Time‐based remote sensing yield estimates of cotton in water‐limiting environments

Abstract

The use of high-throughput phenotyping aids breeding programs in making more informed selections and advancements. This study's objectives were to determine which proximal remote sensing parameters (normalized difference red edge [NDRE], normalized difference vegetation index [NDVI], difference between canopy and air temperatures [∆T], and plant height) are robust estimators of cotton lint yield and to use a time-integrated function of one parameter as a single phenotypic measurement for predicting yield. This study evaluated remote sensing parameters (NDRE, NDVI, ∆T, and plant height) measured weekly from squaring through boll production and development. Of these measurements, NDRE was most consistent in terms of r 2 , slope, and normality in predicting yield. From these findings, a temporal analysis was calculated as NDRE integrated over the season, namely NDRE-days. Significant r 2 values were detected for the individual remote sensing measurements, with the largest r 2 occurring around peak bloom (80 d after planting). An r 2 of 0.81 was identified between ∆T and lint yield in 2015, whereas in 2017 the largest r 2 value with lint yield was with NDRE (r 2 = .71). The temporal analysis showed a significant relationship between NDRE-days and lint yield (P < .0001; r 2 = .58 in 2015 and r 2 = .68 in 2017) that was not cultivar specific. This study presents a suitable method that breeders could use to efficiently evaluate plant growth and estimate yield for variety selections while cutting resource requirements.Abbreviations: ∆T, difference between canopy temperature and air temperature; ET, evapotranspiration; FM, FiberMax; GDD 15.6 , 15.6 • C growing degree days; GPS, global positioning system; NDRE, normalized difference red edge; NDVI, normalized difference vegetation index; SDI, subsurface drip irrigation; ST, Stoneville.

Cotton phenotyping and physiology monitoring with a proximal remote sensing system

“…In a recent study, Thompson et al. (2020) reported the greatest correlations between NDVI, NDRE, T c − T a , and cotton lint yield in side‐by‐side comparisons that we were able to find in the literature. Peak correlations with lint yield reached nearly .8 in one year, depending on the parameter, though correlations were substantially lower in the other, drier season.…”

“…As discussed earlier, broadscale analysis across multiple growing conditions widens the variation in yield responses and leads to better correlations (Arnall et al, 2016;Jiang et al, 2018;Plant et al, 2000; but does not inform on how accurately high-throughput measures can predict lint yield in side-by-side comparisons of cotton germplasm grown in the same conditions. Thompson et al (2020) reported the greatest correlations between NDVI, NDRE, T c − T a , and cotton lint yield that we were able to find in the literature in comparisons of cotton germplasm grown in the same conditions. Peak correlations ranged from about .5 to almost .8 in one year, depending on the parameter, and trends in the rise and fall of R 2 over time were apparent, but correlations were substantially lower in a relatively dry year.…”

“…Ritchie et al (2010) found that NDRE was more sensitive than NDVI to low green ground cover fractions in cotton, but less sensitive to high fractions, with correlations between NDVI and NDRE of .84 and .97. Similarly, in one year of a 2-yr study, Thompson et al (2020) reported earlier correlations between canopy NDRE with cotton lint yield than NDVI when the canopy was less developed. In the present study, NDVI and NDRE measurements were not taken as early as in these other studies, and there was little difference in their performance.…”

“…The relationship between NDVI and canopy height was linear, though a slight shift in the shape of the end of the curve indicated that NDVI started to become “saturated” at a value of about 0.80, failing to reflect further changes in canopy height beyond that point (Figure 2). Others have reported saturation of NDVI in cotton at or somewhat above values of 0.80 (Attia et al., 2015; Ritchie et al., 2010; Sharma & Ritchie, 2015; Thompson et al., 2020; Xu et al., 2019). The linear nature of the relationship between NDVI and canopy height in this study is likely related to the growth habit of cotton, which is bush‐like, growing outward and upward at similar rates in most varieties, and the fact that the physical height of the phenotyping platform did not change among measurement timings.…”

Bottlenecks and opportunities in field-based high-throughput phenotyping for heat and drought stress

“…Remote sensing approaches allow for data collection on much larger studies encompassing a wide genetic diversity in order to phenotype for abiotic stress resilience. Remote sensing has been utilized for a variety of purposes such as measuring canopy height ( Varela et al , 2017 ; Thompson et al ., 2018 , 2020 ; Zhou et al , 2020 ), biomass ( Neumann et al , 2015 ; Padilla-Chacón et al , 2019 ), canopy temperature ( Romano et al , 2011 ; Pauli et al , 2016 ; Graß et al , 2020 ), and leaf area ( Neilson et al , 2015 ; Zhang et al , 2019 ), and predicting yield ( Rischbeck et al , 2016 ; Becker and Schmidhalter, 2017 ; El-Hendawy et al , 2017 ; Zhou et al , 2020 ). Through the use of specialized vegetation indices (VIs) or spectral bands alone, remote sensing can quickly and efficiently collect data on different traits simultaneously, non-destructively, and with a high spatio-temporal frequency.…”