Answers are grounded in this paper and cite their sources.
The rapid estimation of cotton leaf nitrogen content (LNC) requires models that account for growth-stage differences and perform well across years. We analyzed 972 paired canopy spectra and LNC observations collected at six stages during 2023–2025, using 2023–2024 data for model development and 2025 observations for evaluation. Plot-grouped comparisons examined spectral compression, vegetation indices and global versus stage-specific modeling. With partial least squares regression, competitive adaptive reweighted sampling (CARS) reduced 551 spectral inputs to approximately 35 wavelengths, achieving an out-of-fold R2 of 0.813 and reducing the root mean square error (RMSE) from 3.561 to 2.989 g/kg. Stage-specific Stacking achieved the highest R2 during development at full squaring, early flowering, full flowering and early boll. In 2025, spectral Stacking achieved the highest R2 and lowest RMSE at four stages; pooled values were 0.857 and 2.723 g/kg compared with 0.456 and 5.318 g/kg for four-index Stacking. Models selected during development achieved a pooled plot-mean R2 of 0.867. Compact spectral inputs retained predictive information beyond common indices, and stage-specific calibration performed well across years with ensemble benefits varying by stage. These findings support leaf nitrogen monitoring and field sampling at key cotton growth stages.
No takes yet. Share an insight, caveat, or question.
Tang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: