Received: 19 March 2024 Accepted: 15 October 2024

DOI: 10.1002/ppj2.70007

ORIGINAL ARTICLE

Spatial analysis with unoccupied aircraft systems data in wheat

breeding yield trials

Andrew W. Herr1 Kimberly Garland Campbell2 Xianran Li2

Arron H. Carter1

1 Departmentof Crop and Soil Sciences,

Washington State University, Pullman, Abstract
Washington, USA An important aspect of reliable cultivar development is good field trial evaluations.
2 USDA–ARS, Wheat Health, Genetics, and Unoccupied aircraft systems (UAS also known as drones or UAVs) are a popular
Quality Research Unit, Pullman,
high-throughput phenotyping tool that has been used to successfully evaluate plant
Washington, USA
stress and other canopy characteristics in a field. In precision agriculture applications,
Correspondence UAS imagery has been used to identify spatial variability in field settings. Here, we
Arron H. Carter, Department of Crop and
Soil Sciences, Washington State University,
use UAS spectral imagery to improve field trial spatial analysis, better control spatial
Pullman, WA 99164-6420, USA. variability, and reduce errors for more reliable selections. UAS imagery data were
Email: [email protected] collected across 47 breeding trials planted in an augmented complete block design
Assigned to Associate Editor Mohsen (ACBD) or alpha-lattice replicated designs from 2020 through 2023. Trials were eval-
Yoosefzadeh-Najafabadi. uated using three spatial analysis strategies: linear models incorporating block effect,
row-column effect, or 2D splines. UAS-derived spectral reflectance indices (SRI)
Funding information
O.A. Vogel Research Endowment at were combined with each model as covariates. Modeling strategies were used across
Washington State University; National all trials and evaluated for autocorrelation, model fitness, and coefficient of variation
Institute of Food and Agriculture,
(CV). Akaike information criterion (AIC) was used to assess model fitness. For spa-
Grant/Award Numbers: 2022-67013-36426,
2022-68013-36439 tial analysis trials, SRIs significantly lowered model AIC by an average of 38.4 for
alpha-lattice trials and 69.1 for ACBD trials. CV scores were also lowered when SRIs
were utilized, with average CV values being 2.6 lower for alpha-lattice and 2.1 for
ACBD trials. This study highlights the potential for SRIs to improve the analyses of
field breeding trials despite extreme environmental variables and climate conditions,
improving experiment reliability and changing the way breeders plan and implement
breeding experiments.

1 INTRODUCTION

Plant breeding is an essential aspect of reliable agricultural

Abbreviations: %CC, percentage canopy cover; ACBD, augmented
complete block design; AIC, Akaike information criterion; AY, advanced production and improvement. Plant breeders work to maintain
yield; CV, coefficient of variation; HTP, high-throughput phenotyping; and increase the genetic output of cultivars, meeting market
NDRE, normalized difference red-edge; NDVI, normalized difference needs by developing and providing reliable trait characteris-
vegetation index; SRI, spectral reflectance index; UAS, unoccupied aircraft tics. Regardless of the crop of interest, breeding goals will
systems.

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided
the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2024 The Author(s). The Plant Phenome Journal published by Wiley Periodicals LLC on behalf of American Society of Agronomy and Crop Science Society of America.

The Plant Phenome J. 2024;7:e70007. wileyonlinelibrary.com/journal/ppj2 1 of 12

https://doi.org/10.1002/ppj2.70007
2 of 12 HERR ET AL.

focus on several target traits generally defined by grower and

consumer needs. Many of these traits are polygenic, making Core Ideas
selection more complex and resource intensive. One of the
most critical traits in this category is crop yield. The quanti- ∙ Unoccupied aircraft systems (UAS) index data can
tative nature and moderate heritability of crop yield requires account for spatial variation in field experiments.
extensive replicated multi-year field trials across target envi- ∙ UAS index data improves spatial model fit of
ronments (Bernardo, 2002). Environmental variability is a winter wheat yield trials.
significant factor that must be addressed when working with ∙ Including UAS indices in spatial modeling reduces
agricultural field experiments like those utilized by breed- the need for replicated trial design.
ers in crop yield evaluation (Smith et al., 2005). Spatial
variability introduced by soil heterogeneity, slope, elevation,
and management practices will reduce a breeder’s ability to
distinguish genetic effects from local environmental effects, and are reliable methods for dealing with unwanted field
limiting selection efficiency and overall genetic gain (Morales trial spatial heterogeneity (Borges et al., 2019; Hinkelmann,
et al., 2022; van Es & van Es, 1993). It is imperative to estab- 2012). Row and column effect, block effect, and nearest-
lish strategies to minimize the bias of treatment estimates neighbor correlations are popular statistical methods for
and maximize the reliability and efficiency of field-based spatial modeling that have been developed to account for and
breeding yield trial experiments (Stringer et al., 2012). minimize unwanted within-trial spatial effects (Boer et al.,
Local spatial variability in yield trial experiments is most 2020; Gilmour et al., 1997; Piepho & Williams, 2010; Piepho
often addressed through the experimental design, which cor- et al., 2008, 2015). Additionally, advanced spatial models,
rects for variability using randomization and replication of such as splines, can be utilized to model several covariates
experimental units. In plant breeding field trials, the plot is concurrently (Piepho et al., 2022). One such model focuses
the experimental unit, which is part of a larger block. The on P-Splines, a class of splines developed by Eilers and Marx
randomized complete block design, a popular design strategy, (1996). Recently, Rodríguez-Álvarez et al. (2018) highlighted
utilizes blocking and full replication of experimental units the utility of using two-dimensional P-Splines for spatial anal-
across blocks to minimize spatial variability within a block ysis of field trials by modeling covariance as a smooth spatial
and maximize it across blocks (Fisher et al., 1966; Piepho trend. New approaches, such as field-based high-throughput
et al., 2015; Stringer et al., 2012). phenotyping (HTP), can also be used to improve spatial anal-
However, homogeneity within blocks is challenging to ysis in breeding trials but have not been thoroughly explored.
maintain when the number of genotypes evaluated increases In the field of precision agriculture, however, field-based
(Brownie et al., 1993; R.-C. Yang et al., 2004). Alpha-lattice HTP strategies for remote sensing have become promising
designs are able to handle a large number of genotypes by and valuable in identifying spatial trends (Cohen et al., 2013;
utilizing unreplicated incomplete subblocks of a larger com- Geipel et al., 2014; Zhang & Kovacs, 2012).
plete replication to minimize experimental error (Patterson For field-based breeding applications, HTP can be imple-
& Thompson, 1971; R.-C. Yang et al., 2004). Resources mented using unoccupied aircraft systems (UAS), an ideal
are generally limited in the early stages of a breeding combination of scalability while maintaining data resolution
cycle, so replicated multi-location trials can be impractical. (Araus & Cairns, 2014; W. Guo, Carroll, et al., 2021; Herr &
Unreplicated augmented complete block designs (ACBDs) Carter, 2023; Reynolds et al., 2020). Breeders use UAS spec-
are commonly used when many genotypes need to be eval- tral data to provide trait information to aid in the selection
uated, and resources limit the replication of observations of critical traits through the capture of spectral reflectance
(Federer & Raghavarao, 1975). In an ACBD design, repli- indices (SRIs) to estimate routine traits of interest like plot
cated check plots are utilized systematically throughout the quality, biotic, and abiotic stresses (Costa et al., 2013; A.
trial in a block pattern and then augmented with unreplicated Guo, Huang, et al., 2021; Sankaran et al., 2015; Shakoor
test plots. et al., 2017; G. Yang et al., 2017). SRIs collected with UAS
The trial design is only part of the process of accounting provide standardized data focused on essential plant physio-
for spatial variability in yield trial experiments. Corrections logical characteristics by targeting important wavelengths of
through spatial statistical analysis can further increase data light associated with various components of the plant canopy
reliability (Borges et al., 2019). Without spatial analysis, (Myneni et al., 1995; Xue & Su, 2017). One of the early
reaching the desired level of precision in the dataset may be and most popular SRIs is normalized difference vegetation
challenging due to field-level spatial variation and autocorre- index (NDVI), which has been shown to accurately measure
lation present in the experiment (Piepho & Williams, 2010; photosynthetic capacity and overall plant stress in crops, espe-
Piepho et al., 2008). cially in wheat (Cabrera-Bosquet et al., 2011; Herr et al.,
Field trial design and spatial analysis strategies are well- 2023; Lozada et al., 2020; Rouse et al., 1974). Another pop-
established tools used in the field of plant breeding research ular vegetation index used in wheat research is normalized
HERR ET AL. 3 of 12

