Next Article in Journal
Study of Various Mechanical Properties of Maize (Zea mays) as Influenced by Moisture Content
Previous Article in Journal
Research on Rapeseed Above-Ground Biomass Estimation Based on Spectral and LiDAR Data
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agronomy
Volume 14
Issue 8
10.3390/agronomy14081612
agronomy-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

The Use of Low-Cost Drone and Multi-Trait Analysis to Identify High Nitrogen Use Lines for Wheat Improvement

by
Liyan Shen
1,†,
Greg Deakin
2,†,
Guohui Ding
1,†,
Mujahid Ali
1,
Jie Dai
1,
Zhenjie Wen
1,
Felipe Pinheiro
2,
Ji Zhou
1,2,* and
Robert Jackson
2,*
1
College of Engineering, Plant Phenomics Research Centre, Academy for Advanced Interdisciplinary Studies, Nanjing Agricultural University, Nanjing 210095, China
2
Data Sciences Department, Crop Science Centre, National Institute of Agricultural Botany (NIAB), Cambridge CB3 0LE, UK
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this work as first co-authors.
Agronomy 2024, 14(8), 1612; https://doi.org/10.3390/agronomy14081612
Submission received: 25 June 2024 / Revised: 16 July 2024 / Accepted: 18 July 2024 / Published: 23 July 2024
(This article belongs to the Special Issue Study on Agronomic Measures and Physiological–Ecological Mechanisms for High-Yielding, High-Quality, and Resource-Efficient Utilization in Wheat Production)
Browse Figures
Versions Notes

Abstract

:
Breeding for nitrogen use efficiency (NUE) is becoming more important as global uncertainty makes the production and application of nitrogen (N) fertilizers more expensive and environmentally unfriendly. Despite this, most cereal breeding programs still use yield-related components as proxies for NUE, likely due to the prohibitive cost and time of collecting and analyzing samples through traditional lab-based methods. Drone-based NUE phenotyping provides a viable and scalable alternative as it is quicker, non-destructive, and consistent. Here, we present a study that utilized financially accessible cost-effective drones mounted with red-green-blue (RGB) image sensors coupled with the open-source AirMeasurer platform and advanced statistical analysis to exclude low-NUE lines in multi-seasonal field experiments. The method helped us to identify high N agronomic use efficiency lines but was less effective with a high N recovery efficiency line. We found that the drone-powered approach was very effective at 180 kg N per hectare (N180, an optimized N-rate) as it completely removed low-NUE wheat lines in the trial, which would facilitate breeders to quickly reduce the number of lines taken through multi-year breeding programs. Hence, this encouraging and scalable approach demonstrates its ability to conduct NUE phenotyping in wheat. With continuous refinements in field experiments, this method would be employable as an openly accessible platform to identify NUE lines at different N-rates for breeding and resource use efficiency studies in wheat.

1. Introduction

Wheat is one of the major calorie-providing crops globally and plays a vital role in feeding our ever-growing population [1]. The availability of soil-borne inorganic nitrogen (N) plays a key role in achieving maximum yield potential for wheat varieties [2]. However, with global instability limiting access to N fertilizers and concerns over the environmental impacts of its use, it is essential that nitrogen use efficiency (NUE) in wheat is increased for wheat improvement [3,4]. Plant breeding plays an essential role in this but the assessment of NUE in large-scale trials is impractical through traditional methods and, to date, has shown limited success in breeding programs [5,6]. It has been reported that only 33% of applied N is absorbed in crops, which is financially wasteful for farmers and has increased environmental damage [7,8]. As NUE characteristics are heavily correlated with varied wheat varieties [9,10], the interaction between variety and N uptake rate has been proved to be highly significant for N absorption and use efficiency, facilitating the development of strategies to screen high N use varieties that have a higher N absorption [11].
One of the main obstacles to accelerating NUE varietal screening in breeding is the method for assessing NUE features in wheat. Peron-Danaher et al. [12] noted traditional in-field approaches were time consuming, complicated, expensive, and non-scalable. Still, field-based trials have suggested many ways of measuring NUE-related phenotypic features that are affected by the volume of N applied to the crop [10]. For example, shoot biomass is known to correlate with N uptake in wheat [13]; wheat cultivars with more spikes, larger spikes, or more kernels have a higher potential for N uptake [14]. Above-ground biomass, vegetative organ biomass, canopy structure, spike number, grain number per spike, and grain yield are physical traits that are known to increase in response to N fertilizer applications and could, therefore, be used as proxies for assessing NUE-related features [13,14,15,16]. One of the main issues with the above methods in breeding programs, is they need to remove numerous plants from nursery plots with an already small number of plants. Additionally, not all the physical proxies are reliable across varieties and environments. For instance, shoot biomass was not correlated with N uptake at higher fertilizer application rates, whilst the relationship between yield components and NUE could vary based on different varieties [13,17].
Chemical analysis has also shown that above-ground N content, vegetative organ N accumulation, grain N content, above-ground N accumulation, N agronomic use efficiency (NAE), and N recovery efficiency (NRE) correlated with increased fertilizer applications [9,18]. The identification of all these biochemical traits can be quite complicated and, for many, requires the quantification of N content using destructive approaches, either by the chemical Kjeldahl method or the combustion-based Dumas method [19,20]. Both of these methods have drawbacks; the Kjeldahl method requires the use of corrosive and hazardous chemicals, whereas the Dumas method needs the production of NOx gas and the use of high temperatures. In addition, they require the removal of plants from field plots and thus compromise final yield assessment. More importantly, both analyses are performed at specific timepoints during the plant life cycle, which often misrepresent dynamics of N–plant interactions in the field [21].
Because N uptake and utilization are genetically complex and heavily influenced by environmental factors, it, therefore, requires a more dynamic approach to study this topic [21,22]. Due to the large scale of breeding populations and land required to grow and test them, a rapid selection for NUE-related traits would allow for a reduction in cost and human resources in breeding trials by reducing lines carried through in each breeding cycle [23]. Whilst the ideal would be to only select the best performing line, eliminating underperforming lines in NUE would also produce a sizable reduction in cost and selection efforts for breeders. Hence, a quick, non-destructive, and repeatable methodology is needed to phenotype traits that are highly correlated with NUE in wheat plots across the growing season, which can also enable the breeding for NUE in other crops [24].
Compared to ground-based measurements, drones (sometimes referred to as unmanned aerial vehicles, UAVs) represent an economical, non-destructive, and high-throughput tool for monitoring different crop species at key growth stages [25]. Most information collected by drone sensors can be converted into vegetation indices (VIs) through mathematical transformations of target reflectance bandwidths into a single figure to characterize a specific property from crop canopies [26,27]. To date, the adaptation of remote sensing techniques to large-scale plant breeding programs is still limited, with most of the focus still placed on replacing traditional ground-based measurements (e.g., height and canopy coverage) or yield prediction [28,29]. Along with the development of this field and associated analytic solutions based on computer vision and machine learning techniques, the potential for breeders to utilize the latest methodological advances to improve resource use efficiency or response to stress is dramatically increasing [26,30].
The use of drones in quantifying resource use in crops has already been reported. For example, Yang et al. [31] who monitored NUE in wheat using multispectral sensors, stated that the normalized difference red edge index (NDRE) was the most accurate and consistent VI for predicting NUE, particularly at the later developmental phases. Similarly, Liu et al. [32] used multispectral indices to create models for NUE prediction throughout a growing season, with modified normalized difference blue index (NDBI) being the most effective VI for NUE, particularly at the reproductive stage. Whilst both these studies showed the potential for sensing NUE in wheat, they both required relatively expensive multispectral drones with proprietary processing toolkits, which are still prohibitive to many breeders internationally. Other analytic platforms such as UAV-HTPPs [33], UAStools [34], AirMeasurer [25], and FIELDimageR for growth pattern assessment [35] are capable of processing time-series images acquired by drones mounted with regular cameras, converting images into phenotypic information in a few processing steps. For these methods, the field experiments were captured using predetermined flight plans, through which aerial images were processed to generate two-dimensional (2D) orthomosaics and three-dimensional (3D) point cloud datasets for plot segmentation, the analysis of VIs, and canopy-level structural information. For example, two vegetation indices produced by the AirMeasurer platform were the visible atmospherically resistance index (VARI), a method of visualizing vegetation using portions of the visible spectrum whilst accounting for cross site differences in illumination and atmospheric effects [36], and the normalized difference yellowness index (NDYI), developed to identify canola flowering intensity [37] and correlated to yield and NUE in wheat [32].
Building upon our previous work studying N-response traits after N fertilization in wheat, this study aims to demonstrate how drone-based remote sensing could be employed to identify wheat germplasm with improved N use efficiency. In this paper we demonstrated that low-cost drone-acquired visible spectrum could be used to derive spectral traits to identify low-, medium-, and high-NUE of lines on a large scale and to a level of accuracy comparable with lab-based analysis. In addition, we showed that these measures performed consistently throughout multiple seasons, which would allow for the dynamic analysis of NUE in wheat in future trials. The above attempts opened up the possibility of dissecting NUE development under field conditions, on a scale not commonly seen, which allowed us to classify NUE lines to estimate how effectively plants utilize N for growth and reproduction. This encompassed both nitrogen agronomic efficiency (NAE) and N recovery efficiency (NRE), both of which are valuable for crop breeding and precision crop management.

