Optical sensor characterization for rapid detection of nitrogen phosphorus and potassium concentrations in soil

Abstract

Accurate and timely monitoring of soil nutrient concentrations, particularly nitrogen (N), phosphorus (P), and potassium (K), is crucial for optimizing agricultural productivity and minimizing environmental impact. In this study, we present the characterization of an optical sensor designed for the rapid and non-destructive assessment of N, P, and K levels in soil samples. The sensor employs Visible-Near Infrared (VIS–NIR) spectroscopy to analyse the spectral signatures of soil samples and extract quantitative information regarding nutrient concentrations. A total of 30 experimental samples classified into six distinct categories, comprising various soil-fertilizer mixtures, were prepared to calibrate the sensor's performance. These samples were made from four primary constituents: black soil (sourced from a farm in the Auvergne-Rhone-Alpes region of France), NPK (6:3:12) fertilizer, ammonium nitrate, and urea. The six categories were grouped into two sets: sample type 1 consisted of pure samples, while sample types 2 through 6 were mixtures of soil and fertilizer. The mixed samples were prepared with nutrient concentrations higher than those typically found in agricultural soils to enable controlled calibration of sensor performance. The pure sample, specifically Sample10, consisting of humus soil without fertilizer amendment, was included as a reference to approximate real-soil conditions. Laboratory experiments involved illuminating the samples with specific wavelengths of light and measuring the intensity of reflected light within a controlled environment. Calibration curves and equations specific to each macronutrient were developed through regression analysis using data collected from the experiments. Performance evaluation metrics, including Root Mean Square Error (RMSE), Coefficient of Determination (R2), Predicted Residual Error Sum of Squares (PRESS), and Standard Error (s), were employed to assess model accuracy. Results demonstrated satisfactory performance across all sample types, with average R2, RMSE, PRESS, and s values of 0.7835 V, 0.0097 V, 0.0003 V, and 0.0003 V, respectively. Statistical analyses further validated the sensor, indicating average standard errors of 0.003 V for repeatability and 0.001 V for reproducibility, along with a sensitivity of 7% and a linearity of R2 = 0.8. These findings validate the effectiveness of the optical sensor in accurately quantifying soil macronutrient concentrations.