difference red-edge index (NDRE). NDRE is similar to NDVI Walla Walla, Farmington, and Prescott, WA. In this region,
in its ability to identify plant stress, but instead of focusing winter wheat is rotated annually with spring crops.
on red absorption, it evaluates red-edge absorption relative to This study utilized field trials from three of the main
NIR (Gitelson & Merzlyak, 1996). phases of large plot observation in the breeding program:
Spectral data collected from UAS have been utilized early generation, preliminary, and advanced yield (AY) tri-
in many different ways to improve the breeding strategy als. The early-generation trials of genotypes in a large plot
by improving the collection of commonly utilized traits setting were evaluated in ACBD trials derived from either F5
(Sankaran et al., 2015; Tirado et al., 2020), allowing for the (F5 single plots) or double haploid (DH single plots) lines.
collection of previously infeasible traits (Adak, Conrad, et al., These trials utilized a set of checks replicated throughout the
2021; Yin et al., 2022), fueling machine learning modeling trial, while test genotypes were a single observation. Due to
(Etienne et al., 2021; Ferreira et al., 2019), and aiding in seed limitations, one or two of these trials were evaluated
genomic selection (Crain et al., 2018; Rutkoski et al., 2016; annually in each production region. Genotypes selected from
Sun et al., 2019). Morales et al. (2022) provide promising the ACBD single-plot trials were advanced to a three-block,
results, where the use of NDVI enhanced the detection of three-replication alpha-lattice trial, denoted as a preliminary
spatial heterogeneity and genomic selection among inbred trial. Each year, preliminary yield trials were evaluated at one
and hybrid maize field evaluations over traditional methods. to two locations per production region. The final phase for
However, this study was more preliminary in nature and does breeder evaluation occurred in AY trials. Like preliminary tri-
not fully explore the utilization of trial design, spatial model, als, AY trials were designed in a three-block, three-replication
and SRI across a large set of years and experiments. Here, alpha-lattice. Trials denoted as AY1 were targeted for test-
we aimed to extensively test the potential of UAS-derived ing varieties in the lower rainfall region, whereas AY2 is an
SRI data utilized in different spatial models to reduce single- advanced trial targeted for the higher rainfall region. Depend-
location field variability across various locations and years for ing on the year, each AY trial was evaluated in three to five
use in a wheat breeding application. locations in the targeted production region. Plots in the lower
rainfall region were planted using a four-row deep furrow
planter at a seed density of 190 seeds m−2 . The higher rainfall
locations were planted using a double-disc, eight-row small
2 MATERIALS AND METHODS plot planter at a seed density of 220 seeds m−2 . The total plot
size for all locations was 1.5 m wide by 3.5 m long.
2.1 Study population

UAS spectral data and grain yield were collected from Wash- 2.2 Phenotypic data collection
ington State University soft white winter wheat (Triticum
aestivum L.) breeding trials from 2020 through 2022. Grain yield data for all trials was collected with a Zurn 150
Data were collected across 47 individual trial experiments harvester (Zurn Harvesting GmbH & Co. KG) equipped with
(trial/location/year) encompassing 8660 plots. Table S1 out- the Harvest Master Grain Gauge. UAS spectral data were
lines the study population attributes, including location, year, collected with a modified Sentera Quad Multispectral Sen-
trial design, number of unique genotypes, and number of plot sor (Sentera) consisting of four sensors covering eight target
observations. Experimental breeding trials were evaluated bands (450– 970 nm). These spectral bands were then used to
across nine locations ranging in average annual precipitation calculate the SRIs used in this study as outlined in Table 1,
from 257 mm (Lind, WA) to 518 mm (Pullman, WA). Across including their abbreviation, equation, and reference. Of the
years, the trials examined experienced extreme environmental many SRIs available, the ones outlined in Table 1 were chosen
variability. In Pullman, WA, for example, total precipita- because of their historic success in our breeding program and
tion during the key growing season (May–July) ranged from in general wheat research (Gizaw et al., 2016a, 2016b; Herr
20 mm in 2021 to 155 mm in 2022. et al., 2024; Sankaran et al., 2015).
In Washington State, winter wheat production is generally An example of how these SRI phenotypic traits look when
divided into two agronomic production regions. The first is a displayed in a trial layout can be seen in Figure 1. For data col-
low-rainfall wheat fallow region (average annual precipitation lection, the UAS was flown at an approximate altitude of 45 m
of 125–400 mm) consisting of trials in the towns of Daven- with an 85% bidirectional overlap of georeferenced images.
port, Lind, Ritzville, Harrington, and Kahlotus, WA. In this Flights were conducted at anthesis due to the established rela-
region, the wheat crop is planted every other year into stored tionship between reflectance data collected at that maturity
soil moisture retained from the fallow year. The second target and grain yield (Duan et al., 2017; Lozada et al., 2020). UAS
production region has a higher annual rainfall (average annual flights were only conducted within a 5-h window of solar noon
precipitation of 400–660 mm), consisting of trials in Pullman, and on days with clear skies.
4 of 12 HERR ET AL.

TA B L E 1 Spectral reflectance index equations.

Spectral reflectance index Abbreviation Equation Reference

800 nm − 680 nm
Normalized difference vegetation NDVI 800 nm + 680 nm
(Rouse et al., 1974)
index
800 nm − 710 nm
Normalized difference red-edge NDRE 800 nm + 710 nm
(Gitelson & Merzlyak, 1996)
1 ∑𝑁 800 nm − 560 nm ∑𝑁
Percentage canopy cover %CC 𝑁 𝑖=1 ( 800 nm + 560 nm )𝑖 𝑖=1 𝐺𝑁𝐷𝑉 𝐼𝑖 (Sankaran et al., 2015)

F I G U R E 1 Heat map of average plot values for (A) NDVI, (B) grain yield in kilograms per hectare, and (C) percentage of plot canopy cover in
the 2021 Pullman F5 single plot trial. This figure highlights the potential patterns of spatial variability that can be experienced in a single field
experiment.

