Capability of short Vis-NIR band tandem with machine learning to rapidly predict NPK content in tropical farmland: A case study of Aceh Province agricultural soil dry land, Indonesia

Abstract

Developing a rapid, accurate, and cost-effective method for predicting constituents related to soil nitrogen (N), phosphorus (P), and potassium (K) content is considered substantial. One procedure that can work with these criteria and is currently popularly researched is the visible and near-infrared (Vis-NIR) method. Furthermore, machine learning algorithms have been proven to increase the accuracy and robustness of calibration models developed based on short Vis-NIR spectra. Therefore, the potential of combination using short Vis–NIR band (450–1050 nm) in tandem with machine learning algorithm (k-nearest neighbors (kNN), adaboost, random forest) was investigated and evaluated deeply in this study to predict the soil content of N, P, and K using 122 tropical dry farmland samples of Aceh Province origin, Indonesia. Within the investigated variables, the parameter RPD achieved excellent classification (RPD>4.1) for any application for predicted tropical dry farmland soil samples N, P, and K using min-max normalization followed by the kNN algorithm. The short Vis–NIR tandem with machine learning provided promising calibration predictions of N, P, and K content for agricultural soil dry land, especially from Aceh Province, Indonesia.

Keywords

• Chemometrics
• Non-destructive measurement
• Soil nutrients
• Spectroscopy