Zonal Application of Plant Growth Regulator in Cotton to Reduce

University of Nebraska - Lincoln University of Nebraska - Lincoln

DigitalCommons@University of Nebraska - Lincoln DigitalCommons@University of Nebraska - Lincoln

Agronomy & Horticulture -- Faculty Publications Agronomy and Horticulture Department

2023

Zonal Application of Plant Growth Regulator in Cotton to Reduce Zonal Application of Plant Growth Regulator in Cotton to Reduce

Variability and Increase Yield in a Highly Variable Field Variability and Increase Yield in a Highly Variable Field

Carlos M.P. Vaz

Júlio C. Franchini

Eduardo A. Speranza

Ricardo Y. Inamasu

Lúcio A. de C. Jorge

See next page for additional authors

Follow this and additional works at: https://digitalcommons.unl.edu/agronomyfacpub

Part of the Agricultural Science Commons, Agriculture Commons, Agronomy and Crop Sciences

Commons, Botany Commons, Horticulture Commons, Other Plant Sciences Commons, and the Plant

Biology Commons

This Article is brought to you for free and open access by the Agronomy and Horticulture Department at

DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Agronomy & Horticulture --

Faculty Publications by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln.

Authors Authors

Carlos M.P. Vaz, Júlio C. Franchini, Eduardo A. Speranza, Ricardo Y. Inamasu, Lúcio A. de C. Jorge,

Ladislau M. Rabello, Ivani de O.N. Lopes, Sérgio das Chagas, Jefferson L.R. de Souza, Márcio de Souza,

Amandio Pires, and James Schepers

The Journal of Cotton Science 27:60–73 (2023)

http://journal.cotton.org, © The Cotton Foundation 2023

AGRONOMY AND SOILS

Zonal Application of Plant Growth Regulator in Cotton to Reduce Variability

and Increase Yield in a Highly Variable Field

Carlos M.P. Vaz*, Júlio C. Franchini, Eduardo A. Speranza, Ricardo Y. Inamasu,

Lúcio A. de C. Jorge, Ladislau M. Rabello, Ivani de O.N. Lopes, Sérgio das Chagas,

Jeerson L.R. de Souza, Márcio de Souza, Amandio Pires, and James Schepers

C.M.P. Vaz*, R.Y. Inamasu, L.A.de C. Jorge, and L.M. Rabello,

Brazilian Agricultural Research Corporation, Embrapa

Instrumentation, São Carlos, SP, Brazil; J.C. Franchini and I.

de O.N. Lopes, Brazilian Agricultural Research Corporation,

Embrapa Soybean, Londrina, PR, Brazil; E.A. Speranza,

Brazilian Agricultural Research Corporation, Embrapa Digital

Agriculture, Campinas, SP, Brazil; S. das Chagas and J.L.R. de

Souza, Amaggi Group, Sapezal, MT, Brazil; M. de Souza and

A. Pires, Mato Grosso Cotton Institute, Cuiabá, MT, Brazil;

and J. Schepers, University of Nebraska, Agronomy and

Horticulture Department, Lincoln, NE.

*Corresponding author: [email protected]

ABSTRACT

Variable-rate application has great potential

to reduce variability and increase yield by spa-

tially optimizing agricultural inputs. In cotton,

plant growth regulators (PGRs) control exces-

sive growth and provide suitable plant height for

harvest operations. This study evaluates the eect

of variable-rate PGR application compared to

constant-rate application to reduce yield spatial

variability and increase yield. The variable-rate

approach was carried out in 2020 based on zonal

applications dened by clustering analysis using

soil electrical conductivity, vegetation indexes,

and yield maps. Application doses and timings

were determined by integrating plant height

measurements for the whole eld in 2019 and by

zone in 2020. To compare the two procedures, cul-

tivar and plant populations were kept constant;

fertilization and accumulated rain were similar

in both seasons. A reduction in yield spatial vari-

ability due to the zonal application was observed,

with yield coecient of variation (CV) decreasing

from 18% in 2019 to 12% in 2020. Spatial and

temporal analysis of Normalized Dierence Veg-

etation Index satellite images showed higher CV

values in 2019 (constant-rate) reaching 30% at the

end of the season, whereas in 2020 (variable-rate)

CV was constant (approximately 10%). Cotton

yield increased from 3.5 to 4.3 t ha

-1

between

2019 and 2020, which can be partially attributed

to the variable-rate approach. The variable-rate

approach based on application zones and plant

height measurements was a viable strategy for

reducing yield spatial variability and likely in-

creasing yield in a highly variable cotton eld.

razil is an important world cotton producer,

ranking fourth in cotton lint yield and second in

exports in the 2021/2022 season (USDA, 2023). The

major production areas are the Brazilian savanna-like

Cerrado, which can be characterized as having highly

weathered and acidic soils, low cation exchange

capacity, low natural fertility, flat landscape, and

average annual precipitation of 800 to 2,000 mm (rainy

season from October to March). The most common

production system on large farms is rainfed, double-

crop cotton after soybean with intensive high-input

management. In general, large commercial eld units of

approximately 100 to 400 ha are individually managed

to maximize their production potential for the specic

soil type, topography, climate, and cultivar. At this

scale, eld units exhibit large natural variability (soil

texture, organic matter, topography, climatic variables,

etc.), causing large in-field yield variations when

treated as homogeneous units. In this scenario, variable-

rate application (VRA) technology has potential

for reducing variability on the production units and

increasing yield and protability by spatially optimizing

the agricultural input applications, considering both

natural and anthropic spatial variations.

Variable-rate application technology has been gain-

ing prominence in the last few decades. Advances in the

agricultural machinery industry have provided methods

for site-specic application of liquid and granular fer-

tilizers, lime, pesticides, seed, plant growth regulators

(PGRs), defoliants, and ripeners based on predened

georeferenced application maps (Martins et al., 2020;

Nawar et al., 2017) or by on-the-go approaches us-

ing sensors (Stamatiadis et al., 2020; Yu et al., 2019).

