Data for: Hyperspectral sensing to identify Orobanche minor parasitism in red clover

Description

This dataset contains hyperspectral reflectance measurements from red clover (Trifolium pratense L.) plants, used to detect parasitism by small broomrape (Orobanche minor Sm.). Spectral readings were collected post-emergence, after O. minor inflorescences had appeared aboveground, with an ASD FieldSpec 4 (Malvern Panalytical, Malvern, United Kingdom) spectroradiometer. The hypothesis was that parasitized and non-parasitized red clover plants differ in leaf reflectance, and that these differences can be used to discriminate between the two groups using hyperspectral sensing. Data collection had two parts. Greenhouse studies (April 2024 and August 2025) involved pots inoculated with or without O. minor seed, with spectral readings taken from leaves once O. minor inflorescences had emerged aboveground. Field studies (June 2024 and June 2025) took place in two red clover seed production fields near Rickreall, Oregon, with known O. minor infestation. Parasitized plants were identified by visible inflorescence, and the haustorial connection to the host was visually confirmed at excavation. Non-parasitized plants were excavated from the same fields as controls. Data were collected at approximately 1,400 growing degree day (GDD) (2024) and 1,600 GDD (2025), calculated from January 1 with a base temperature of 0°C. Each file contains raw reflectance values from 350 to 2500 nm, at 1 nm intervals. Each value is the average of 2–3 readings taken per leaf sample with the instrument. The dataset covers four groups: 2024 greenhouse (n = 12 per group), 2025 greenhouse (n = 23 per group), 2024 field (n = 12 per group), and 2025 field (n = 10 per group). Plants are identified by sample ID, with treatment and treatment code columns marking parasitized versus healthy status. This data was used to visualize spectral signatures, run PLS regression on both raw and first-derivative spectra, and calculate VIP scores to identify wavelengths associated with post-emergence O. minor parasitism. First-derivative models gave classification accuracies between 93.5% and 100% across the four groups, with the near-infrared region contributing the most discriminating wavelengths, followed by the red-edge, visible, and short-wave infrared regions. To reuse this data, apply first-derivative transformation before PLS modeling, since that step improved classification accuracy over raw reflectance in every group. Reflectance values below 400 nm and above 2400 nm carry more instrument noise and are typically excluded or treated with caution in analysis.

Institutions

• Oregon State University

  Oregon, Corvallis

Categories

Licence