R.; supervision, J.R. and U.Ž.; funding acquisition, J.R. and U.Ž. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Data available from the author.

Acknowledgments

The authors would like to thank Barbara Pipan, Aleš Kolmanič, Hans-Josef Schroers, Špela Modic, Melita Theuerschuh, David Snoj and Tadej Galič for their advice and help in data collection.

Conflicts of Interest

The authors declare no conflict of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. Hyperspectral data analysis pipeline: (a) hyperspectral image in radiance units from the VNIR sensor (400–950 nm); (b) image segmentation into five classes (plants, reference plate and three classes for background); (c) separation into individual plants.

Figure 2. Physiological parameters of two maize hybrids subjected to herbivory (no wireworms (N) vs. wireworm-infested (Y)) and to two watering regimes (well-watered (W) vs. water-deficient (D)) on day 21. Data are presented as mean values ± SE (n = 4). Different letters indicate a significant difference between treatments (p < 0.05).

Figure 3. Scatter plot of the first two principal components. The dots are colour-coded to plant infestation status. Although the first two PCA components explain more than 86% variance, no particular patterns can be observed.

Figure 4. Variable importance in projection (VIP) analysis. Wavelengths with VIP values above 1 are considered to be relevant for differences between classes. The number next to Day/Drought/Pest indicates the imaging session (i.e., day 14, 21 or 28 after adding wireworms and changing the watering regime).

Figure 5. Mean spectra of plants from two analysis groups. (A) Non-infested and wireworm-infested plants; (B) well-watered and water-deficient plants. Note the very similar spectral signatures in the visible part of the spectrum (400–750 nm) and increased differences in the NIR (750–1000 nm) and SWIR (900–2500 nm) regions.

Figure 6. Morphological parameters of two maize hybrids subjected to herbivory (no wireworms (N) vs. wireworm-infested (Y)) and to two water regimes (well-watered (W) vs. water-deficient (D)). Data are presented as mean values ± SE (n = 7). Different letters indicate a significant difference between treatments (p < 0.05).

Figure 7. Morphological characterisation by principal component analysis (PCA) for overall comparison of maize plants of different hybrids (ZP341 vs. FuturiXX), under different watering regimes (well-watered vs. water-deficient) and subjected to herbivory (no wireworms vs. wireworm-infested). The first axis (Dim1) explains 64.2% of the accumulated variance, and the second axis (Dim2) explains 17.8% of accumulated variance.

Figure 8. Physiological characterisation by principal component analysis (PCA) comparing maize plants of different hybrids (ZP341 vs. FuturiXX in the middle column), under different watering regimes (well-watered vs. water-deficient in the left column) and subjected to herbivory (no wireworms vs. wireworm-infested in the right column). The first axis (Dim1) explains 53.5% of the accumulated variance, and the second axis (Dim2) explains 24.4% of accumulated variance on day 14 (first row); the first axis (Dim1) explains 71.4% of the accumulated variance, and the second axis (Dim2) explains 13.7% of the accumulated variance on day 21 (second row); and lastly, the first axis (Dim1) explains 64.4% of the accumulated variance, and the second axis (Dim2) explains 19.4% of the accumulated variance on day 28 (third row). rwc: relative water content; yield: effective quantum yield of PSII; cond: stomatal conductivity; Trmmol: transpiration rate; photo: net photosynthesis; SPAD: chlorophyll content.

Table 1

Setup of factorial pot experiment, where two maize hybrids were subjected to two watering regimes (water-deficient [D] vs. well-watered [W]) and to belowground herbivory (no wireworms [N] vs. wireworm-infested [Y]).

Treatment	Hybrid	Pest	Watering Regime
N/D	ZP341	No	D
N/W	ZP341	No	W
N/D	FuturiXX	No	D
N/W	FuturiXX	No	W
Y/D	ZP341	Yes	D
Y/W	ZP341	Yes	W
Y/D	FuturiXX	Yes	D
Y/W	FuturiXX	Yes	W

Table 2

Identification of the phenological development stages (BBCH) of maize for each experimental step.

Date	DPES a	DPSI b	BBCH c	Activity
9 April 2020	0	0	0	Planting the maize
6 May 2020	27	0	34	Adding wireworms and changing watering regime
20 May 2020	41	14	35	Acquisition of physiological parameters and hyperspectral imaging
27 May 2020	48	21	35	Acquisition of physiological parameters and hyperspectral imaging
3 June 2020	55	28	35	Acquisition of physiological parameters and hyperspectral imaging
4 June 2020	56	29	35	Acquisition of morphological parameters and termination of the experiment

^a DPES: days post experiment start; ^b DPSI: days post stress induction; ^c BBCH: Biologische Bundesanstalt, Bundessortenamt and Chemical industry.

Table 3

Physiological parameters of two maize hybrids subjected to herbivory and to two watering regimes on day 14, day 21 and day 28. Data are presented as mean values ± SE (n = 4). Treatment labels are defined in Table 1; different letters indicate a significant difference among treatments (p < 0.05).

