Using Ground and UAV Vegetation Indexes for the Selection of Fungal-Resistant Bread Wheat Varieties

Abstract

The productivity of wheat in the Mediterranean region is under threat due to climate-change-related environmental factors, including fungal diseases that can negatively impact wheat yield and quality. Wheat phenotyping tools utilizing affordable, high-throughput plant phenotyping (HTPP) techniques, such as aerial and ground RGB images and quick canopy and leaf sensors, can aid in assessing crop status and selecting tolerant wheat varieties. This study focused on the impact of fungal diseases on wheat productivity in the Mediterranean region, considering the need for a precise selection of tolerant wheat varieties. This research examined the use of affordable HTPP methods, including imaging and active multispectral sensors, to aid in crop management for improved wheat health and to support commercial field phenotyping programs. This study evaluated 40 advanced lines of bread wheat (Triticum aestivum L.) at five locations across northern Spain, comparing fungicide-treated and untreated blocks under fungal disease pressure (Septoria, brown rust, and stripe rust observed). Measurements of leaf-level pigments and canopy vegetation indexes were taken using portable sensors, field cameras, and imaging sensors mounted on unmanned aerial vehicles (UAVs). Significant differences were observed in Dualex flavonoids and the nitrogen balance index (NBI) between treatments in some locations (p < 0.001 between Elorz and Ejea). Measurements of canopy vigor and color at the plot level showed significant differences between treatments at all sites, highlighting indexes such as the green area (GA), crop senescence index (CSI), and triangular greenness index (TGI) in assessing the effects of fungicide treatments on different wheat cultivars. RGB vegetation indexes from the ground and UAV were highly correlated (r = 0.817 and r = 0.810 for TGI and NGRDI). However, the Greenseeker NDVI sensor was found to be more effective in estimating grain yield and protein content (R² = 0.61–0.7 and R² = 0.45–0.55, respectively) compared to the aerial AgroCam GEO NDVI (R² = 0.25–0.35 and R² = 0.12–0.21, respectively). We suggest as a practical consideration the use of the GreenSeeker NDVI as more user-friendly and less affected by external environmental factors. This study emphasized the throughput benefits of RGB UAV HTPPs with the high similarity between ground and aerial results and highlighted the potential for HTPPs in supporting the selection of fungal-disease-resistant bread wheat varieties.

1. Introduction

Crop productivity can be significantly impacted by climate change, reducing food production in many parts of the world. Climate change impacts are not limited to the effects of abiotic stresses but also affect outbreaks of biotic stresses. The projected rapid pace of climatic change without mitigation will have increasingly serious implications for the world economy and food security [1]. Crop scientists are faced with the dual challenge of increasing production while also maintaining long-term sustainability and food security by reducing the negative impacts of biotic and abiotic stressors, among others. As in many other Mediterranean countries, wheat is mostly grown under rainfed conditions in Spain. This is especially true in the north of Spain, where yields can be unpredictable due to erratic rainfall patterns; the area is characterized by cold winters and moderate precipitation throughout the year [2,3,4]. The main abiotic variables limiting bread wheat (Triticum aestivum L.) productivity in this region are thus drought and heat stress; however, fungal outbreaks (stem rust, Fusarium head blight, powdery mildew, etc.) are frequent, predicted to accompany climate change, and can have a significant effect on grain productivity and quality [5,6,7]. Fungal pathogens can spread quickly when temperature and humidity are favorable for disease development. In fact, temperature and relative humidity control the most crucial stages of the pathogen life cycle, including spore germination, infection, latent period, sporulation, spore survival, and host resistance, all of which influence epidemic onset [8]. Fungal pathogens reduce grain yield and quality by different mechanisms, including limiting root growth, reducing water and nutrient uptake, and inducing chlorosis and necrosis of photosynthetic tissues, which affects photosynthesis and causes widespread foliar senescence and poor grain filling [9,10,11].

The most economical and environmentally beneficial technique of disease management is the development of resistant cultivars. However, the creation of new cultivars that are more productive and offer greater yield stability across a range of environmental situations and disease pressures requires tremendous effort through breeding programs and phenotypic selection [12]. To that end it is necessary to explore the available genetic variability for developing modern high-yielding varieties that are disease-resistant and flexible to changing environmental conditions [13]. Furthermore, it will be imperative to find cost-effective and high-throughput technologies in order to better support and speed up these advancements in crop breeding [14].

The use of high-throughput phenotyping platforms (HTPPs), such as unmanned aerial vehicles (UAVs) equipped with high-resolution cameras, permits the rapid and nondestructive screening of crops to distinguish between abiotic and disease-related symptoms, such as vigor or crop color characteristics, at the crop canopy level [15]. In order to identify plant diseases in wheat and track their progression, a variety of imaging and spectroscopic approaches have been investigated. These techniques vary in their complexity and data requirements (e.g., [16,17,18]). For instance, at an early point in the development of winter stem rust, changes in spectral reflectance using in-field spectral images between healthy and diseased wheat plants allowed for the identification of infected plants [16].

Furthermore, vegetation indexes derived from conventional RGB (red, green, and blue) cameras have been demonstrated as robust indicators of biotic stress in wheat [19,20]. In fact, genotypic variability in the susceptibility to fungal diseases has been observed at the field level using conventional RGB images, where, based on the combination of RGB spectral bands, wheat affected by leaf rust and stripe rust were identified [15]. The findings of Heidarian and colleagues [15] further mentioned that two foliar fungal illnesses (stem rust and Fusarium head blight) altered the reflectance of wheat leaves between the green and red spectral channels. Other researchers have focused on the extraction of color parameters and the use of alternative color spaces for quantifying fungal infections [20]. An additional advantage of RGB indexes is their affordability and quality color calibration [19].

Among multispectral indexes, the normalized difference vegetation index (NDVI) has been to date one of the most commonly used multispectral near-infrared indexes when phenotyping resistance to fungal diseases [21,22,23], though other indexes have been found to have similar performance. Still, RGB sensors offer increased image resolution compared to other spectral sensors, which may be important to detect early symptoms of fungal stress [18,20,24], and often equal or surpass the capacity of NDVI for assessing fungal infection and estimating grain yield losses [20]. Besides the spectral nature of the sensors (RGB cameras versus multispectral field sensors and imagers), spatial resolution and the distance between camera and subject are relevant issues. This means that both ground-based cameras and sensors placed on a UAV are approaches amenable to high-throughput wheat phenotyping in multi-environment setups where several trials differing in growing conditions are assessed, though each may offer some advantages over the other [25].

Our study enters this context by comparing the effectiveness of RGB indexes acquired at both the ground level and from a UAV aerial platform. We compared their relative performance against a baseline consisting of measurements using an active NDVI canopy sensor at the ground level and final yield and protein content. We compared a large set of advanced (F8) lines of bread wheat growing in five different locations that varied in water availability and temperature, which may strongly affect the impact yield and fungal disease pressure. Moreover, we also compared the value of assessing photosynthetic and protective pigments at the leaf level using an advanced portable leaf-clip sensor. For each location we compared the same set of genotypes in trials untreated and treated with fungicide. Finally, genotypic differences in grain yield and susceptibly to fungal attacks were assessed by comparing relative differences between treated and untreated plots to vegetation indexes and yield and protein content at harvest.

