<link rel="stylesheet" href="styles.cf390c6bfbe7555f.css">

UAV-based proximal remote sensing for nitrogen stress detection and grain yield and protein prediction in winter wheat

Loading...
Thumbnail Image

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Nitrogen (N) management influences both grain yield and grain protein concentration in hard red winter wheat, but crop response to fertilizer N varies across environments. Unmanned aerial vehicle (UAV) multispectral imagery can monitor crop status during the growing season. Its usefulness for N management and pre-harvest prediction, however, depends on the environmental context and on whether models transfer to new locations. This thesis evaluated UAV multispectral sensing of winter wheat across seven Kansas environments during the 2024–2025 growing season. At each location, eight cultivars were grown under seven N rates (0 to 202 kg N ha⁻¹) in a split-plot design. Imagery was collected at up to six growth stages and summarized into eight vegetation indices across three developmental stages: tillering (Feekes 3–5), stem elongation (Feekes 6–9), and grain formation (Feekes 10–11.4). The first study evaluated whether grain yield and protein could be predicted at locations not used for model training, using nested leave-one-location-out validation. Partial least squares regression (PLSR) outperformed random forest and XGBoost. GNDVI at stem elongation and N rate were the dominant yield predictors. MTCI at grain formation, N rate, and plant height were the most important protein predictors. The final PLSR models achieved R² = 0.55, concordance correlation coefficient (CCC) = 0.72, and RMSE = 516.4 kg ha⁻¹ for grain yield, and R² = 0.47, CCC = 0.67, and RMSE = 1.09 percentage points for grain protein. Removing cultivar information reduced protein R² from 0.47 to 0.14.

The second study evaluated UAV detectability of fertilizer-N response across environments with contrasting background soil N supply. Profile soil NO₃-N ranged from 29 to 270 kg N ha⁻¹, and grain-yield response to fertilizer N ranged from −0.7% to 105%. N detectability increased with crop development. Red-edge-based indices were more sensitive than NDVI during stem elongation (Feekes 6–9). Greater soil nitrate tended to correspond with weaker yield response and weaker red-edge detectability (ρ = −0.68, P = 0.094). Red-edge detectability ranked environments in the same order as relative yield response (ρ = 1.00). Normalizing vegetation indices to a local N-rich reference strengthened their relationship with yield response; for NDVI, |ρ| increased from 0.36–0.79 to 0.89–0.96.

Together, these results show that UAV multispectral imagery provides useful information for winter wheat N management and pre-harvest yield and protein prediction. That information was most reliable when interpreted alongside crop developmental stage, local soil N supply, an N-rich reference, and known management and cultivar information.

Description

Keywords

precision nitrogen management, leave-one-location-out validation, nitrogen rich sufficiency index, grain protein concentration, hard red winter wheat

Graduation Month

December

Degree

Master of Science

Department

Department of Agronomy

Major Professor

Major Professor Not Listed

Date

Type

Thesis

Citation

Endorsement

Review

Supplemented By

Referenced By