	Hybrid	Treatment	Rwc(%) a	gs (mmol m−2 s−1) b	E (mmol H2O m−2 s−1) c	PN (µmol/m2/s) d	SPAD e	Fv’/Fm’ f	Fv/Fm g
DAY 14	ZP341	N/D	92.68 ± 1.49 a	0.11 ± 0.02 a	1.35 ± 0.20 a	18.74 ± 2.75 a	49.08 ± 1.55 a	0.58 ± 0.02 a	0.79 ± 0.00 a
		N/W	96.06 ± 0.50 a	0.16 ± 0.02 a	2.04 ± 0.25 a	26.50 ± 1.99 a	49.13 ± 1.18 a	0.58 ± 0.02 a	0.78 ± 0.00 ab
		Y/D	91.77 ± 2.88 a	0.13 ± 0.02 a	1.76 ± 0.12 a	23.81 ± 2.09 a	46.93 ± 0.95 a	0.58 ± 0.02 a	0.76 ± 0.01 abc
		Y/W	96.34 ± 1.35 a	0.15 ± 0.01 a	1.99 ± 0.29 a	23.79 ± 3.99 a	44.30 ± 2.47 a	0.61 ± 0.01 a	0.78 ± 0.01 ab
	FuturiXX	N/D	96.36 ± 10.31 a	0.09 ± 0.03 a	1.22 ± 0.27 a	15.37 ± 4.53 a	46.70 ± 2.86 a	0.50 ± 0.07 a	0.78 ± 0.00 ab
		N/W	97.73 ± 0.59 a	0.13 ± 0.02 a	1.71 ± 0.30 a	19.73 ± 2.40 a	41.33 ± 2.00 a	0.60 ± 0.02 a	0.77 ± 0.01 abc
		Y/D	102.65 ± 3.69 a	0.14 ± 0.01 a	1.90 ± 0.23 a	19.60 ± 2.39 a	43.70 ± 1.78 a	0.63 ± 0.01 a	0.75 ± 0.01 bc
		Y/W	101.67 ± 4.31 a	0.14 ± 0.01 a	1.96 ± 0.25 a	20.37 ± 2.02 a	42.98 ± 1.46 a	0.61 ± 0.02 a	0.74 ± 0.01 c
DAY 21	ZP341	N/D	76.23 ± 2.22 b	0.05 ± 0.01 b	0.85 ± 0.14 b	9.70 ± 1.41 a	47.08 ± 0.39 a	0.48 ± 0.03 ab	0.79 ± 0.00 a
		N/W	81.68 ± 4.01 ab	0.07 ± 0.02 ab	1.23 ± 0.38 ab	13.02 ± 4.71 a	46.88 ± 1.53 a	0.47 ± 0.06 ab	0.79 ± 0.00 a
		Y/D	82.30 ± 1.37 ab	0.05 ± 0.02 b	0.90 ± 0.28 b	10.34 ± 3.40 a	43.45 ± 1.05 a	0.46 ± 0.07 ab	0.79 ± 0.01 a
		Y/W	87.35 ± 1.79 ab	0.15 ± 0.02 a	2.28 ± 0.22 a	19.74 ± 2.86 a	37.73 ± 6.41 a	0.64 ± 0.01 ab	0.79 ± 0.01 a
	FuturiXX	N/D	77.97 ± 4.36 b	0.06 ± 0.01 b	1.02 ± 0.14 ab	11.73 ± 1.58 a	43.28 ± 3.20 a	0.44 ± 0.05 b	0.79 ± 0.00 a
		N/W	91.08 ± 0.92 a	0.13 ± 0.01 ab	2.00 ± 0.06 ab	20.39 ± 1.23 a	44.23 ± 2.22 a	0.63 ± 0.01 ab	0.79 ± 0.00 a
		Y/D	92.07 ± 1.68 a	0.10 ± 0.02 ab	1.59 ± 0.34 ab	14.71 ± 2.59 a	37.50 ± 3.02 a	0.60 ± 0.05 ab	0.78 ± 0.00 a
		Y/W	91.22 ± 0.98 a	0.11 ± 0.03 ab	1.75 ± 0.43 ab	15.24 ± 3.84 a	38.63 ± 1.53 a	0.65 ± 0.01 a	0.78 ± 0.01 a
DAY 28	ZP341	N/D	83.11 ± 5.88 a	0.08 ± 0.03 a	1.12 ± 0.39 a	15.85 ± 5.40 a	46.70 ± 1.00 a	0.47 ± 0.08 a	0.79 ± 0.00 a
		N/W	83.72 ± 4.29 a	0.08 ± 0.03 a	1.22 ± 0.38 a	16.13 ± 4.13 a	47.03 ± 1.56 a	0.47 ± 0.05 a	0.78 ± 0.00 a
		Y/D	85.08 ± 3.19 a	0.08 ± 0.01 a	1.39 ± 0.23 a	15.40 ± 1.44 a	44.00 ± 1.74 a	0.49 ± 0.04 a	0.79 ± 0.00 a
		Y/W	89.94 ± 2.32 a	0.10 ± 0.02 a	1.84 ± 0.38 a	17.71 ± 2.88 a	45.58 ± 1.82 a	0.53 ± 0.05 a	0.78 ± 0.01 a
	FuturiXX	N/D	83.96 ± 3.64 a	0.07 ± 0.01 a	1.17 ± 0.29 a	14.40 ± 3.30 a	44.28 ± 2.19 a	0.47 ± 0.05 a	0.79 ± 0.00 a
		N/W	89.47 ± 3.61 a	0.13 ± 0.03 a	2.26 ± 0.47 a	19.20 ± 2.53 a	41.33 ± 3.68 a	0.56 ± 0.05 a	0.79 ± 0.01 a
		Y/D	91.09 ± 1.99 a	0.09 ± 0.01 a	1.49 ± 0.13 a	16.23 ± 1.87 a	38.03 ± 2.01 a	0.56 ± 0.04 a	0.78 ± 0.00 a
		Y/W	90.76 ± 1.66 a	0.11 ± 0.01 a	2.04 ± 0.38 a	20.04 ± 1.59 a	41.88 ± 2.09 a	0.58 ± 0.04 a	0.78 ± 0.00 a

^a Rwc: relative water content; ^b g_s: stomatal conductivity; ^c E: transpiration rate; ^d P_N: net photosynthesis; ^e SPAD: chlorophyll content; ^f Fv’/Fm’: effective quantum yield of PSII; ^g Fv/Fm: maximum quantum efficiency of PSII.