2. Materials and Methods

2.1. Study Sites

The research was carried out during the 2020–2021 growing season in five experimental field trials located across central–NE Spain that the Limagrain Seed Company uses for its annual evaluation of accessions: Tordómar in the municipality of Tordómar, 42°2′9.76″ N (Burgos, Spain), Quintanilla de San Garcia in the municipality of Briviesca, 42°33′51.39″ N (Burgos, Spain), Elorz in the municipality of Noáin, 42°43′8.12″ N (Navarra, Spain), Sos del Rey Católico in the municipality of Sos del Rey Católico, 42°32′39.66″ N (Zaragoza, Spain), and Ejea de los Caballeros in the same municipality at 42°6′53.67″ N (Zaragoza, Spain). The geographic location and province of each field site are shown in Figure 1.

2.2. Climate Conditions of Experimental Sites

The five experimental sites were characterized by generally continental Mediterranean climate conditions, with a large variability in ranges of temperature and precipitation. Climatic data from 2021 were recorded through the Spanish platform SIAR (Servicio de Información Agroclimática para el Regadio, www.siar.es (accessed on 12 September 2022) from meteorological stations close to the field sites. All the five field sites exhibited a strong trend from wet, cool to dry, warmer, moving from the more western (especially Tordómar, Briviesca and Elorz) to the eastern (Sos del Rey Católico and Ejea de los Caballeros) study sites. The data of climate condition variation for temperature and precipitation during the field season in 2021 at the five experimental sites are presented in Figure 2 and in Supplementary material Table S1.

2.2.1. Tordómar

In Tordómar, the wintertime precipitation is noticeably heavier than the summertime precipitation, though summer precipitation events can be frequent. Rainfall during field season 2021 was 437 mm. April was the wettest month of the year after February in 2021, while March was relatively less wet. Less than 2 mm of precipitation was recorded as the least amount in July with minimal precipitation in August as well.

2.2.2. Briviesca

The Briviesca site (near the town of Quintanilla de San Garcia) generally receives a considerable amount of rain all year long. The average rainfall during the filed season is 768 mm. Even in the driest months, there can still appear a sizable amount of rainfall. Only the two driest months, July and August, recorded less than 6 mm of precipitation each. In 2021, precipitation increased from March until June, then dropped off sharply in July.

2.2.3. Elorz

It can rain heavily during most of the year in Noáin (Elorz), with 714 mm of rainfall on average each year. July and August were the driest months of the year in the field season 2021, as usual; however, March and May saw relatively low precipitation, while June experienced some heavy rains.

2.2.4. Sos Del Rey Católico

The average rainfall of the field season in Sos del Rey Católico was 532 mm. The driest months were July and August, with less than 2 mm of precipitation. Rainfall in Sos del Rey Católico was moderate for March, May, and June in 2021.

2.2.5. Ejea de Los Caballeros

Low rainfall is a year-round occurrence in Ejea de los Caballeros, with a total of 350 mm of rainfall on average during the field season. These three months were the driest in 2021: May, June, and July. The smallest amount of precipitation, less than 1 mm, was recorded in July.

2.3. Experimental Trial Design

Forty advanced (F8) winter wheat (Triticum aestivum. L.) accessions were assessed at the five experimental sites. A randomized complete block design with 3 untreated replicates and 1 treated replicate of plots made up the experimental setup for a total of 160 separate 8.5 m² plots at each site. Due to their geographical and climatological differences, different sowing dates were followed for each site as detailed at the top of Figure 3. The control trial did not receive any antifungal treatments, to determine varietal susceptibility to fungal attack, and the treated trials received fungicide applications by spraying (Prosaro^®, prothioconazole and tebuconazole, Bayer Crop Science, Leverkusen, Germany) three times at tillering, flag-leaf emergence, and heading/flowering. Fungal diseases identified as present at the trials during the 2021 field season included Septoria (Mycosphaerella graminicola and Stagonospora nodorum), brown rust (Puccinia triticina), and stripe rust (Puccinia striiformis). The insecticide Decis (Bayer Crop Science, Germany) was applied to all trial sites and blocks at the same time as the fungicide applications. Ejea de los Caballeros additionally received 3 supplemental irrigations by inundation, each 40–50 mm, and Sos del Rey Católico received sprinkler supplemental irrigation 5 times, 18 mm each time. Meanwhile, the other three sites at Tordómar, Briviesca, and Elorz were rainfed only.

During the five field samplings, the following physiological stages match approximately the research visit timing: 1. March (stem elongation), 2. April (anthesis), 3. mid-May (heading), 4. late-May (grain filling), and 5. June (maturity), as detailed in Figure 3. The warmer sites, Ejea de los Caballeros followed by Sos del Rey Católico and Elorz, developed faster than Tordómar and Briviesca, due to their higher temperatures. As shown in the summary flowchart in Figure 3, the portable NDVI Greenseeker, RGB ground camera, and UAV RGB and NDVI cameras were employed during all five visits. Dualex leaf sensor measurements were conducted twice during the third and fourth visits, and the isotope analysis grain samples were gathered during the fifth field visit.

2.4. Imaging and Sensor Field Measurements

2.4.1. Dualex

The Dualex is an optical portable leaf-clip sensor (Force-A, Orsay, France) for determining flavonoids, anthocyanins, and chlorophyll levels in leaves that was used for the additional assessment of leaf chlorophyll (Chl), epidermal flavonoids (Flav), and nitrogen balance of the study sites during the 3rd and 4th visits (Figure 3). The Dualex offers the advantages of high-throughput, nondestructive, and real-time measurements to be made of these relevant physiological crop components and is factory calibrated to provide leaf Chl content in µg cm⁻² units with a linear response function and a documented error of 2.4 and 3.4% compared to chemical composition analyses and for Chl and Flav, respectively [26]. Furthermore, the Dualex determines the nitrogen balance index (NBI), which is the chlorophyll/flavonoids ratio, and informs on nitrogen and carbon allocation [27].

The Dualex manages high-precision and linear response functions using a unique combination of a UV excitation beam at 357 nm, which corresponds to the maximum absorption for flavonoids, with a green LED for anthocyanins, a red reference beam at 650 nm, which corresponds to chlorophyll absorption, and two other near-infrared reference wavelengths (710 nm and 850 nm) [28,29]. The Dualex displayed high accuracy in Chl content assessment when specifically tested on wheat in a study that assessed fluorescence techniques for estimations of phenolic compounds positively [28]. As a field device, it adjusts the level of fluorescence caused by a reference red light to the level of fluorescence caused by UV light to provide quick measurements from attached leaves in field conditions; furthermore, it is water-resistant, stores data digitally, and has an integrated GPS sensor for both location and accurate time stamping. It has furthermore been used to estimate soil nitrogen and wheat productivity estimations [30].

The measurements with the Dualex sensor were performed on the leaf adaxial side and repeated five times for each plot in order to sample a representative measurement of the various pigments on both sides of the plot. In each case, the last fully expanded leaf was measured.