2.3 Phenotypic data processing and analysis (2015), where soil masking was skipped in order to iden-
tify the percentage of pixels in a plot area that are plant
Orthomosaic reflectance images were created from each sen- canopy.
sor per location and flight using Pix4Dmapper (Pix4D Inc.).
After stitching, images were georeferenced, and individual
plots were identified using geographic information system 2.4 Spatial modeling and analysis
(QGIS). Image standardization, soil masking, index calcula-
tion, and mean plot data extraction of all spectral data were The study population consists of 47 field trials in which geno-
performed utilizing the R packages “FIELDimageR” (Matias types were evaluated in an experimental unit (plot) in one of
et al., 2020) and “raster” (Hijmans et al., 2015). Image stan- two experimental designs: alpha-lattice or ACBD. These field
dardization across flights was done using a set of target panels trials were used to compare three spatial analysis strategies
(five panels ranging from 2% to 85% reflectance, MosaicMill with and without SRI covariates.
Oy). Using these panels, all eight raw band layers collected The control method for this study was to utilize linear mixed
were adjusted based on the relationship as follows: models and the trial design alone to account for spatial vari-
ability. This will be referred to as the “Block Effect” method.
SR = 𝑎 × CR + 𝑏 (1) For ACBD trials, the block effect was adjusted using the
model:
where the slope (a) and intercept (b) are based on the regres-
sion of the observed reflectance in target panels, collected 𝑌𝑖𝑗𝑘 = Gen𝑖(𝑗) + Block𝑗 + Check𝑘 + 𝜀𝑖𝑗𝑘 (2)
reflectance (CR) is the raw observed pixel values, and sur-
face reflectance (SR) is the true reflectance value (Iqbal that Yijk is the phenotypic value for grain yield of the ith geno-
et al., 2018). The methodology for calculating the percent- type in the jth block; Geni(j) is the random effect of genotype
age canopy cover differed from NDVI and NDRE and more i in the jth block; Blockj is the random effect of the jth block;
closely followed the methods outlined by Sankaran et al. Checkk is the fixed effect of the kth replicated cultivar check;
HERR ET AL. 5 of 12

and εijk is the residual error. For alpha-lattice trials, the block where Yhijklm is the phenotypic value for grain yield; ƒ(uh , vh )
effect was adjusted for with the model: is the P-spline framework denoted by row and column position
uh and vh , respectively; Geni is the fixed effect of genotype
𝑌𝑖𝑗𝑘 = Gen𝑖 + Block𝑗 + Sub𝑘(𝑗) + 𝜀𝑖𝑗𝑘 (3) i; Blockj is the random effect of the jth block; Subk(j) is the
random effect of the sub-block k in the jth block; Rowl is the
where Yijk is the phenotypic value for grain yield of the ith random effect of the lth row; Colm is the random effect of the
genotype and the jth block; Geni is the fixed effect of genotype mth column; and εhijklm is the residual error.
i; Blockj is the random effect of the jth block; Subk(j) is the Each of these models, denoted in Equations (2–7), was
random effect of the sub-block k in the jth block; and εijk is evaluated alone and with the inclusion of an SRI as a fixed
the residual error. effect. Each model was evaluated with NDVI, NDRE, and
The second method for spatial adjustments used in this percentage canopy cover (%CC). Each iteration of the model
study is based on the block effect models shown in Equa- is outlined in Table 2. All models were generated using R’s
tions (2) and (3) but will incorporate row and column of plots “sommer” package (Covarrubias-Pazaran, 2016).
as random effects in the model. These will be referred to as the The first statistical method used to evaluate the validity
“row-column effect” method. For ACBD trials, this analysis of using SRIs in spatial analysis models was to conduct a
is represented as: Moran I test to identify the different spatial analyses that
best reduced trial autocorrelation. The Moran I test utilizes
𝑌𝑖𝑗𝑘𝑙𝑚 = Gen𝑖(𝑗) + Block𝑗 + Check𝑘 + Row𝑙 + Col𝑚 + 𝜀𝑖𝑗𝑘𝑙𝑚 model residuals and plot position to determine overall exper-
(4) iment spatial observation trends and distribution (clustered,
where Rowl is a random effect of the lth row and Colm is random, or dispersed) (Arlinghaus, 2020). Models were com-
the random effect of the mth column added to Equation (2). pared for model fitness using the trial data from which they
Similarly, in alpha-lattice trials, the row-column effect is were generated. With the structure of the utilized population
incorporated as: and nature of the spatial model, the most appropriate method
for comparison of models is in model fitness. Akaike infor-
𝑌𝑖𝑗𝑘𝑙𝑚 = Gen𝑖 + Block𝑗 + Sub𝑘(𝑗) + Row𝑙 + Col𝑚 + 𝜀𝑖𝑗𝑘𝑙𝑚 mation criterion (AIC) and root mean square error (RMSE)
(5) were chosen and calculated for each model as a function
where Rowl is the random effect of the lth row and Colm is associated with the model output using the R package “som-
the random effect of the mth column added to Equation (3). mer” (Covarrubias-Pazaran, 2016). AIC was chosen over
The final method utilized for spatial analysis is 2D other methods (Bayesian information criterion or Log Likeli-
splines. Two-dimensional anisotropic tensor product P- hood) because of its ability to deal with and compare models
splines can create a spatial field indicating spatial dependence containing a different number of parameters while not bias-
(Rodríguez-Álvarez et al., 2018). This spatial trend can be ing against model complexity yet still avoiding overfitting
used in conjunction with standard mixed model effects. For (Chakrabarti & Ghosh, 2011). Another method used in the
ACBD trials, the model is written as: comparison of spatial model success was the calculation of
( ) the coefficient of variation (CV) based on the model-adjusted
𝑌ℎ𝑖𝑗𝑘𝑙𝑚 = 𝑓 𝑢ℎ , 𝑣ℎ + Gen𝑖(𝑗) + Block𝑗 + Check𝑘 + Row𝑙 standard deviation of grain yield divided by the mean trial
yield:
+ Col𝑚 + 𝜀ℎ𝑖𝑗𝑘𝑙𝑚 (6)
𝜎
CV = (8)
where Yhijklm is the phenotypic value for grain yield; ƒ(uh , vh ) 𝜇
is the P-spline framework denoted by row and column posi-
tion uh and vh , respectively; Blockj is the random effect of the where CV is the coefficient of variation, σ is the standard devi-
jth block; Checkk is the fixed effect of the kth replicated cul- ation of the experiment grain yield, and μ is the experiment
tivar check; Geni(j) is the random effect of genotype i in the mean grain yield (Everitt, 1998). CV as a statistical measure
ith check; Rowl is the random effect of the lth row; Colm is does have some drawbacks, as will be addressed in the discus-
the random effect of the mth column; and εhijklm is the resid- sion. Nonetheless, CV is a popular method for determining
ual error. More information on how the P-Spline framework the reliability and overall precision of the data produced in
is modeled can be found in Rodríguez-Álvarez et al. (2018). a field experiment and has, therefore, been included as a
Alpha-lattice trials are similarly modeled as: statistical measure evaluating model performance (Fasahat
et al., 2015).
( ) Models AIC and CV were compared for statistical differ-
𝑌ℎ𝑖𝑗𝑘𝑙𝑚 = 𝑓 𝑢ℎ , 𝑣ℎ + Gen𝑖 + Block𝑗 + Sub𝑘(𝑗) + Row𝑙
ences among both alpha-lattice and ACBD designed exper-
+ Col𝑚 + 𝜀ℎ𝑖𝑗𝑘𝑙𝑚 (7) iments. Statistical difference was evaluated using Tukey’s
honest statistical difference (HSD) test (Tukey, 1949).
6 of 12 HERR ET AL.

