Early-season estimation and mapping of the nitrogen and phosphorus contents are essential for enhancing sugarcane production and preventing yield loss. Timely data acquisition is needed to estimate and map the nitrogen and phosphorus contents to facilitate decision-making and precision agriculture in sugarcane production. However, the traditional methods used to estimate and map the nitrogen and phosphorus contents are time-consuming, laborious, and expensive.

This study’s aims were to map the spatial and temporal variation in the nitrogen and phosphorus contents early in the sugarcane season using UAVs, machine learning algorithms, and soil and vegetation indices. The sugarcane plantations were in Emangweni in the Nkomazi Local Municipality, Mpumalanga, South Africa.

Soil and leaf samples, computed vegetation indices, and ground survey data were used as inputs for the machine learning algorithms. The performance of Random Forest, Support Vector Machines, and Partial Least Squares Regression was compared based on the accuracy of the models in estimating and mapping the nitrogen and phosphorus contents in sugarcane plantations. The Pearson Correlation Coefficient (R), p-Value (p), Coefficient of Determination (R²), and Root Mean Square Error (RMSE) were used to validate the accuracy of the machine learning algorithms.

Based on our results, Random Forest is expected to outperform Support Vector Machines and Partial Least Squares Regression in estimating and mapping the nitrogen and phosphorus contents. The Normalized Difference Red Edge is expected to perform better in estimating and mapping the nitrogen contents; however, a combination of vegetation indices will be required to estimate and map the phosphorus contents in sugarcane plantations.

Future studies employing unmanned aerial vehicles should focus on the estimation and mapping of the nitrogen, phosphorus, and potassium contents over the entire sugarcane season.