Table 4

sPLS-SVM classification results of pooled data from all three imaging sessions (days 14, 21 and 28). Treatment labels are defined in Table 1.

Group a	N b	LV c	Var (%) d	Mahalanobis e	c f	Sigma g	OA h	Kappa i		Class j	Accuracy k	Sensitivity	Specificity	PPV l	NPV m
Imaging session	49	19	95.30	0.0176	10	0.011187	1	1		14	1	1	1	1	1
										21	1	1	1	1	1
										28	1	1	1	1	1
Treatment	46	29	97.90	0.699	10	0.001	0.63	0.577	ZP341	N/D	0.808	0.667	0.95	0.667	0.95
										N/W	0.917	0.83	1	1	0.976
										Y/D	0.975	1	0.95	0.75	1
										Y/W	0.66	0.4	0.927	0.4	0.927
									FuturiXX	N/D	0.78	0.667	0.9	0.5	0.947
										N/W	0.76	0.6	0.927	0.5	0.95
										Y/D	0.58	0.167	1	1	0.889
										Y/W	0.796	0.667	0.925	0.57	0.949
Pest	49	19	95.30	0.248	10	0.001	0.98	0.959		/	0.979	1	0.958	0.96	1
Drought	49	13	90.50	0.379	10	0.001	0.959	0.918		/	0.96	0.92	1	1	0.92

^a Group: classification group (identification of imaging session, all treatments, pest infestation and drought stress); ^b N: number of samples in the test set; ^c LV: number of selected latent variables; ^d Var: cumulative explained variance of the selected LVs; ^e maximum/centroids/Mahalanobis: distance measures for sPLS-DA classification error rates; ^f c: SVM cost function value; ^g Sigma: SVM sigma value; ^h OA: overall accuracy of sPLS-SVM classification; ⁱ Kappa: kappa statistic of repeated cross validation; ^j Class: Class designation by imaging sessions (Day 14/21/28) and treatment (ZP341 N/D, N/W, Y/D and Y/W; FuturiXX N/D, N/W, Y/D and Y/W); ^k Accuracy: per-class accuracy of classification; ^l PPV: positive predictive value; ^m NPV: negative predictive value.

Table 5

sPLS-SVM classification results of each imaging session. Treatment labels are defined in Table 1.

Group a	N b	LV c	Var (%) d	Mahalanobis e	c f	Sigma g	OA h	Kappa i		Class j	Accuracy k	Sensitivity	Specificity	PPV l	NPV m
DAY 14	16	19	97.90	0.689	10	0.01018	0.867	0.847	ZP341	N/D	0.808	0.667	0.95	0.667	0.95
										N/W	0.917	0.83	1	1	0.976
										Y/D	0.975	1	0.95	0.75	1
										Y/W	0.66	0.4	0.927	0.4	0.927
									FuturiXX	N/D	0.78	0.667	0.9	0.5	0.947
										N/W	0.76	0.6	0.927	0.5	0.95
										Y/D	0.58	0.167	1	1	0.889
										Y/W	0.796	0.667	0.925	0.57	0.949
DAY 21	12	11	92.40	0.655	10	0.009059	0.833	0.81	ZP341	N/D	0.7	0.5	0.9	0.5	0.9
										N/W	0.7	0.5	0.9	0.5	0.9
										Y/D	1	1	1	1	1
										Y/W	1	1	1	1	1
									FuturiXX	N/D	1	1	1	1	1
										N/W	1	1	1	1	1
										Y/D	1	1	1	1	1
										Y/W	1	1	1	1	1
DAY 28	15	8	91.40	0.841	10	0.01	0.67	0.61	ZP341	N/D	0.5	0	1	0	0.867
										N/W	1	1	1	1	1
										Y/D	0.67	0.5	0.846	0.333	0.92
										Y/W	0.92	1	0.846	0.5	1
									FuturiXX	N/D	0.96	1	0.92	0.67	1
										N/W	0.5	0	1	0	0.93
										Y/D	0.75	0.5	1	1	0.93
										Y/W	1	1	1	1	1
pest_14	16	7	86.50	0.254	1000	0.0001	1	1			1	1	1	1	1
pest_21	15	6	82	0.402	100	0.001	1	1			1	1	1	1	1
pest_28	16	17	93.80	0.374	1000	0.0001	1	1			1	1	1	1	1
drought _14	16	9	90.70	0.468	1000	0.0001	1	1			1	1	1	1	1
drought _21	15	16	94.80	0.448	100	0.001	1	1			1	1	1	1	1
drought _28	17	9	85	0.462	1000	0.0001	1	1			1	1	1	1	1

^a Group: day 14/21/28 identification of all treatment groups for each imaging session (the number designates the imaging session); pest 14/21/28: identification of infested plants; drought 14/21/28: identification of plants under drought stress; ^b N: number of samples in the test set; ^c LV: number of selected latent variables; ^d Var: cumulative explained variance of the selected LVs; ^e Maximum/centroids/Mahalanobis: distance measures for sPLS-DA classification error rates; ^f c: SVM cost function value; ^g Sigma: SVM sigma value; ^h OA: overall accuracy of sPLS-SVM classification; ⁱ Kappa: kappa statistic of repeated cross validation; ^j Class: class designation by treatment group (ZP341 N/D, N/W, Y/D and Y/W; FuturiXX N/D, N/W, Y/D and Y/W) for each imaging session; ^k Accuracy: per-class accuracy of classification; ^l PPV: positive predictive value; ^m NPV: negative predictive value.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/agronomy13010178/s1, Table S1: Eigenvalues and proportion of variance of each principal component; Table S2: PERMANOVA.