2.4.2. Trimble GreenSeeker

The Trimble Greenseeker (Greenseeker, Trimble, Sunnyvale, CA, USA) uses an active light source to measure the normalized difference vegetation index (NDVI) at the canopy level, which is as an indicator of crop vigor understood as a combination of photosynthetic capacity and biomass. NDVI measures the difference between near-infrared (which vegetation biomass strongly reflects) and red light to quantify vegetation (which photosynthetic vegetation absorbs). Measurements were performed with the portable crop canopy sensor measured at ground level, with the sensor placed over the middle of each wheat plot at a constant height of 0.5 m above the canopy [19]. The sensor sends out brief bursts of red and infrared light (656 nm and 774 nm, respectively) and then quantifies light reflectance for the plant for each wavelength. For as long as the trigger is pressed, it continues to sample the scanned area to provide the average value measured in terms of an NDVI index reading (normalized index between 0 and 1) for the whole plot.

2.4.3. RGB Images and Derived Vegetation Indexes

A selection of different vegetation indexes was extracted from the RGB photos acquired during each of the visits. The ground images were taken in the middle of each plot, with the camera placed zenithally above the canopy at a height of 80 cm. The camera utilized in this operation was a Panasonic Lumix DMC GX7 with 16 MP (Panasonic, Osaka, Japan). The CerealScanner plugin (https://gitlab.com/sckefauver/cerealscanner (accessed on 25 September 2021) [31] was then used to analyze the images implementing Breedpix 2.0 (https://bio-protocol.org/e1488, (accessed on 25 September 2021) IRTA, Lleida, Spain), which creates RGB vegetation indexes using RGB and alternative color space attributes, such as the hue, saturation, and intensity (HSI) model, which is included in this package, to quantify plant properties of interest. The percentage of pixels classed as green or very green were used to calculate indexes such as GA (green area) and GGA (greener green area). The percentage of pixels in photos with a hue range of 60 to 180, including yellow to bluish green, was used to determine GA. GGA, on the other hand, has a narrower range, ranging from 80 to 180, which means it excludes yellowish-green tones. The crop senescence index (CSI) was created by combining the two earlier indexes [32], which use a scaled ratio of yellow and green pixels to determine the number of senescent plants.

Other indexes were calculated with the digital values of the red, green, and blue bands obtained from the RGB images directly. Thus, the triangular greenness index (TGI) is an index that estimates chlorophyll content based on the area of a triangle with three points corresponding to the red, green, and blue bands following

Table 1. Select vegetation indexes derived from the RGB cameras.

Index	Formula
Green Area (GA)	60 < Hue < 180 [35]
Greener Area (GGA)	80 < Hue < 180 [35]
Crop Senescence Index (CSI)	(GA-GGA)/GA [36]
Normalized Green–Red Difference Index (NGRDI)	(R550 − R670)/(R550 + R670) [37]
Triangular Greenness Index (TGI)	−0.5[190(R670 − R550) − 120(R670 − R480)] [38]

. The normalized green–red difference index (NGRDI) is similar to the NDVI, but instead of red and near-infrared (NIR) bands, it employs green and red [33]. Meanwhile, two vegetation index adaptations, the NGRDIveg and the TGIveg, were filtered for vegetation pixels only (the selection of vegetated pixels was NGRDI > 0) to create a vegetation mask, thus removing the effects of the soil background. As a result, they recorded values higher than the whole-plot-level calculations, which included some soil background effects [34].

2.4.4. Aerial Images

A DJI Mavic Pro 2 UAV (DJI, Shenzhen, China) was used to capture aerial photographs. The Mavic 2 Pro UAV weighs 907 g and features an integrated, gimbal-stabilized, Hasselblad optics, 20 MP camera with a high-quality adjustable aperture f/2.8-f/11 lens (28 mm full-frame equivalent). The UAV was programmed to take pictures every 2 s from a height of 50 m for an overlap of 80% along and between flight paths guided by its internal GPS system. A separate GPS receiver was mounted on top of the UAV and connected to the AgroCam GEO NDVI (AgroCam, Debrecen, Hungary) camera mounted below the UAV. The AgroCam is a modified RGB camera with the near-infrared (NIR) blocking filter removed and replaced with a special “NDVI 7” filter that allows the camera to capture NIR–red–blue, thus enabling it to calculate the NDVI-blue variant of NDVI ((NIR − blue)/(NIR + blue)). The AgroCam is furthermore equipped with a special signal sensor that triggers its image-capture function whenever the Mavic 2 Pro camera captures an image, thus effectively providing separately georeferenced yet simultaneous image captures. Agisoft Metashape Professional software (Agisoft, St. Petersburg, Russia) was used to produce the orthomosaic images from each camera for the flights at each site for each visit (2 × 5 × 5 total flights). Then, using the MosaicTool plugin [39], regions of interest corresponding to each variety microplot image were segmented and exported using the image analysis platform FIJI (Fiji is Just ImageJ; http://fiji.sc/Fiji (accessed on 25 September 2021). The MosaicTool also includes integrated processing tools for calculating the different RGB vegetation indexes in the same way as the field RGB images. The AgroCam NDVI-blue was calculated using a custom code implemented in FIJI. An overview of the process is also detailed in Figure 3.

2.5. Additional Analyses

2.5.1. Determination of Stable Isotopes: δ¹³C and δ¹⁵N of the Grains

The determination of stable isotopes was conducted to further validate whether the plants suffered from water stress over the whole of the crop season; these traits can be seen as an integral measurement of water stress. From each plot, mature grains were collected at harvest from the two sites Ejea de los Caballeros and Elorz. These samples were dried at 60 °C for a minimum of 48 h and pulverized to a fine powder, from which 1 mg was enclosed in tin capsules and analyzed using an elemental analyzer (Flash 1112 EA; Thermo Finnigan, Schwerte, Germany) coupled with an isotope ratio mass spectrometer (Delta C IRMS, Thermo Finnigan) operating in continuous flow mode at the Scientific and Technical facilities of the University of Barcelona. Different secondary standards were utilized for carbon (IAEA CH₇, IAEA CH₆ and IAEA-600, and USGS 40) and nitrogen (IAEA-600, N₁, N₂, NO₃, urea, and acetanilide) isotope studies. Nitrogen content in grains was expressed in percentages (%), and the corresponding isotope compositions in parts per thousand (‰), with an analytical precision (standard deviation) of 0.2‰ for δ¹³C and 0.2‰ for δ¹⁵N and following:

δ¹³C (‰)/δ¹⁵N (‰) = [(R_sample/R_standard) − 1] × 1000,

(1)

where R_standard is the molar abundance ratio of the secondary standard calibrated against the primary standard, Pee Dee Belemnite in the case of carbon (δ¹³C) and N₂ from air in the case of nitrogen (δ¹⁵N) [40].

2.5.2. Grain Yield and Protein Content (CPRO)

At maturity, each plot was mechanically harvested, and the grain yield was obtained. After that, the grain yield (GY, T ha⁻¹) was determined considering the final total surface area of each plot. Furthermore, total protein content was also determined by the Limagrain company. For experimental purposes, a NIRS (near-infrared spectrometer) integrated into the combine harvester (BAURAL SP-2100, Blois, France) was used for the determination of protein content in situ during harvest, which, being incorporated into the same harvesting process, therefore provided increased efficiency for large trials [41].

