The study was conducted at the Conservation and Production Research Laboratory (CPRL) in Bushland, Texas (35° 10’2.42” N, 102°5’32.5” W, elevation 1142 m). The site is in a semi-arid region with mean annual rainfall of approximately 400 mm, and variable rainfall during the cropping season (May – September). The soils were Pullman clay loam fine, superactive, mixed, thermic torretic Paleustoll, (Unger and Pringle, 1981). The soil is slowly permeable with a hard pan layer of calcium carbonate at approximately 1.5 m and the plant-available water holding capacity to this depth is roughly 210 mm (Evett et al., 2019). Mean bulk density ranges from approximately 1. 4 to 1.7 Mg m^-3 (Evett et al., 2022). The source of irrigation water is the Ogallala Aquifer. Practicing deficit irrigation in the surrounding region is important as saturated thickness of the aquifer in this region is depleted between 7 to 15 m ( Figure 1 ).

Cotton (Gossypium hirsutum L.), variety Phytogen 210 W3FE, was planted on June 4, 2021, day of year (DOY 155) and on May 13, 2022, (DOY 133) under one-half of a center pivot field (10.5 ha) and irrigated with a six-span center pivot sprinkler system. The cotton variety was an early maturing genotype. Critical growth stages for the two years were offset by the difference in planting dates ( Table 1 ).

The center pivot irrigation system was a six-span Valley 8000 series with a pivot lateral length of 286 m. The sprinkler was equipped with a commercial zone control variable rate irrigation (VRI) hardware package, a GPS unit at the end tower, and the Irrigation Supervisory Control and Data Acquisition (ISSCADA) system, patented by ARS scientists (Evett et al., 2014) and described in detail in Andrade et al. (2020). The irrigation scheduling methods used by the ISSCADA system are summarized under “Irrigation Scheduling Methods” later in this section.

For this study, two sprinkler zones were combined to operate as one control zone. Each control zone contained 12 drop hoses and measured a total of 18.3 m wide. Irrigation was applied using low elevation spray application (LESA) with low-drift nozzles and multi-trajectory spray plate assemblies (3030 Series, 3NV, Nelson, Walla Walla, WA) elevated approximately 0.46 m above the ground. Irrigation buffers in the shape of pie-slices were established at the beginning and end of each half of the cropped field to ensure the VRI system had time to synchronize the operation of its zone-controlled sprinkler banks at the start of an irrigation and to provide a cropped buffer area for the treatment plots.

An embedded rugged computer (model MXE-1401, ADLINK Technology, San Jose, CA) with a cellular modem was located at the pivot point and connected to the control panel of the center pivot sprinkler with an RS-232 cable. Also connected to the computer were two radio modems- the SAPIP coordinator (Dynamax, Inc., Houston, TX), and a spread spectrum radio (model RF-407, 900 MHz, Campbell Scientific, Logan, UT). A total of 10 infrared thermometers (IRTs) (model- SAPIP-IRT, Dynamax, Inc., Houston, TX) were mounted on the pivot lateral. Two pairs of IRTs were mounted on brackets near the borders of each control zone and pointed towards the center of the zone. In addition, one IRT was located in each of the M₁₀₀ treatment plots. These data were averaged and used as the reference temperature curve to scale data from the moving network of IRTs on the pivot lateral. The IRTs measured object temperature, sensor body temperature, and battery voltage every 6 s. One-minute averaged data packets were transmitted to the computer at the pivot point. More details on the IRTs and the IRT network communication protocol are found in O’Shaughnessy et al. (2011).

A soil water sensing station was located in each type of ISSCDA-hybrid treatment plot (H₇₅, H₅₀ and H₂₅) to provide input of soil water depletion levels. The soil water sensing stations were established as a distributed wireless sensor network, where each station contained a datalogger (CR300 with an internal RF-407 radio, Campbell Scientific, Logan, UT) and transmitted data hourly to the base station RF-407 radio connected to the embedded computer at the pivot point. Each datalogger was powered by a sealed lead acid battery that was recharged with a 10-W solar panel. A trench was dug in a furrow and four soil water sensors, time domain reflectometers (TDRs), (model TDR-315L or 315H, Acclima, Meridian, ID) were installed horizontally near the cotton plants at depths of 10 cm, 20 cm, 30 cm, and 50 cm. Data from the TDRs were acquired every 1-minute, averaged every 15-minutes. Each soil water sensing station was within 1 m of the neutron access tube installed in the treatment plot.