However, compared to other precision agriculture tools

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

such as Global Navigation Satellite System (GNSS)

guidance, sprayer boom control, and planter row or

section shuto, VRA technology is less popular among

farmers (Fountas et al., 2005; Lowenberg-DeBoer

and Erickson, 2019; Zhou et al., 2017). This is prob-

ably because farmers like the idea of using VRA in

general, but they are not completely convinced of its

value (Lowenberg-DeBoer and Erickson, 2019). One

important challenge is establishing inexpensive and

technically ecient protocols to generate prescription

maps and create application strategies to deliver the

right doses varying spatially and timely, for specic

crops, regions, and production systems (Campanella,

2000; Jin et al., 2019; Nawar et al., 2017).

Plant growth regulators are commonly applied

in cotton to restrict excessive vegetative growth, re-

directing photosynthates to reproductive growth, and

providing benets as early owering and increasing

boll retention lower on the plant and setting plant ar-

chitecture favorable for mechanized harvesting (Fang et

al., 2019; Samples et al., 2015). Some studies indicate

that PGR also can provide yield increments (Leal et al.,

2020; Sawan, 2018; Tung et al., 2020), whereas others

report negative or no eects on cotton yield and ber

quality (O’Berry et al., 2009; Vistro et al., 2017).

Several studies have evaluated the agronomic

and economic benets of VRA of PGR in cotton to

control excessive vegetative growth (Sawan, 2018;

Tung et al., 2020). These applications reduced spatial

variability in plant height and yield and increased total

cotton lint yield and protability. Spatial and temporal

variability of cotton plant height, height-to-node ratio

(HNR) and length of top ve internodes, were evalu-

ated by Thurman and Heiniger (1999b) who showed

that uniform application of PGR on highly variable

elds increased plant height and HNR variability. The

study concluded that spatial analysis of plant growth

improved the eectiveness of PGR application, and

that large eld variability justies VRA of PGR. A

procedure for VRA of PGR based on plant height

using a tractor-mounted infrared light sensor, a crop

simulation model, and relationships between plant

height and total plant weight for eight cotton cultivars

was developed by Landivar et al. (1999). Results

showed a reduction in the plant height coecient of

variation (CV) (from 12.6% before the VRA to 7.6%

after two PGR applications). However, yield incre-

ments due to the VRA of PGR were negligible and

the lack of response was mainly attributed to the dry

season experienced during the reproductive period,

thus masking possible yield benets of the VRA.

Baio et al. (2018) applied PGR and fruit ripener

at variable rates in a large commercial cotton eld

based on vegetation index (VI) maps acquired with an

optical canopy sensor and phenological measurements.

Three homogeneous application zones were dened

according to VI variability, delineating low, average,

and high VI zones. Plant height and growth rate were

then monitored during the growing seasons for each

zone to support PGR application decisions (timing and

doses). The VRA procedure increased the uniformity

of plant height and fruit opening among application

zones, resulting in seed cotton yield and net revenue

increments of 265 kg ha

-1

and $152 USD ha

-1

, respec-

tively (averaged over two growing seasons).

Trevisan et al. (2018) evaluated two optical

canopy and ultrasound sensors to detect spatial

variability of plant height and generated prescrip-

tion maps for VRA of PGR. The applied procedure

reduced PGR cost by 17% but had no eect on cotton

yield. Similarly, Bethel et al. (2003) obtained PGR

application rate reductions varying from 10 to 53%

using a variable-rate procedure based on Normal-

ized Dierence Vegetation Index (NDVI) maps to

establish application zones.

In summary, these studies have shown benets

of using VRA of PGR, which include control of plant

growth for harvesting and decrease of in-eld yield

variability (Baio et al., 2018; Landivar et al., 1999;

Thurman and Heiniger, 1999a), reduction in the

amount of applied PGR (Bethel et al., 2003; Trevisan

et al., 2018), and improved yields (Baio et al., 2018;

Thurman and Heiniger, 1999b) when compared to

PGR applied at constant-rates. However, although

some studies of VRA of PGR have shown positive ef-

fects on reducing in-eld variability, the total amount

of PGR applied, yield, and revenues; in some experi-

ments, no improvements were observed in yield or

in the reduction of production costs (Bethel et al.,

2003; Landivar et al., 1999; Nelson, 2006; Trevisan

et al., 2018). Additional studies are needed to evaluate

the agronomic and economic gains of VRA of PGR,

understand soil, topography, climate and plant vari-

ability eects on the VRA performance, and establish

eective protocols for VRA of PGR at the farm level,

considering regional and local specicities and dier-

ent production systems. Contributing to the diculty,

the present study evaluates the performance of zonal

application of PGR in a highly variable cotton eld

unit (soil clay content varying from 7 to 37%) at farm

level. Spatial variability was assessed by soil apparent

electrical conductivity maps, VI images, and cotton

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

yield maps to establish the application zones. The

performance of the variable-rate approach applied

to three dierent zones was compared to uniform

PGR application.

MATERIALS AND METHODS

The study was conducted in a commercial cot-

ton eld of 169 ha (13

35’ S, 58

53’ W) located in

Sapezal, Mato Grosso State, Brazil (Tucunaré Farm,

Amaggi Group) during two growing seasons (2019

and 2020). In the 2019 season, PGR was applied at

constant rate according to the procedure adopted by

the farm technical team that included plant height

measurements and the historical management of the

area. In 2020, PGR was applied with dierent doses

in three delineated zones based on soil, crop imagery,

and yield maps acquired in the 2019 season. Doses

and application dates of PGR in 2019 and 2020 are

presented in Tables 1 and 2.

