A 2-haplot (33.448360˚ latitude, −90.869928˚ longitude) located at the United States Department of Agriculture, Agriculture Research Service, Mechanization Farm was used for this study. The farm is in Washington County, MS and is located approximately 5.8 km northeast of the city of Stoneville, MS (Figure 1). The study plot is referred to as Field 14 (Figure 1). The field consisted of Sharkey clay, (0.5% to 2% slope, very-fine, smectitic, thermic Chromic Epiaquerts) and Tunica clay (0% to 2% slope, Clayey over loamy, smectitic over mixed, superactive, nonacid, thermic Vertic Epiaquepts).The average rainfall, maximum temperature, and minimum temperature of Washington County, MS are 135 cm, 33.6 degrees Celsius in July, and 2.2 degrees Celsius in January, respectively [13]. Major crops grown in Washington County include soybeans, corn, cotton, rice (Oryza sativa L.), and wheat [14].

The soybean cultivars planted in the field for the 2018 and 2019 growing seasons were Dyna-Gro “31RY45” and Asgrow “AG45X”, respectively. The following are the agronomic traits of Dyna-Gro “31RY45”: 1) maturity group—4.5, 2) herbicide tolerance—roundup ready 2 yield trait, 3) flower color—purple, 4) pubescence color—light tawny, 5) pod color—brown, 6) hilum color—black, and 7) growth habit—indeterminate. Asgrow “AG45X8” has the following plant characteristics: 1) maturity group—4.5, 2) herbicide tolerance—roundup ready 2

xtend (i.e., dicamba and glyphosate herbicide tolerance), 3) flower color—purple, 4) pubescence color—light tawny, 5) pod color—brown, 6) hilum color—black, and 7) growth habit—indeterminate. The planting dates were May 2, 2018 and 2019. For the 2018 growing season, the field contained a twelve-row buffer of corn planted on the east and west sides of the plot and a 15-m buffer of corn planted on the north end of the plot. Therefore in 2018, plant mapping was limited to the row area planted to soybeans.

The sensor described by [5] was used in this study. It consisted of a battery-powered, microcontroller-based system containing a Bluetooth radio, global positioning receiver, four infrared temperature sensors (plant canopy temperature measurements), four ultrasonic distance sensors (plant height measurement), micro secure digital card, and voltage regulation components. The sensors were installed in plastic enclosures, with one temperature and one ultrasonic sensor in each enclosure, with enclosures mounted at 1-m spacings along a horizontal aluminum channel. The system was then mounted on the front of an agricultural spray vehicle to collect data simultaneously from four crop rows. Prior to data collection in the field, the sensor system was driven to a bare soil area adjacent to the study plot; the distance between the plant height sensors and bare ground area was measured to establish the maximum height of the sensor above ground level, the reference distance. This information was used to calculate the plant height measurements obtained with the ultrasonic sensor in the field. Additionally, the distance between the sensor and bare ground level was measured with a tape measure to evaluate the accuracy of plant height sensors. The plant height sensors measurements were ±2 cm of the manual measurements. In post-processing of the data, the plant height was determined by subtracting the measured distance obtained from the plant canopy sensor during field measurements from the bare ground reference-distance measurement.

The study plot was divided into four-row transects for data collection. Figure 2 shows an example of the sensor system collecting data from one of the

four-row transects. Note the sensing system consists of four sensors; each sensor was collecting plant height and plant temperature data from one of the rows within the transect. Data were collected on July 26, 2018 and August 8, 2019 between 13:48 to 14:56 hrs. and 9:17 to 10:16 hrs. daylight savings time, respectively.

For further analysis, we only used data collected from the two center sensors. Measurements from these sensors coincide with other field sensors that are being tested on this plot. Data were cleaned prior to further analysis including removal of inaccurate values and points that were collected outside of the field boundary. Then, the plant height and plant canopy temperature data were averaged for the two center sensors.