References

1. Bento, V.A.; Ribeiro, A.F.S.; Russo, A.; Gouveia, C.M.; Cardoso, R.M.; Soares, P.M.M. The impact of climate change in wheat and barley yields in the Iberian Peninsula. Sci. Rep.; 2021; 11, 15484. [DOI: https://dx.doi.org/10.1038/s41598-021-95014-6] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34326411]

2. Jennings, S.A.; Koehler, A.K.; Nicklin, K.J.; Deva, C.; Sait, S.M.; Challinor, A.J. Global Potato Yields Increase Under Climate Change with Adaptation and CO₂ Fertilisation. Front. Sustain. Food Syst.; 2020; 4, [DOI: https://dx.doi.org/10.3389/fsufs.2020.519324]

3. Potopová, V.; Lhotka, O.; Možný, M.; Musiolková, M. Vulnerability of hop-yields due to compound drought and heat events over European key-hop regions. Int. J. Climatol.; 2021; 41, pp. E2136-E2158. [DOI: https://dx.doi.org/10.1002/joc.6836]

4. Jacobs, C.; Berglund, M.; Kurnik, B.; Dworak, T.; Marras, S.; Mereu, V.; Michetti, M. Climate Change Adaptation in the Agriculture Sector in Europe (4/2019); European Environment Agency: Copenhagen, Denmark, 2019.

5. Chen, J.; Xu, W.; Velten, J.; Xin, Z.; Stout, J. Characterization of maize inbred lines for drought and heat tolerance. J. Soil Water Conserv.; 2012; 67, pp. 354-364. [DOI: https://dx.doi.org/10.2489/jswc.67.5.354]

6. Westgate, M.E.; Boyer, J.S. Osmotic adjustment and the inhibition of leaf, root, stem.pdf. Planta; 1985; 164, pp. 540-549. [DOI: https://dx.doi.org/10.1007/BF00395973]

7. Skendžić, S.; Zovko, M.; Živković, I.P.; Lešić, V.; Lemić, D. The Impact of Climate Change on Agricultural Insect Pests. Insects; 2021; 12, 440. [DOI: https://dx.doi.org/10.3390/insects12050440]

8. Veres, A.; Wyckhuys, K.A.G.; Kiss, J.; Tóth, F.; Burgio, G.; Pons, X.; Avilla, C.; Vidal, S.; Razinger, J.; Bazok, R. et al. An update of the Worldwide Integrated Assessment (WIA) on systemic pesticides. Part 4: Alternatives in major cropping systems. Environ. Sci. Pollut. Res.; 2020; 27, pp. 29867-29899. [DOI: https://dx.doi.org/10.1007/s11356-020-09279-x]

9. Lohse, G.A. Familie: Elateridae. Die Käfer Mitteleuropas, Bd. 6: Diversicornia (Lycidea-Byrrhidae); Freude, H. Spektrum Akademischer Verlag: Heidelberg, Germany, 1979; pp. 103-186.

10. Subklew, W. Agriotes lineatus L. und Agriotes obscurus L. (Ein Beitrag zu ihrer Morphologie und Biologie.). Zeitschrift für Angew. Entomol.; 1934; 21, pp. 96-122. [DOI: https://dx.doi.org/10.1111/j.1439-0418.1935.tb00394.x]

11. Furlan, L. The biology of Agriotes ustulatus Schaller (Col., Elateridae). II. Larval development, pupation, whole cycle description and practical implications. J. Appl. Entomol.; 1998; 122, pp. 71-78. [DOI: https://dx.doi.org/10.1111/j.1439-0418.1998.tb01464.x]

12. Furlan, L. The biology of Agriotes sordidus Illiger (Col., Elateridae). J. Appl. Entomol.; 2004; 128, pp. 696-706. [DOI: https://dx.doi.org/10.1111/j.1439-0418.2004.00914.x]

13. Saussure, S.; Plantegenest, M.; Thibord, J.B.; Larroudé, P.; Poggi, S. Management of wireworm damage in maize fields using new, landscape-scale strategies. Agron. Sustain. Dev.; 2015; 35, pp. 793-802. [DOI: https://dx.doi.org/10.1007/s13593-014-0279-5]

14. Furlan, L.; Contiero, B.; Chiarini, F.; Colauzzi, M.; Sartori, E.; Benvegnù, I.; Fracasso, F.; Giandon, P. Risk assessment of maize damage by wireworms (Coleoptera: Elateridae) as the first step in implementing IPM and in reducing the environmental impact of soil insecticides. Environ. Sci. Pollut. Res.; 2017; 24, pp. 236-251. [DOI: https://dx.doi.org/10.1007/s11356-016-7692-z]

15. Kozina, A.; Lemic, D.; Bazok, R.; Mikac, K.M.; Mclean, C.M.; Ivezić, M.; Igrc Barčić, J.; Aukema, B. Climatic, edaphic factors and cropping history help predict click beetle (Coleoptera: Elateridae) (Agriotes spp.) abundance. J. Insect Sci.; 2015; 15, pp. 1-8. [DOI: https://dx.doi.org/10.1093/jisesa/iev079]

16. Staudacher, K.; Schallhart, N.; Pitterl, P.; Wallinger, C.; Brunner, N.; Landl, M.; Kromp, B.; Glauninger, J.; Traugott, M. Occurrence of Agriotes wireworms in Austrian agricultural land. J. Pest Sci.; 2013; 86, pp. 33-39. [DOI: https://dx.doi.org/10.1007/s10340-011-0393-y]

17. Čamprag, D. Skočibube (Elateridae) i Integralne Mere Suzbijanja; Poljoprivredni Fakultet, Institut za Zaštitu Bilja, Design Studio Stanišić: Novi Sad, Serbia, 1997.