Application zones in 2020 were established

based on cluster analysis, considering soil ap-

parent electrical conductivity (EC

) maps, VI

images (NDVI and Normalized Dierence Red

Edge [NDRE]), and the 2019 cotton yield map. In

2019, plant height was measured at nine locations

(Fig. 1c shows geographic locations) for seven

dates (44, 61, 67, 80, 88, 102, and 123 days after

sowing [DAS]) and averaged for the whole eld

on the dierent dates. In 2020, cotton plant height

was measured and averaged by zone, approxi-

mately every 5 d (19 dates from 35 to 135 DAS)

at locations shown in Fig. 1d (approximately 40

points by zone). One plant height was determined

for each point and date in both cases (2019 and

2020). Plant growth regulator (Mepiquat chloride,

Sponsor 250 g L

-1

) (FMC Química do Brasil Ltda,

Brazil) was applied 94 DAS in 2019, and 57 and

65 DAS in 2020 (Table 2), when plant heights

were approximately 80 to 90 cm.

Table 1. Cotton management and production parameters for the 2019 and 2020 seasons in the experimental eld

Parameter Unit 2019 2020

Cultivar - TMG 81 WS

Row spacing m 0.9

Plant population plant m

-1

Post-planting Nitrogen kg ha

-1

202 248

Number of N applications - 6 7

Post-planting Potassium kg ha

-1

246 245

Number of K applications - 3 3

Total PGR applied

g ai ha

-1

12.5 42.5

Number of PGR applications - 1

Cumulative precipitation

mm 1,070 1,171

Sowing date - 21 Jan 2019 27 Dec 2019

Harvesting date - 06 Aug 2019 13 July 2020

Cotton Yield kg ha

-1

3,399 4,152

Mepiquat chloride, Sponsor (250 g active ingredient L

-1

)

Uniform application

Variable-rate application

For the whole crop season

Table 2. Mepiquat chloride (Sponsor, 250 g L

-1

) doses and application data for the two growing seasons. Z1, Z2, and Z3 are

the application zones dened by clustering analysis (Fig. 2)

Season

PGR Applications

DAS

Dose (g ai ha

-1

)

Type Date Z1 Z2 Z3

2019 uniform 24 April 2019 94 12.5 12.5 12.5

2020

variable 21 Feb 2020 57 20 25 12.5

variable 29 Feb 2020 65 20 30 20

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

Cotton cultivar, linear seed density, and row

spacing were kept the same for the two growing

seasons to allow comparisons between the constant-

rate and variable-rate PGR approaches. In-season

potassium application and accumulated precipitation

for both seasons were similar, whereas in-season

nitrogen application was 23% higher in 2020 than

in 2019 (Table 1).

Soil and Vegetation Index Maps for Delinea-

tion of Application Zones. The EC

maps (0-30

and 0-90 cm depths) were acquired on 19 January

2019 using the Veris 3100 system (Veris Technol-

ogy, Salina, KS) pulled by a tractor at 10 km h

-1

15-m spaced transects and acquisition intervals of

1 s, georeferenced using a GNSS receiver model

AG114 DGPS (Trimble, Sunnyvale, CA). The soil

clay content map was obtained by collecting and

analyzing 170 georeferenced soil samples (0-20 cm

depth) at a regular grid sampling (100 m x 100 m)

and interpolated by analysis using the software VES-

PER (University of Sydney, Australia). The cotton

canopy spatial variability was assessed by NDVI and

NDRE images acquired by a Matrice 200 drone (DJI,

Shenzhen, China) with a high resolution RedEdge-M

camera (MicaSense, Seattle, WA) on 15 May and 7

June 2019 (end of owering/boll development and

beginning open boll, respectively). The acquired

images were georeferenced by eld-dened ground

control points (six points on the boundary and three

in the middle) using a model AG114 DGPS GNSS

receiver and the orthomosaics were generated in the

pix4DMapper platform (Pix4D S.A., Switzerland).

Cluster analyses were performed with seven ac-

quired maps: two EC

(0-30 and 0-90 cm) maps, two

NDVI (May and June) maps, two NDRE (May and

June) maps, and the 2019 cotton yield map. For this,

the agglomerative hierarchical Ward’s unsupervised

clustering method (Ward-Junior, 1963) was applied,

implemented in the R programming environment, as

illustrated in Fig. 2. Prior to the analysis, the data were

normalized and re-sampled to a regular grid of 10 m x

10 m, because each attribute was mapped with dier-

ent sampling densities. Additionally, data distribution

in a regular spatial grid is a prerequisite for the correct

execution of the clustering algorithm so that it does

not follow a bias focused only on geographic location.

This resolution is sucient to identify spatial vari-

ability and delineate management zones for a 169-ha

plot and compatible with the platform width of the

machinery used by the farmer for PGR interventions.

The Ward’s method provides a tree of clusters, known

as a dendrogram, by fusion of similar groups in each

level, based on the lower increment of the mean

square error. To avoid small area clusters, a hierarchi-

cal clustering initialization method known as initial

tessellation (Ruβ and Kruse, 2011) was applied and

number of clusters was selected using the silhouette

width internal validation criteria (Rousseeuw, 1987).

NDVI Temporal Series from Satellite Images.

NDVI images obtained from Sentinel-2 (10-m spatial

resolution) and Landsat-8 (30-m resolution) satellites

were used to assess the cotton plant growth spatial

variability in the two harvesting seasons, evaluating

and comparing the eect of the two PGR application

methods. The Earth Engine API (Google, Menio Park,

CA) was used to evaluate, select, and download the

NDVI images for the two cotton seasons, providing

NDVI temporal series for the spatial variability evalu-

ations. Only images without clouds were selected for

the analysis. In that region, the rainiest period for the

cotton season is from December to February; there-

fore, few NDVI images without clouds were obtained

in the rst three growing months in both years. In 2019

cotton was sown on 21 January 2019 and harvested

on 6 August 2019; whereas in 2020, cotton was sown

on 27 December 2019 and harvested on 13 July 2020.

