Utilizing machine learning to optimize agricultural inputs for improved rice production benefits

Abstract

Summary: Lower efficiency of agricultural inputs in the four conventional rice planting methods limits productivity and environmental benefits in Southwest China. Thus, we developed a machine-learning-based decision-making system for achieving optimal comprehensive benefits during rice production. Based on conventional benefits for achieving optimal benefits, implemented strategies in these planting methods: reducing N fertilizer by 16% while increasing seed inputs by 9% in mechanical transplanting (MT) method improved yield and environmental benefits; reducing N fertilizer and seed inputs by 10–12% in mechanical direct seeding (MD) method decreased environmental impacts; increasing N-K fertilizers and seed inputs by 15–33% in manual transplanting (MAT) method improved its comprehensive benefits by 7–14%; applying N-P-K fertilizer ratio of 2:1:2 in manual direct seeding (MAD) method enhanced yield. Our study provides strategies for improving benefits in these planting methods, with MT method being more beneficial for optimizing comprehensive benefits, especially in yield and environmental impacts, in Southwest China.

Keywords

• Agricultural science
• machine learning