2.6. Statistical Processing

To determine which indexes exhibited the best correlation and assess the complementarity between sensors, a comparison of aerial and ground data was conducted as part of our study, in addition to comparisons to yield and protein as our principal metrics of varietal performance. Statgraphics Centurion XVI (Statpoint Technologies, Warrenton, VA, USA) software was used to perform ANOVA analyses. For simple data analysis, such as mean and correlations between different indexes, we used Statgraphics Centurion XVI (Statpoint Technologies, Warrenton, VA, USA). Finally, the graphics were created by Sigmaplot (Systat Software Inc., Chicago, IL, USA).

3. Results

3.1. Isotopic Composition of Sites with Contrasting Water Conditions

Figure 4 shows that both δ¹³C and δ¹⁵N revealed a highly significant difference between the two sites (p < 0.001), with Elorz having nitrogen composition values compared to Ejea de los Caballeros and less negative value of carbon composition. This may be a general assessment of the difference in site conditions between the two most different sites in terms of water and nutrient status.

3.2. Comparison of Leaf Pigments between Treated and Untreated Plants

Differences between treated and untreated trial blocks were only significant for three leaf traits (flavonoids; the nitrogen balance index (NBI); and the ratio between chlorophyll (Chl) and flavonoids (Flav)) at two sites (Elorz and Ejea de los Caballeros). In addition, differences in Chl content were also evidenced at Ejea de los Caballeros (

Table 2. ANOVA analyses comparing the effects of the two treatments (Treated, T; and Untreated, U) for all sites during the third visit during the period of 10–14 May.

	Tordómar		Briviesca		Elorz		Sos		Ejea
	Mean	p-Value	Mean	p-Value	Mean	p-Value	Mean	p-Value	Mean	p-Value
Chl	T = 48.69	0.81	T = 35.90	0.08	T = 1.50	0.5	T = 34.92	0.97	T = 48.00	0.011 *
	U = 48.69		U = 35.20		U = 1.54		U = 34.94		U = 45.78
Flav	T = 1.54	0.83	T = 1.553	0.19	T = 1.510	0.002 **	T = 1.46	0.2	T = 1.58	0.005 *
	U= 1.55		U = 1.552		U = 1.592		U = 1.44		U = 1.62
Anth	T = 0.056	0.94	T = 0.049	0.8	T = 0.023	0.28	T = 0.026	0.93	T = 0.023	0.088
	U = 0.054		U = 0.047		U = 0.021		U = 0.026		U = 0.021
NBI	T = 31.65	0.75	T = 23.38	0.14	T = 35.69	0.006 **	T = 24.29	0.65	T = 30.4	0.000 ***
	U = 31.85		U = 22.82		U = 34.56		U = 24.64		U = 28.08

* p < 0.05, ** p < 0.01, *** p < 0.001. Chl, chlorophyll leaf content; Flav, flavonoid leaf content; Anth, anthocyanin leaf content; and NBI, nitrogen balance index.

). The untreated plots reported values of Flav that were higher than the treated ones at Elorz ($\overline{\mathrm{x}}$ = 1.510 Treated vs. $\overline{\mathrm{x}}$ = 1.592 Untreated, p < 0.01) and Ejea de los Caballeros ($\overline{\mathrm{x}}$ = 1.58 Treated vs. $\overline{\mathrm{x}}$ = 1.62 Untreated, p < 0.05), whereas for the NBI, the treated wheat plots had significantly higher values than the untreated plots at both sites (p < 0.01 for Elorz and p < 0.001 for Ejea de los Caballeros). In the case of Ejea de los Caballeros, Chl was significantly higher in the treated than the untreated genotypes. No significant differences were found for the fourth visit later that same month. Due to differences in site location and planting times, no comparisons were made directly between sites, only between treated and untreated blocks at each site.

3.3. Comparison of Treated and Untreated Trials Using Different RGB Ground Vegetation Indexes for All Experimental Sites and Visits

The evolution of the vegetation indexes TGIveg, GA, and CSI for each of the treatments across the different visits is shown in Figure 5, Figure 6 and Figure 7, respectively. Differences in vegetation indexes at early stages may inform about differences in emergence and fractional vegetation cover, while after-anthesis vegetation indexes might be saturated by high biomass at higher quality sites, and at grain filling and onward the indexes might indicate water stress or early onset of senescence (whether due to fungal infection or not). In the case of TGIveg (Figure 5), the differences in Tordómar were only observed for the first two visits (9 March and 20 April, p < 0.001), and the same occurred for Ejea de los Caballeros (12 March, stem elongation and 14 May, heading, p < 0.05). Meanwhile, there were no differences between treatments for Elorz at any of the visits. On 21 April and 12 May, in Briviesca, a significant difference (p < 0.001) was discovered between untreated and treated plants. In the case of Sos del Rey Católico, TGIveg values of treated were significantly higher than those of untreated on 11 March, 22 April, and 13 May (p < 0.05).

Figure 6 shows that differences in GA between treatments existed within each site for at least for one of the visits (e.g., Briviesca), while for two other sites, differences were recorded for two visits (e.g., Tordómar and Elorz), and at one site (Sos del Rey Católico) differences were recorded for three visits; finally, at Ejea de los Caballeros, differences were recorded for four separate visits. In almost all cases, GA was significantly higher in treated versus untreated plants, and for all the sites, differences were significant during the last visit (p < 0.001). CSI (Figure 7) was the index that performed the best, with Tordómar, Briviesca, and Ejea de los Caballeros showing significant differences for all the visits, Elorz in four of the five visits, and Sos del Rey Católico in two of the five (p < 0.001). For all the sites, the two last visits showed significant differences. In most cases, treated plants exhibited lower values than untreated ones. The three vegetation indexes detailed above (TGIveg, GA, and CSI) in Figure 5, Figure 6 and Figure 7 exhibited the greatest number of significant differences between treated and untreated wheat, considering all of the sites and field visit dates of the study.

3.4. Relationships between Ground Vegetation Indexes with Grain Yield and Protein

To see how the different vegetation indexes can be used to estimate yield, the relationship across treated and untreated genotypes between NDVI and both grain yield (Figure 8a, upper part) and grain protein content (Figure 8b, lower part) were assessed. There was variability in the strength of the relationship among different sites and across the various dates inside the same site. Treated trials exhibited higher correlations with grain yield and protein content than untreated ones, with Ejea de los Caballeros and Tordómar exhibiting the highest correlations, and Briviesca exhibiting the lowest correlation (Figure 8b, lower part).

Subsequently, Figure 9 shows the relationships between the ground-level RGB index NGRDI against grain yield (Figure 9a, upper part) and protein (Figure 9b, lower part) within sites and across visits. The best correlations with grain yield and protein were found in Ejea de los Caballeros, while the lowest correlations were found in Briviesca. Moreover, the correlations increased during consecutive visits (from tillering to booting and anthesis), while correlations declined during the last visit, coinciding with grain filling and the onset of crop senescence. Among the vegetation indexes, NDVI and NGRDI are featured here for the most consistently high correlations with grain yield across sites and dates.

3.5. Comparisons of Aerial and Ground RGB Images and Analysis of Aerial Images for the Assessment of Treated and Untreated Trials at the Five Different Study Sites

According to

Table 3. Comparison between vegetation indexes extracted at the ground level, using RGB images taken by camera Panasonic Lumix GX7, and from the aerial level, using the UAV DJI Mavic 2 Pro during the last visit (June) to the different sites of Limagrain company: Ejea de los Caballeros June 7, Sos del Rey Católico June 7, Briviesca June 8, Elorz June 8, and Tordómar June 10.

	Aerial	GA	GGA	CSI	NGRDI	NGRDIveg	TGI	TGIveg
Ground
Tordómar		0.599	0.508	0.329	0.793	0.639	0.741	0.627
Briviesca		0.568	0.543	0.355	0.810	0.699	0.677	0.583
Elorz		0.048	0.578	0.552	0.646	0.706	0.619	0.639
Sos		−0.077	0.313	0.250	0.773	0.546	0.800	0.349
Ejea		0.247	0.139	0.041	0.559	0.167	0.817	0.256

, most of the sites observing the link between ground and aerial level had the highest level of compatibility with NGRDI indexes; Briviesca registered the highest correlation (r) of this value (r = 0.810). Incorporating the mask (i.e., NGRDIveg, including plot values only where NGRDI > 0) only weakened the relationship. In the same context, TGI also performed strongly, with a strong correlation registered at all the five locations, with Ejea de los Caballeros exhibiting the highest aerial–ground correlation (r = 0.817). These results further indicate the promise of some of the RGB indexes, and, in particular, TGI and NGRDI (r = 0.817 and r = 0.800, respectively) for their use from the aerial platform, where higher throughput can be managed for large trials.

This is in addition to the strong performance of these band math RGB vegetation indexes in separating treatments, as shown for TGI in Figure 5, while the high correlations between NGRDI and grain yield and protein are demonstrated in Figure 9. In contrast, GA and CSI only exhibited strong ground–aerial correlations at some sites, being potentially more susceptible to the related changes in spatial resolution when capturing the image from farther away.

In Figure 10, based on the best performing vegetation indexes during each field visit, we present the utility of different UAV-acquired RGB and NDVI vegetation indexes for separating treated and untreated plots captured using the DJI Mavic 2 Pro Hasselblad 20 MP RGB camera and the AgroCam NDVI 12 MP modified camera showing NDVI-blue. TGI showed significant differences between treatments for Briviesca and Elorz in April (p < 0.001 and p < 0.05, respectively), while the TGIveg detected significant differences between treated and untreated only for Tordómar (p < 0.001). The CSI demonstrated differences between treatments in May for all sites except for Tordómar, one of the coldest sites, which also had the lowest levels of senescence. GA detected differences for all sites in June (p < 0.001). Meanwhile, the AgroCam NDVI-blue only detected differences in Sos del Rey Católico and Ejea de los Caballeros in May (p < 0.05).

3.6. Evaluation of Ground- and Aerial-Acquired Vegetation Indexes Assessing Genotypic Variability in the Response to Disease Treatments

We compared the relationships across genotypes through regressions between the ratios of treated versus untreated vegetation indexes (VIs) (NDVI and NGRDI at the ground and aerial levels (R VI = treated/untreated VI values)) against the ratios of treated versus untreated grain yield (RGY = GY treated/GY untreated).

When NDVI was assessed at ground level (i.e., with the GreenSeeker), most of the correlations shown against GY (Supplemental Table S2) were significant. Nevertheless, we can observe a variation in the value of correlation across genotypes between the ratio of treated versus untreated Greenseeker NDVI (R NDVI ground) and the ratio of treated/untreated grain yield (RGY = GY treated/GY untreated). Elorz registered the highest value during the visits on 22 April (R² = 0.700, p < 0.001). These results demonstrated a strong correlation between the relative sensibility of the lines in terms of leaf color and the calculated sensitivity to grain yield. In Supplemental Table S3, we detailed that the correlation value across genotypes recorded at the majority of the sites was low for the ratio of treated to untreated aerial NDVI (R NDVI aerial) and the ratio of treated to untreated grain yield (RGY = GY treated/GY untreated), which contrasts with results at the ground level. The highest value of determination coefficient between aerial NDVI and grain yield was recorded at Tordómar on 12 May (R² = 0.231, p < 0.001). Meanwhile, the coefficient is higher in the NDVI at ground level, for example, Elorz on 22 April (R² = 0.700, p < 0.001).

Furthermore, as detailed in Supplemental Table S4, the link between the ground NGRDI ratio (R NDVI ground) and the grain yield ratio (RGY = GY treated/GY untreated) varied across the visits, but except for Sos del Rey Católico, the highest correlation was achieved in the last visits. The greatest value was achieved on 11 March in Elorz (R² = 0.956, p < 0.001). The association between the ratios of NGRDI at the aerial level (R NGRDI aerial) and grain yield ratio (RGY = GY treated/GY untreated) was quite good and comparable (or even somewhat better) than the correlations observed between the two ratios at ground level. In this case Elorz on 27 May exhibited the highest determination coefficient value (R² = 0.94, p < 0.05), further detailed in Supplemental Table S5.

3.7. Combinations of NDVI, Grain Yield and Treatments for Guiding the Selection of Genotypes

Some of the relationships at the five sites between the ratios of treated versus untreated NDVI at ground level and the ratios of treated versus untreated grain yield is depicted in Figure 11. The strongest correlation seen in Figure 11 was on 21 April at Briviesca (R² = 0.65, p < 0.001), demonstrating the strength of the correlation between the ratio of NDVI from the ground and the ratio of grain yield. This makes it clear that the treatment played a part in determining the index value. Yet, the ratio of the treatments for the same index (NDVI and grain yield) were well-correlated.

4. Discussion

Bread wheat is central to the nourishment of the world population and is under threat from a number of sources, including biotic and abiotic stress [3,4]. Our study investigated the dynamics of crop growth and fungal resistance in advanced (F8) bread lines growing at five different sites across northern Spain with a range of different local climates. We showed the differences in temperature and precipitation patterns across the geographic distribution of the selected sites in Figure 1 and Figure 2. Sowing dates also varied between sites (Figure 3), and site visits were adjusted slightly accordingly (Figure 3). We furthermore detailed other relevant differences with the δ¹³C and δ¹⁵N stable isotope comparison between two of the most extreme sites, Elorz and Ejea de los Caballeros, in Figure 4, showing significant differences in site water and nutrient conditions. In addition, the vegetation indexes shown in Figure 5, Figure 6 and Figure 7 can help to identify the effect of the treatment.

Our main objective was to better comprehend fungal resistance across different environmental conditions, as well as which cost-effective, affordable remote sensing phenotyping tools work best in different growing conditions and crop growth stages. The evaluation of fungicide treatment efficacy and varietal resistance in protecting bread wheat was compared using ground- and aerial-level remote sensing tools. Further insights were explored regarding how to best leverage this information for optimizing varietal selection. To better structure the different times, locations, and scales of observation, we have organized our discussion around four specific questions related to the stated objectives:

• What were the differences in crop status between the different sites on dates of visit?
• Are RGB and NDVI vegetation indexes able to detect the treatments and fungal tolerance?
• When is the best phenological time to assess vegetation indexes in order to screen for fungal resistance?
• Can proximal imaging or aerial imaging be used accurately to select cultivars with a greater fungal resistance?

