Analysis of the temporal evolution of the spectral and spatial sensitivities of a support vector regression maize yield algorithm
Year: 2025
Type: article
Abstract: Knowledge of crop yield is essential for evaluating alternative management practices. Support vector regression (SVR) may be a successful algorithm for corn ( Zea mays , L.) grain and silage yield estimation from spectral imagery at multiple scales, i.e., within-field, farm, and regional. The objective of this study was to evaluate time series of spatial and spectral dependencies for SVR-based corn grain and silage yield estimation. Data were collected from two silage fields and one grain field with a 272-band imaging spectrometer, mounted on an unmanned aerial system (UAS). The original 0.06 m ground sample distance (GSD) imagery was spatially resampled to multiple resolutions, up to 30 m GSD. Spectral band selection algorithms guided the creation of ten different spectrally down-sampled datasets to reduce data dimensionality, while maintaining yield estimation accuracy. Both silage and grain yield estimations were most accurate at the tassel growth stage (approximately 80 days after sowing [DAS]), followed closely by the soft dough stage (approximately 110 DAS) and at 16 m and 4 m GSD, respectively. Spectrally, sampling at 450, 550, 670, 770, 830, and 920 nm was most optimal for both grain and silage yield, suggesting that these band centers may be ideal for use in vegetation index development for yield estimation. These results suggest a robust sensing solution for multi-season applications can be developed based on only a select set of wavelengths.
Institutions: Rochester Institute of Technology, Cornell University
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: hybrid
Estimated APC paid: $3,870
OpenAlex ID: W4413817080