2. Material and Methods

2.1. Experimental Site and Design

A panel of 54 NUE responsive winter wheat lines from the middle and lower reaches of the Yangtze River and other main wheat production regions in China were selected as previously described [21]. Trials were conducted at the Zhenjiang Agricultural Science and Technology Innovation Center (31°57′ N, 119°18′ E; Jiangsu, China) due to its stable weather conditions, between 2019 and 2021 wheat growing seasons. The trials were planted in November of both seasons, and the field was irrigated after completing the planting. After previous crop rotation and fertilizer applications, soil N was classed as consistent across the trial site.
The field experiment was conducted in a two-factor split-area experimental design, i.e., at three N fertilizer levels: 0, 180, and 270 kg·ha−1, denoted by N0, N180, and N270, respectively. N180 and N270 were the optimal and excess N levels for the experimental site, and the amount of N applied was converted pure N. The 54 wheat lines were used with three replications, a total of 486 (54 lines, 3 replicates, 3 treatments) plots per trial. Within the field trial, each plot had an area of 5.25 m2 (1.75 m × 3 m), planting density of 240 plants per m2, row spacing of 25 cm, and 7 rows of 3 m row length in each plot. Urea was used as the N fertilizer, and the application rate was based on 60% of the basal fertilizer applied before sowing, and the nodulation fertilizer was set at 40% applied when the 54 lines largely entered the nodulation stage, e.g., 130 days after sowing (DAS). The basal fertilizers included phosphorus (P2O5) and potassium (K2O) at a rate of 120 kg-ha−1. Fungicides and insecticides were also sprayed at the nodulation and flowering stages to control pests and diseases according to the cultivation and management conditions in the region.

2.2. Sampling and N-Related Measurements

At the beginning of flowering, 40 individual tillers of wheat with uniform growth were selected from each plot for sampling. Twenty individual tillers were cut at ground level with scissors at both flowering and then maturity. Wheat tillers, leaves and spikes (divided into glumes and seeds at maturity) were separated and dried at 80 °C until constant weight was reached, and the weights of different organs (e.g., leaves, stems, glumes, and seeds) were recorded. The N content of different organs was quantified using an Automatic Discrete Analyzer Cleverchem 380 (DeChem-Tech, Hamburg, Germany).
During grain ripening, two rows from each plot were selected and the aboveground biomass was harvested, so that the number of wheat spikes could be counted manually. The whole aboveground plants were oven-dried at 80 °C to a constant weight. Then, samples were manually threshed to obtain the grain, which were tested using Toppan Yunnong’s TPKZ-3 Intelligent Seed Testing System (Zhejiang Topu Yunnong Technology, Hangzhou, China) to measure seed dimensions (e.g., seed size and number). From the acquired samples, we produced a range of N-related measurements, including above-ground biomass at anthesis (AGBa), above-ground N content at anthesis (ANCa), vegetative organ N accumulation at anthesis (VONAa), spike number (SN), grain number per spike (GN), thousand grain weight (TGW), grain yield (GY), vegetative organ biomass (VOB), above-ground biomass (AGB), grain N content (GNC), vegetative organ N content (VONC), above-ground N content (ANC), grain N accumulation (GNA), vegetative organ N accumulation (VONA), above-ground N Accumulation (AGNA), N translocation amount (NTA), N translocation efficiency (NTE), contribution rate of N translocation amount to grain (CNTA), N agronomic use efficiency (NAUE), and N recovery efficiency (NRE). Table 1 shows how each of the manually acquired N-related measures was calculated from the samples.

2.3. Drone Phenotyping and Image Processing

To provide accurate geographic coordinates for drone-collected red-green-blue (RGB) image series to be pre-processed, 22 semi-permanent ground control points (GCPs) were set up in the field. Global Navigation Satellite System (GNSS) information of each GCP was collected using RTK (real-time kinematic; Figure 1a). This was used to calibrate the geo-coordinates for image processing (Figure 1b) in accordance with a previously published method [38]. The wheat images were collected using a Mavic2 Pro UAV (DJI, Shenzhen, China). Each aerial photography mission took place on a sunny, windless day from either 9 to 11 am or from 2 to 4 pm. The flight missions were performed using the DJI GS Pro flight control software (version 2.0.17.), with the flight altitude set to 14 m to maximize the image resolution of crop canopies. The camera gimbal angle was set to 80 degrees, and the image overlap rate was set to 85% in the heading direction and 80% in the side direction (Figure 1c). Three dimensional point clouds (in .LAS format) and two dimensional orthomosaics (in .TIFF format) were generated (Figure 1d) using the drone-collected image series using the DJI Terra software (DJI, Shenzhen, China, version 3.0.0). Eight flights were conducted at key growth stages throughout the season, from 10 to 170 DAS (e.g., 42, 109, 124, 145, 155, 161, 174, 196 DAS).
We followed a number of algorithmic steps to pre-process 3D point clouds to generate a canopy height model (CHM) within a region of interest (ROI) for plot-based NUE-related trait analysis, including outlier removal from the 3D point clouds using the statistical outlier removal algorithm [39]; the application of Cloth Simulation Filter algorithm to differentiate ground-level and aboveground 3D points [40]; the identification of ROIs based on GCPs with geo-coordinates (e.g., GNSS) recorded; the generation of CHMs based on ground-level and above-ground 3D points; and, the iterative self-organizing data thresholding [41] to produce a field-level mask to identify plots [42]. All of the above 2D/3D image processing steps were derived from the AirMeasurer method published previously [25].

2.4. Automated Phenotypic Analysis

The AirMeasurer analysis platform was applied to automate trait analysis based on the processed 2D/3D image series tagged with RTK information for producing plot-level trait and vegetative indices (Figure 1e). Following previously published processing steps (https://github.com/The-Zhou-Lab/UAV-AirMeasurer/releases, accessed on 20 November 2022), the analysis included: (1) the selection of input 2D orthomosaics, 3D point clouds, and .shp file for geo-coordinates; (2) the verification of ROIs and the CHM for initial plot segmentation; (3) the rectification of the plot delineation to produce plot masks; (4) batch-processing to generate a performance matrix for all the plots. In our study, plot-based canopy height, canopy cover, VARI, and NDYI from the AirMeasurer-derived results were employed for further advanced data analysis. Because dynamic traits that are more informative when investigating plant–environment interactions than static traits [22], the AirMeasurer-derived results were converted to dynamic traits such as growth rates of canopy coverage, canopy height, VARI, and NDYI by applying several fitting functions as per [21,25].