3 RESULTS 3.2 Model fit

A unique factor that plays a significant role in the results of Evaluation of model fit was the best way to identify the mod-
this study were the years and growing conditions evaluated. eling strategy that works in a particular scenario and with a
In 2021, growing conditions were hot, with below-average specific dataset (Chakrabarti & Ghosh, 2011). Spatial analysis
precipitation. In 2020 and 2022, growing conditions led to was conducted on 47 trials using models outlined in Equa-
less plant stress with cooler temperatures and above-average tions (2–7), both with and without SRIs (NDVI, NDRE, and
precipitation. For example, during the key growing season %CC) as covariates. These models were then evaluated for
(May–July) in Pullman, WA, the average daily temperature model fitness by calculating AIC. The distribution of AIC of
was 14.6˚C in 2020, 17.8˚C in 2021, and 14.7˚C in 2022. the models is found in Figure 3, where a lower AIC indicated
Similarly, total precipitation in this growing window was a better model fit.
113 mm in 2020, 20 mm in 2021, and 155 mm in 2022. As The alpha-lattice produced lower AIC values than the com-
expected, this weather variation impacted crop performance parable location ACBD trials, with an overall average AIC
across years dramatically. In the Pullman AY2 trial, the aver- of 100.93 and 132.14, respectively, across modeling strate-
age grain yield was 11,184 kg ha−1 in 2020, 5138 kg ha−1 gies with and without SRI variables. This trend was expected
in 2021, and 9375 kg ha−1 in 2022. As will be discussed due to the replication of genotypes in the more robust alpha-
later, the environmental variation found in this study popula- lattice design. However, comparisons of these designs were
tion across years substantially impacts how results should be limited due to a differing set of genotypes within the experi-
interpreted. It highlights the inconsistency or stability of the ments. The robustness of the alpha-lattice was also shown in
methodologies tested across diverse experimental conditions. the limited ability of SRIs to improve model fitness for tri-
als over the control models. SRIs, however, have a significant
effect on model fit when used for ACBD trials. This was most
3.1 Autocorrelation apparent in the ability of SRIs to minimize poor outlier AIC
scores in augmented and alpha lattice trials regardless of the
After spatial analysis, trials were evaluated for a clustering base model used, an indication of improved model reliability
effect of the adjusted yield. The distribution of the Moran I and overall robustness. This is well outlined when looking at
statistic across the study population highlights the minimal the tails of the box and whisker plots in Figure 3. Identify-
impact of spatial modeling (Figure S1). However, evaluat- ing and decreasing the influence of outliers is critical when
ing autocorrelation does highlight the impact trial design and trait data are used in selection indices, especially genomic
yearly environmental effects can have on the correlation of selection indices. Among SRIs, there was little difference in
neighboring plot values within an experimental trial. Figure model fitness. Each index evaluated improved overall model
2B shows the variability year to year and within experiments fitness similarly regardless of base model or trial design. As
with a mean trial grain yield in 2020, 2021, and 2022 of 7519, shown in Table S3, the best spatial modeling strategy aver-
3791, and 6335 kg ha−1 , respectively. Additionally, trials pro- aged across this study population was the row-column effect
duced an average standard deviation of 908, 489, and 876 in model with NDRE as a covariate with an AIC average of 73.74
2020, 2021, and 2022, respectively. The distribution of the and 88.27 for alpha-lattice and ACBD, respectively. The poor-
Moran I statistic for all trials and model iterations, as shown est fitting models were the block-effect control models, where
in Figure 2, was centered around 0, indicating that spatial alpha-lattice had an average AIC of 141.07 and ACBD with
analysis maintained a complete random spatial distribution. 214.92. Table S3 also highlights this pattern of model fitness
Omitting some 2021 outliers with Moran I’s greater than 0.5, with RMSE.
alpha-lattice produced autocorrelation values that point to no
clustering within the trial datasets, with most having a value
lower than 0. The alpha-lattice outliers produced were from 3.3 Coefficient of variation
the 2021 growing season, which saw unprecedented water
stress in many locations, as highlighted in mean grain yield. CV followed a similar trend to that found with model fitness,
In general, 2021 saw a shift in trial values that indicated more where alpha-lattice trials produce a more optimal average CV
clustering and spatial correlation of plots, which was unsur- value across all models (10.38) than the comparable models
prising considering the year’s growing conditions. Moran for ACBD trials (14.02), again this comparison was limited
distributions centered around 0 for 2020 and 2022 growing by the difference in genotypes utilized in each respective trial
seasons with few water stress events. design. It is important to note that, like with AIC, a lower
HERR ET AL. 7 of 12

F I G U R E 2 Distribution of trial autocorrelation by (A) trial design and (B) year. A Moran I value of 1 indicates plots are entirely clustered
(high correlations present), 0 indicates a completely random distribution, and −1 indicates no correlations with a perfect distribution pattern. ACBD,
augmented complete block design.

CV score is preferred. As seen in Figure 4, the difference 1994; Parmley et al., 2019; Xie & Yang, 2020), and biotic
in CV performance did not change between alpha-lattice and stress (Dammer et al., 2011; A. Guo, Huang, et al., 2021;
ACBD in the same model strategy, even when utilizing SRIs. Tetila et al., 2017), have been well-established as observations
However, SRIs lowered CV scores by an average of 2.55 for that can be captured by a UAS and implemented in breeding
alpha-lattice and 2.10 for ACBD trials. As was expected, the strategy. Outside of breeding research, UAS are more com-
poorest average CV was produced by the Block Effect con- monly utilized for the identification of spatial trends in fields
trol model, with a CV of 12.69 for alpha-lattice and 15.77 like urban green space development (Moreno-Armendáriz
for ACBD. Dissimilar to AIC, the best improvement in aver- et al., 2019), forest management (Lee et al., 2016; Minařík &
age CV for alpha-lattice (8.84) was using the 2D Spline with Langhammer, 2016), and field scale agronomics (Cohen et al.,
NDVI as a covariate, while for ACBD (12.73), the 2D Spline 2013; Geipel et al., 2014; Han et al., 2019). Han et al. (2019)
with NDRE had the lowest average CV. and Zhang and Kovacs (2012) highlight a popular use for
CV works well to highlight the improvement SRI-enabled remote sensing technology in field-level agronomic research.
spatial modeling can have on the data collected. The trends The study utilizes satellite data from a region of China to build
found in the distribution of model CV match those found with crop models that utilize canopy cover, water management,
AIC and RMSE, as highlighted in Table S3. SRIs improved a and relative biomass, capturing spatial trends to improve a
trial’s uniformity and reduced deviation around the trial mean, standard crop prediction model.
creating a better-fit model that produced more optimal, lower Similar analysis with UAS is lacking for plot-level breed-
AIC and CV values. ing research, which is surprising considering the importance
of identifying spatial variability to ensure data quality from
field experiments. Morales et al. (2022), in a short preprint,
4 DISCUSSION highlighted the potential for UAS spatial analysis to improve
breeding and genomic selection. However, regarding practi-
The introduction of UAS research in plant breeding has cre- cal breeding applications, it was limited to a narrow set of
ated many new avenues for efficient data collection as well as trial designs, environments, as well as population diversity
increased the number of traits available to breeders in selec- and size.
tion (Herr et al., 2023). Important traits, such as emergence This study aimed to evaluate these complicating factors
(Chen et al., 2018; Sankaran et al., 2015), plant height (Adak, more comprehensively by looking at the validity of utilizing
Murray, et al., 2021; Anthony et al., 2014), maturity (McFar- UAS-derived SRI data to improve spatial models for the eval-
land et al., 2020; Xu et al., 2018), plant health (Munden et al., uation of grain yield in individual field experiments across
8 of 12 HERR ET AL.