A standalone weather station was located nearby the center pivot field. Air temperature, relative humidity, solar irradiation, wind speed and wind direction were collected with the base-station datalogger (CR1000X, Campbell Scientific, Logan, UT). All micrometeorological data were measured every 5 s, averaged every 1-minute and transmitted hourly to the computer at the pivot point. The embedded computer was accessed in the office through the cellular connection using the ‘Remote Desktop’ application. Accessing the embedded computer allowed for monitoring data collection, reviewing and uploading prescription maps and starting an irrigation or scan (running the sprinkler dry across the field). All center pivot operations were conducted with the ARSPivot software, which resided on the embedded computer.

Fertilizer was applied using a knife rig to meet a yield goal of 682 kg ha^-1 of cotton based on test results from composite soil samples taken in March of 2021 and 2022 from the W-SW and N-NE halves of the field, respectively ( Table 1 ). Test results were from a commercial soil testing laboratory. Pre-plant irrigations were applied to help enable germination. Two pre-plant irrigations were applied in 2021 in late April, and four pre-plant irrigations were applied in 2022 due to a dry winter and spring season. Each irrigation event was 25 mm. In both years, cotton was planted at a rate of 21 seeds m^-2 and 2.5 cm deep in circular rows spaced 0.76 m apart using a GPS guided planter. The W-SW half of the center pivot field was cropped in 2021, while in 2022, the N-NE half of the field was cropped; in both years, planting was on ground fallowed in the previous season. Post-plant herbicides were applied to control weeds, mainly pigweed (Amaranthus spp.) and Devil’s Claw (Harpagophytum spp.). The pesticide, Govern 4E, was applied to avert extensive damage from thrips (Thysanoptera) ( Table 1 ).

For each growing season, the field was divided into 44 treatment plots (MZs), where each treatment- M₁₀₀, M₇₅, M₅₀, M₂₅, M₀, H₇₅, H₅₀, H₂₅, C₇₅, C₅₀ and C₂₅ was replicated four times and arranged in a complete randomized design ( Figure 2 ). The labels “M”, “H”, and “C” designate the irrigation scheduling methods- manual (M), the ISSCADA-hybrid (H), and the ISSCADA-plant feedback I system, respectively. The numeric values designate the irrigation treatment level, I₁₀₀, I₇₅, I₅₀, and I₂₅. The length of each plot varied depending on the radial distance of the control zone from the pivot point. The width of each plot was 18.2 m and plot lengths varied from 32.0 m to 57.3 m at half width, depending on their distance from the pivot point. A neutron access tube was installed near the center of each plot to a depth of 3.0 m. Plant height and width measurements were recorded starting in the second week of July and continued biweekly through the last week in August. All agronomic measurements and hand-harvested samples were taken in each treatment plot from a 10 m² area (4 rows x 328 cm) centered around the access tube.

Heat units (HU, °C) were calculated to characterize the growing environment for cotton using equation 3:

where T_max , T_min are the maximum and minimum daily air temperature, respectively and 15.6°C was used as the base temperature (T_b ), typical for the study region (Mahan et al., 2014; Masasi et al., 2020).

Seasonal evapotranspiration (ET_c) was calculated for the growing season using the soil water balance equation 4:

where I was cumulative irrigation, P was seasonal precipitation, ΔS was the change in soil water content from the first to the last NP measurement to a depth of 2.3 m, F was flux across the lower boundary of the control volume, and R was runoff, all units were in mm. Values of F were assumed to be zero as NP measurements did not indicate a change in flux in soil water content below 170 cm (data not shown).