18. Radcliffe, E.B.; Lagnaoui, A. Insect pests in potato. Potato Biology and Biotechnology: Advances and Perspectives; Vreugdenhil, D. Elsevier B.V.: Amsterdam, The Netherlands, 2007; pp. 543-567. ISBN 9780444510181

19. Jones, E.W. Laboratory Studies on the Moisture Relations of Limonius (Coleoptera: Elaterida). Ecology; 1951; 32, pp. 284-293. [DOI: https://dx.doi.org/10.2307/1930420]

20. Voigt, C.A. Callose-mediated resistance to pathogenic intruders in plant defense-related papillae. Front. Plant Sci.; 2014; 5, pp. 1-6. [DOI: https://dx.doi.org/10.3389/fpls.2014.00168]

21. Susič, N.; Žibrat, U.; Sinkovič, L.; Vončina, A.; Razinger, J.; Knapič, M.; Sedlar, A.; Širca, S.; Gerič Stare, B. From Genome to Field—Observation of the Multimodal Nematicidal and Plant Growth-Promoting Effects of Bacillus firmus I-1582 on Tomatoes Using Hyperspectral Remote Sensing. Plants; 2020; 9, 592. [DOI: https://dx.doi.org/10.3390/plants9050592]

22. Brugger, A.; Behmann, J.; Paulus, S.; Luigs, H.-G.; Kuska, M.T.; Schramowski, P.; Kersting, K.; Steiner, U.; Mahlein, A.-K. Extending Hyperspectral Imaging for Plant Phenotyping to the UV-Range. Remote Sens.; 2019; 11, 1401. [DOI: https://dx.doi.org/10.3390/rs11121401]

23. Abd El-Ghany, N.M.; Abd El-Aziz, S.E.; Marei, S.S. A review: Application of remote sensing as a promising strategy for insect pests and diseases management. Environ. Sci. Pollut. Res.; 2020; 27, pp. 33503-33515. [DOI: https://dx.doi.org/10.1007/s11356-020-09517-2]

24. Wójtowicz, M.; Wójtowicz, A.; Piekarczyk, J. Application of remote sensing methods in agriculture. Commun. Biometry Crop Sci.; 2016; 11, pp. 31-50.

25. Acker, J.; Williams, R.; Chiu, L.; Ardanuy, P.; Miller, S.; Schueler, C.; Vachon, P.W.; Manore, M. Remote Sensing from Satellites. Reference Module in Earth Systems and Environmental Sciences; Elsevier: Amsterdam, The Netherlands, 2014.

26. Terentev, A.; Dolzhenko, V.; Fedotov, A.; Eremenko, D. Current State of Hyperspectral Remote Sensing for Early Plant Disease Detection: A Review. Sensors; 2022; 22, 757. [DOI: https://dx.doi.org/10.3390/s22030757] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35161504]

27. Susič, N.; Žibrat, U.; Širca, S.; Strajnar, P.; Razinger, J.; Knapič, M.; Vončina, A.; Urek, G.; Gerič Stare, B. Discrimination between abiotic and biotic drought stress in tomatoes using hyperspectral imaging. Sens. Actuators B Chem.; 2018; 273, pp. 842-852. [DOI: https://dx.doi.org/10.1016/j.snb.2018.06.121]

28. Santos, L.B.; Bastos, L.M.; de Oliveira, M.F.; Soares, P.L.M.; Ciampitti, I.A.; da Silva, R.P. Identifying Nematode Damage on Soybean through Remote Sensing and Machine Learning Techniques. Agronomy; 2022; 12, 2404. [DOI: https://dx.doi.org/10.3390/agronomy12102404]

29. Hunt, E.R.; Rondon, S.I. Detection of potato beetle damage using remote sensing from small unmanned aircraft systems. J. Appl. Remote Sens.; 2017; 11, 026013. [DOI: https://dx.doi.org/10.1117/1.JRS.11.026013]

30. Yang, Z.; Rao, M.N.; Elliott, N.C.; Kindler, S.D.; Popham, T.W. Differentiating stress induced by greenbugs and Russian wheat aphids in wheat using remote sensing. Comput. Electron. Agric.; 2009; 67, pp. 64-70. [DOI: https://dx.doi.org/10.1016/j.compag.2009.03.003]

31. Iost Filho, F.H.; de Bastos Pazini, J.; de Medeiros, A.D.; Rosalen, D.L.; Yamamoto, P.T. Assessment of Injury by Four Major Pests in Soybean Plants Using Hyperspectral Proximal Imaging. Agronomy; 2022; 12, 1516. [DOI: https://dx.doi.org/10.3390/agronomy12071516]

32. Zhang, X.; Liu, F.; He, Y.; Li, X. Application of hyperspectral imaging and chemometric calibrations for variety discrimination of maize seeds. Sensors; 2012; 12, pp. 17234-17246. [DOI: https://dx.doi.org/10.3390/s121217234]

33. Williams, P.; Geladi, P.; Fox, G.; Manley, M. Maize kernel hardness classification by near infrared (NIR) hyperspectral imaging and multivariate data analysis. Anal. Chim. Acta; 2009; 653, pp. 121-130. [DOI: https://dx.doi.org/10.1016/j.aca.2009.09.005]