2.5. Advanced Statistical Analysis

Statistical analysis in this study was conducted using R version 4.1.1 [43]. A split-plot analysis of variance (ANOVA) was used to test for N efficiency indicators considering N fertilizer treatment and wheat varieties. The square of the correlation coefficient (R2) and root-mean-square errors (RMSE) were used to assess linear regression between the various phenotypic and agronomic traits produced. These traits were also correlated against the AirMeasurer-produced traits. Cluster analysis was performed on the 54 selected lines to identify high-, medium-, and low-NUE groups under N0, N180, and N270 conditions. NAE and NRE agronomic traits were used for clustering. A similar approach was used for the 54 lines using the canopy cover, canopy height, NDYI, and VARI produced by AirMeasurer. In both cases, all variable measurements were first scaled using the Z-scores. Agglomerative hierarchical clustering using Ward’s method implemented in the R Cluster package [44] was employed to cluster the Euclidean distances between the 54 lines.
AirMeasurer clusters were produced for each of the 8 drone data points extracted from the dynamic phenotypic profiles taken at incremental days after sowing. Principal component analysis (PCA) was performed on the scaled agronomic data and the total variance explained by clusters was calculated by separately fitting linear models to each principal component (PC) and multiplying the explained variance (e.g., sum of squares/total sum of squares) by the variance explained by each PC. Explained variation was plotted against DAS to select an appropriate drone flight to use for further analysis. Finally, in order to estimate the effectiveness of the clusters to predict high, medium, and low NUE cultivars, linear models were fit to the NAE, NRE, and principal component datasets. For all linear models, post hoc estimated marginal means and contrasts were generated using the emmeans 1.7.5 package [45], p-values were adjusted for multiple testing using the Tukey method.

3. Results

3.1. The Identification of N-Efficient Wheat Varieties

The 54 winter wheat varieties were previously selected for their different N responses after N treatments at the tillering stage [21]. In this experiment, all wheat varieties used had a significant (p < 0.0001) effect on all measured NUE-related traits (yield, biomass, AGNC, AGNA, NAE, and NRE) and all NUE-related measured traits, except for NAE and NRE whose differences were not statistically significant (p > 0.05), were significantly affected by different N treatments (p < 0.0001) (Figure 2). Correlation significance of various N-related traits was visualized in a matrix using different sizes of circles, with colors representing the direction of the correlation (e.g., lighter colors represent a negative correlation, darker colors represent a stronger correlation; Figure 3). Based on the correlation matrix, we observed a large number of traits with a strong positive correlation with each other. As NAE and NRE showed the most consistently strong correlations with the most other lab-derived NUE traits in Figure 3, it was decided to use only these two traits in the cluster analysis to divide genotypes into high-, medium-, and low-NUE clusters in Figure 4.
In order to establish a valid lab-based dataset for identifying high-, medium-, and low-N-efficient wheat varieties, we grouped the 54 field-screened varieties using a three-cluster analysis of variety performance for both NAE and NRE at both N180 and N270 treatments (Figure 4a,b). The cluster analysis at N180 showed that 3 clusters contained 25 (Cluster 1), 22 (Cluster 2), and 7 (Cluster 3) wheat varieties, whilst at N270, the 3 clusters contained 22 (Cluster 1), 10 (Cluster 2), and 22 (Cluster 3) lines. Linear models (ANOVA) were then fitted to the NAE and NRE measures at both N180 and N270, testing whether each cluster differed in NUE (Figure 4c–f). Post hoc analysis indicated that, for both N180 and N270, Cluster 2 had higher NRE values compared to Clusters 1 and 3 (p < 0.0001), and Cluster 1 had higher values compared to Cluster 3 (p < 0.0001). For NAE, Cluster 2 had a significantly higher value than the other two clusters (p < 0.0001) at N180, while Cluster 1 and 2 at N270 had significantly different values from Cluster 3 (p < 0.0002). As a result, for both N180 and N270, Cluster 2 (N180 = 22 lines, N270 = 10 lines) was considered to contain high NUE varieties, and Cluster 3 (N180 = 7 lines, N270 = 22 lines) for low NUE varieties. This classification of the clusters was further enforced when cluster performance was compared across all lab-derived NUE-related traits (Figure S1).

3.2. Drone Phenotyping to Dynamically Analyze Target Morphological Traits

To compare clustering based on lab-based approaches with wheat lines to that derived from drone phenotyping, we first used the AirMeasurer platform to perform phenotypic analysis of the morphological traits (e.g., canopy cover and canopy height) and the spectral traits (e.g., NDYI and VARI) using spectral, textural, and spatial data. Among many traits output by the AirMeasurer, these four traits were chosen due to their biological relevance to NUE under differing N treatments [21]. The aerial imaging was performed on all plots across both seasons. A total of 8 flights were recorded and analyzed, covering from 30 DAS (seedling; Growth Stage [GS]15) to 196 DAS (grain filling; GS85). As drone-based phenotyping can be impacted by varied field conditions (illumination, wind speed), the accuracy of static trait analysis could be impacted negatively. Therefore, we used the eight collected data points to develop profile curves using several fitting functions such as Gaussian, Fourier, and Weibull, through which we identified changing profiles of the four traits over a season to study the effects of the three N treatments, with outliers removed (Figure 5).
In both seasons, after the N applications at around 130 DAS, plot-based canopy cover increased to around 150 DAS then stabilized. There was little distinction between the N180 and N270 treatments across both seasons; in the 2020–2021 season, canopy coverage at N180 peaked approximately six days before N270. N0′s canopy coverage was consistently below both N180 and N270, suggesting an optimal application of N (e.g., N180) had a pronounced effect on canopy cover but an excess N application (e.g., N270) gave very little benefit, which was a pattern observed across all the drone-derived traits across the two seasons. Canopy height increased during the season before showing a slight decline at 175 DAS as the crop started to senesce and dry, which resulted in the reduction of canopy cover and height as the ears of the wheat began to curl. In general, there was a steady increase in height and canopy cover from establishment up to 100 DAS, at which point stem elongation and tillering together with N applications at 130 DAS led to a rapid increase up to 150 DAS (flowering), followed by a growth plateau before declining at 170 DAS (senescence). For plant height, there was a distinct difference in performance between the N0 plants and the treated (N180 and N270) plants, with N0 heights reaching a maximum of 60 cm in both seasons, whereas the treated plants reached a maximum of over 100 cm in both seasons. There was very little difference in heights between the N180 and N270 treatments in the 2019–2020 season but in 2020–2021 plants treated with N180 reached a slightly higher maximum height.

3.3. Drone Phenotyping to Dynamically Analyze Target Spectral Traits

We utilized NDYI to quantify vegetive yellowness and flowering and VARI for vegetation greenness while minimizing the impact of atmospheric conditions. Thus, NDYI in our experiments showed an inverse pattern to VARI. Our study suggested that VARI increased up to 130 DAS before declining sharply. N0 treatment showed a distinctly lower value for VARI when compared to the N180 and N270 treatments. There was minimal difference between the N180 and N270 treatments. Unlike canopy cover, there was a rapid decline post-145 DAS in VARI, suggesting it was more sensitive to senescence than simple canopy cover measurements.
In NDYI measurements, N0 plants showed a small decline to around 150 DAS before a rapid increase. For N180 and N270 plants, the decline was much larger (dropping to 0.2) before a rapid recovery. Also, there was a pronounced difference in the pace of the NDYI decline in N180 and N270 plants between the two seasons: (1) in the first season, NDYI stayed steady up to 100 DAS before a rapid decline and recovery at 150 DAS, with both N180 and N270 following a similar pattern; (2) in the second season, the decline in NDYI was much more gradual and occurred from 50 DAS. There was also a distinct difference between plants in the N180 and N270 groups. N180 showed a consistent decline, whereas N270 was initially gradual and then increased over time. The difference in NDYI response between N0 and N180/270 suggests that untreated wheat varieties have a longer vegetive stage compared with the N-treated lines. Across all drone-derived traits, N0 was consistently different from both N180 and N270, but with limited difference between N180 and N270, indicating an optimal application of N (e.g., N180) had a pronounced effect on wheat developmental paces but an excess application of N (e.g., N270) gave very little benefit. This could ultimately have implications for cluster analysis as drone-derived traits might not be able to identify the subtle differences in plant growth and development when the N application (e.g., N270) is excessive.