Cumulative precipitation amounts from May through October of each year were 305 mm and 326 mm for the 2021 and 2022 growing seasons, respectively ( Table 3 ). However, the pattern of rainfall was significantly different between the two years. Rainfall from January through May of 2021 totaled 145 mm. The total rainfall in May occurred prior to planting, thereafter, measurable rainfall occurred approximately every 14 days. Monthly cumulative precipitation for August and September were smaller than the average values for these months. In 2022, total precipitation from January through the first half of May was less than 32 mm. While precipitation in this second growing season was greater than 2021, approximately 42% occurred on July 29 and July 30. Outside of these two events, 86% of all rainfall events for the season were recorded to be less than 10 mm. Except for the month of July, monthly cumulative precipitation was smaller than the 58-year average precipitation amounts for the study location (O’Shaughnessy et al., 2014). Preplant precipitation in 2021 provided adequate moisture in the soil profile, avoiding excessive postplant irrigations. However, the relatively cool air temperatures in June and July slowed crop maturation. Even so, the overall limited precipitation during the growing season aided the ISSCADA system in detecting different levels of crop water stress. In 2022, the paucity of precipitation in the winter and spring required postplant irrigations to germinate and establish a uniform stand.

Postplant irrigations were applied to establish a uniform stand; these irrigations totaled 38 mm and 95 mm in 2021 and 2022, respectively ( Table 4 ). The uniform postplant irrigations were greater in 2022 for two reasons- drought conditions prevailed in the winter and spring months of 2022, and secondly, the VFD for the center pivot pump failed in early June (after planting), and two uniform irrigations were applied prior to replacing the equipment to assure adequate soil moisture during the down time. Irrigations were terminated in the treatment plots designated M₀, when VRI applications were initiated on July 10, 2021, and July 15, 2022. Soil water sensing stations were deployed on July 7, 2021, and July 12, 2022, in the appropriate ISSCADA-hybrid treatment plots.

Plant height is a good indicator for vegetative cotton growth (Ritchie et al., 2010; Zhou and Yin, 2014). Throughout the growing season, mean plant heights in 2021 were closely clustered among irrigation levels for the manual and ISSCADA-hybrid irrigation scheduling treatment plots until the end of the irrigation season. The increase in plant heights at the end of the season could have resulted from the last two irrigations at the end of the season ( Figure SM-1 ). Plant heights in the ISSCADA-plant feedback method showed distinct heights among irrigation levels beginning Aug 12 ( Figure 4 ). In 2022, plant height appeared to increase quickly after the intensive rainfall event of July 30, indicating luxury vegetative growth between Jul 26 and Aug 11 ( Figure 4 ). Apart from the H₇₅ treatment plots, scans executed on Aug 3 and Aug 6 resulted in withholding irrigations on Aug 4 and Aug 7 in all ISSCADA-plant feedback treatment plots and in the ISSCADA-hybrid treatment plots at the I₅₀ and I₂₅ levels ( Figure 1 . SM- 2). The irrigation amount prescribed for the H₇₅ treatment plots resulted from the _iCWSI threshold value as the calculated SWD level was between the minimum and maximum thresholds. Irrigations were terminated on Sep 9, 2021, and on Aug 26, 2022, after observing that most plants had at least five vegetative nodes above first white flower. Although a scan was executed on Aug 11, 2022, irrigations were only recommended for the H₇₅ treatment plot. The last irrigation applied on Aug 26 was erroneously applied as a uniform irrigation of 11.4 mm.

