Estimating yield-contributing physiological parameters of cotton using UAV-based imagery

• Published in: Frontiers in plant science Vol. 14; p. 1248152
• Main Authors: Pokhrel, Amrit, Virk, Simerjeet, Snider, John L., Vellidis, George, Hand, Lavesta C., Sintim, Henry Y., Parkash, Ved, Chalise, Devendra P., Lee, Joshua M., Byers, Coleman
• Format: Journal Article
• Language: English
• Published: Frontiers Media S.A 19.09.2023
• Subjects: fraction of intercepted photosynthetically active radiation; harvest index; multispectral imagery; Plant Science; radiation use efficiency; RGB imagery; unmanned aerial vehicles
• ISSN: 1664-462X; 1664-462X
• DOI: 10.3389/fpls.2023.1248152

Lint yield in cotton is governed by light intercepted by the canopy (IPAR), radiation use efficiency (RUE), and harvest index (HI). However, the conventional methods of measuring these yield-governing physiological parameters are labor-intensive, time-consuming and requires destructive sampling. This study aimed to explore the use of low-cost and high-resolution UAV-based RGB and multispectral imagery 1) to estimate fraction of IPAR (IPAR f ), RUE, and biomass throughout the season, 2) to estimate lint yield using the cotton fiber index (CFI), and 3) to determine the potential use of biomass and lint yield models for estimating cotton HI. An experiment was conducted during the 2021 and 2022 growing seasons in Tifton, Georgia, USA in randomized complete block design with five different nitrogen treatments. Different nitrogen treatments were applied to generate substantial variability in canopy development and yield. UAV imagery was collected bi-weekly along with light interception and biomass measurements throughout the season, and 20 different vegetation indices (VIs) were computed from the imagery. Generalized linear regression was performed to develop models using VIs and growing degree days (GDDs). The IPAR f models had R 2 values ranging from 0.66 to 0.90, and models based on RVI and RECI explained the highest variation (93%) in IPAR f during cross-validation. Similarly, cotton above-ground biomass was best estimated by models from MSAVI and OSAVI. Estimation of RUE using actual biomass measurement and RVI-based IPAR f model was able to explain 84% of variation in RUE. CFI from UAV-based RGB imagery had strong relationship (R 2 = 0.69) with machine harvested lint yield. The estimated HI from CFI-based lint yield and MSAVI-based biomass models was able to explain 40 to 49% of variation in measured HI for the 2022 growing season. The models developed to estimate the yield-contributing physiological parameters in cotton showed low to strong performance, with IPAR f and above-ground biomass having greater prediction accuracy. Future studies on accurate estimation of lint yield is suggested for precise cotton HI prediction. This study is the first attempt of its kind and the results can be used to expand and improve research on predicting functional yield drivers of cotton.