3.4. Comparison of High-NUE Lines Identified by Drone- and Lab-Based Approaches

As the key aim of this research, we trialed the use of drone-based phenotyping to identify high-NUE varieties at a low-cost, in a rapid and scalable manner within in the context of breeding. As such, we used a similar three-cluster analysis, Figure 4. Because conducting a cluster analysis of drone data for each timepoint when drone phenotyping was conducted might not be informative, we, therefore, chose to employ AirMeasurer-derived traits at several key timepoints for comparison when the most variation in NUE based on NAE and NRE was observed, resulting in 145 DAS for N180 and 157 DAS for N270 (Figure S2). The drone-derived traits from the two timepoints were then used for cluster analysis, grouping the 54 lines into 3 clusters according to their performance in canopy cover, canopy height, NDYI, and VARI at both N180 and N270 treatments (Figure 6).
At N180, Cluster I contained 28 lines, Cluster II and III contained 19 and 7 lines, respectively. At N270, Cluster I contained 20 lines, Cluster II and III contained 22 and 12 lines, respectively. As with the clusters produced from lab-based NUE methods, a linear model (e.g., ANOVA) was fitted to the NAE and NRE data to test whether clusters differed in NUE (Figure 6c–f). Post hoc analysis indicated that for NAE at N180, Cluster II had a significantly higher value than the other two clusters (p < 0.05), whilst Cluster I and II had significantly different values from Cluster III (p < 0.02) for N270. For NRE, at N180, Cluster I was significantly higher than Cluster III (p = 0.028), while none of the clusters differed significantly for N270, though Cluster II was close to significant difference to Cluster III (p = 0.059).
The performance of the clusters for NAE and NRE at N180 (Figure 6c,d) showed that Cluster II was the high-NUE grouping, whilst Cluster I was medium, and Cluster III was low. For N270, identifying the high-NUE cluster was more difficult with both Cluster I and II, which showed a similar NAE and NRE performance. Cluster II was marginally higher and, therefore, the high-NUE cluster (Figure 6e,f), whereas Cluster III was most likely the low-NUE cluster. To compare the cluster analysis produced from both lab- and drone-based approaches, PCs from the PCA for both N180 and N270 of NAE and NRE data were plotted (Figure 7), with circles, triangles, and squares representing high, low, and medium lab-derived data, respectively (black, orange, and blue representing the drone-derived Cluster I (medium), II (high), and III (low), respectively). At N180, Cluster II members clustered reasonably well and, crucially, some distance from all the low NUE lab-derived cluster members. At N270, the picture is more confused, with no clear clusters from either lab-derived or drone-derived methods.

4. Discussion

N availability and use efficiency are two of the most important factors affecting crop yield and grain quality [11]. Breeding for better NUE would allow for a reduction in N application by farmers, resulting in reduced costs and environmental footprint [46]. Conventional lab-based methods of NUE quantification are arduous and difficult to implement regularly during the season. Hence, this study demonstrates that the use of cost-effective drone phenotyping could acquire regular datasets from key growth stages, from which plot-level resolution of trait analysis can be generated to enable automated analysis of NUE-relevant VIs and phenotypic traits. Furthermore, this approach is potentially valuable for identifying low-NUE varieties within large scale breeding programs, thus reducing the economic cost of screening a large number of crop varieties. Nevertheless, further verification of economic costs, throughput, processing time, and the effectiveness of the approach for identifying low-NUE varieties under varied N applications is beyond the scope of this study but valuable for future studies.

4.1. The Application of Drone-Derived Static and Dynamic Traits for NUE Assessment

Our previous research has shown that canopy cover, canopy height, VARI, and NDYI are effective proxies for measuring N responses under field conditions [21]. In this study, we used the open-source AirMeasurer platform and cost-effective drones to examine these traits at many key growth stages in winter wheat, which would not be possible to accomplish using lab-based approaches. To our knowledge, this was only partially achievable through expensive multispectral drones and proprietary software [31,32]. In addition to static traits, the estimation of dynamic traits from fitting functions based on time-series measures of static traits allows the understanding of dynamic features (e.g., growth rates and phenotypic changes of target traits) without excessively conducting drone phenotyping [25]. Here, this has enabled us to evaluate the dynamic NUE responses to different N treatments of 54 varieties, instead of focusing on arbitrary time points. In fact, using drone-derived dynamic traits provides a more comprehensive understanding of plant performance under varying conditions, assessing crops’ adaptation to N levels over time and the resilience to climate change [47], as well as tracking target traits at key growth stages for optimizing planting and crop management practices [26]. In our case, the change rates of the NUE-related traits helped us measure and analyze different NUE characteristics of the 54 varieties, which allowed us to identify preferred lines for the selection of high-NUE crop varieties under varied N treatments. Although the present approach primarily worked well under optimized application, we trust that such an approach would also be valuable for the development of sustainable agricultural practices focusing on minimizing resource wastage and environmental impacts. Potentially, the integration of dynamic NUE responses with genetic mapping could lead to new QTL discovery and estimates of QTL x environment interaction in a larger population, with more replicates and over multiple sites. The above indicates the possibility of applying both static and dynamic drone-derived traits to compliment or even replace some lab-based methods for NUE assessment in field trials.

4.2. The Potential Use of the Drone-Based Selection for NUE-Focused Breeding

The selection of high-NUE lines in a breeding population can be difficult due to the static nature of lab-based NUE assessments and the dynamic response of varieties to N. We explored the use of drone-derived traits and a three-cluster analysis for NUE performance to identify lines, showing a promising approach to advance NUE-focused breeding in wheat. For example, by selecting for a high-NUE cluster based on drone-derived traits, a breeder would remove any lab-derived low-NUE varieties from the next generation field experiment. If this were applied across a large-scale breeding program, this could remove a large percentage of unwanted lines, which would be financially and resource-productively beneficial to breeders [48]. In addition, the identification of dynamic NUE proxies has the potential to provide insights into how different varieties interact with key environmental factors. This is important for identifying the performing varieties under specific environmental conditions, leading to the understanding of traits for genetic mapping and development of molecular markers under genotype–environment interactions [49]. Furthermore, by combining physiological [25] with N-responsive [21] measurements, it will be feasible to provide numerous dynamic traits to breeders, who will make more precise and informative decisions for variety selection.

4.3. Limitations of the Drone-Derived NUE Cluster Protocol

As with all drone phenotyping, we encountered some common problems such as weather conditions (e.g., flying with strong wind >15 mph is not possible), rainfall, or reduced visibility due to fog, illuminance, and aviation regulations that required regular contacts with local authorities for regular phenotyping. One of the disadvantages of small drones is the reduced payload capacity which prevented us from using multispectral and hyperspectral image sensors to verify the RGB approach, which were reportedly effective at monitoring N responses [32]. The proprietary software used first to generate the 3D point clouds and 2D orthomosaics using the PIX4DMapper software (version 4.9.0) can also result in prolonged computational time, incorrect geo-referencing, and mismatched 2D/3D patches, which could now be replaced by an Education edition of the software or open-source software implementations such as AliceVision Meshroom (https://alicevision.org/, version 2023.2.0, accessed on 29 June 2024). Finally, one of the biggest issues we faced in this study was the identification of high-NUE lines when a trial was treated with excess N (e.g., N270). At this N rate, drone-derived clusters struggled to identify high-NUE lines, although this might be expected as, the identification of high-NUE became more difficult when excessive N fertilizers were applied as it reduced the resource use efficiency for all lines [50].

4.4. Future Work for NUE-Related Phenotyping

This study demonstrated the effectiveness of drone-based phenotyping and the AirMeasurer platform when assessing complex traits based on NUE-related physiological and spectral features. This suggests that similar methods could be applied to study and estimate other complex traits such as grain ripening or disease infection. By identifying key timepoints of plant growth and disease–plant interactions, dynamic traits such as grain filling rates and disease progression can be established to monitor crop performance in a timely manner, enabling informed decision-making on pest management strategies to breed disease-resistant varieties and precise agriculture activities in terms of irrigation, fertilization, and harvesting, including agronomic indications of the timing for N application.

5. Conclusions

In this study, we assessed the effectiveness of using a financially accessible RGB drone to conduct field-based NUE phenotyping in wheat instead of costly and time-consuming lab-based analysis. The main conclusions are as follows: (1) Drone analysis can be used instead of lab-based analysis for exclusion of NUE lines. We found that the drone-based method was very effective at N180 (optimized N-application rates) as it completely removed low-NUE lines but was less effective at higher and excessive N application. (2) Currently, the only way to assess NUE in the field using drones is with a multispectral camera, for which we have shown that a cost-effective RGB camera could be as effective as the hyperspectral option. (3) The dynamic phenotyping that the drone-based method enabled provided a realistic analysis of NUE development over a season when compared to the static analysis provided by the lab-based method. (4) Combining all these results shows that drone-based phenotyping and dynamic analysis of NUE responses could be applied to accelerate NUE selection within breeding programs and resource use efficiency studies in wheat.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/agronomy14081612/s1, Figure S1: Performance of each cluster, produced in Figure 2, for each traditional N-related phenotype; Figure S2: Total variation in nitrogen use efficiency based on NAE and NRE produced by using AirMeasurer-derived traits across multiple days after sowing (DAS).