4.1. What Were the Differences in Crop Status between the Different Sites on Dates of Visit?

At the leaf level, some pigments, such as flavonoids, chlorophyll, anthocyanins, and the NBI ratio detected significant differences between the two treatments (treated/untreated) at various sites. Particularly, the differences in Elorz and Ejea de los Caballeros are obvious when derived from the various leaf-level pigment measurements. Both flavonoid and NBI serve as plant stress signals, showing the importance of fungicide treatment in the protection of wheat against fungal stressors. In treated wheat, the Dualex sensor recorded higher NBI values than in untreated wheats in both Elorz (p < 0.01) and Ejea de los Caballeros (p < 0.001). NBI is a measure of how well-balanced leaf N is with other crucial macronutrients and, consequently, how well it can support plant functions, particularly photosynthesis. Thus, higher NBI supports higher yield. This is similar to the study conducted by Simón and coworkers [42], whose findings indicated that foliar fungal infections may significantly reduce wheat yield and quality. These biotic stresses could have various effects on crop development rates, altering grain carbohydrate buildup and nitrogen (N) dynamics. Another notable measurement made by the Dualex leaf sensor is the flavonoid concentration, which showed a significant difference between treated and untreated plants at the two study locations (for Elorz and Ejea, p < 0.01 and p < 0.5, respectively). The greater levels of flavonoids seen in the bread wheat that was not treated in our study were likely related to stress signaling and indicated that key photochemical processes were being compromised. Fungicides may be able to mitigate these effects by limiting fungal infection. Our research is in line with Surovy et al.’s findings [43], which claim that wheat infected by the blast fungus Magnapoorpthe oryzae pathotype Triticum reacts at the beginning of the attack by an increase in flavonoid concentration, followed later by a decrease in flavonoids, as this pigment is involved in the plant reaction against fungal attacks and the access of the pathogen to the cell.

The analysis of the carbon and nitrogen stable isotope composition of two sites from the different Limagrain trials showed significant differences between the two sites measured (p < 0.001). Stable carbon and nitrogen isotopes as well as the ratio of carbon and nitrogen in biomass are general indicators for plant water conditions and nutritional status. While stable carbon isotope ratios δ¹³C (‰) immediately reflect the plant’s water state, stable nitrogen isotope ratios δ¹⁵N are more instructive of nutritional status and specific nutrient consumption [40]. Nitrogen composition δ¹⁵N (‰) can be used to measure this reduction in N availability and utilization as the water supply declines. For instance, a more negative result for δ¹³C indicates that the plant is growing under better water conditions. The fact that Elorz had less negative value in δ¹³C (p < 0.001) and higher value of δ¹⁵N (p < 0.001), which significantly exceeded that of Ejea de los Caballeros, indicated that wheat in Ejea benefited from better water status, but the yield in Elorz benefited from a better uptake of nitrogen from the soil. Consequently, the yield in Elorz was higher than in Ejea de los Caballeros. Rezzouk et al. [44] concluded for durum wheat that a greater δ¹⁵N resulted in more nitrogen uptake and, thus, more effective photosynthesis and final production. This can be explained by the heavy precipitation in Ejea de los Caballeros during the month of April (75 mm), in contrast to Elorz, which received only 50 mm; this can affect the speed and the intensity of the disease invasion. In addition, Ejea de los Caballeros received three irrigations by inundation, each one about 40–50 mm. In the same sense, a more negative δ¹³C value, especially in mature grains, implies that the crop received more water, and therefore produced more grain yield [45].

The experimental site distribution map reveals that the trials were spread apart from one another, creating a variety of climatic conditions between them, which is further accentuated by the topography of the region. Because the field season runs from sowing in October or November to heading and maturity from the months of March to June, these characteristics have the potential to influence growth patterns, senescence, or fungal developments. We observed that some locations, such as Briviesca, were characterized by regular precipitation and high temperatures, whereas other locations were characterized by inconsistent precipitation and high temperatures. For RGB vegetation indexes assessed from the ground, we noted that bread wheat in Briviesca registered better performance than Ejea de los Caballeros, which was characterized by greater precipitation during the month of April, which, given the mild temperature, may be the best moment for the spread of the disease in the wheat crop. Differences between the sites were observed at different times following seasonal site differences. RGB vegetation indexes, such as TGIveg and GA, were higher at Ejea de los Caballeros and Sos del Rey Católico in March early in the season, whereas the same chlorophyll- and vigor-related indexes were higher in Tordómar and Briviesca during the visits in May and June. This may indicate that the environment has negatively influenced crop status at Ejea de los Caballeros, with an earlier onset of crop infection during April when the crop experienced the highest precipitation out of all of the sites, combined with moderate temperatures. Considering the higher heat at Sos del Rey Católico and Ejea de los Caballeros, earlier onset of senescence was also expected, but differences between treated and untreated were also observed in June.

In this context, Mielniczuk and coworkers [46] found that cereals in the flowering and post-flowering stages were the most vulnerable to fungal infection, especially in warm, humid weather, and during this time of year there was a lot of persistent rainfall at Ejea de los Caballeros, more so than other sites. It is seen from the metrological data that Ejea de los Caballeros, especially, received heavy precipitation during the month of April. In addition, this site was provided supplemental irrigation by flooding, which would increase the humidity of the canopy, favoring the development and spread of the disease. Relative humidity can affect the rate of plant disease spread as spore germination and infection increases with sufficient moisture (RH ≥ 60%) [47]. Additionally, rainfall has been identified as a key requirement for the development of STB (Septoria leaf blotch), as it allows for the swelling of the fungal pycnidia and subsequently aids in the dispersal of spores across the upper leaves of wheat [48].

Indexes measured at both visits in April and May were close to the GA saturation limit, suggesting that the experimental plots achieved their maximum vegetation cover at Ejea de los Caballeros earlier in the season. On the other hand, the climatic circumstances at the locations at Tordómar, Briviesca, and Elorz were more favorable for both crop and fungus growth in general later on in the season, that is, in May. However, despite the possibility of greater yields under rainfed conditions, there was also more potential for late-season fungal infection compared to Sos del Rey Católico and Ejea de los Caballeros. However, Ejea de los Caballeros exhibited one of the greatest differences between treated and untreated for the two visits during the month of May (see GA and CSI in Figure 6 and Figure 7), possibly related to the heavy rains that it experienced in late April (Figure 2e), which may have done more to intensify the fungal infection for the remainder of the season. Thus, we suggest that the severity of fungal attack on wheat crops was determined by the combination of the environment, timing of the infection with the wheat cycle, and geographic location [49].

Comparing the meteorological data between the different sites, we note a difference in the month that recorded the highest precipitation. The majority of the sites were characterized by heavy precipitation early in the season (November), which presented a short time after the sowing dates. In contrast, Ejea de los Caballeros received the highest amount of precipitation in the month of April (at anthesis). This can help explain the effect of the precipitation date on the severity of the disease. While the four sites (Tordómar, Elorz, Briviesca, and Sos del Rey Católico) were less affected by the disease impacts and the risk of disease spread later in the season, they were more affected during the early phases of emergence; furthermore, the risk of wheat fungal disease infection in Ejea de los Caballeros was higher at the anthesis stage due to the facility of pathogen spread from the precipitation. These conditions create a humid microclimate with moderate temperature favorable to fungal invasion of wheat. A moderate temperature with humidity has been found to facilitate the penetration of Fusarium pseudograminearum and the infection of wheat in Australia at around 15 °C and reduces the wheat resistance to the disease [50]. The same was demonstrated in other studies, where they also indicated that the most pathogenic conditions for cereals were in warm temperatures and under high humidity [49,51].