The M₁₀₀ and M₀ treatment plots were established as control treatments for determining maximum irrigation amounts and evaluating IWUE, respectively. Individual comparisons of mean cumulative irrigation and mean crop response of seed cotton yield and IWUE were made between the M₁₀₀ control treatment ( Table 4 ) and means from each type of deficit irrigation level X irrigation scheduling treatment. Due to differences in climatic conditions, data were analyzed separately by year. Of interest when comparing mean differences were reductions and overages in irrigation applied, gains and losses in seed cotton yield, and improvements in IWUE. In 2021, the comparisons showed that mean cumulative irrigations in the deficit irrigated treatments were significantly less as compared with the M₁₀₀ treatment, yet mean seed cotton yields were not significantly different at p< 0.05. This indicated that no significant yield loss had occurred between any of the deficit irrigation treatments and the fully irrigated treatment. Mean IWUE values were numerically greater in all the deficit irrigated treatments, but not significantly different from the M₁₀₀ treatment (at p< 0.05). In 2022, mean cumulative irrigations for the deficit irrigated treatment plots were again significantly less than the mean for the M₁₀₀ treatment. Mean seed cotton yields were statistically similar as compared with the M₁₀₀ treatment, however, mean yields were not sustained in the M₂₅, C₅₀, C₂₅, H₅₀ and H₂₅ treatments. Mean IWUE values were not significantly different from the M₁₀₀ treatment, though IWUE means were numerically greater in all I₇₅ irrigation level by irrigation scheduling treatments. The mean seed cotton yield in the M₀ treatment was only 11% less than that in the M₁₀₀ treatment and was relatively large, due to the seed cotton yield measured in plot 1, which was at the 90^th quantile of the distribution of harvested values within the M₀ treatment. Removing this value as an outlier, increased mean IWUE values for all treatments, although the adjusted differences were not significantly different compared with the M₁₀₀ treatment. Means in Tables 4 – 6 include all measured values.

Pairwise means were compared to investigate differences between irrigation level, irrigation method and their interaction at the I₇₅, I₅₀ and I₂₅ levels. Data were analyzed separately for each growing season. In 2021, when data were grouped by irrigation level, cumulative irrigation amounts and ET_c were significantly different across the three irrigation levels ( Table 5 ). Overall seed cotton yield was significantly less in the I₂₅ treatment level. Mean IWUE was not significantly different among irrigation levels. Grouping the data by irrigation scheduling method demonstrated that scheduling methods had a significant impact on cumulative irrigation amount and ET_c in the ISSCADA-plant feedback treatment plots, resulting in significantly less water applied to and significantly smaller ET_c values in these plots ( Table 5 ). However, mean seed cotton yield, CWP and IWUE were statistically identical among scheduling methods.

Analyzing all sets of pairwise means for irrigation level by irrigation scheduling method indicated that mean cumulative irrigation for the C₇₅ treatment plots was significantly less compared with the M₇₅ and H₇₅ treatments. At the I₅₀ and I₂₅ levels, mean cumulative irrigations were similar among irrigation scheduling methods. Mean differences in ET_c response followed the same statistical trend as differences in cumulative irrigation. Mean seed cotton yields were only significantly greater between the M₇₅ treatment and all irrigation scheduling methods at the I₂₅ level. Mean values of CWP and IWUE were not statistically different among all irrigation level X irrigation scheduling treatments.

In 2022, as in the previous season, results showed that when data were grouped by irrigation level, overall mean cumulative irrigation and ET_c values were significantly different. However, overall mean seed cotton yield, CWP and IWUE were not affected by irrigation level ( Table 6 ). Aggregating data by irrigation scheduling method showed that mean cumulative irrigations applied to the ISSCADA-hybrid treatment plots were less compared with the manual and ISSCADA-plant feedback treatment plots. However, mean seed cotton yield, ET_c, CWP and IWUE were statistically identical among irrigation scheduling methods.

Analyses of the pairwise means for irrigation level by irrigation method showed that the mean cumulative irrigations at the I₇₅ and the I₂₅ irrigation levels were similar among irrigation scheduling methods. In the I₅₀ treatment, cumulative irrigation for the ISSCADA-hybrid irrigation scheduling method was significantly smaller as compared with the manual and ISSCADA-plant feedback methods. Mean seed cotton yield, CWP and IWUE were similar for all irrigation level by irrigation scheduling treatments. Results were mixed for differences among mean ET_c values, where ET_c values for the M₂₅ and H₂₅ treatments were significantly less than all irrigation scheduling methods at the I₇₅ level ( Table 6 ).