Average, standard deviation (SD), and coecient of

variation (CV) of NDVI for the whole eld and for the

application zones were determined and compared in

the two growing seasons to evaluate the eect of the

VRA in reducing plant spatial variability.

Figure 1. Average cotton plant height measured for (a) the

two growing seasons and (b) by zone (Z1, Z2, and Z3) in

the 2020 season. Data tted with polynomial equation of

second degree. DAS: days after sowing. In 2019, plant

height was measured and averaged in (c) the whole eld

for the dierent dates and (d) by zone in 2020. Dots indicate

the coordinates of measured plant height.

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

Cotton Management and Harvesting. The cot-

ton cultivar TMG 81WS (Tropical, Melhoramento &

Genética, Cambé, Brazil) was sowed at 0.9-m spaced

rows with 9 seeds m

-1

density in the two growing

seasons, using a Hercules 10000 distributor (Stara,

Não-Me-Toque, RS, Brazil). Plant growth regulator

was applied at one uniform dose on 24 April 2019

(12.5 g ai ha

-1

) and at two variable doses applied on

21 February and 29 February 2020, in the previously

established application zones (total applied by zone

was 40 g ai ha

-1

in Z1; 55 g ai ha

-1

in Z2; and 32.5 g

ai ha

-1

in Z3), using a self-propelled sprayer PV 4730

(John Deere, Moline, IL) in 2019 and the Uniport

3030 (Jacto, Pompéia, SP, Brazil) in 2020. Granular

fertilizers were applied (Table 1) with the Hercules

10000 distributor. Cotton yield maps were obtained

using the cotton picker model CP690 (John Deere,

Moline, IL) that was properly calibrated before use.

RESULTS AND DISCUSSION

All parameters mapped in the experimental eld

are presented in Fig. 2. Soil clay content correlated

relatively well to EC

(linear determination coecient,

= 0.65 and 0.62 for depths 0-30 and 0-90 cm, respec-

tively). Seed cotton yield, EC

, NDVI, and NDRE maps

displayed similar spatial variability patterns in 2019 and

were selected for clustering analysis to generate the

PGR application zones. Elevation and slope were not

included in the clustering analysis due to low in-eld

variations and low spatial correlation with the other rel-

evant parameters, although clay content was discarded

due to its intimate relation with EC

observed in the

experimental eld, thus avoiding redundancy. Three

application zones were dened using Ward’s method

along with the criteria to select the most appropriate

number of clusters (Fig. 2). The border between ap-

plication zone 3 (Z3) and the others were straightened,

as shown on the map of the delimitated zones in Fig. 2,

to facilitate PGR application in that zone.

The mapped parameters shown in Fig. 3 reveal

large spatial variability of soil and plant features

(EC

, clay content, NDVI, and NDRE), which in turn

impacted the cotton yield spatial variability, as also

indicated by frequency distribution graphs (Fig. 4).

Clay content, EC

, and cotton yield histograms present

three peaks (obtained by deconvolution of peaks us-

ing the Multiple Peak Fit tool of the Origin software),

whereas NDVI and NDRE acquired by drone in June

2019 exhibit two peaks, which is in accordance to the

number of clusters selected (three zones).

To complement the zonal application of PGR in

the 2020 season, plant heights were measured in the

three zones up to approximately 140 DAS. Fig. 1a

shows plant height averaged along the entire eld for

each collection dates during the 2020 and 2019 sea-

sons, and Fig. 1b shows the plant height averaged by

collected date for each zone in 2020 (no measurement

was made by zone in the 2019 season). A signicant

dierence (p < 0.001) in plant height was observed

between the two growing seasons (Fig. 1a), according

to the ANOVA test applied to the quadratic model t-

ting data, showing an average plant height dierence

of 0.27 m between the two crop seasons. Dierences

in plant growth and cotton yield were likely inuenced

by the higher dose of post-planting fertilizer N applied,

better rain distribution in the 2020 season (Fig. 5) and

by the zonal application of PGR, contributing to the

improved cotton yield in 2020, which was 22% higher

than in 2019 (Table 1).

Figure 2. Illustration of the clustering procedure employed to generate the application zones (Z1, Z2, and Z3). The border

of Z3 with Z1 and Z2 (dotted line) was rectied (solid line), as shown on the map of the delimitated zones, to facilitate the

zonal application of the plant growth regulator.

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

Figure 3. Soil parameters maps (apparent electrical conductivity, EC

, at 0-30 and 0-90 cm depths, and clay content),

vegetation indexes images (NDVI and NDRE obtained by drone), topographic parameters (elevation and slope obtained

by drone measurements), and cotton yields maps obtained with a harvesting monitor system in the 2019 and 2020 seasons.

Figure 4. (a) Frequency distribution of cotton yields in the

2019 and 2020 seasons, (b) apparent electrical conductivity,

and (d) soil clay content. Peaks 1, 2, and 3 were obtained

by deconvolution of peaks using Origin software.

Figure 5. Monthly accumulated precipitation in the

experimental eld from sowing to harvesting for the two

growing seasons.

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

Mepiquat chloride was applied at 12.5 g ai ha

-1

in 2019 (one uniform application) and an average

of 42.5 g ai ha

-1

in 2020 (two applications, vari-

able by zone, detailed in Table 2). The higher PGR

dose in 2020 was necessary due to larger plants

growing in that year, compared to 2019 (Fig. 1a).

Dierent doses for each zone in 2020 (Table 2)

were dened based on plant height measurements

in the three zones taken between 35 and 60 DAS

(Fig. 6) and on previous experiences in that area

regarding plant growth due to in-season climate

variation, application of fertilizers, and cultivar

following the State of Mato Grosso Best Manage-

ment Practices Manual (Echer et al., 2020), result-

ing in 32.5 g ai ha

-1

for Z3 (lowest dose applied),

55 g ai ha

-1

for Z2 (highest dose applied), and 40

g ai ha

-1

for Z1.

