Multispectral UAV imaging and machine learning for estimating wheat nitrogen nutrition index.

Area of Science:

• Agricultural Science
• Remote Sensing
• Crop Physiology

Background:

• Accurate monitoring of wheat nitrogen nutrition is vital for optimizing fertilizer application and ensuring crop yield.
• Existing methods for assessing nitrogen status can be labor-intensive and time-consuming.
• Developing efficient, non-destructive techniques for nitrogen nutrition index (NNI) estimation is crucial for precision agriculture.

Purpose of the Study:

• To estimate the wheat nitrogen nutrition index (NNI) using UAV-based multispectral imagery.
• To investigate the impact of varying planting densities and nitrogen application rates on NNI.
• To construct a robust model for NNI estimation in spring wheat.

Main Methods:

• UAVs captured multispectral canopy imagery of wheat at key growth stages.
• Vegetation indices were selected based on correlation and feature importance analysis.
• A Bayesian optimized random forest model was developed to estimate NNI.

Main Results:

• Several vegetation indices (DVI, MDD, NGI, MEVI, NDVI, EVI, ENDVI) showed high robustness for NNI estimation.
• The optimal NNI estimation model achieved R² of 0.785 and RMSE of 0.137 under specific nitrogen application rates (N2).
• The best NNI model at a planting density of P1 (1 million plants/hm²) yielded R² of 0.716 and RMSE of 0.158.

Conclusions:

• The study successfully developed an accurate NNI estimation model using UAV multispectral data.
• Findings provide insights into how planting density and nitrogen levels influence wheat NNI.
• The model serves as a valuable tool for assessing wheat growth and guiding optimal management practices.