Unmanned Aerial Remote Sensing for Estimating Cotton Yield
Abstract
Unmanned aerial systems (UAS) allow collection of imagery with unprecedented temporal, spatial, and spectral resolutions suitable for specialized purposes. Crop yield data is critical both for precision agriculture management purposes and crop breeding programs. However, collecting yield data at fine scales necessary for small plot research is labor-intensive. UAS could be leveraged to quantify yield variability while limiting labor requirements. Therefore, the objectives of this dissertation were to examine the relationship between cotton yield and derivatives from UAS multispectral and thermal infrared imagery and to determine optimal in-season timing of UAS flights for the strongest relationship with cotton yield. The experimental design was a 3x8 factorial within a completely randomized design arrangement with four repetitions and the study was conducted over four growing seasons (2017-2020). One treatment factor was three irrigation levels applied as a percentage of the estimated crop evapotranspiration (ETc) requirement: 0%, 40%, and 80% ET replacement while the other factor was eight commercial cotton cultivars. UAS imagery was acquired at biweekly intervals to produce high resolution multispectral and thermal infrared orthomosaics. Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge (NDRE), a pixel-based classification of cotton bolls termed Boll Area Index (BAI), and canopy temperature were derived from the orthomosaics and analyzed to determine suitability for cotton yield estimation. NDVI had a positive linear relationship with yield, which was strongest at approximately 1200 heat units (R² = 0.61, 0.78, 0.49, and 0.78 in 2017, 2018, 2019 and 2020, respectively). There were strong positive linear relationships between BAI and yield each year (R² = 0.61, 0.79, 0.67, and 0.73). Multiple linear regression using vegetation indices, boll area index, and/or canopy temperature from two flight dates produced better yield estimates (Adjusted R² = 0.79, 0.89, 0.84, and 0.81 for 2017, 2018, 2019 and 2020). Vegetation indices, BAI, and canopy temperature could differentiate variation among irrigation levels. Results suggest that derivatives from just two or three UAS flights presents a detailed dataset for cotton yield prediction while limiting labor, risk, requisite computational resources, and equipment wear.
Citation
Siegfried, Jeffrey Alan (2021). Unmanned Aerial Remote Sensing for Estimating Cotton Yield. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /196253.