Random forest machine learning for maize yield and agronomic efficiency prediction in Ghana.

Area of Science:

• Agricultural Science
• Machine Learning
• Agronomy

Background:

• Maize (Zea mays) is vital for Sub-Saharan African food security, but production increases often rely on land expansion, causing environmental issues.
• Yield improvement per unit area is crucial for sustainable maize production in Ghana.
• Accurate prediction of maize yields and nutrient use efficiency is essential for informed decision-making.

Purpose of the Study:

• To develop and evaluate a random forest machine learning model for predicting maize yield and agronomic efficiency in Ghana.
• To identify key soil, climate, environmental, and management factors influencing maize production.
• To provide insights for improving fertilizer recommendations for sustainable maize production.

Main Methods:

• Trained a random forest machine learning algorithm using data from 482 maize field trials (3136 plots) in Ghana (1991-2020).
• Employed a 5x10-fold nested cross-validation approach for model calibration and evaluation.
• Analyzed the importance of predictor variables (soil, climate, environment, management) for yield and agronomic efficiency.

Main Results:

• The random forest model demonstrated good prediction performance for maize yield (MEC=0.81).
• Moderate prediction performance was achieved for agronomic efficiency (AE-N: MEC=0.63, AE-P: MEC=0.55, AE-K: MEC=0.54).
• Soil variables were more important than climatic variables for yield prediction; temperature was key for yield, and rainfall for agronomic efficiency.

Conclusions:

• Random forest models enhance understanding of maize yield and agronomic efficiency drivers in tropical climates.
• The findings offer valuable insights for optimizing fertilizer recommendations to support sustainable maize production.
• This approach contributes to improving food security in Sub-Saharan Africa through data-driven agricultural practices.