F I G U R E 3 Comparison of the distribution of model fit across individual experiments for augmented complete block design (ACBD) trials and
alpha-lattice trials relative to spatial model strategy. The letters above bars indicate significant differences amongst models evaluating the same trial
design using Tukey’s honest statistical difference (HSD). AIC, akaike information criterion; %CC, percentage canopy cover; NDRE, normalized
difference red-edge; NDVI, normalized difference vegetation index.

F I G U R E 4 Comparison of the coefficient of variation (CV) distribution across individual experiments for (A) augmented complete block
design (ACBD) trials and (B) alpha-lattice trials utilizing different spatial model strategies The letters above bars indicate significant differences
amongst models evaluating the same trial design using Tukey’s honest statistical difference (HSD). AIC, akaike information criterion; %CC,
percentage canopy cover; NDRE, normalized difference red-edge; NDVI, normalized difference vegetation index.
HERR ET AL. 9 of 12

an extensive set of locations and years. All SRIs evaluated, the cost to the overall breeding effort while improving breeder
NDVI, NDRE, and %CC, were found to enhance average efficiency.
experiment model fitness for trial design (Block-Effect), row- The novel nature of this technology and methodology does
column effect, and 2D Spline model trends. These results were leave room for more questions to be answered. A concern
found amidst experiments that were evaluated in a diversity of these methods present is the introduction of bias from SRI
environments and growing conditions. SRIs improved model values in selecting cultivars. Besides the potential for biasing
fitness and accounted for more significant amounts of spatial field experiment data, consideration would be needed before
variability in both alpha-lattice and ACBD trials. ACBD trials comparing SRI spatially adjusted field data, like yield, with
saw much more significant gains in model performance from SRI collected traits. For example, utilizing SRIs in both spa-
SRIs. tial analysis and genomic selection introduces a bias that will
The study population evaluated was made up of trials uti- skew genomic predictions. Another area of future research
lizing one of two experiment designs: alpha-lattice or ACBD. could be identifying the more valuable place for SRI data in
These trials were observed over three extreme environmen- that pipeline, aiding genomic selection or spatial adjustment
tal years, testing both water stress tolerance and limits of models.
yield potential under ideal growing conditions. Because of This study highlights the powerful potential of SRI to
this, the trials evaluated have distinct characteristics and chal- improve the overall efficiency of field trial design and analysis
lenges that must be dealt with in spatial analysis. In 2021, a in a breeding program, providing decision-makers with more
severe drought year, the trials exhibited increased autocorre- reliable field trial datasets. While the results of this study are
lation and clustering, indicating potential patterns of spatial promising, it is worth further exploring how indices utilized in
variability within experimental trials. Less spatial clustering spatial modeling impact phenotypic values and genotype rank.
was observed in the more ideal growing seasons of 2020 and It is feasible that the improved models, while reducing spatial
2022. The diverse study population highlights the plasticity variation, could introduce unwanted bias into the cultivar data
of the spatial modeling strategies utilized. Despite extreme utilized in breeder selections. Before implementation of this
variation year to year and across geography, spatial analysis methodology at scale, more research needs to be done assess-
utilizing SRIs produces more reliable experimental data. ing how these spatial models impact selection results and
The design of a field trial and how it is analyzed play a the introduction of potential bias created by using a specific
significant role in the quality and reliability of the data it SRI.
produces. The reality of field experimentation is that spatial
variability can be present and introduce errors and out- AU T H O R C O N T R I B U T I O N S
liers into the trial trait data. Trial design and, secondarily, Andrew Herr: Conceptualization; data curation; formal
spatial analysis work to mitigate this error (Borges et al., analysis; investigation; methodology; project administration;
2019; Gilmour et al., 1997; Stringer et al., 2012). Tradition- visualization; writing—original draft; writing—review and
ally, trial design and spatial analysis have provided breeders editing. Kimberly Garland-Campbell: Conceptualization;
with increased confidence that collected trait data represents writing—review and editing. Xianran Li: Conceptualiza-
genetic potential. However, there is still room for improve- tion; writing—review and editing. Arron H. Carter: Con-
ment in how breeders deal with spatial variation and error in ceptualization; funding acquisition; resources; supervision;
their field experiments. UAS-collected SRIs have a unique writing—review and editing.
ability to capture plant stress and other trial characteristics
like canopy cover, which, as outlined in this study, shows AC K N OW L E D G M E N T S
great potential as a tool in a breeder’s effort to mitigate field The research was partially supported by the Agriculture and
variation. Food Research Initiative Competitive Grant awards 2022-
While UAS-collected spectral data are considered valuable 67013-36426 and 2022-68013-36439 (WheatCAP) from the
for many agricultural sectors, including plant breeding, it is National Institute of Food and Agriculture, Hatch project
still expensive and requires specialized labor and equipment 1014919, and the O.A. Vogel Research Endowment at Wash-
(Lachowiec et al., 2024). It is important to find as many uses ington State University. Thank you to Dr. Julia Piaskowski,
as possible to maximize the value of the data collected by who ran the workshop “Incorporating Spatial Analysis into
UAS, maximizing overall utility and resource efficiency. The Agricultural Field Experiments” (https://idahoagstats.github.
strategies outlined in this research are designed to be uti- io/guide-to-field-trial-spatial-analysis/), demonstrating many
lized by breeders already collecting UAS data for other traits of the statistical methods utilized in this study.
of interest. Spectral data captured within the breeding pro-
gram for other traits, used in spatial analysis and adjustment, C O N F L I C T O F I N T E R E ST STAT E M E N T
increases the value and viability of the resource and justifies The authors declare no conflicts of interest.
10 of 12 HERR ET AL.

