A machine learning approach for estimating forage maize yield and quality in NW Spain

Area of Science:

• Agricultural Science
• Agronomy
• Crop Modeling

Background:

• Crop models are essential tools for simulating crop growth under various environmental conditions.
• Forage maize yield and quality are critical for livestock production and require accurate prediction methods.

Purpose of the Study:

• To adapt the CSM-CERES-Maize model for simulating forage maize yield and quality.
• To identify key environmental and management variables influencing forage maize production.
• To develop accessible predictive models for non-specialist users.

Main Methods:

• Calibrated genetic parameters of six forage maize cultivars using the CSM-CERES-Maize model (DSSAT) across multiple sites and years.
• Utilized historical meteorological data (2000-2022) for yield and quality simulations.
• Employed LightGBM (a machine learning technique) to model non-linear relationships and identify influential variables.

Main Results:

• Achieved high prediction accuracy: 94.7% for yield, 94.0% for energy, and 93.0% for protein.
• Identified Growing Season and Radiation as the most influential predictor variables.
• Longer-cycle cultivars (cycle 400) demonstrated superior yield and quality metrics.

Conclusions:

• The adapted crop models accurately simulate forage maize yield and quality.
• Growing season length and solar radiation significantly impact forage maize production.
• The developed models provide a valuable tool for predicting forage maize performance and aiding agricultural decision-making.