Combining machine learning and remote sensing-integrated crop modeling for rice and soybean crop simulation.

Area of Science:

• Agricultural Science
• Remote Sensing Technology
• Machine Learning Applications

Background:

• Accurate crop growth prediction is vital for food security.
• Leaf Area Index (LAI) is a key variable for crop modeling.
• Integrating remote sensing data with crop models enhances prediction accuracy.

Purpose of the Study:

• Develop a machine learning (ML) method for estimating rice and soybean Leaf Area Index (LAI) using proximal sensing data.
• Evaluate the performance of a Remote Sensing-Integrated Crop Model (RSCM) with ML algorithms.
• Identify optimal ML algorithms for LAI estimation from vegetation indices.

Main Methods:

• Collected and analyzed rice and soybean datasets.
• Employed various ML regression models: ridge, lasso, support vector machine, random forest, and extra trees.
• Modeled the relationship between LAI and vegetation indices derived from canopy reflectance.

Main Results:

• The extra trees regression model showed the best performance for LAI estimation (test scores: 0.86 for rice, 0.89 for soybean).
• The ML-integrated model accurately replicated observed LAI values under varying nitrogen treatments (Nash-Sutcliffe efficiencies: 0.93 for rice, 0.97 for soybean).
• Demonstrated effective capture of seasonal LAI variations across different management practices.

Conclusions:

• ML techniques significantly enhance the integration of remote sensing data into crop models.
• The developed ML-based LAI estimation method improves crop growth prediction accuracy.
• This approach offers substantial potential for advanced crop monitoring and productivity assessment.