D A T A AVA I L A B I L I T Y S T A T E M E N T Chakrabarti, A., & Ghosh, J. K. (2011). AIC, BIC and recent advances
All code and data used in the study can be found at https:// in model selection. In P. S. Bandyopadhyay & M. R. Forster (Eds.),
github.com/AW-Herr/Large-scale-breeding-applications-of- Philosophy of statistics (pp. 583–605). Elsevier.
Chen, R., Chu, T., Landivar, J. A., Yang, C., & Maeda, M. M.
UAS-enabled-genomic-prediction.
(2018). Monitoring cotton (Gossypium hirsutum L.) germination
using ultrahigh-resolution UAS images. Precision Agriculture, 19(1),
ORCID 161–177. https://doi.org/10.1007/s11119-017-9508-7
Andrew W. Herr https://orcid.org/0000-0001-5111-2342 Cohen, S., Cohen, Y., Alchanatis, V., & Levi, O. (2013). Combining
Kimberly Garland Campbell https://orcid.org/0000-0003- spectral and spatial information from aerial hyperspectral images for
4747-3270 delineating homogenous management zones. Biosystems Engineer-
Xianran Li https://orcid.org/0000-0002-4252-6911 ing, 114(4), 435–443. https://doi.org/10.1016/j.biosystemseng.2012.
Arron H. Carter https://orcid.org/0000-0002-8019-6554 09.003
Costa, J. M., Grant, O. M., & Chaves, M. M. (2013). Thermography
to explore plant–environment interactions. Journal of Experimental
REFERENCES Botany, 64(13), 3937–3949. https://doi.org/10.1093/jxb/ert029
Adak, A., Conrad, C., Chen, Y., Wilde, S. C., Murray, S. C., Anderson Covarrubias-Pazaran, G. (2016). Genome-assisted prediction of quanti-
II, S. L., & Subramanian, N. K. (2021). Validation of functional poly- tative traits using the R package sommer. PLoS One, 11(6), e0156744.
morphisms affecting maize plant height by unoccupied aerial systems http://doi.org/10.1371/journal.pone.0156744
discovers novel temporal phenotypes. G3 Genes|Genomes|Genetics, Crain, J., Mondal, S., Rutkoski, J., Singh, R. P., & Poland, J.
11(6), jkab075. https://doi.org/10.1093/g3journal/jkab075 (2018). Combining high-throughput phenotyping and genomic infor-
Adak, A., Murray, S. C., Božinović, S., Lindsey, R., Nakasagga, S., mation to increase prediction and selection accuracy in wheat
Chatterjee, S., Anderson, S. L., & Wilde, S. (2021). Temporal veg- breeding. The Plant Genome, 11(1), 170043. http://doi.org/10.3835/
etation indices and plant height from remotely sensed imagery can plantgenome2017.05.0043
predict grain yield and flowering time breeding value in maize via Dammer, K.-H., Möller, B., Rodemann, B., & Heppner, D. (2011).
machine learning regression. Remote Sensing, 13(11), 2141. https:// Detection of head blight (Fusarium ssp.) in winter wheat by color
doi.org/10.3390/rs13112141 and multispectral image analyses. Crop Protection, 30(4), 420–428.
Anthony, D., Elbaum, S., Lorenz, A., & Detweiler, C. (2014). On crop https://doi.org/10.1016/j.cropro.2010.12.015
height estimation with UAVs. IEEE. dos Santos Ferreira, A., Freitas, D. M., da Silva, G. G., Pistori, H., &
Araus, J. L., & Cairns, J. E. (2014). Field high-throughput phenotyping: Folhes, M. T. (2019). Unsupervised deep learning and semi-automatic
The new crop breeding frontier. Trends in Plant Science, 19(1), 52–61. data labeling in weed discrimination. Computers and Electronics
https://doi.org/10.1016/j.tplants.2013.09.008 in Agriculture, 165, 104963. https://doi.org/10.1016/j.compag.2019.
Arlinghaus, S. (Ed.). (2020). Practical handbook of spatial statistics. 104963
CRC Press. Duan, T., Chapman, S. C., Guo, Y., & Zheng, B. (2017). Dynamic mon-
Bernardo, R. (2002). Breeding for quantitative traits in plants. Stemma itoring of NDVI in wheat agronomy and breeding trials using an
Press Woodbury. unmanned aerial vehicle. Field Crops Research, 210, 71–80. https://
Boer, M. P., Piepho, H.-P., & Williams, E. R. (2020). Linear vari- doi.org/10.1016/j.fcr.2017.05.025
ance, P-splines and neighbour differences for spatial adjustment in Eilers, P. H. C., & Marx, B. D. (1996). Flexible smoothing with B-splines
field trials: How are they related? Journal of Agricultural, Biologi- and penalties. Statistical Science, 11(2), 89–121. https://doi.org/10.
cal and Environmental Statistics, 25(4), 676–698. https://doi.org/10. 1214/ss/1038425655
1007/s13253-020-00412-4 Etienne, A., Ahmad, A., Aggarwal, V., & Saraswat, D. (2021). Deep
Borges, A., González-Reymundez, A., Ernst, O., Cadenazzi, M., Terra, learning-based object detection system for identifying weeds using
J., & Gutiérrez, L. (2019). Can spatial modeling substitute for exper- UAS imagery. Remote Sensing, 13(24), 5182. https://doi.org/10.3390/
imental design in agricultural experiments? Crop Science, 59(1), rs13245182
44–53. https://doi.org/10.2135/cropsci2018.03.0177 Everitt, B. (1998). The Cambridge dictionary of statistics. Cambridge
Brownie, C., Bowman, D. T., & Burton, J. W. (1993). Estimating spatial University Press.
variation in analysis of data from yield trials: A comparison of meth- Fasahat, P., Rajabi, A., Mahmoudi, S. B., Noghabi, M. A., & Rad, J.
ods. Agronomy Journal, 85(6), 1244–1253. https://doi.org/10.2134/ M. (2015). An overview on the use of stability parameters in plant
agronj1993.00021962008500060028x breeding. Biometrics & Biostatistics International Journal, 2(5), 149–
Cabrera-Bosquet, L., Molero, G., Stellacci, A., Bort, J., Nogués, S., & 159.
Araus, J. (2011). NDVI as a potential tool for predicting biomass, Federer, W. T., & Raghavarao, D. (1975). On augmented designs.
plant nitrogen content and growth in wheat genotypes subjected Biometrics, 31(1), 29–35. https://doi.org/10.2307/2529707
to different water and nitrogen conditions. Cereal Research Com- Fisher, R. A., Genetiker, S., & Genetician, S. (1966). The design of
munications, 39(1), 147–159. https://doi.org/10.1556/crc.39.2011.1. experiments. Oliver and Boyd Edinburgh.
15 Geipel, J., Link, J., & Claupein, W. (2014). Combined spectral and spa-
Castelao Tetila, E., Brandoli Machado, B., Belete, N. A., Guimaraes, D. tial modeling of corn yield based on aerial images and crop surface
A., & Pistori, H. (2017). Identification of soybean foliar diseases using models acquired with an unmanned aircraft system. Remote Sensing,
unmanned aerial vehicle images. IEEE Geoscience and Remote Sens- 6(11), 10335–10355. https://doi.org/10.3390/rs61110335
ing Letters, 14(12), 2190–2194. http://doi.org/10.1109/LGRS.2017. Gilmour, A. R., Cullis, B. R., Verbyla, A. P., & Verbyla, A. P. (1997).
2743715 Accounting for natural and extraneous variation in the analysis of field
HERR ET AL. 11 of 12

