Date of Graduation

5-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Geosciences (PhD)

Degree Level

Graduate

Department

Geosciences

Advisor/Mentor

Tullis, Jason

Committee Member

Poncet, Aurelie

Second Committee Member

Peter, Brad

Keywords

NDVI; Planet; Site-specific zoning; Soybean; Spatial autocorrelation; Time series

Abstract

Remote sensing and land surface phenology provide a strong basis for characterizing crop development, but their usefulness for within-field yield-zone delineation depends on whether crop development is analyzed as a phenological trajectory and whether the resulting zones are evaluated with spatially defensible statistics. These questions are especially relevant in soybean production because soybean is one of the most important row crops in the United States and the country’s dominant oilseed crop. This relevance is strong in Arkansas, where soybean is a major component of agricultural production. In soybean production, within-field yield variability reflects the combined influence of soil properties, stand establishment, and in-season crop stress. In farmer-managed fields, however, these controls vary sharply over short distances and do not act uniformly through time. Despite the availability of multitemporal imagery, many remote-sensing studies still reduce seasonal crop development to selected image dates or summary phenological metrics, rather than examining the full canopy trajectory and identifying when field locations begin to diverge in agronomically meaningful ways. This dissertation addresses that limitation by developing a phenology-based, high-spatial-resolution framework for soybean yield-zone analysis using Planet imagery. Vegetation-index time series were aligned by cumulative growing degree days (cGDD) so that crop development was represented by its phenological progress. Functional data analysis was then used to group field locations with similar canopy-development trajectories and translate those temporal patterns into spatially explicit management clusters. Candidate critical stages were identified from portions of the seasonal signal where divergence was greatest, with major windows occurring around 1100–1300 and 1900–2200 cGDD. Semivariogram-guided aggregation was subsequently used to support statistically defensible comparisons while accounting for spatial dependence. The results showed that the agronomic value of remote sensing was both stage-dependent and index-specific. Several stage-specific vegetation-index models produced strong explanatory performance, with adjusted R² values approaching 0.70. Model comparisons further showed that adding NDVI-derived zones increased the explained variation in yield deviation from 33.2% to 67.3% beyond soil and treatment alone. These findings show that phenology-based clustering of high-spatial-resolution satellite imagery can provide a scientifically rigorous and operationally relevant framework for yield zoning, scouting, and site-specific management in precision agriculture.

Download

Share

COinS