34. Oglesby, C.; Fox, A.A.A.; Singh, G.; Dhillon, J. Predicting In-Season Corn Grain Yield Using Optical Sensors. Agronomy; 2022; 12, 2402. [DOI: https://dx.doi.org/10.3390/agronomy12102402]

35. Wang, S.; Guan, K.; Wang, Z.; Ainsworth, E.A.; Zheng, T.; Townsend, P.A.; Liu, N.; Nafziger, E.; Masters, M.D.; Li, K. et al. Airborne hyperspectral imaging of nitrogen deficiency on crop traits and yield of maize by machine learning and radiative transfer modeling. Int. J. Appl. Earth Obs. Geoinf.; 2021; 105, 102617. [DOI: https://dx.doi.org/10.1016/j.jag.2021.102617]

36. Feng, X.; Zhao, Y.; Zhang, C.; Cheng, P.; He, Y. Discrimination of transgenic maize kernel using NIR hyperspectral imaging and multivariate data analysis. Sensors; 2017; 17, 1894. [DOI: https://dx.doi.org/10.3390/s17081894]

37. Dhau, I.; Adam, E.; Mutanga, O.; Ayisi, K.K. Detecting the severity of maize streak virus infestations in maize crop using in situ hyperspectral data. Trans. R. Soc. S. Afr.; 2017; 73, pp. 8-15. [DOI: https://dx.doi.org/10.1080/0035919X.2017.1370034]

38. Del Fiore, A.; Reverberi, M.; Ricelli, A.; Pinzari, F.; Serranti, S.; Fabbri, A.A.; Bonifazi, G.; Fanelli, C. Early detection of toxigenic fungi on maize by hyperspectral imaging analysis. Int. J. Food Microbiol.; 2010; 144, pp. 64-71. [DOI: https://dx.doi.org/10.1016/j.ijfoodmicro.2010.08.001]

39. Zhao, X.; Wang, W.; Chu, X.; Li, C.; Kimuli, D. Early detection of Aspergillus parasiticus infection in maize kernels using near-infrared hyperspectral imaging and multivariate data analysis. Appl. Sci.; 2017; 7, 90. [DOI: https://dx.doi.org/10.3390/app7010090]

40. Williams, P.J.; Geladi, P.; Britz, T.J.; Manley, M. Investigation of fungal development in maize kernels using NIR hyperspectral imaging and multivariate data analysis. J. Cereal Sci.; 2012; 55, pp. 272-278. [DOI: https://dx.doi.org/10.1016/j.jcs.2011.12.003]

41. Ausmus, B.S.; Hilty, J.W. Reflectance studies of healthy, maize dwarf mosaic virus-infected, and Helminthosporium maydis-infected corn leaves. Remote Sens. Environ.; 1971; 2, pp. 77-81. [DOI: https://dx.doi.org/10.1016/0034-4257(71)90080-0]

42. Garcia Furuya, D.E.; Ma, L.; Faita Pinheiro, M.M.; Georges Gomes, F.D.; Gonçalvez, W.N.; Junior, J.M.; de Castro Rodrigues, D.; Blassioli-Moraes, M.C.; Furtado Michereff, M.F.; Borges, M. et al. Prediction of insect-herbivory-damage and insect-type attack in maize plants using hyperspectral data. Int. J. Appl. Earth Obs. Geoinf.; 2021; 105, 102608. [DOI: https://dx.doi.org/10.1016/j.jag.2021.102608]

43. Chávez-Arias, C.C.; Ligarreto-Moreno, G.A.; Ramírez-Godoy, A.; Restrepo-Díaz, H. Maize Responses Challenged by Drought, Elevated Daytime Temperature and Arthropod Herbivory Stresses: A Physiological, Biochemical and Molecular View. Front. Plant Sci.; 2021; 12, 702841. [DOI: https://dx.doi.org/10.3389/fpls.2021.702841]

44. Kölliker, U.; Jossi, W.; Kuske, S. Optimised protocol for wireworm rearing. IOBC/WPRS Bull.; 2009; 45, pp. 457-460.

45. van Herk, W.G.; Vernon, R.S. Categorization and numerical assessment of wireworm mobility over time following exposure to bifenthrin. J. Pest Sci.; 2013; 86, pp. 115-123. [DOI: https://dx.doi.org/10.1007/s10340-011-0381-2]

46. Hartung, J.; Wagener, J.; Ruser, R.; Piepho, H.P. Blocking and re-arrangement of pots in greenhouse experiments: Which approach is more effective?. Plant Methods; 2019; 15, 143. [DOI: https://dx.doi.org/10.1186/s13007-019-0527-4] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31798669]

47. Lichtenthaler, H.K.; Buschmann, C.; Knapp, M. How to correctly determine the different chlorophyll fluorescence parameters and the chlorophyll fluorescence decrease ratio RFd of leaves with the PAM fluorometer. Photosynthetica; 2005; 43, pp. 379-393. [DOI: https://dx.doi.org/10.1007/s11099-005-0062-6]

48. Kramer, P.J.; Boyer, J.S. Water Relations of Plants and Soils; Academic Press, Inc.: New York, NY, USA, 1995; ISBN 978-0124250604

49. Piqueras, S.; Burger, J.; Tauler, R.; de Juan, A. Relevant aspects of quantification and sample heterogeneity in hyperspectral image resolution. Chemom. Intell. Lab. Syst.; 2012; 117, pp. 169-182. [DOI: https://dx.doi.org/10.1016/j.chemolab.2011.12.004]

50. Park, E.; Cho, M.; Ki, C.S. Correct use of repeated measures analysis of variance. Korean J. Lab. Med.; 2009; 29, pp. 1-9. [DOI: https://dx.doi.org/10.3343/kjlm.2009.29.1.1] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19262072]

51. Kassambara, A. rstatix: Pipe-Friendly Framework for Basic Statistical Tests. R Package Version 0.7.0 2021. Available online: https://rpkgs.datanovia.com/rstatix/ (accessed on 15 September 2021).