experiments. Journal of Agricultural, Biological, and Environmental Lozada, D. N., Godoy, J. V., Ward, B. P., & Carter, A. H. (2020).
Statistics, 2(3), 269–293. https://doi.org/10.2307/1400446 Genomic prediction and indirect selection for grain yield in US Pacific
Gitelson, A. A., & Merzlyak, M. N. (1996). Signature analysis of leaf Northwest winter wheat using spectral reflectance indices from
reflectance spectra: Algorithm development for remote sensing of high-throughput phenotyping. International Journal of Molecular
chlorophyll. Journal of Plant Physiology, 148(3), 494–500. https:// Sciences, 21(1), 165. https://doi.org/10.3390/ijms21010165
doi.org/10.1016/S0176-1617(96)80284-7 Matias, F. I., Caraza-Harter, M. V., & Endelman, J. B. (2020). FIELDim-
Gizaw, S. A., Garland-Campbell, K., & Carter, A. H. (2016a). Evalua- ageR: An R package to analyze orthomosaic images from agricultural
tion of agronomic traits and spectral reflectance in Pacific Northwest field trials. The Plant Phenome Journal, 3(1), e20005. https://doi.org/
winter wheat under rain-fed and irrigated conditions. Field Crops 10.1002/ppj2.20005
Research, 196, 168–179. https://doi.org/10.1016/j.fcr.2016.06.018 Mcfarland, B. A., Alkhalifah, N., Bohn, M., Bubert, J., Buckler, E.
Gizaw, S. A., Garland-Campbell, K., & Carter, A. H. (2016b). Use of S., Ciampitti, I., Edwards, J., Ertl, D., Gage, J. L., Falcon, C. M.,
spectral reflectance for indirect selection of yield potential and sta- Flint-Garcia, S., Gore, M. A., Graham, C., Hirsch, C. N., Holland,
bility in Pacific Northwest winter wheat. Field Crops Research, 196, J. B., Hood, E., Hooker, D., Jarquin, D., Kaeppler, S. M., . . . De
199–206. https://doi.org/10.1016/j.fcr.2016.06.022 Leon, N. (2020). Maize genomes to fields (G2F): 2014–2017 field
Guo, A., Huang, W., Dong, Y., Ye, H., Ma, H., Liu, B., Wu, W., Ren, Y., seasons: Genotype, phenotype, climatic, soil, and inbred ear image
Ruan, C., & Geng, Y. (2021). Wheat yellow rust detection using UAV- datasets. BMC Research Notes, 13(1), Article 71. http://doi.org/10.
based hyperspectral technology. Remote Sensing, 13(1), 123. https:// 1186/s13104-020-4922-8
doi.org/10.3390/rs13010123 Minařík, R., & Langhammer, J. (2016). Use of a multispectral UAV
Guo, W., Carroll, M. E., Singh, A., Swetnam, T. L., Merchant, N., photogrammetry for detection and tracking of forest disturbance
Sarkar, S., Singh, A. K., & Ganapathysubramanian, B. (2021). UAS- dynamics. The international archives of the photogrammetry. Remote
based plant phenotyping for research and breeding applications. Plant Sensing and Spatial Information Sciences, 41, 711–718.
Phenomics, 2021, Article 9840192. https://doi.org/10.34133/2021/ Morales, N., Gore, M. A., Mueller, L. A., & Robbins, K. R. (2022).
9840192 Spatio-temporal modeling of high-throughput multi-spectral images
Han, C., Zhang, B., Chen, H., Wei, Z., & Liu, Y. (2019). Spatially dis- improves agronomic trait genomic prediction in hybrid maize.
tributed crop model based on remote sensing. Agricultural Water Authorea Preprints.
Management, 218, 165–173. https://doi.org/10.1016/j.agwat.2019. Moreno-Armendáriz, M., Calvo, H., Duchanoy, C., López-Juárez, A.,
03.035 Vargas-Monroy, I., & Suarez-Castañon, M. (2019). Deep green diag-
Herr, A. W., & Carter, A. H. (2023). Remote sensing continuity: A nostics: Urban green space analysis using deep learning and drone
comparison of HTP platforms and potential challenges with field images. Sensors, 19(23), 5287. http://doi.org/10.3390/s19235287
applications. Frontiers in Plant Science, 14, 1233892. https://www. Munden, R., Curran, P., & Catt, J. (1994). The relationship between red
frontiersin.org/articles/10.3389/fpls.2023.1233892 edge and chlorophyll concentration in the Broadbalk winter wheat
Herr, A. W., Schmuker, P., & Carter, A. H. (2024). Large-scale breeding experiment at Rothamsted. Remote Sensing, 15(3), 705–709. https://
applications of unoccupied aircraft systems enabled genomic predic- doi.org/10.1080/01431169408954110
tion. The Plant Phenome Journal, 7(1), e20101. https://doi.org/10. Myneni, R. B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The
1002/ppj2.20101 interpretation of spectral vegetation indexes. IEEE Transactions on
Herr, A. W., Adak, A., Carroll, M. E., Elango, D., Kar, S., Li, C., Jones, S. Geoscience and Remote Sensing, 33(2), 481–486. https://doi.org/10.
E., Carter, A. H., Murray, S. C., Paterson, A., Sankaran, S., Singh, A., 1109/TGRS.1995.8746029
& Singh, A. K. (2023). Unoccupied aerial systems imagery for pheno- Parmley, K. A., Higgins, R. H., Ganapathysubramanian, B., Sarkar, S.,
typing in cotton, maize, soybean, and wheat breeding. Crop Science, & Singh, A. K. (2019). Machine learning approach for prescriptive
63, 1722–1749. https://doi.org/10.1002/csc2.21028 plant breeding. Scientific Reports, 9(1), Article 17132. https://doi.org/
Hijmans, R. J., Van Etten, J., Cheng, J., Mattiuzzi, M., Sumner, M., 10.1038/s41598-019-53451-4
Greenberg, J. A., Lamigueiro, O. P., Bevan, A., Racine, E. B., Patterson, H. D., & Thompson, R. (1971). Recovery of inter-block infor-
Shortridge, A., & Hijmans, M. R. J. (2015). Package ‘raster’. R mation when block sizes are unequal. Biometrika, 58(3), 545–554.
Package, 734, 473. https://doi.org/10.1093/biomet/58.3.545
Hinkelmann, K. (2012). Design and analysis of experiments. Volume 3, Piepho, H.-P., Boer, M. P., & Williams, E. R. (2022). Two-dimensional
Special designs and applications. Wiley-Interscience. P-spline smoothing for spatial analysis of plant breeding trials.
Iqbal, F., Lucieer, A., & Barry, K. (2018). Simplified radiometric cali- Biometrical Journal, 64(5), 835–857. https://doi.org/10.1002/bimj.
bration for UAS-mounted multispectral sensor. European Journal of 202100212
Remote Sensing, 51(1), 301–313. https://doi.org/10.1080/22797254. Piepho, H.-P., Richter, C., & Williams, E. (2008). Nearest neigh-
2018.1432293 bour adjustment and linear variance models in plant breeding trials.
Lachowiec, J., Feldman, M. J., Matias, F. I., LeBauer, D., & Gregory, Biometrical Journal, 50(2), 164–189. https://doi.org/10.1002/bimj.
A. (2024). Adoption of unoccupied aerial systems in agricultural 200710414
research. The Plant Phenome Journal, 7(1), e20098. https://doi.org/ Piepho, H. P., & Williams, E. R. (2010). Linear variance models for plant
10.1002/ppj2.20098 breeding trials. Plant Breeding, 129(1), 1–8. https://doi.org/10.1111/
Lee, Y. S., Lee, D. G., Yu, Y. G., & Lee, H. J. (2016). Application of j.1439-0523.2009.01654.x
drone photogrammetry for current state analysis of damage in forest Piepho, H.-P., Williams, E. R., & Michel, V. (2015). Beyond latin
damage areas. Journal of Korean Society for Geospatial Information squares: A brief tour of row-column designs. Agronomy Journal,
Science, 24(3), 49–58. https://doi.org/10.7319/kogsis.2016.24.3.049 107(6), 2263–2270. https://doi.org/10.2134/agronj15.0144
12 of 12 HERR ET AL.