Author Contributions

R.J., J.Z. and G.D. (Greg Deakin) wrote the manuscript with inputs from all the authors; G.D. (Guohui Ding), L.S., M.A. and Z.W. conducted lab and field experiments together with drone phenotyping; L.S., J.D., F.P. and R.J. performed trait analysis; G.D. (Greg Deakin), R.J. and J.Z. performed statistical analysis; L.S. and R.J. conducted dynamic phenotypic analysis under J.Z. supervision. L.S., G.D. (Greg Deakin) and G.D. (Guohui Ding) contributed equally. All authors have read and agreed to the published version of the manuscript.

Funding

The drone-based phenotyping, yield, and N measures were supported by the National Natural Science Foundation of China (32070400 to Ji Zhou). Ji Zhou and Felipe Pinheiro were supported by the Allan & Gill Gray Philanthropies’ sustainable productivity for crops programme (G118688 to the University of Cambridge and Q-20-0370 to NIAB). Robert Jackson and Greg Deakin were supported by the United Kingdom Research and Innovation’s (UKRI) Biotechnology and Biological Sciences Research Council’s (BBSRC) International Partnership Grant (BB/Y514081/1). Ji Zhou and Robert Jackson were also supported by the One CGIAR’s Seed Equal Research Initiative for wheat varietal research (5507-CGIA-07 to Ji Zhou).

Data Availability Statement

The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding authors.

Conflicts of Interest

The authors declare no conflict of interest.

Abbreviations

Above-ground biomass (AGB), above-ground biomass at anthesis (AGBa), above-ground N accumulation (AGNA), above-ground N content (ANC), above-ground N content at anthesis (ANCa), analysis of variance (ANOVA), canopy height model (CHM), contribution rate of N translocation amount to grain (CNTA), days after sowing (DAS), Global Navigation Satellite System (GNSS), grain N accumulation (GNA), grain N content (GNC), grain number per spike (GN), grain yield (GY), ground control points (GCPs), N agronomic use efficiency (NAE), N recovery efficiency (NRE), N translocation amount (NTA), N translocation efficiency (NTE), nitrogen use efficiency (NUE), normalized difference red edge index (NDRE), normalized difference yellowness index (NDYI), normalized difference blue index (NDBI), principal component (PC), principal component analysis (PCA), real-time kinematic (RTK), red-green-blue (RGB), region of interest (ROIs), root-mean-square errors (RMSE), spike number (SN), thousand grain weight (TGW), vegetation indices (VIs), vegetative organ biomass (VOB), vegetative organ N accumulation (VONA), vegetative organ N accumulation at anthesis (VONAa), vegetative organ N content (VONC), visible atmospherically resistance index (VARI).