Descriptive statistics (i.e., minimum, maximum, mean, standard deviation, skewness), histograms, and frequency distribution plots were used to evaluate the normality and variability of the plant height and the plant canopy temperature datasets. Pearson correlation (p ≤ 0.05) [15] was employed to determine the relationship between the plant height and plant canopy temperature measurements. The R software [“Action of the Toes” [16] ] was used to calculate the descriptive statistics and the Pearson correlation coefficient.

Geostatistical techniques including variogram models and kriging were explored to determine the spatial correlation of plant height and plant canopy temperature measurements and to derive maps of plant height and plant canopy temperature, respectively. Isotropic and anisotropic (0, 45, 90, and 135 degrees) variograms were evaluated to derive the model for describing and mapping the spatial variation of the plant parameters. Spherical, linear, exponential, and gaussian variogram models were tested. They were selected by using a combination of residual sums of squares, coefficient of determination, and manual evaluation of the model fit to the calculated variogram. Model accuracy was tested with the leave-one-out cross-validation method provided by the software. After selecting the appropriate variogram model, we created maps via block kriging (i.e., ordinary kriging) with a window size of 8 × 8 meters. Variogram model selection, evaluation, and block kriging were completed with the Geostatistics for Environmental Science software (GS+, version 10, Gamma Design, Plainwell, MI). Geostatistical techniques are commonly used to model and map agricultural variables; for a more detailed description of geostatistical methods, see Oliver and Webster [17] [18]. Block kriged data were imported into the QGIS software to develop the final maps (Madeira, Version 3.4, Long term release, [19]).

Descriptive statistics are summarized in Table 1 for the 2018 plant height and plant canopy temperature measurements. One thousand eighty-two points were used for the data analysis. A difference of 65.1 cm was observed in the plant height; whereas, a difference of 11.7˚C occurred for the plant canopy temperature.

In 2019, 965 points were used for data analysis (Table 1). The minimum and maximum values indicated an in-field variability of 40 cm in plant height. Plant canopy temperature was less variable; only a 2˚C difference between the minimum and maximum values was observed. Overall, the 2019 dataset was less variable than the 2018 dataset.

Moderate to weak statistically significant (p ≤ 0.05) negative correlations were observed between the plant height and the plant canopy temperature (Table 2). The best correlation occurred for the 2018 data measurements, indicating the strength of the relationship can change from one year to the next.

The plant height and the plant canopy temperature data were spatially correlated (Table 3). Isotropic exponential models best described the plant height spatial correlation for the 2018 and 2019 field measurements and the plant canopy temperature measurements for the 2019 dataset. An isotropic spherical model best described the spatial variability of the plant canopy temperature in 2018.

^aPLH—Plant height; PLCT—plant canopy temperature; Min—minimum; Max—maximum; SD—standard deviation; Var—variance; and Skew—skewness.

^a*Statistically significant at p ≤ 0.05.

^aPLH—Plant height; PLCT—plant canopy temperature; RSS—residual sums of squares; r²—how well the model fits the variogram.

The spatial correlation of plant height was 102.8 m and 71.4 m for 2018 and 2019, respectively. The spatial correlation for temperature was 92.2 m and 21 m for the 2018 and 2019 datasets, respectively. Plant height was spatially correlated over a longer distance than the plant canopy temperature.

Plant height and plant canopy temperature maps derived from block kriging are shown in Figure 3. In 2018, the tallest soybean plants were observed in the southwest region of the plot; shorter plants were detected in the mid central and southeast sections of the field (Figure 3(a)). The plant canopy temperature in-field variability pattern was similar to the plant height measurements’ in-field variability pattern (Figure 3(b)). The lowest canopy temperature occurred in the southwest region of the plot where the tallest soybean plants were mapped.

For the 2019 results, the tallest soybean plants occurred in the eastern and the southwestern sections of the plot (Figure 3(c)). There was a general agreement in the variability patterns of plant height and plant canopy temperature in that increases in plant height were associated with decrease in plant canopy temperature (Figure 3(c) and Figure 3(d)). The plant height differences observed in the southwest section of the field were consistent with the plant height differences observed in 2018. Furthermore, in 2019, the shortest plants were observed in the northern section of the field plot.