A combined model of shoot phosphorus uptake based on sparse data and active learning algorithm.

Area of Science:

• Agricultural Science
• Environmental Science
• Remote Sensing

Background:

• Excessive phosphorus fertilizer application severely damages soil ecosystems, hindering fertility restoration and sustainable agriculture.
• Shoot phosphorus uptake (SPU) is crucial for crop monitoring and nutrient management, but large-scale, non-destructive measurement is challenging.
• Accurate spatiotemporal estimation of SPU at regional scales remains a significant hurdle for effective agricultural management.

Purpose of the Study:

• To develop and validate a hyperspectral prediction model for estimating maize shoot phosphorus uptake (SPU) at the canopy scale.
• To explore the potential of combining spectral data with advanced analytical methods for robust SPU prediction.
• To provide an effective tool for monitoring soil phosphorus and promoting sustainable agricultural development.

Main Methods:

• A combination prediction model was developed, integrating predicted P uptake from maize leaves, stems, and grains.
• Hyperspectral data from Henan Province was utilized to explore prediction capabilities.
• Advanced techniques including first-order differentially enhanced two-dimensional correlation spectroscopy (1Der-2DCOS) and two-trace 2DCOS (filling-milk-2T2DCOS) were employed for spectral analysis.
• A hybrid model incorporating a Newton-Raphson-based optimizer with active learning was used to refine predictions.

Main Results:

• The combined prediction model significantly improved accuracy (R² = 0.87, RMSE = 2.39 kg/ha, RPD = 2.71) compared to simple empirical models.
• Two-dimensional correlation spectroscopy methods demonstrated robustness in extracting spectral trait relationships, enabling efficient prediction with small sample sizes.
• The hybrid model effectively filtered localized simulation data, enhancing prediction accuracy for practical applications across different regions.

Conclusions:

• Hyperspectral remote sensing, coupled with advanced analytical models and small, representative sample sets, can accurately predict maize SPU.
• The developed method offers a valuable tool for evaluating sustainable phosphorus use in agriculture.
• This approach shows significant potential for monitoring global soil phosphorus surplus and advancing sustainable agricultural development.