Reynolds, M., Chapman, S., Crespo-Herrera, L., Molero, G., Mondal, van Es, H. M., & van Es, C. L. (1993). Spatial nature of random-
S., Pequeno, D. N. L., Pinto, F., Pinera-Chavez, F. J., Poland, J., ization and its effect on the outcome of field experiments. Agron-
Rivera-Amado, C., Saint Pierre, C., & Sukumaran, S. (2020). Breeder omy Journal, 85(2), 420–428. https://doi.org/10.2134/agronj1993.
friendly phenotyping. Plant Science, 295, 110396. https://doi.org/10. 00021962008500020046x
1016/j.plantsci.2019.110396 Xie, C., & Yang, C. (2020). A review on plant high-throughput pheno-
Rodríguez-Álvarez, M. X., Boer, M. P., Van Eeuwijk, F. A., & Eilers, P. typing traits using UAV-based sensors. Computers and Electronics
H. C. (2018). Correcting for spatial heterogeneity in plant breeding in Agriculture, 178, 105731. https://doi.org/10.1016/j.compag.2020.
experiments with P-splines. Spatial Statistics, 23, 52–71. https://doi. 105731
org/10.1016/j.spasta.2017.10.003 Xu, R., Li, C., Paterson, A. H., Jiang, Y. U., Sun, S., & Robertson, J.
Rouse, J. W., Hass, R. H., Schell, J. A., Deering, D. W., & Harlan, J. S. (2018). Aerial images and convolutional neural network for cotton
C. (1974). Monitoring the vernal advancement and retrogradation bloom detection. Frontiers in Plant Science, 8, Article 2235. https://
(green wave effect) of natural vegetation. NASA. doi.org/10.3389/fpls.2017.02235
Rutkoski, J., Poland, J., Mondal, S., Autrique, E., Pérez, L. G., Crossa, Xue, J., & Su, B. (2017). Significant remote sensing vegetation indices:
J., Reynolds, M., & Singh, R. (2016). Canopy temperature and vege- A review of developments and applications. Journal of Sensors, 2017,
tation indices from high-throughput phenotyping improve accuracy Article 1353691. https://doi.org/10.1155/2017/1353691
of pedigree and genomic selection for grain yield in wheat. G3: Yang, G., Liu, J., Zhao, C., Li, Z., Huang, Y., Yu, H., Xu, B., Yang, X.,
Genes|Genomes|Genetics, 6(9), 2799–2808. https://doi.org/10.1534/ Zhu, D., Zhang, X., Zhang, R., Feng, H., Zhao, X., Li, Z., Li, H., &
g3.116.032888 Yang, H. (2017). Unmanned aerial vehicle remote sensing for field-
Sankaran, S., Khot, L. R., & Carter, A. H. (2015). Field-based crop based crop phenotyping: Current status and perspectives. Frontiers in
phenotyping: Multispectral aerial imaging for evaluation of win- Plant Science, 8, Article 1111. https://www.frontiersin.org/journals/
ter wheat emergence and spring stand. Computers and Electronics plant-science/articles/10.3389/fpls.2017.01111
in Agriculture, 118, 372–379. https://doi.org/10.1016/j.compag.2015. Yang, R.-C., Ye, T. Z., Blade, S. F., & Bandara, M. (2004). Efficiency of
09.001 spatial analyses of field pea variety trials. Crop Science, 44(1), 49–55.
Shakoor, N., Lee, S., & Mockler, T. C. (2017). High throughput pheno- https://doi.org/10.2135/cropsci2004.4900
typing to accelerate crop breeding and monitoring of diseases in the Yin, C., Lv, X., Zhang, L., Ma, L., Wang, H., Zhang, L., & Zhang, Z.
field. Current Opinion in Plant Biology, 38, 184–192. https://doi.org/ (2022). Hyperspectral UAV images at different altitudes for monitor-
10.1016/j.pbi.2017.05.006 ing the leaf nitrogen content in cotton crops. Remote Sensing, 14(11),
Smith, A. B., Cullis, B. R., & Thompson, R. (2005). The analysis of crop 2576. http://doi.org/10.3390/rs14112576
cultivar breeding and evaluation trials: An overview of current mixed Zhang, C., & Kovacs, J. M. (2012). The application of small unmanned
model approaches. The Journal of Agricultural Science, 143(6), 449– aerial systems for precision agriculture: A review. Precision Agri-
462. https://doi.org/10.1017/S0021859605005587 culture, 13(6), 693–712. https://doi.org/10.1007/s11119-012-9274-
Stringer, J. K., Smith, A. B., & Cullis, B. R. (2012). Spatial analysis of 5
agricultural field experiments. Design and analysis of experiments:
Special designs and applications. Wiley-Interscience.
S U P P O RT I N G I N F O R M AT I O N
Sun, J., Poland, J. A., Mondal, S., Crossa, J., Juliana, P., Singh, R. P.,
Rutkoski, J. E., Jannink, J.-L., Crespo-Herrera, L., Velu, G., Huerta- Additional supporting information can be found online in the
Espino, J., & Sorrells, M. E. (2019). High-throughput phenotyping Supporting Information section at the end of this article.
platforms enhance genomic selection for wheat grain yield across
populations and cycles in early stage. Theoretical and Applied Genet-
ics, 132(6), 1705–1720. https://doi.org/10.1007/s00122-019-03309-
0 How to cite this article: Herr, A. W., Campbell, K.
Tirado, S. B., Hirsch, C. N., & Springer, N. M. (2020). UAV-based imag- G., Li, X., & Carter, A. H. (2024). Spatial analysis
ing platform for monitoring maize growth throughout development. with unoccupied aircraft systems data in wheat
Plant Direct, 4(6), e00230. https://doi.org/10.1002/pld3.230 breeding yield trials. The Plant Phenome Journal, 7,
Tukey, J. W. (1949). Comparing individual means in the analysis of
e70007. https://doi.org/10.1002/ppj2.70007
variance. Biometrics, 5(2), 99–114. https://doi.org/10.2307/3001913