4.2. Are RGB and NDVI Vegetation Indexes Able to Detect the Treatments and Fungal Tolerance?

RGB and NDVI vegetation indexes and their changes over the growing season can be used as tools to learn more about the state of a plant and, in this case, whether the fungicide treatment was beneficial towards the crop status for different varieties. Almost all of the different vegetation indexes differentiated fungicide-treated plants compared to untreated ones for at least one site or visit time, but TGIveg, GA, and CSI were the most consistent from the ground level. This can be explained by the vegetation index sensitivity at different plant growth stages (some indexes may lose sensitivity at higher biomass stages). Wheat goes through senescence more quickly at hotter sites, resulting in less chlorophyll and vigor, which lowers NDVI and NGRDI values later in the season. Additionally, fungal attacks act as a form of stress on plants by causing a drop in chlorophyll content and crop vigor. According to Dammer et al. [49], NGRDI can be a useful index for identifying plant stress in general. This index has a strong correlation with grain yield, which decreases under stressful conditions. Combining various vegetative indexes, such as the NDVI and NGRDI, can help to estimate the amount of wheat that will be lost due to fungal infection. The findings were in line with the study conducted by Vergara-Diaz et al. [20], which found that indexes retrieved from RGB images are extremely helpful for estimating wheat losses due to fungal (yellow rust) assault.

For aerial RGB indexes, particularly for GA, the differences were evident, showing that the value of treated wheat is generally higher than that of untreated wheat at most sites. This conclusion was also supported by research by Vergara-Diaz and coworkers [20], who discovered that GA had a strong correlation with grain yield at the anthesis stage, allowing for the detection of the yellow rust effect on durum wheat losses, keeping in mind that NDVI at the aerial level has also been proven effective as a technique to identify the impact of fungal attack on the crop of bread wheat later in the wheat cycle. Vergara-Diaz et al. [20] also demonstrated a strong correlation between ground-level NDVI and grain yield during the tillering and jointing stages, allowing for the quantification of the direct impact of yellow rust on durum wheat production and quality. In the same context, CSI extracted from aerial images also aligned with the results from the ground, where we note that untreated wheat exhibited higher and earlier senescence. Senescence is the natural final phase of the plant season and may be induced by a number of different factors, from heat and drought to other stressors such as fungal infection. The results showed that there were significant differences between treatments in leaf and canopy senescence parameters associated to stay-green expression, which as a varietal trait can lead to increased yields. Additionally, it should be noted that in the majority of sites, a significant difference between treated and untreated plants was observed in the CSI index near the end of the season, suggesting that fungal pressure played an important role in our trials. In the same context, we note that CSI values derived from RGB aerial images were similar to CSI values from ground, where the untreated wheat had higher values than treated, which means that untreated wheat suffered more from fungal aggression. CSI detection early in the season where the infection affects the absorption of the incident light by the chlorophyll further indicates fungal pressure. The same conclusion was found by Heidarian Dehkordi et al. [15], where these authors indicated that winter wheat infected by leaf rust showed an early decrease in the absorption of incident light by the plant with the slope flattened from the green to the red spectral bands.

The difference between the values of the various indexes measured by the various sensors demonstrates the ability of these sensors to detect wheat cultivar susceptibility to fungal attack under untreated conditions as well as the benefits reaped from the fungicide treatments, with some notable differences at the different locations. The reaction of the wheat varieties varied between treated and untreated, and the majority of the indexes favored a greater performance for the treated wheat, indicating the wheat that has not been treated with fungicide was more affected by the fungal impacts. According to another study, indexes such as the normalized difference index (NDI), green index (GI), green leaf index (GLI), and ground NDVI in winter wheat infected by leaf rust can all be effective tools to assess fungal foliar disease in wheat, where the indexes correlated well with the coefficient infection that measured the severity of infection in the crop [52]. Meanwhile, others used different spectral vegetation indexes, including NDVI, such as NBNDVI, PRI, GI, and RVSI, to quantify the severity of rust disease on the wheat crop developed using more advanced hyperspectral imaging sensors [22].

4.3. When Is the Best Time to Assess Vegetation Indexes in Order to Screen for Fungal Resistance?

The correct sequence of images obtained during the field season as well as the period of image processing can be confirmed by comparing vegetation indexes at the aerial and ground levels. Knowledge about plant status can be gleaned through the examination of indexes taken from ground images. Otherwise, depending on the hue shifts shown by the mosaic image, aerial pictures can give an overall view of the area and an estimate of its condition. The variations in vegetation indexes during the last stage (maturity) can serve to help us comprehend the phenological state of the plant. Thus, the decline in the values of the vegetation indexes at that time indicated senescence onset but could also have occurred earlier as a consequence of the fungal biotic stress. Additionally, these vegetation indexes can provide us with information regarding the state of the plants at the many experiment sites, as shown previously, where each site is characterized by a particular climate condition that may help to prevent or lessen the development and spread of a fungal infection.

The vegetation indexes performed well at detecting differences between treatments during April (booting–heading), which aligned with some prior studies which demonstrated that the RGB indexes such as GA and GGA were able to detect the effect of yellow rust on the losses of durum wheat early during the anthesis stage [20]. Post anthesis/early grain filling is the optimum stage to assess the vegetation indexes because this stage has been usually considered optimal for inferring the potential yield achievable by the crop; this is in spite of the fact many vegetation indexes may exhibit saturation patterns and the consequent loss of precision when evaluated between heading to anthesis. Moreover, depending on the kind of fungal disease, the pathogen can move through exposed florets and extruded anthers on wheat spikes during that time [53]. By providing the nutrients required for germination and penetration, extruded anthers can also trap fungi and aid in their growth. The low weight and deformed kernels, together with a decrease in kernel quality of diseased plants, are the consequence of disease impact on grain filling. Early detection of fungus attacks helps shield plants from the fungus’ damaging impacts. Whereas some research claims that anthesis is the crop’s most susceptible stage for Fusarium head blight, early detection of the disease can minimize losses to wheat crops [53]. Yet, according to one study, it was possible to detect Fusarium head blight early in the grain’s development and ripening stage by employing technologies such as multispectral imaging [54], to some extent comparable to the approximations of the Dualex and AgroCam NDVI sensors in our study.

4.4. Can Proximal Imaging or UAVs Be Used Accurately to Select Cultivars with a Greater Fungal Resistance?

While NDVI assessed at ground level correlated quite efficiently with grain yield, the NDVI aerial measurements showed overall poor association. This leads us to favor the Trimble GreenSeeker over the AgroCam for the NDVI measurements, the first being a more controlled and consistent active field sensor even if more time-consuming to use. These results disagree with those of Moazzam et al. [55], who used images obtained from an AgroCam camera to distinguish a sesame crop from weeds in order to maintain a precise application of herbicide in the field. When we compare the aerial and ground RGB photos, we can see that the findings obtained from both are similar in terms of the way that the treatments are separated and how grain yield is correlated; the capacity of RGB indexes to distinguish between the two treatments may be employed to forecast the disease’s impact on grain yield. For example, we demonstrated that the NGRDI index showed a greater association between ground and aerial levels of observation (r= 0.810). This suggests using the UAVs may offer benefits because aerial photographs can provide improved efficiency in information collection due to their ability to swiftly cover a large region. Therefore, when favorable conditions for flying as well as imaging are available, UAV platforms carrying RGB sensors or the AgroCam should have the advantage of allowing fast measurements of large crop trials [56].