52. Wickham, H.; Averick, M.; Bryan, J.; Chang, W.; McGowan, L.; François, R.; Grolemund, G.; Hayes, A.; Henry, L.; Hester, J. et al. Welcome to the Tidyverse. J. Open Source Softw.; 2019; 4, 1686. [DOI: https://dx.doi.org/10.21105/joss.01686]

53. Lenth, R.V. emmeans: Estimated Marginal Means, aka Least-Squares Means. R Package Version 1.7.1-1. 2021; Available online: https://CRAN.R-project.org/package=emmeans (accessed on 15 September 2021).

54. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2019; Available online: https://www.R-project.org/ (accessed on 1 March 2021).

55. Le, S.; Josse, J.; Husson, F. FactoMineR: An R Package for Multivariate Analysis. J. Stat. Softw.; 2008; 25, [DOI: https://dx.doi.org/10.18637/jss.v025.i01]

56. Kassambara, A.; Mundt, F. factoextra: Extract and Visualize the Results of Multivariate Data Analyses. R Package Version 1.0.7. 2020; Available online: https://rpkgs.datanovia.com/factoextra/index.html (accessed on 16 September 2021).

57. Anderson, M.J. A new method for non-parametric multivariate analysis of variance. Austral Ecol.; 2001; 26, pp. 32-46. [DOI: https://dx.doi.org/10.1111/J.1442-9993.2001.01070.PP.X]

58. Salazar, G. EcolUtils: Utilities for Community Ecology Analysis. R Package Version 0.1. 2022; Available online: https://rdrr.io/github/GuillemSalazar/EcolUtils/man/EcolUtils.html (accessed on 22 September 2021).

59. Oksanen, J.; Guillaume Blanchet, F.; Friendly, M.; Kindt, R.; Legendre, P.; McGlinn, D.; Minchin, P.R.; O’Hara, R.B.; Simpson, G.L.; Solymos, P. et al. vegan: Community Ecology Package. R Package Version 2.5-7. 2020; Available online: https://CRAN.R-project.org/package=vegan (accessed on 17 September 2021).

60. Chang, C.I. An information-theoretic approach to spectral variability, similarity, and discrimination for hyperspectral image analysis. IEEE Trans. Inf. Theory; 2000; 46, pp. 1927-1932. [DOI: https://dx.doi.org/10.1109/18.857802]

61. Agricultural Institute of Slovenia. SiaPy: A Tool for Efficient Processing of Spectral Images with Python. Python Package v1.0.0. 2022; Available online: https://zenodo.org/record/7409194#.Y5sgP0nMIuU (accessed on 1 September 2022).

62. Lê Cao, K.-A.; Boitard, S.; Besse, P. Sparse PLS discriminant analysis: Biologically relevant feature selection and graphical displays for multiclass problems. BMC Bioinform.; 2011; 12, 253. [DOI: https://dx.doi.org/10.1186/1471-2105-12-253]

63. Rohart, F.; Gautier, B.; Singh, A.; Lê Cao, K.A. mixOmics: An R package for ‘omics feature selection and multiple data integration. PLoS Comput. Biol.; 2017; 13, e1005752. [DOI: https://dx.doi.org/10.1371/journal.pcbi.1005752]

64. Karatzoglou, A.; Smola, A.; Hornik, K.; Zeileis, A. kernlab - An S4 Package for Kernel Methods in R. J. Stat. Softw.; 2004; 11, pp. 1-20. [DOI: https://dx.doi.org/10.18637/jss.v011.i09]

65. Kuhn, M. Package ‘caret’: Classification and Regression Training 2021. Available online: https://CRAN.R-project.org/package=caret (accessed on 24 January 2022).

66. Stevens, A.; Ramirez-Lopez, L. An Introduction to the Prospectr Package. R Package Vignette R Package Version 0.2.4. R Foundation for Statistical Computing: Vienna, Austria, 2022.

67. Wickham, H. ggplot2; Springer: New York, NY, USA, 2009; ISBN 978-0-387-98140-6

68. Microsoft, C.; Weston, S. doParallel: Foreach Parallel Adaptor for the “Parallel” Package. R Package Version 1.0.17. 2022; Available online: https://rdocumentation.org/packages/doParallel/versions/1.0.17 (accessed on 24 January 2022).

