Developing a soil spectral library using a low-cost NIR spectrometer for precision fertilization in Indonesia

Precision fertilization aims to apply fertilizer according to the nutrient variability of the soil and crop-specific nutrient requirement. Formulating fertilizer recommendations requires laboratory analysis of soil samples, which is expensive, time-consuming, and involves the use of chemical extractants. Near-infrared (NIR) spectroscopy has been recognized and used widely as a rapid method to predict various soil properties with comparable accuracy to conventional soil analysis. However, most of the research was conducted using a laboratory-grade visible-near-infrared (vis-NIR) spectrometer to predict specific soil properties. This study aims to investigate the efficacy of a miniature NIR spectrometer (NeoSpectra), which operates in the wavelength range of 1300–2600 nm, for soil analysis to prescribe fertilizer recommendations for various food crops in Indonesia. Legacy soil samples (N = 1601) were collected from various crop-growing regions in Indonesia and were scanned with the NeoSpectra. Regression models for various soil properties analyzed (including soil texture, pH, carbon, and soil nutrients) were developed using the Cubist regression model; and categorical models were also developed for soil properties with large variability (available P and K) using the C5.0 decision tree model. Mitscherlich-Bray (MB) equations to formulate fertilizer nutrient requirements were developed using field trial data. The predictions from the categorical model were used as input in the MB equation to provide fertilizer recommendations for various crops in Indonesia, particularly for rice, soybean, and corn. The recommendations were then further corrected based on various soil properties that affected the nutrient dynamics. The study demonstrated the feasibility of using NIR spectrometer as a rapid tool for fertilizer recommendations.