Thermal Imaging Detects Early Drought Stress in Turfgrass Utilizing Small Unmanned Aircraft Systems

Recent advances in aerial platforms and thermal imaging provide opportunities to improve water management in turfgrass, but research on this topic is limited. Our objectives were to: (i) evaluate the ability of canopy temperature (Tc) imaging from small unmanned aircraft systems (sUAS) to detect drought stress early in turfgrass; (ii) compare early drought-stress detection ability of Tc measurements with that of sUAS-mounted and handheld optical sensors; and (iii) evaluate thermal data’s relationship to spectral reflectance from sUAS-mounted and handheld sensors, soil volumetric water content (VWC), soil temperature, turfgrass visual quality (VQ), and percentage green cover (PGC). The study was conducted during summer 2017 on creeping bentgrass (Agrostis stolonifera L.) irrigated with 15 to 100% evapotranspiration (ET) replacements to impose a gradient of drought stress. Airborne spectral reflectance measurements included three bands (near infrared [NIR, 680–780 nm], green and blue bands [overlapped, 400–580 nm]) and eight derived vegetation indices. Results indicated Tc measurements via the sUAS detected rises of Tc in 15 and 30% compared with 100% ET plots, corresponding with declines in VWC, before drought stress became visible. This was comparable to the best spectral parameters on companion flights, and Tc was closely correlated with spectral data from sUASmounted (|r| = 0.52–0.69) and handheld sensors (|r| = 0.75–0.82). Thermal data were more strongly correlated with turfgrass VQ (r = -0.60 to -0.77) and PGC (r = -0.58 to -0.78) than with VWC (r = -0.43 to -0.63) and soil temperature (|r| = 0.27–0.41).


Summary
Plots of fairway-height creeping bentgrass were watered differently to create a gradient of drought stress from severe deficit irrigation to well-watered, under an automatic rainout shelter in Manhattan, KS. Canopy temperature (Tc) measured by a small unmanned aerial system (sUAS) predicted drought stress approximately 5 days or more before drought symptoms were evident in either turfgrass visual quality (VQ) or percentage green cover (PGC). The ability of Tc to predict drought stress was comparable to the best spectral parameters acquired by sUAS on companion flights [i.e., near infrared (NIR) and GreenBlue VI], and slightly better than with spectral data obtained from handheld sensors. Better drought-prediction ability combined with faster data collection using sUAS indicates significant potential for sUAS-based compared with ground-based drought stress monitoring methods.

Rationale
Recent advances in aerial platforms and thermal imaging provide opportunities to improve turfgrass management, including early drought detection and water conservation, but research on this topic has been limited.

Objectives
The objectives of this study were to 1) evaluate the ability of Tc imaging from sUAS to detect drought stress early in turfgrass; and 2) compare early drought-stress 1 GD Animal Health, Deventer, The Netherlands.
Kansas State University Agricultural Experiment Station and Cooperative Extension Service detection ability of Tc imaging with that of spectral reflectance measurements from sUAS-mounted and handheld optical sensors.

Study Description
This 1-year study was a companion study of Hong et al. (2018) conducted from June 7 to August 31, 2017, under an automatic rainout shelter at the Kansas State University Rocky Ford Research Center, Manhattan, KS. 'Declaration' creeping bentgrass was mowed at 5/8-inch height and treated from severe deficit irrigation to well-watered with 15, 30, 50, 65, 80, and 100% reference evapotranspiration (ET) replacement. A thermal camera (FLIR VUE PRO R 336, 35º FOV, 9-mm focal length), mounted on an IRIS+ (3D Robotics), was used to attain Tc weekly when weather permitted. One thermal image containing all the plots was selected for each measurement date, and average Tc was retrieved from the center of each plot using the FLIR tool (v. 5.13.), which also produced color-enhanced thermal maps. A modified Canon S100 camera was used to take airborne spectral reflectance measurements, including reflectance of three individual bands (NIR, green and blue bands), and their eight derived vegetation indices (VI). Traditional measurements included soil volumetric water content (VWC), VQ, and PGC using digital image analysis, and spectral reflectance measurements from handheld optical sensors.

Results
Stress in deficit-irrigation treatments was detected with Tc measurements from sUAS as the dry down progressed (Figure 1). Thermal imaging detected rises of Tc in 15% and 30% compared to 100% ET plots on June 15, corresponding to declines in VWC, before evident decreases in VQ and PGC (Table 1; Figure 1A). However, declines of PGC and VQ in 30 and 15% ET treatments compared to 100% ET were not observed until 5 days and 16 days later on June 20 and July 1, respectively. It is not certain that Tc detected drought stress symptoms a full 16 days early, and it is possible that drought symptoms became visible before July 1, because VQ was not evaluated on days between sUAS flights (e.g., between June 20 and July 1). This indicated the ability of Tc imaging from sUAS to detect drought stress more than 5 days before it became visible in turfgrass.
Interestingly, the early drought detection by thermal imaging was as early as NIR and GreenBlue VI [(green-blue)/(green+blue)] measured on a companion flights, and normalized difference vegetation index (NDVI) and Red band of a handheld active optical sensor (Table 1). The NIR and GreenBlue VI were the best spectral parameters reported in the companion study that consistently predicted drought stress throughout the 3-yr study (Hong et al., 2018). Moreover, Tc, NIR, and GreenBlue VI from the sUAS were more sensitive at detecting early drought stress in the lower irrigation levels in that they differentiated 100% ET plots from 15% and 30% ET plots, whereas the handheld sensor only detected early drought stress in 15% ET (Table 1). Better drought-prediction ability combined with faster data collection Kansas State University Agricultural Experiment Station and Cooperative Extension Service using sUAS indicates significant potential for sUAS-based compared with groundbased drought stress monitoring methods. ‡Visual quality based on a 1 to 9 scale, with 1 = dead; 6 = minimally acceptable; and 9 = uniform, green, dense turfgrass. §GreenBlue = (G-B)/(G+B); G = green spectral reflectance, and B = blue spectral reflectance overlap between 400-580 nm. ¶NDVI = (NIR-Red)/(NIR+Red); NIR peaks at 780 nm, near infrared spectral reflectance; Red peaks at 670 nm.