The relatively comparable efficiency of ground- versus aerial-assessed RGB vegetation indexes to evaluate genotypic susceptibility to fungal disease may be based on two facts. First, the high resolution of RGB images makes them less prone to loss efficiency when the indexes are formulated from aerial platforms instead of from the ground. The second factor is the size of the canopy sampled. While the aerial images gather information about the whole plot, images acquired from the ground only cover a fraction (ca. 1 m²) of the entire plot. By contrast, the NDVI camera used from the drone had a lower resolution than the RGB cameras, while the assessment of NDVI from the ground implies covering the entire plot by foot in a more time-consuming manner.

The study of indexes generated from aerial and ground images can be useful when there is a high degree of correlation between the two levels of data collection. Nevertheless, we can now foresee any alterations to the plant or dangers to its development. Using RGB and multispectral imagery taken by a UAV, Mateen and Zhu [57] showed in their research that it is feasible to find weed patches in wheat. In contrast to other studies, which focused on a specific topic, Vergara et al. [20] conducted a study which served as the foundation for our research, tackling the estimation of grain yield losses in yellow-rusted durum wheat using digital and conventional tools. In the same sense, Francesconi et al. [58] carried out of a study that demonstrated the utility of integrating thermal data and RGB indexes to detect the Fusarium head blight more effectively in durum wheat. The combination of RGB images and temperature data at this level allowed for differentiation between the biotic stress induced by Fusarium and abiotic stress caused by dry weather, which they also mentioned as possibly a decrease in stomatal conductance in response to Fusarium infection. Moreover, it may also be worthwhile to further explore a wider range of vegetation indexes. Thus, the normalized difference red edge (NDRE), green normalized vegetation index (GNDVI), and NDVI extracted from five bands of a multispectral image are the most significant factors influencing grain yield and worked well in studies of wheat phenotyping, and gave better results than RGB images in some cases where they worked well early in the growth stage [59].

The strength of the correlations across genotypes between the ratios of treated versus untreated vegetation indexes (R VI) against the corresponding ratio for grain yield (R GY) provides an insightful perspective for selecting top-performing wheat phenotypes. By plotting such relationships (e.g., Figure 11), it can be easier to observe which specific genotypes performed best in terms of tolerance to fungicides and their implication in terms of grain yield. Thus, having the grain yield ratio in the horizontal axis and considering a grain yield ratio of 1 as a threshold, all the genotypes placed on the left side would be amenable for selection based on yield. From these genotypes and considering the vertical axis depicting the ratio of treated to untreated vegetation indexes, all the genotypes placed below 1 should be further selected based on their vegetation index performance.

Some practical applications can be used to reduce the severity of the fungal disease, such as the date of sowing. If the sowing date is such that flowering coincides with spore release, then more frequent and severe attacks are likely. Therefore, the choice of an early sowing date can make early maturing wheat cultivars tend to be less impacted via an avoidance strategy [60]. In addition, the density of the canopy can be an important factor affecting the invasion, as the high density can increase the humidity of the canopy, favoring spore germination and decreasing the efficacy of the fungicide treatment. A reasonable density should be applied at the time of sowing [61]. Furthermore, besides precipitation, the timing of irrigation of a field can influence its microclimate and may encourage the development of the pathogen. Regardless of whether the climate conditions are favorable for the disease in a given year, irrigation timing and application techniques can increase the frequency and severity of the disease [62]. Irrigation should be limited to supplemental irrigation in dry weather. The effects of crop rotation have been studied in detail [63]. They depend on the preceding crop, whether that crop is a potential host for the pathogens responsible for fungal disease, and the frequency of the crop concerned in the rotation. The shorter the rotation, the higher the frequency of disease. A large period of rotation reduces risk of infestation by the same pathogen [63]. The optimal fungicide application timing has also been found to be crucial for the protection of the wheat against fungal attack [64].

5. Conclusions

Combining the information from different growing sites, remote sensing approaches (NDVI versus RGB indexes), sensor placement (ground versus aerial), and times during crop cycle together with the effect of fungicide (Prosaro) applications have allowed us to conclude some general recommendations in terms of phenotyping, the likelihood of aggressiveness from fungus impacts, and the need for early fungal treatments. The differences between the several sites were apparent from the various indexes in our research, emphasizing the susceptibility of the wheat to fungal assault in different environmental conditions. In considering the environmental conditions of the Ejea de los Caballeros location, which was characterized by heavy precipitations and warm temperatures during the month of April, both of these factors were critical for the onset of infection and had a significant impact on spread of the fungus. The observed conditions may help to establish a strong fungal presence early in the season and require rapid treatment.

Additionally, HTPP sensor assessments early in the wheat cycle can help to protect the wheat crop from a serious fungus attack and ultimately enhance production in terms of quantity and quality. Leaf-pigment measurements proved the importance of fungicide treatment in order to protect wheat crops in terms of quantity and quality. Exploring the genetic variability of fungal resistance as the first line of defense against fungal attack needs to be explored further and can be supported with the use of HTPP technology, as fungicide applications may have additional side effects to the crop, besides the associated added costs. In summary, for now, it is clear that the application of fungicide can make a significant difference in how the wheat crop is protected from fungal disease. However, the environmental and health implications of spraying fungicides and the concern of customers make it more urgent to push forward to field crop breeding strategies with HTPPs.

For NDVI, our results suggest greater reliability from the use of the field sensor GreenSeeker NDVI as a tool due to its better overall performance in comparison to the UAV AgroCam NDVI, which is an RGB camera modified to measure NIR, green, and blue light to provide NDVI-blue, a modification of the original NDVI formulation. For the rest of the RGB indexes, our results indicate benefits from the UAV, which worked well, adapted well to weather conditions, and were quicker to capture images. However, this decision must be supported by early threat detection capacity in order to intervene with fungicide treatment that shields the wheat from attack and encourages the plant to grow in its preferred environment without experiencing any kind of stress. The combination of vegetation indexes (TGIveg, GA, CSI, or NGRDI from RGB images or NDVI) and production (grain yield and protein) data provides a useful approach to selecting the variety with the highest yield potential and that is more resistant to the fungal disease.

The generation of new wheat varieties using HTPPs may also provide a more cost-effective approach [14]. In the current study sites, in spite of the fact that treated plants performed far better than untreated ones, there were also cases when a lack of effect of fungicide or even a negative effect were evidenced, for example, when plotting the ratios for GY and the vegetation indexes between the treated and untreated across the set of advanced lines. When using fungicides to prevent fungus penetration in wheat plants, their application should be timed for efficacy and to minimize residue buildup. Applying fungicides quickly after a pathogen has been detected by a sensor can also be effective in preventing pathogen penetration inside plants, thereby ensuring greater production with improved grain quality. The benefits provided by the various sensors and their capacity to distinguish between treatments applied at various locations, each of which is defined by a unique climatic state, will make them effective tools for identifying pathogen-resistant varieties.