An ensemble framework for farmland quality evaluation based on machine learning and physical models.

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Farmland quality (FQ) evaluation is vital for preventing agricultural land misuse and promoting ecological nitrogen management in North China.
• Accurate spatial distribution of nitrogen emissions is currently lacking, hindering precise FQ estimation.

Purpose of the Study:

• To develop a Machine Learning (ML) - Nitrogen Export Verification (NEV) ensemble framework for precise FQ evaluation.
• To address the challenge of obtaining verified spatial nitrogen emission data for accurate FQ assessment.

Main Methods:

• Employed physical models for precise spatial estimation of Nitrogen Export (NE) values.
• Utilized ML methods to compute the spatial distribution of FQ using Farmland Quality Evaluation System (FQES) indicators.
• Applied the framework to the Beijing-Tianjin-Hebei 200 km traffic zone as a case study.

Main Results:

• The ML-NEV framework demonstrated high accuracy, with NEV method relative error below 5.25% and ML method Determination coefficient exceeding 0.84.
• Identified good-quality farmland areas (~47.25%) concentrated in southwest-northeast regions, showing significant improvement potential.
• Determined Fractal Dimension, NE values, and unbalanced irrigation/drainage capabilities as key drivers of FQ.

Conclusions:

• The developed ML-NEV ensemble framework provides a robust method for precise FQ evaluation.
• Findings offer decision support for refining FQ assessments and implementing targeted improvements.
• The study contributes to safeguarding grain yield and revitalizing agribusiness in developing countries.