69. Brereton, R.G.; Lloyd, G.R. Partial least squares discriminant analysis: Taking the magic away. J. Chemom.; 2014; 28, pp. 213-225. [DOI: https://dx.doi.org/10.1002/cem.2609]

70. Workman, J.; Weyer, L.J. Practical Guide and Spectral Atlas for Interpretative Near-Infrared Spectroscopy; CRC Press: New York, NY, USA, 2012; ISBN 9781439875254

71. Abd-Elgawad, M.M.M.; Kerlan, M.-C.; Molinari, S.; Abd-El-Kareem, F.; Kabeil, S.S.A.; Mohamad, M.M.; El-Nagdi, W.A. Histopathological Changes and Enzymatic Activities Induced by Meloidogyne incognita on Resistant and Susceptible Potato. Int. J. Phytopathol.; 2012; 1, pp. 62-72. [DOI: https://dx.doi.org/10.33687/phytopath.001.01.0017]

72. Anami, S.; De Block, M.; MacHuka, J.; Van Lijsebettens, M. Molecular improvement of tropical maize for drought stress tolerance in Sub-Saharan Africa. CRC. Crit. Rev. Plant Sci.; 2009; 28, pp. 16-35. [DOI: https://dx.doi.org/10.1080/07352680802665305]

73. Çakir, R. Effect of water stress at different development stages on vegetative and reproductive growth of corn. F. Crop. Res.; 2004; 89, pp. 1-16. [DOI: https://dx.doi.org/10.1016/j.fcr.2004.01.005]

74. Álvarez-Iglesias, L.; Roza-Delgado, B.D.L.; Reigosa, M.J.; Revilla, P.; Pedrol, N. A simple, fast and accurate screening method to estimate maize (Zea mays L.) tolerance to drought at early stages. Maydica; 2017; 62, 34.

75. Ye, J.; Wang, S.; Deng, X.; Yin, L.; Xiong, B.; Wang, X. Melatonin increased maize (Zea mays L.) seedling drought tolerance by alleviating drought-induced photosynthetic inhibition and oxidative damage. Acta Physiol. Plant.; 2016; 38, 48. [DOI: https://dx.doi.org/10.1007/s11738-015-2045-y]

76. Taupin, P. Panorama des Une lutte de longue haleine. Perspect. Agric.; 2007; 339, pp. 27-31.

77. Staley, J.T.; Hodgson, C.J.; Mortimer, S.R.; Morecroft, M.D.; Masters, G.J.; Brown, V.K.; Taylor, M.E. Effects of summer rainfall manipulations on the abundance and vertical distribution of herbivorous soil macro-invertebrates. Eur. J. Soil Biol.; 2007; 43, pp. 189-198. [DOI: https://dx.doi.org/10.1016/j.ejsobi.2007.02.010]

78. Furlan, L. The biology of Agriotes ustulatus Schäller (Col., Elateridae). I. Adults and oviposition. J. Appl. Entomol.; 1996; 120, pp. 269-274. [DOI: https://dx.doi.org/10.1111/j.1439-0418.1996.tb01605.x]

79. Miles, H.W. Wireworms and Agriculture, With Special Reference To Agriotes obscurus L. Ann. Appl. Biol.; 1942; 29, pp. 176-180. [DOI: https://dx.doi.org/10.1111/j.1744-7348.1942.tb07585.x]

80. Mattson, W.J.; Haack, R.A. The Role of Drought in Outbreaks of Plant-Eating Insects. Bioscience; 1987; 37, pp. 110-118. [DOI: https://dx.doi.org/10.2307/1310365]

81. Copolovici, L.; Kännaste, A.; Remmel, T.; Niinemets, Ü. Volatile organic compound emissions from Alnus glutinosa under interacting drought and herbivory stresses. Environ. Exp. Bot.; 2014; 100, pp. 55-63. [DOI: https://dx.doi.org/10.1016/j.envexpbot.2013.12.011] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29367790]

82. Nguyen, D.; D’Agostino, N.; Tytgat, T.O.G.; Sun, P.; Lortzing, T.; Visser, E.J.W.; Cristescu, S.M.; Steppuhn, A.; Mariani, C.; van Dam, N.M. et al. Drought and flooding have distinct effects on herbivore-induced responses and resistance in Solanum dulcamara. Plant Cell Environ.; 2016; 39, pp. 1485-1499. [DOI: https://dx.doi.org/10.1111/pce.12708] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/26759219]

83. Bansal, S.; Hallsby, G.; Löfvenius, M.O.; Nilsson, M.C. Synergistic, additive and antagonistic impacts of drought and herbivory on Pinus sylvestris: Leaf, tissue and whole-plant responses and recovery. Tree Physiol.; 2013; 33, pp. 451-463. [DOI: https://dx.doi.org/10.1093/treephys/tpt019]

84. Bauerfeind, S.S.; Fischer, K. Testing the plant stress hypothesis: Stressed plants offer better food to an insect herbivore. Entomol. Exp. Appl.; 2013; 149, pp. 148-158. [DOI: https://dx.doi.org/10.1111/eea.12118]

85. Rosa, E.; Minard, G.; Lindholm, J.; Saastamoinen, M. Moderate plant water stress improves larval development, and impacts immunity and gut microbiota of a specialist herbivore. PLoS ONE; 2019; 14, e0204292. [DOI: https://dx.doi.org/10.1371/journal.pone.0204292]

86. Furlan, L.; Kreutzweiser, D. Alternatives to neonicotinoid insecticides for pest control: Case studies in agriculture and forestry. Environ. Sci. Pollut. Res.; 2015; 22, pp. 135-147. [DOI: https://dx.doi.org/10.1007/s11356-014-3628-7]

87. Traugott, M.; Benefer, C.M.; Blackshaw, R.P.; van Herk, W.G.; Vernon, R.S. Biology, Ecology, and Control of Elaterid Beetles in Agricultural Land. Annu. Rev. Entomol.; 2015; 60, pp. 313-334. [DOI: https://dx.doi.org/10.1146/annurev-ento-010814-021035]

88. Furlan, L.; Contiero, B.; Chiarini, F.; Benvegnù, I.; Tóth, M. The use of click beetle pheromone traps to optimize the risk assessment of wireworm (Coleoptera: Elateridae) maize damage. Sci. Rep.; 2020; 10, 8780. [DOI: https://dx.doi.org/10.1038/s41598-020-64347-z]