References

  1. Curtis, T.; Halford, N.G. Food Security: The Challenge of Increasing Wheat Yield and the Importance of Not Compromising Food Safety. Ann. Appl. Biol. 2014, 164, 354–372. [Google Scholar] [CrossRef]
  2. Bhardwaj, A.K.; Rajwar, D.; Yadav, R.K.; Chaudhari, S.K.; Sharma, D.K. Nitrogen Availability and Use Efficiency in Wheat Crop as Influenced by the Organic-Input Quality under Major Integrated Nutrient Management Systems. Front. Plant Sci. 2021, 12, 634448. [Google Scholar] [CrossRef]
  3. Nguyen, G.N.; Kant, S. Improving Nitrogen Use Efficiency in Plants: Effective Phenotyping in Conjunction with Agronomic and Genetic Approaches. Funct. Plant Biol. 2018, 45, 606–619. [Google Scholar] [CrossRef]
  4. Sharma, L.K.; Bali, S.K. A Review of Methods to Improve Nitrogen Use Efficiency in Agriculture. Sustainability 2017, 10, 51. [Google Scholar] [CrossRef]
  5. Sylvester-Bradley, R.; Kindred, D.R. Analysing Nitrogen Responses of Cereals to Prioritize Routes to the Improvement of Nitrogen Use Efficiency. J. Exp. Bot. 2009, 60, 1939–1951. [Google Scholar] [CrossRef]
  6. Govindasamy, P.; Muthusamy, S.K.; Bagavathiannan, M.; Mowrer, J.; Jagannadham, P.T.K.; Maity, A.; Halli, H.M.; Sujayananad, G.K.; Vadivel, R.; Das, T.K.; et al. Nitrogen Use Efficiency—A Key to Enhance Crop Productivity under a Changing Climate. Front. Plant Sci. 2023, 14, 1121073. [Google Scholar] [CrossRef]
  7. Raun, W.R.; Johnson, G.V. Improving Nitrogen Use Efficiency for Cereal Production. Agron. J. 1999, 91, 357–363. [Google Scholar] [CrossRef]
  8. Chen, Z.; Wang, H.; Liu, X.; Lu, D.; Zhou, J. The Fates of 15N-Labeled Fertilizer in a Wheat-Soil System as Influenced by Fertilization Practice in a Loamy Soil. Sci. Rep. 2016, 6, 34754. [Google Scholar] [CrossRef]
  9. Hawkesford, M.J. Reducing the Reliance on Nitrogen Fertilizer for Wheat Production. J. Cereal Sci. 2014, 59, 276–283. [Google Scholar] [CrossRef]
  10. Monostori, I.; Szira, F.; Tondelli, A.; Gierczik, K.; Cattivelli, L.; Galiba, G.; Vágújfalvi, A. Genome-Wide Association Study and Genetic Diversity Analysis on Nitrogen Use Efficiency in a Central European Winter Wheat (Triticum aestivum L.). PLoS ONE 2017, 2, e0189265. [Google Scholar] [CrossRef]
  11. Barraclough, P.B.; Howarth, J.R.; Jones, J.; Lopez-Bellido, R.; Parmar, S.; Shepherd, C.E.; Hawkesford, M.J. Nitrogen Efficiency of Wheat: Genotypic and Environmental Variation and Prospects for Improvement. Eur. J. Agron. 2010, 33, 1–11. [Google Scholar] [CrossRef]
  12. Peron-Danaher, R.; Russell, B.; Cotrozzi, L.; Couture, J.J.; Mohammadi, M. Incorporating Multi-Scale, Spectrally Detected Nitrogen Concentrations into Assessing Nitrogen Use Efficiency for Winter Wheat Breeding Populations. Remote Sens. 2021, 13, 3991. [Google Scholar] [CrossRef]
  13. Kamiji, Y.; Pang, J.; Milroy, S.P.; Palta, J.A. Shoot Biomass in Wheat Is the Driver for Nitrogen Uptake under Low Nitrogen Supply, but Not under High Nitrogen Supply. Field Crops Res. 2014, 165, 92–98. [Google Scholar] [CrossRef]
  14. Mann, C.C. Crop Scientists Seek a New Revolution. Science 1999, 283, 310–316. [Google Scholar] [CrossRef]
  15. Voss-Fels, K.P.; Stahl, A.; Wittkop, B.; Lichthardt, C.; Nagler, S.; Rose, T.; Chen, T.W.; Zetzsche, H.; Seddig, S.; Majid Baig, M.; et al. Breeding Improves Wheat Productivity under Contrasting Agrochemical Input Levels. Nat. Plants 2019, 5, 706–714. [Google Scholar] [CrossRef]
  16. Zhu, Y.; Sun, G.; Ding, G.; Zhou, J.; Wen, M.; Jin, S.; Zhao, Q.; Colmer, J.; Ding, Y.; Ober, E.S.; et al. Large-Scale Field Phenotyping Using Backpack LiDAR and CropQuant-3D to Measure Structural Variation in Wheat. Plant Physiol. 2021, 187, 716–738. [Google Scholar] [CrossRef]
  17. Sarker, U.K.; Romij Uddin, M.; Salahuddin Kaysar, M.; Alamgir Hossain, M.; Somaddar, U.; Saha, G. Exploring Relationship among Nitrogen Fertilizer, Yield and Nitrogen Use Efficiency in Modern Wheat Varieties under Subtropical Condition. Saudi J. Biol. Sci. 2023, 30, 103602. [Google Scholar] [CrossRef]
  18. Fuertes-Mendizábal, T.; González-Murua, C.; González-Moro, M.B.; Estavillo, J.M. Late Nitrogen Fertilization Affects Nitrogen Remobilization in Wheat. J. Plant Nutr. Soil Sci. 2012, 175, 115–124. [Google Scholar] [CrossRef]
  19. Kjeldahl, J. Neue Methode Zur Bestimmung Des Stickstoffs in Organischen Körpern. Z. Anal. Chem. 1883, 22, 366–382. [Google Scholar] [CrossRef]
  20. Dumas, J.B.A. Procedes de l’analyse Organic. Ann. Chem. Phys. 1831, 247, 198–213. [Google Scholar]
  21. Ding, G.; Shen, L.; Dai, J.; Jackson, R.; Liu, S.; Ali, M.; Sun, L.; Wen, M.; Xiao, J.; Deakin, G.; et al. The Dissection of Nitrogen Response Traits Using Drone Phenotyping and Dynamic Phenotypic Analysis to Explore N Responsiveness and Associated Genetic Loci in Wheat. Plant Phenomics 2023, 5, 0128. [Google Scholar] [CrossRef]
  22. Adak, A.; Murray, S.C.; Washburn, J.D. Deciphering Temporal Growth Patterns in Maize: Integrative Modeling of Phenotype Dynamics and Underlying Genomic Variations. New Phytol. 2024, 242, 121–136. [Google Scholar] [CrossRef]
  23. Jackson, R.; Buntjer, J.B.; Bentley, A.R.; Lage, J.; Byrne, E.; Burt, C.; Jack, P.; Berry, S.; Flatman, E.; Poupard, B.; et al. Phenomic and Genomic Prediction of Yield on Multiple Locations in Winter Wheat. Front. Genet. 2023, 14, 1164935. [Google Scholar] [CrossRef]
  24. Fradgley, N.S.; Bentley, A.R.; Swarbreck, S.M. Defining the Physiological Determinants of Low Nitrogen Requirement in Wheat. Biochem. Soc. Trans. 2021, 49, 609. [Google Scholar] [CrossRef]
  25. Sun, G.; Lu, H.; Zhao, Y.; Zhou, J.; Jackson, R.; Wang, Y.; Xu, L.-X.; Wang, A.; Colmer, J.; Ober, E.; et al. AirMeasurer: Open-Source Software to Quantify Static and Dynamic Traits Derived from Multiseason Aerial Phenotyping to Empower Genetic Mapping Studies in Rice. New Phytol. 2022, 236, 1584–1604. [Google Scholar] [CrossRef]
  26. Gano, B.; Bhadra, S.; Vilbig, J.M.; Ahmed, N.; Sagan, V.; Shakoor, N. Drone-Based Imaging Sensors, Techniques, and Applications in Plant Phenotyping for Crop Breeding: A Comprehensive Review. Plant Phenome J. 2024, 7, e20100. [Google Scholar] [CrossRef]
  27. Jiang, Z.; Tu, H.; Bai, B.; Yang, C.; Zhao, B.; Guo, Z.; Liu, Q.; Zhao, H.; Yang, W.; Xiong, L.; et al. Combining UAV-RGB High-Throughput Field Phenotyping and Genome-Wide Association Study to Reveal Genetic Variation of Rice Germplasms in Dynamic Response to Drought Stress. New Phytol. 2021, 232, 440–455. [Google Scholar] [CrossRef]
  28. Atkinson, J.A.; Jackson, R.J.; Bentley, A.R.; Ober, E.; Wells, D.M. Field Phenotyping for the Future. In Annual Plant Reviews Online; Wiley: Hoboken, NJ, USA, 2018; pp. 719–736. [Google Scholar]
  29. Yang, M.; Hassan, M.A.; Xu, K.; Zheng, C.; Rasheed, A.; Zhang, Y.; Jin, X.; Xia, X.; Xiao, Y.; He, Z. Assessment of Water and Nitrogen Use Efficiencies Through UAV-Based Multispectral Phenotyping in Winter Wheat. Front. Plant Sci. 2020, 11, 927. [Google Scholar] [CrossRef]
  30. Dhanya, V.G.; Subeesh, A.; Kushwaha, N.L.; Vishwakarma, D.K.; Nagesh Kumar, T.; Ritika, G.; Singh, A.N. Deep Learning Based Computer Vision Approaches for Smart Agricultural Applications. Artif. Intell. Agric. 2022, 6, 211–229. [Google Scholar] [CrossRef]
  31. Yang, G.; Liu, J.; Zhao, C.; Li, Z.Z.; Huang, Y.; Yu, H.; Xu, B.; Yang, X.; Zhu, D.; Zhang, X.; et al. Unmanned Aerial Vehicle Remote Sensing for Field-Based Crop Phenotyping: Current Status and Perspectives. Front. Plant Sci. 2017, 8, 1111. [Google Scholar] [CrossRef] [PubMed]
  32. Liu, J.; Zhu, Y.; Tao, X.; Chen, X.; Li, X. Rapid Prediction of Winter Wheat Yield and Nitrogen Use Efficiency Using Consumer-Grade Unmanned Aerial Vehicles Multispectral Imagery. Front. Plant Sci. 2022, 13, 1032170. [Google Scholar] [CrossRef]
  33. Wang, J.; Li, X.; Guo, T.; Dzievit, M.J.; Yu, X.; Liu, P.; Price, K.P.; Yu, J. Genetic Dissection of Seasonal Vegetation Index Dynamics in Maize through Aerial Based High-Throughput Phenotyping. Plant Genome 2021, 14, e20155. [Google Scholar] [CrossRef]
  34. Anderson, S.L. R/UAStools::Plotshpcreate: Create Multi-Polygon Shapefiles for Extraction of Research Plot Scale Agriculture Remote Sensing Data. Front. Plant Sci. 2020, 11, 511768. [Google Scholar] [CrossRef]
  35. Matias, F.I.; Caraza-Harter, M.V.; Endelman, J.B. FIELDimageR: An R Package to Analyze Orthomosaic Images from Agricultural Field Trials. Plant Phenome J. 2020, 3, e20005. [Google Scholar] [CrossRef]