The eect of PGR applied by zone (2020

season) on cotton yield spatial variability reduc-

tion can be evaluated by the yield CV obtained

for the entire eld and for each zone (Table 3). In

the whole eld, yield CV decreased from 17.9%

in 2019 to 12.4% in 2020 (5.5% reduction), but

inside the application zones the dierences be-

tween the two seasons were much lower (about

1% reduction). This is due to the strategy adopted

(dierent PGR doses per zone), which reduced

whole-eld yield variability, but had less eect

on reducing internal zone variability in that PGR

was applied at constant-rate within each zone.

Alternatively, another approach would be to apply

PGR at continuously variable-rate, based on plant

height maps obtained by sensors as a potentially

more eective way to reduce spatial variability

(Bethel et al., 2003). However, this VRA approach

requires special sprayers that were not available

Figure 6. Average plant height measured at the three

application zones at 35-50 and 50-60 DAS in the 2020.

Measurements to subsidize PGR doses in the dierent

zones in 2020.

in this study. The approach adopted here for the

zonal application was implemented by adjusting

PGR doses prior to the applications and turning

the system on when inside and o when outside.

In this case, three sprayer passes were necessary

in the rst PGR application (doses of 12.5, 20, and

25 g ai ha

-1

) and two in the second (doses of 20

and 30 g ai ha

-1

) to accomplish the variable-rate

application by zone in 2020 (Table 2).

Yield is the primary metric used by producers

when evaluating changes in management practices.

However, producers cannot ignore the impact of

soil properties, nutrient availability, water supply,

and climatic interactions. Water typically has a

great inuence on yield, so factors that inuence

soil water holding capacity like clay content and

inltration rate are primary considerations when

comparing yields. Zonal delineation in this study

incorporated many of these considerations in terms

of their absolute values (range across zones) more

so than the variability within zones. Lowest yields

in both years were attained in Z1 and highest in

Z2 (Table 3). Positive correlations with yield were

obtained with clay content, EC

, and the previous

year’s VIs (NDRE and NDVI).

Data analysis within zones showed that the CV

for yield was larger for Z1 (approximately 17%)

and lower for Z3 (approximately 7%) and Z2 (about

11%) in both years. Lower yields in Z1 can be par-

tially attributed to lower clay content and lower EC

(i.e., contributed to lower water holding capacity

and likely to be less fertile) compared to Z2 and Z3.

Zone 2 was intermediate to Z1 and Z3 in terms of

clay content and EC

but received the largest amount

of PGR and had the highest yields in 2019 and 2020.

The need for the highest dose of PGR in Z2 implies

that plant height had a major role in determining the

PGR rate. Cotton grown in this zone was unique in

that lush plant growth was present in 2020 (taller

plants) that could not be explained by soil clay con-

tent, EC

or any of the other considerations. This

observation illustrates the opportunity for real-time

sensing of crop biomass to help guide the applica-

tion rate of PGR.

The problem with whole-eld research is that

it does not lend itself to replications unless the area

is divided, which can be dicult considering eld

shape, topography, and soil type. To complicate mat-

ters further, this type of research requires a great deal

of background information before one can develop

a management strategy for the next crop.

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

In addition to the analysis on cotton yield CV

to assess spatial variability reduction due to the

zonal application of PGR, NDVI satellite images,

such as the ones from Sentinel-2 and Landsat-8

platforms, provide a good tool to evaluate plant

vigor dynamics spatially. In the experimental

field, 21 and 13 NDVI images without cloud

interference were obtained in the 2019 and 2020

cotton seasons, respectively (Table 4) and used for

spatial and temporal analysis. Fig. 7 shows NDVI

images (Sentinel-2) in four dierent DAS during

the growing seasons. The internal lines delineate

the application zones. NDVI frequency distribu-

tions for the two seasons are also given in Fig. 7.

A visual analysis of these images indicates larger

NDVI variability in 2019, especially in the 159

and 189 DAS images, which is conrmed by the

presence of two well-dened and spaced peaks

in the frequency distribution graphs in that year

(Figs. 7k and 7l).

Table 3. Mean, standard deviation (SD), and coecient of variation (CV) values in the whole eld (W) and the application

zones (Z1, Z, Z3) for cotton yield, soil clay content, apparent electrical conductivity (EC

), NDVI-drone (June 2019),

NDRE-drone (June 2019), land elevation, and slope

Parameter Zone Mean SD CV

Yield – 2019

W 3.5 0.6 17.9

Z1 2.8 0.05 17.5

Z2 4.0 0.5 12.3

Z3 3.5 0.3 7.4

Yield – 2020

W 4.3 0.5 12.4

Z1 3.9 0.6 16.8

Z2 4.5 0.5 10.1

Z3 4.3 0.3 6.8

Clay Content

(%)

W 16.4 5.5 33.5

Z1 11.8 2.1 18.1

Z2 16.2 3.9 24.2

Z3 20.9 5.3 25.5

0-30 cm

(mS m

-1

)

W 3.2 0.9 29.2

Z1 2.2 0.3 14.1

Z2 3.2 0.8 26.3

Z3 3.7 0.7 19.8

NDVI – Drone

W 0.87 0.04 4.9

Z1 0.82 0.05 6.0

Z2 0.88 0.02 1.8

Z3 0.89 0.01 1.1

NDRE – Drone

W 0.50 0.05 9.5

Z1 0.44 0.03 7.7

Z2 0.51 0.03 5.4

Z3 0.55 0.01 2.2

Land Elevation

(m)

W 504 7 1.3

Z1 501 6 1.2

Z2 500 5 1.0

Z3 511 3 0.5

Slope

(%)

W 2.0 0.6 29.4

Z1 2.1 0.7 31.8

Z2 2.2 0.6 25.6

Z3 1.6 0.2 15.1

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

NDVI variability (SD and CV) with DAS (Fig.

