ML improves fertilizer efficiency and maize yields – study

Efficient fertilizer application remains critical for enhancing productivity and profitability in Sub-Saharan Africa, where soil fertility varies significantly across regions.

However, conventional approaches used to determine fertilizer requirements are often expensive, time-consuming, and susceptible to unpredictable climatic conditions.

Addressing this, a study by researchers, including the College of Agriculture and Natural Resources’ (CANR) Vincent Logah, has demonstrated the potential of Machine Learning to improve fertilizer recommendations and boost maize yields in Ghana.

The study explored the use of machine learning (ML) as a data-driven alternative for generating site-specific fertilizer recommendations.

Researchers developed a random forest model trained on 482 maize yield experiments, comprising 3,136 observations collected between 1991 and 2020.

The findings showed that the machine learning-based approach generally recommended lower rates of phosphorus and potassium compared to conventional methods.

The machine learning model produced the highest yields at six of the 14 sites, outperforming the Quantitative Evaluation of the Fertility of Tropical Soils (QUEFTS), Conventional Fertilizer Dose Response (CFDR), and Updated Conventional Fertilizer Dose Response (UCFDR) approaches.

Although differences in yield across the approaches were not always statistically significant, the study highlights the potential of machine learning as a tool for improving efficiency of fertilizer.

The study also calls for the integration of socio-economic factors to enhance the practical adoption of machine learning-based solutions in Ghana and across Sub-Saharan Africa.

The study was published in the Agronomy Journal.