Abstract
A hyperspectral fertility signal appeared in maize before any ear could be measured, but it did not carry through to yield. Six hybrids were grown at four rates of a slow-release NPK fertiliser (20–80% of full rate, three reps, 72 plants) in an automated phenotyping greenhouse and scanned six times between 29 and 69 days after planting (400–2500 nm). The red-edge index separated every pair of NPK rates from day 38 (p<1e-22), 26 days before the first measurable ear, and ear length rose 9.6 mm per rate step (p<0.0001). But full-spectrum models predicting ear length (R²=0.44) did not beat treatment and hybrid alone (R²=0.53), and collapsed for kernel number (R²=−0.21); ear length itself did not predict kernel number (r=0.05), which was three times as variable as grain filling or ear structure. Unmeasured late-season water stress and poor pollination confounded every kernel-derived trait, and scanning stopped before grain fill. The facility resolves treatment effects early and reliably; it does not yet resolve individual plants or reproductive outcomes, a gap the next trial needs to design around.
Keywords
Hyperspectral imaging; High-throughput phenotyping; Yield prediction
DOI
10.5703/1288284318727
Phenotyping Maize Fertility Response in an Automated Greenhouse: A Pilot Evaluation
A hyperspectral fertility signal appeared in maize before any ear could be measured, but it did not carry through to yield. Six hybrids were grown at four rates of a slow-release NPK fertiliser (20–80% of full rate, three reps, 72 plants) in an automated phenotyping greenhouse and scanned six times between 29 and 69 days after planting (400–2500 nm). The red-edge index separated every pair of NPK rates from day 38 (p<1e-22), 26 days before the first measurable ear, and ear length rose 9.6 mm per rate step (p<0.0001). But full-spectrum models predicting ear length (R²=0.44) did not beat treatment and hybrid alone (R²=0.53), and collapsed for kernel number (R²=−0.21); ear length itself did not predict kernel number (r=0.05), which was three times as variable as grain filling or ear structure. Unmeasured late-season water stress and poor pollination confounded every kernel-derived trait, and scanning stopped before grain fill. The facility resolves treatment effects early and reliably; it does not yet resolve individual plants or reproductive outcomes, a gap the next trial needs to design around.