8) was similar in both years up to approximately

130 DAS and deviated signicantly from that point

forward. A sharp increase in CV at the end of the

crop season in 2019 was present, whereas this factor

was nearly constant in 2020. Close to harvesting (189

DAS), CV was approximately 30% in 2019 and 10%

in 2020 (Fig. 8b). The lower NDVI spatial variability

in 2020 was likely caused by the zonal application of

Figure 7. NDVI images from Sentinel-2 along the (a, b, c, d) 2019 and (e, f, g, h) 2020 cotton seasons at dierent days after

sowing (DAS). Internal lines delineate the application zones (Fig. 2). NDVI frequency distributions (i, j, k, l) are compared

at similar DAS for the two seasons.

Table 4. Number of satellite images without cloud interference obtained from Sentinel-2 and Landsat-8 at the experimental

eld during the 2019 and 2020 seasons

Season Satellite

Number of NDVI images

0-90 DAS

90-140 DAS 140-190 DAS Total

2019

Sentinel 1 6 9 16

Landsat 0 2 3 5

2020

Sentinel 2 1 6 9

Landsat 1 1 2 4

DAS, days after sowing

PGR in that year. Regarding variations of SD with

DAS (Fig. 8a), a noticeable decrease of SD after ap-

proximately 175 DAS was observed in both years,

likely caused by the application of defoliant and

ripener (applied at 170-175 DAS) that homogenized

cotton maturation. Such decrease was not observed

in the CV graph (Fig. 8b) because CV expresses

the ratio between SD and the mean value and mean

reduction was more intense than SD.

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

NDVI temporal series for each zone are

presented in Fig. 9, showing larger dierences

among the zones in 2019 (Fig. 9a), as expected,

with Z3 exhibiting the highest NDVI values and

Z1 the lowest. In 2020, dierences were much

lower, having a similar trend in NDVI among

zones, especially later in the growing season, but

slightly larger NDVI for Z3 and lower for Z1

from approximately 120 to 170 DAS. The large

NDVI variations among zones in 2019 and low

variations in 2020 (Fig. 9) agrees with the yield

CV reduction in 2020 (Table 3) and highlights the

eect of the zonal application of PGR on spatial

variability reduction in 2020 as compared to the

uniform PGR application in 2019.

Plant vigor, expressed by NDVI, has been

correlated to yield for several crops (Baio et al.,

2019; Huang et al., 2013; Johnson, 2016). In cot-

ton, an additional aspect that must be considered

is PGR use, due to the indeterminate growth habit

of cotton plants that aects plant height, vigor, and

yield. To evaluate the correlations between plant

vigor and cotton yield spatially and temporally,

yield was plotted against NDVI for dierent DAS

(Fig. 10) in the two seasons. In such analysis, data

were re-sampled to 40-x-40-m pixels, due to the

dierent spatial resolutions of these two variables

and, additionally, pixels in the borders between

two zones were not included to allow a better

evaluation by zone.

Figure 8. (a) Standard deviation and (b) coecient of variation of NDVI from Sentinel-2 and Landsat-8 satellite images

determined for the whole eld. DAS: days after sowing.

Figure 9. NDVI temporal series from Sentinel-2 satellite images averaged for the three zones (Z1, Z2, and Z3) in the (a) 2019

and (b) 2020 growing seasons. DAS: days after sowing.

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

Figure 10. NDVI (Sentinel-2) at dierent DAS, correlated spatially to cotton yield in (a, c, e, g) 2019 and (b, d, f, h)

2020. Distinct symbols are used for the application zones.

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

Linear trends between yield and NDVI were

observed for all DAS in 2020 when PGR was applied

at dierent doses by zone, although determination

coecients (r

) decreased in the late season (after

approximately 130 DAS) due to the eects of PGR,

defoliant, and ripener. In 2019 (uniform PGR ap-

plication), the data for Z3 gradually deviated from

the other two zones as DAS increased and NDVI

values in Z3 obtained close to harvest shifted sig-

nicantly from the others. This behavior is probably

an eect of the PGR application at constant-rate. As

Z3 has the highest average soil clay content (Table

3), it likely retained more water leading to higher

plant height comparatively to the other two zones,

demanding larger amounts of PGR compared to

the other zones. Additionally, in a year where rain

distribution was uneven, as it was in 2019 (almost

no rain after 100 DAS, as shown in Fig. 5), such

dierences in plant growth among zones tends to

be exacerbated. Although no plant height measure-

ment was made by zone in 2019, the NDVI satellite

images clearly indicates the larger NDVI values in

Z3, which can be associated with higher vegetative

growth and insucient application of PGR. Average

NDVI values by zone at 189 DAS were 0.46 (Z3),

0.29 (Z2), and 0.25 (Z1) in 2019 and 0.35 (Z3), 0.38

(Z2), and 0.34 (Z1) in 2020.

CONCLUSIONS

The zonal application of PGR provided a de-

crease in the yield CV across the eld from 17.9% in

2019 to 12.4% in 2020 (5.5% reduction), although

for each zone reductions were much lower (about

1% CV reduction from 2019 to 2020). This likely

happened because inside the zones PGR was ap-

plied at constant rates. In 2020, Z1 presented the

larger yield CV (approximately 16.8%) compared

to Z2 (10.1%) and Z3 (6.8%), indicating the need to

re-evaluate and re-dene the zones for future VRA

of PGR in this eld, aiming to reduce even further

the cotton yield variability. Cotton yield increased

from approximately 3.5 t ha

-1

in 2019 to 4.3 t ha

-1

in 2020 (increment of 0.764 t ha

-1

or 22%), when

PGR was applied at variable-rate by zone. Inside

the zones, increments were 1.101 t ha

-1

(40%) in Z1,

0.535 t ha

-1

(13%) in Z2, and 0.728 t ha

-1

(21%) in

Z3. These increments cannot be attributed solely to

the PGR variable-rate approach. Although cultivar,

row spacing, and seed density were kept constant,

23% more post-planting N fertilizer was applied

in 2020 and the better precipitation distribution

in that season could have a signicant impact on

cotton yield. Nevertheless, the data and the analy-

sis presented herein suggest that the applied PGR

variable-rate procedure reduced the yield spatial

variability. Satellite images and subsequent NDVI

data oer a powerful tool to evaluate the spatial

variability dynamically, especially for the second

crop in Brazil, which is generally sowed in the

summer (rainy season) and harvested in the winter

(dry season). In this case, a considerable number

of satellite images without cloud interference can

be accessed. In 2019, when PGR was applied at

constant rate, the NDVI coecient of variation

increased signicantly from approximately 130

DAS, reaching 30% close to harvesting; whereas

in 2020 (PGR applied at variable-rate), CV was es-

sentially constant (approximately 10%). Temporal

series NDVI data by zone showed large dierences

among zones in 2019 (highest values for Z2 and

lowest values for Z1), from 130 to 190 DAS and

minor dierences in the 2020 season. The spatial

and temporal analysis performed with the NDVI

satellite images agrees and complements the

spatial variability analysis made with the cotton

yield maps, allowing eective evaluation of the

agronomic impacts and benets of the zonal ap-

plication of PGR compared to uniform application

on a highly variable eld. This study indicates that

VRA of PGR based on application zones and plant

height measurements can be a viable strategy for

reducing spatial variability in highly variable cotton

elds. Nevertheless, more research is necessary to

evaluate the eect of VRA of PGR in agricultural

elds with dierent levels of variability and under

variable climatic conditions among years.

ACKNOWLEDGMENTS

The authors acknowledge the financial and

technical supports from the Brazilian Agricultural

Research Corporation [41.14.09.001], the Mato

Grosso Cotton Institute, and the Amaggi Group.

REFERENCES

Baio, F.H.R., D.C. Neves, and P.E. Teodoro. 2019. Soil

chemical attributes, soil type, and rainfall eects on

normalized dierence vegetation index and cotton ber

yield variability. Agron. J. 111:2910–2919. https://doi.

org/10.2134/agronj2019.05.0375

JOURNAL OF COTTON SCIENCE, Volume 27, Issue 2, 2023

Baio, F.H.R., D.C. Neves, H.B. Souza, A.J.F. Leal, R.C. Leite,

J.P. Molin, and S.P. Silva. 2018. Variable rate spraying

application on cotton using an electronic ow control-

ler. Precis. Agric. 19:912–928. https://doi.org/10.1007/

s11119-018-9564-7

Bethel, M., T. Gress, S. White, S. Johnson, T. Sheely, B. Rob-

erts, M. Paggi, and N. Groenenberg. 2003. Image-based,

variable rate plant growth regulator application in cotton

at Sheely Farms in California. p. 1755–1766 In Proc.

Beltwide Cotton Conf., Nashville, TN. 6-10 Jan. 2003,

Natl. Cotton Counc. Am., Memphis, TN.

Campanella, R. 2000. Testing components toward a remote-

sensing-based decision support system for cotton produc-

tion. Photogramm. Eng. Remote Sens. 66(10):1219–

1227.

Echer, F.R., P.R. Mello, and C.A. Rosolem. 2020. Manage-

ment of plant growth regulators. In J.L. Belot and

P.M.C.A. Vilela (ed.), Best Management Practices for

Cotton in Mato Grosso State (in Portuguese), 4

edition.

IMAmt-AMPA, Cuiabá, MT.

Fang, S., K. Gao, W. Hu, S. Wang, B. Chen, and Z. Zhou.

2019. Foliar and seed application of plant growth regula-

tors aects cotton yield by altering leaf physiology and

oral bud carbohydrate accumulation. Field Crops Res.

231:105–114. https://doi.org/10.1016/j.fcr.2018.11.012

Fountas, S., S. Blackmore, D. Es, S. Hawkins, G. Blumho,

J. Lowenberg-deBoer, and C.G. Sorensen. 2005. Farmer

experience with precision agriculture in Denmark and

the US Eastern Corn Belt. Precis. Agric. 6:121–141.

https://doi.org/10.1007/s11119-004-1030-z

Huang, Y., R. Sui, S.J. Thomson, and D.K. Fisher. 2013. Esti-

mation of cotton yield with varied irrigation and nitrogen

treatments using aerial multispectral imagery. Int. J.

Agric. Biol. Eng. 6(2):37–41. https://doi.org/10.3965/j.

ijabe.20130602.005

Jin, Z., V.S. Archontoulis, and D.B. Lobell. 2019. How much

will precision nitrogen management pay o? An evalu-

ation based on simulating thousands of corn elds over

the US Corn-Belt. Field Crops Res. 240:12–22. https://

doi.org/10.1016/j.fcr.2019.04.013

Johnson, D.M. 2016. A comprehensive assessment of the

correlations between eld crop yields and commonly

used MODIS products. Int. J. Appl. Earth Obs. Geoinf.

52:65–81. https://doi.org/10.1016/j.jag.2016.05.010

Landivar, J.A., S. Searcy, and J. Stewart. 1999. A variable-

rate chemical application system for cotton production in

south Texas. p. 581–583 In Proc. Beltwide Cotton Conf.,

Orlando, FL. 3-7 Jan. 1999. Natl. Cotton Counc. Am.,

Memphis, TN.

Leal, A.J.F., G.L. Piati, R.C. Leite, M.S. Zanella, C.R.W.S.

Osorio, and S.F. Lima. 2020. Nitrogen and mepi-

quat chloride can aect ber quality and cotton yield.

Rev. Bras. de Eng. Agricola e Ambient. 24(4):238–

243. https://doi.org/10.1590/1807-1929/agriambi.

v24n4p238-243

Lowenberg-DeBoer, J., and B. Erickson. 2019. Setting

the record straight on precision agriculture adoption.

Agron J. 111(4):1552–1569. https://doi.org/10.2134/

agronj2018.12.0779

Martins, R.N., F.A.C. Pinto, A.D. Moura, W.C. Siqueira, and

F.M. Villar. 2020. Nitrogen variable rate fertilization

in corn crop prescribed by optical sensor. J. Plant Nutr.

43(11):1681–1688. https://doi.org/10.1080/01904167.20

20.1729805

Nawar, S., R. Corstanje, G. Halcro, D. Mulla, and

A.M. Mouazen. 2017. Delineation of soil manage-

ment zones for variable-rate fertilization: A review.

Adv. Agron. 143:175–245. https://doi.org/10.1016/

bs.agron.2017.01.003

Nelson, J.R. 2006. Site-specic management and remote sens-

ing based plant growth regulator application decisions

in cotton. Ph.D. diss. North Carolina State University,

Raleigh. http://www.lib.ncsu.edu/resolver/1840.16/734

O’Berry, N.B., J.C. Faircloth, M.A. Jones, D.A. Herbert-

Jr, A.O. Abaye, T.E. McKemie, and C. Brownie. 2009.

Dierential responses of cotton cultivars when applying

mepiquat pentaborate. Agron. J. 101(1):25–31. https://

doi.org/10.2134/agronj2007.0333

Rousseeuw, P.J. 1987. Silhouettes: a graphical aid to the in-

terpretation and validation of cluster analysis. J. Comput.

Appl. Math. 20:53–65. https://doi.org/10.1016/0377-

0427(87)90125-7

Ruβ G., and R. Kruse. 2011. Exploratory hierarchical clus-

tering for management zone delineation in precision

agriculture. p. 161–173 In P. Perner (ed.), Advances

in Data Mining. Applications and Theoretical Aspects.

ICDM 2011. Lecture Notes in Computer Science, Book

Series, v. 6870. Springer, Berlin, Heidelberg. https://doi.

org/10.1007/978-3-642-23184-1_13

Samples, C., D.M. Dodds, A.L. Catchot, B.R. Golden, J. Gore,

and J.J. Varco. 2015. Determining optimum plant growth

regulator application rates in response to fruiting struc-

ture and ower bud removal. J. Cotton Sci. 19:359–367.

Sawan, Z.M. 2018. Mineral fertilizers and plant growth

retardants: Its eects on cottonseed yield; its quality and

contents. Cogent Biol. 4:1–9. https://doi.org/10.1080/233

12025.2018.1459010

VAZ ET AL.: ZONAL PGR APPLICATION IN COTTON TO REDUCE VARIABILITY

Stamatiadis, S., J.S. Schepers, E. Evangelou, A. Glampedakis,

M. Glampedakis, N. Dercas, C. Tsadilas, N. Tserlikakis,

and E. Tsadila. 2020. Variable-rate application of high

spatial resolution can improve cotton N-use eciency

and protability. Precis. Agric. 21:695–712. https://doi.

org/10.1007/s11119-019-09690-6

Thurman, M.E., and R.W. Heiniger. 1999b. Evaluation of

variable rate Pix (mepiquat chloride) application by

soil type. p. 524–526 In Proc. Beltwide Cotton Conf.,

Orlando, FL, 3-7 Jan. 1999. Natl. Cotton Counc. Am.,

Memphis, TN.

Thurman, M.E., and R.W. Heiniger. 1999a. Using GPS to

scout cotton for variable rate Pix (mepiquat chloride)

application. p. 953–961 In P.C. Robert, R.H. Rust, and

W.E. Larson (ed.). Proc. Fourth Int. Conf. on Precision

Agric., St. Paul, MN, 19–22 Jul. 1998. ASA, CSSA, and

SSSA Books.

Trevisan, R.G., N.S. Vilanova-Júnior, M.T. Eitelwein, and J.P.

Molin. 2018. Management of plant growth regulators

in cotton using active crop canopy sensors. Agriculture

8(7):101. https://doi.org/10.3390/agriculture8070101

Tung, S.A., Y. Huang, A. Hafeez, S. Ali, A. Liu, M.S. Chattha,

S. Ahmad, and G. Yang. 2020. Morpho-physiological ef-

fects and molecular mode of action of mepiquat chloride

application in cotton: A review. J. Soil Sci. Plant Nutr.

20:2073–2086. https://doi.org/10.1007/s42729-020-

00276-0

United States Department of Agriculture [USDA]. 2023. Cot-

ton: World Markets and Trade. Available at https://apps.

fas.usda.gov/psdonline/circulars/cotton.pdf (veried 15

Feb. 2023).

Vistro, R., Q.I. Chachar, S.D. Chachar, N.A. Chachar, A. La-

ghari, S. Vistro, and I. Kumbhar. 2017. Impact of plant

growth regulators on the growth and yield of cotton. Int.

J. Agric. Technol. 13:353–362.

Ward-Junior, J.H. 1963. Hierarchical grouping to optimize an

objective function. J. Am. Stat. Assoc. 58(301):236–244.

https://doi.org/10.1080/01621459.1963.10500845

Yu, J., X. Yin, T.B. Raper, S. Jagadamma, and D. Chi. 2019.

Nitrogen consumption and productivity of cotton

under sensor-based variable-rate nitrogen fertilization.

Agron. J. 111(6):3320–3328. https://doi.org/10.2134/

agronj2019.03.0197

Zhou, X.V., B.C. English, J.A. Larson, D.M. Lambert, R.K.

Roberts, C.N. Boyer, M. Velandia, L.L. Falconer, and

S.W. Martin. 2017. Precision farming adoption trends in

the Southern U.S. J. Cotton Sci. 21:143–155. https://doi.

org/10.56454/GGKY6251