Improving Estimation of Winter Wheat Nitrogen Status Using Random Forest by Integrating Multi-Source Data Across...

Area of Science:

• Agronomy
• Plant Science
• Remote Sensing

Background:

• Accurate estimation of plant nitrogen (N) status is vital for precision N management in agriculture.
• Non-destructive estimation of crop N status across diverse agro-ecological zones (AZs) remains a significant challenge.

Purpose of the Study:

• To enhance the estimation of winter wheat (Triticum aestivum L.) N status across two AZs.
• To evaluate the efficacy of random forest regression (RFR) models utilizing multi-source data for improved N status estimation.

Main Methods:

• Conducted 15 site-year experiments across two AZs (2015-2020) with varying N rates and cultivars.
• Developed RFR models integrating vegetation index, climatic, and management factors.
• Employed a variable selection strategy based on Pearson correlation coefficient to identify key predictive variables.

Main Results:

• RFR models with integrated data (R² = 0.72-0.86) significantly outperformed models using only vegetation index (R² = 0.36-0.68).
• Selected 6-7 key variables achieved performance comparable to using all variables.
• The contribution of climatic and management factors varied by AZ and N indicator; climatic factors were more critical in higher latitudes, especially water-related factors.

Conclusions:

• Integrating multi-source data with RFR models substantially improves winter wheat N status estimation across AZs.
• Future research should focus on developing advanced remote sensing-based machine learning models with multi-source data for broad-scale crop N monitoring.