  36. Gitelson, A.A.; Stark, R.; Grits, U.; Rundquist, D.; Kaufman, Y.; Derry, D. Vegetation and Soil Lines in Visible Spectral Space: A Concept and Technique for Remote Estimation of Vegetation Fraction. Int. J. Remote Sens. 2002, 23, 2537–2562. [Google Scholar] [CrossRef]
  37. Sulik, J.J.; Long, D.S. Spectral Considerations for Modeling Yield of Canola. Remote Sens. Environ. 2016, 184, 161–174. [Google Scholar] [CrossRef]
  38. Watanabe, K.; Guo, W.; Arai, K.; Takanashi, H.; Kajiya-Kanegae, H.; Kobayashi, M.; Yano, K.; Tokunaga, T.; Fujiwara, T.; Tsutsumi, N.; et al. High-Throughput Phenotyping of Sorghum Plant Height Using an Unmanned Aerial Vehicle and Its Application to Genomic Prediction Modeling. Front. Plant Sci. 2017, 8, 421. [Google Scholar] [CrossRef]
  39. Hodge, V.J.; Austin, J. A Survey of Outlier Detection Methodologies. Artif. Intell. Rev. 2004, 22, 85–126. [Google Scholar] [CrossRef]
  40. Zhang, W.; Qi, J.; Wan, P.; Wang, H.; Xie, D.; Wang, X.; Yan, G. An Easy-to-Use Airborne LiDAR Data Filtering Method Based on Cloth Simulation. Remote Sens. 2016, 8, 501. [Google Scholar] [CrossRef]
  41. Irvin, B.J.; Ventura, S.J.; Slater, B.K. Fuzzy and Isodata Classification of Landform Elements from Digital Terrain Data in Pleasant Valley, Wisconsin. Geoderma 1997, 77, 137–154. [Google Scholar] [CrossRef]
  42. Singh, T.R.; Roy, S.; Singh, O.I.; Sinam, T.; Singh, K.M. A New Local Adaptive Thresholding Technique in Binarization. Int. J. Comput. Sci. Issues 2012, 271–277. [Google Scholar]
  43. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2021; Available online: https://www.R-project.org/ (accessed on 10 August 2021).
  44. Maechler, M.; Rousseeuw, P.; Struyf, A.; Hubert, M.; Hornik, K. Cluster: Cluster Analysis Basics and Extensions. R Package Version 2.1.6. 2023. Available online: https://CRAN.R-project.org/package=cluster (accessed on 2 May 2024).
  45. Lenth, R.V. Emmeans: Estimated Marginal Means, Aka Least-Squares Means. R Package Version 1.7.5. 2022. Available online: https://CRAN.R-project.org/package=emmeans (accessed on 22 June 2022).
  46. Lammerts van Bueren, E.T.; Struik, P.C. Diverse Concepts of Breeding for Nitrogen Use Efficiency. A Review. Agron. Sustain. Dev. 2017, 37, 50. [Google Scholar] [CrossRef]
  47. Balasundram, S.K.; Shamshiri, R.R.; Sridhara, S.; Rizan, N. The Role of Digital Agriculture in Mitigating Climate Change and Ensuring Food Security: An Overview. Sustainability 2023, 15, 5325. [Google Scholar] [CrossRef]
  48. Jeon, D.; Kang, Y.; Lee, S.; Choi, S.; Sung, Y.; Lee, T.H.; Kim, C. Digitalizing Breeding in Plants: A New Trend of next-Generation Breeding Based on Genomic Prediction. Front. Plant Sci. 2023, 14, 1092584. [Google Scholar] [CrossRef]
  49. Reynolds, M.; Chapman, S.; Crespo-Herrera, L.; Molero, G.; Mondal, S.; Pequeno, D.N.L.; Pinto, F.; Pinera-Chavez, F.J.; Poland, J.; Rivera-Amado, C.; et al. Breeder Friendly Phenotyping. Plant Sci. 2020, 295, 110396. [Google Scholar] [CrossRef]
  50. Anas, M.; Liao, F.; Verma, K.K.; Sarwar, M.A.; Mahmood, A.; Chen, Z.L.; Li, Q.; Zeng, X.P.; Liu, Y.; Li, Y.R. Fate of Nitrogen in Agriculture and Environment: Agronomic, Eco-Physiological and Molecular Approaches to Improve Nitrogen Use Efficiency. Biol. Res. 2020, 53, 47. [Google Scholar] [CrossRef]
Agronomy 14 01612 g001
Figure 1. A general workflow of drone-based field phenotyping and phenotypic analysis for collection of orthomosaics, 3D point clouds, and AirMeasurer inputs and outputs. (a) RTK location of ground control points (GCPs) within the trial. (b) Trial images using a DJI Mavic 2 flying a preset DJI GS Pro protocol. Images were processed using DJI Terra. (c) A 2D orthomosaic image of the field trial. (d) 2D orthomosaic and 3D point clouds were combined for plot segmentation and trait analysis using the AirMeasurer platform. (e) Pseudo-colored images of plots showing crop height, crop cover, VARI, and NDYI at three different N-rates (N0, N180, N270).
Figure 1. A general workflow of drone-based field phenotyping and phenotypic analysis for collection of orthomosaics, 3D point clouds, and AirMeasurer inputs and outputs. (a) RTK location of ground control points (GCPs) within the trial. (b) Trial images using a DJI Mavic 2 flying a preset DJI GS Pro protocol. Images were processed using DJI Terra. (c) A 2D orthomosaic image of the field trial. (d) 2D orthomosaic and 3D point clouds were combined for plot segmentation and trait analysis using the AirMeasurer platform. (e) Pseudo-colored images of plots showing crop height, crop cover, VARI, and NDYI at three different N-rates (N0, N180, N270).
Agronomy 14 01612 g001
Agronomy 14 01612 g002
Figure 2. Varieties by N application interaction at three different N-rates (N0, N180, and N270). An ANOVA was applied to compare the varieties by N-application interaction at N0, N180, and N270, for the traits such as above-ground biomass (AGB), above-ground N content (ANC), above-ground N accumulations (AGNA), N agronomic use efficiency (NAE), and N recovery efficiency (NRE).
Figure 2. Varieties by N application interaction at three different N-rates (N0, N180, and N270). An ANOVA was applied to compare the varieties by N-application interaction at N0, N180, and N270, for the traits such as above-ground biomass (AGB), above-ground N content (ANC), above-ground N accumulations (AGNA), N agronomic use efficiency (NAE), and N recovery efficiency (NRE).
Agronomy 14 01612 g002
Agronomy 14 01612 g003
Figure 3. Correlation of various N-related traits. Size of circle represents the strength of the correlation with the color of the circle representing the direction of the correlation, lighter colors represent a negative correlation, darker colors represent a stronger correlation. Above-ground biomass at anthesis (AGBa), above-ground N content at anthesis (ANCa), vegetative organ N accumulation at anthesis (VONAa), spike number (SN), grain number per spike (GN), thousand grain weight (TWG), grain yield (GY), vegetative organ biomass (VOB), above-ground biomass (AGB), grain N content (GNC), vegetative organ N content (VONC), above-ground N content (ANC), grain N accumulation (GNC), vegetative organ N accumulation (VONC), above-ground N accumulation (AGNA), N translocation amount (NTA), N translocation efficiency (NTE), contribution rate of N translocation amount to grain (CNTA), N agronomic use efficiency (NAE), and N recovery efficiency (NRE).
Figure 3. Correlation of various N-related traits. Size of circle represents the strength of the correlation with the color of the circle representing the direction of the correlation, lighter colors represent a negative correlation, darker colors represent a stronger correlation. Above-ground biomass at anthesis (AGBa), above-ground N content at anthesis (ANCa), vegetative organ N accumulation at anthesis (VONAa), spike number (SN), grain number per spike (GN), thousand grain weight (TWG), grain yield (GY), vegetative organ biomass (VOB), above-ground biomass (AGB), grain N content (GNC), vegetative organ N content (VONC), above-ground N content (ANC), grain N accumulation (GNC), vegetative organ N accumulation (VONC), above-ground N accumulation (AGNA), N translocation amount (NTA), N translocation efficiency (NTE), contribution rate of N translocation amount to grain (CNTA), N agronomic use efficiency (NAE), and N recovery efficiency (NRE).
Agronomy 14 01612 g003
Agronomy 14 01612 g004
Figure 4. More balanced three-cluster analysis conducted to define high-, medium-, and low-NUE wheat lines using lab-based N content measurements NAE and NRE. (a,b) Cluster analyses of lines under N180 and N270 conditions on the 54 selected lines to identify high, medium, and low-NUE groups based on traits NAE and NRE. Agglomerative hierarchical clustering of NAE and NRE data with Euclidean distance calculation was applied. Performance of clusters were tested with ANOVA at N180 for NAE (c) and NRE (d), and at N270 for NAE (e) and NRE (f), showing that Cluster 2 (blue) contained the high-NUE lines at both treatments, whilst Cluster 1 (pink) represented medium-NUE, and Cluster 3 (orange) represented low-NUE lines.
Figure 4. More balanced three-cluster analysis conducted to define high-, medium-, and low-NUE wheat lines using lab-based N content measurements NAE and NRE. (a,b) Cluster analyses of lines under N180 and N270 conditions on the 54 selected lines to identify high, medium, and low-NUE groups based on traits NAE and NRE. Agglomerative hierarchical clustering of NAE and NRE data with Euclidean distance calculation was applied. Performance of clusters were tested with ANOVA at N180 for NAE (c) and NRE (d), and at N270 for NAE (e) and NRE (f), showing that Cluster 2 (blue) contained the high-NUE lines at both treatments, whilst Cluster 1 (pink) represented medium-NUE, and Cluster 3 (orange) represented low-NUE lines.
Agronomy 14 01612 g004
Agronomy 14 01612 g005
Figure 5. AirMeasurer-derived canopy cover, height, VARI, and NDYI development of the crop through the growing season at three N-rates (e.g., N0, N180, and N270 with three replicates across 54 varieties; colored orange, blue and green, respectively). (a) Eight plot-based measurements were used to create the growth curves for canopy cover, height, VARI, and NDYI under each N rate in the 2019–2020 season. (b) Eight plot-based measurements were used to create the curves for canopy cover, canopy height, and VARI under each N rate in the 2020–2021 season.
Figure 5. AirMeasurer-derived canopy cover, height, VARI, and NDYI development of the crop through the growing season at three N-rates (e.g., N0, N180, and N270 with three replicates across 54 varieties; colored orange, blue and green, respectively). (a) Eight plot-based measurements were used to create the growth curves for canopy cover, height, VARI, and NDYI under each N rate in the 2019–2020 season. (b) Eight plot-based measurements were used to create the curves for canopy cover, canopy height, and VARI under each N rate in the 2020–2021 season.
Agronomy 14 01612 g005
Agronomy 14 01612 g006
Figure 6. Three-cluster analysis for NUE performance based on AirMeasurer-derived target traits (canopy cover, canopy height, VARI, and NDYI) based on the 54 lines to identify high-, medium-, and low-NUE groups under (a) N180 and (b) N270 treatments. All variable measurements were standardized by the use of Z-scores. Agglomerative hierarchical clustering with Euclidean distance calculation was applied. Performance of cluster at N180 for NAE (c) and NRE (d), and at N270 for NAE (e) and NRE (f), shows that Cluster II (blue) contains the high-NUE lines at both treatments, Cluster I (pink) represents medium-NUE, and Cluster III (orange) represents low-NUE lines.
Figure 6. Three-cluster analysis for NUE performance based on AirMeasurer-derived target traits (canopy cover, canopy height, VARI, and NDYI) based on the 54 lines to identify high-, medium-, and low-NUE groups under (a) N180 and (b) N270 treatments. All variable measurements were standardized by the use of Z-scores. Agglomerative hierarchical clustering with Euclidean distance calculation was applied. Performance of cluster at N180 for NAE (c) and NRE (d), and at N270 for NAE (e) and NRE (f), shows that Cluster II (blue) contains the high-NUE lines at both treatments, Cluster I (pink) represents medium-NUE, and Cluster III (orange) represents low-NUE lines.
Agronomy 14 01612 g006
Agronomy 14 01612 g007
Figure 7. Principal component plot comparing NUE predictive power of clusters produced from drone-derived data verses lab-derived data for each variety for each N regime. At the optimal N application (N180) Cluster II did not select for low-NUE varieties but at N270 (excessive N), selection of potentially high-NUE lines was not clear. Shapes represent high (circle), medium (triangle) and low (square) NUE based on lab data whilst colors represent the drone derived clustering for the same line. Black represents Cluster I the medium-NUE from drone. Orange represents Cluster II the high-NUE from drone. Blue represents Cluster III the low-NUE from drone.
Figure 7. Principal component plot comparing NUE predictive power of clusters produced from drone-derived data verses lab-derived data for each variety for each N regime. At the optimal N application (N180) Cluster II did not select for low-NUE varieties but at N270 (excessive N), selection of potentially high-NUE lines was not clear. Shapes represent high (circle), medium (triangle) and low (square) NUE based on lab data whilst colors represent the drone derived clustering for the same line. Black represents Cluster I the medium-NUE from drone. Orange represents Cluster II the high-NUE from drone. Blue represents Cluster III the low-NUE from drone.
Agronomy 14 01612 g007
Table 1. Overview of NUE-related traits collected in this study and the equations for their calculation.
Table 1. Overview of NUE-related traits collected in this study and the equations for their calculation.
PhenotypeAbbreviationTimingCalculation
Above-ground biomass at anthesis AGBaAnthesis A G B = W e i g h t   o f   L e a v e s + W e i g h t   o f   S t e m s + W e i g h t   o f   E a r s
Above-ground N content at anthesis ANCaAnthesis A N C = N i t r o g e n   c o n t e n t   o f   l e a v e s + N i t r o g e n   c o n t e n t   o f   s t e m s + N i t r o g e n   c o n t e n t   o f   r e p r o d u c t i v e   s t r u c t u r e s T o t a l   a b o v e − g r o u n d   b i o m a s s × 100
Vegetative organ nitrogen accumulation at anthesis VONAa Anthesis V O N A a = Nitrogen content of leaves at anthesis + Nitrogen content of stems at anthesis
Spike number SN Harvest S N = N u m b e r   o f   s p i k e s   p e r   m 2
Grain number per spike GN Harvest G N = a v e r a g e   n u m b e r   o f   g r a i n   p e r   s p i k e   f r o m   40   p l a n t s   p e r   p l o t
Thousand grain weight TWGHarvest T G W = W e i g h t   o f   g r a i n N u m b e r   o f   g r a i n × 1000
Grain yield GYHarvest G Y = w e i g h t   o f   s e e d   p e r   m 2   c a l c u l a t e d   m a n u a l l y   f r o m   2   m e t e r   l o n g   r o w s   f r o m   a   p l o t
Vegetative organ biomass VOBHarvest V O B = W e i g h t   o f   L e a v e s + W e i g h t   o f   S t e m s
Above ground biomass AGBHarvest A G B = W e i g h t   o f   L e a v e s + W e i g h t   o f   S t e m s + W e i g h t   o f   E a r s
Grain N content GNCHarvest G r a i n ,   C a l c u l a t e d   u s i n g   K h j d a h l   m e t h o d
Vegetative organ N content VONCHarvest S t e m s + L e a v e s ,   C a l c u l a t e d   u s i n g   K h j d a h l   m e t h o d
Above-ground N content ANCHarvest S t e m s + L e a v e s + E a r s ,   C a l c u l a t e d   u s i n g   K h j d a h l   m e t h o d
Grain nitrogen accumulation GNCHarvest G N C = G N C × G r a i n   w e i g h t
Vegetative organ nitrogen accumulation VONCHarvest V O N C = V O N C × V O B
Above-ground nitrogen accumulation AGNAHarvest A G N A = A N C × A G B
N translocation amount NTAHarvest N T A = V e g a t i v e   o r g a n   N   c o n t e n t   a t   a n t h e s i s − V e g a t i v e   o r g a n   N   c o n t e n t   a t   h a r v e s t + R e p r o d u c t i v e   o r g a n   N   c o n t e n t   a t   a n t h e s i s − R e p r o d u c t i v e   o r g a n   N   c o n t e n t   a t   h a r v e s t
N translocation efficiency NTEHarvest N T E = N T A V O N A a × 100
Contribution rate of N translocation amount to grain CNTAHarvest C N T A = N T A G N C × 100
Nitrogen agronomic use efficiency NAEHarvest N U E = G Y N 0 − G Y N x N   a p p l i e d × 10
Nitrogen recovery efficiencyNREHarvest N R E = A N C N 0 − A N C N x N   a p p l i e d × 10
Canopy coverCCHarvest C a l c u l a t e d   b y   A i r M e a s u r e r
Canopy heightCHHarvest C a l c u l a t e d   b y   A i r M e a s u r e r
Normalized Differential Yellowness IndexNDYIHarvest N D Y I = R e f l e c t a n c e ( G ) − R e f l e c t a n c e ( R ) R e f l e c t a n c e G + R e f l e c t a n c e ( R )
Visible Atmospherically Resistant IndexVARIHarvest V A R I = R e f l e c t a n c e ( G r e e n ) − R e f l e c t a n c e ( R e d ) R e f l e c t a n c e G r e e n + R e f l e c t a n c e R e d − R e f l e c t a n c e ( B l u e )
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.

Share and Cite

MDPI and ACS Style

Shen, L.; Deakin, G.; Ding, G.; Ali, M.; Dai, J.; Wen, Z.; Pinheiro, F.; Zhou, J.; Jackson, R. The Use of Low-Cost Drone and Multi-Trait Analysis to Identify High Nitrogen Use Lines for Wheat Improvement. Agronomy 2024, 14, 1612. https://doi.org/10.3390/agronomy14081612

AMA Style

Shen L, Deakin G, Ding G, Ali M, Dai J, Wen Z, Pinheiro F, Zhou J, Jackson R. The Use of Low-Cost Drone and Multi-Trait Analysis to Identify High Nitrogen Use Lines for Wheat Improvement. Agronomy. 2024; 14(8):1612. https://doi.org/10.3390/agronomy14081612

Chicago/Turabian Style

Shen, Liyan, Greg Deakin, Guohui Ding, Mujahid Ali, Jie Dai, Zhenjie Wen, Felipe Pinheiro, Ji Zhou, and Robert Jackson. 2024. "The Use of Low-Cost Drone and Multi-Trait Analysis to Identify High Nitrogen Use Lines for Wheat Improvement" Agronomy 14, no. 8: 1612. https://doi.org/10.3390/agronomy14081612

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop