Optical Sensor Characterization Enables Rapid Soil Macronutrient Detection in Laboratory Tests

A research team led by Mangeh Elsie Jaja, Valery Nkemeni, Pierre Tsafack, Pierre Brosselard, Nicole Gilon, and Patrick Jame has characterized a low-cost, non-destructive optical sensor system designed for the rapid assessment of nitrogen, phosphorus, and potassium concentrations in prepared soil samples. Detailed in a study published in Discover Sensors, the investigation evaluates an optical design based on visible and near-infrared spectroscopy to capture the characteristic reflectance signatures of primary macronutrients. Utilizing light-emitting diodes paired with a photodiode detector in a free-space arrangement, the laboratory prototype demonstrated the capacity to measure spectral responses and predict nutrient levels in less than one minute per sample, offering an operational contrast to traditional wet-chemistry soil testing methods that typically require multiple hours of processing.

Maintaining appropriate concentrations of nitrogen, phosphorus, and potassium is fundamental to crop development, influencing essential physiological processes ranging from chlorophyll synthesis to energy transfer and structural resilience against plant diseases. Traditional approaches to soil nutrient profiling rely on manual field sampling using augers or push probes, followed by laboratory extraction, chemical reagent treatment, and analytical quantification. In agricultural settings and resource-constrained environments, such conventional protocols present notable logistical barriers due to labor intensity, chemical waste, and extended turnaround times. To address these limitations, researchers have explored both electrochemical and optical sensors. While electrochemical sensors employ transducers to detect potential differences or current changes, they often face issues of surface drift and electrode fouling that necessitate frequent recalibration. Optical spectroscopy offers a non-destructive alternative by monitoring light absorption and reflectance across specific electromagnetic bands.

In the reported investigation, the research team focused on the visible-near infrared spectral region, specifically targeting optical wavelengths where macronutrients exhibit prominent absorption properties. Previous experimental determinations established that nitrogen demonstrates substantial optical absorption in the range of 800 to 980 nanometers, phosphorus exhibits peak absorbance between 620 and 630 nanometers, and potassium shows primary absorbance between 460 and 470 nanometers. To exploit these optical characteristics without relying on complex multi-component spectrophotometers, the researchers built an emission unit comprising three discrete commercially available light-emitting diodes. An infrared LED emitting within the 800 to 980 nanometer band was chosen for nitrogen evaluation, a red LED was designated for phosphorus monitoring, and a blue LED operating near 460 to 470 nanometers was assigned for potassium measurement.

The optical transmission unit was paired with a single silicon photodiode detector capable of absorbing the light reflected from the sample surface and converting that optical energy into an electrical current. During preliminary electrical characterization, the authors established the peak current and voltage limits for each light source to ensure maximum radiant intensity without causing thermal degradation. Specifically, the infrared diode was operated at 30 milliamperes and 1.6 volts, the red diode was operated at 20 milliamperes and 2.2 volts, and the blue diode was driven at 20 milliamperes and 3.4 volts. The generated photocurrent from the photodiode was measured as reflected voltage using an oscilloscope, which permitted real-time signal inspection and precise voltage logging across repeated measurement cycles.

To overcome common structural challenges associated with optical fiber systems, such as fragility, connector misalignment, environmental vulnerability, and high fabrication expenses, the team designed the sensor around free-space optical transmission. However, unguided light rays in free-space configurations risk causing direct stray-light exposure from the emitter to the detector, which skews analytical readings. To resolve this geometric challenge, the investigators systematically dimensioned the sensor architecture according to the optical law of reflection. By evaluating angles of incidence from 0 to 90 degrees in increments of 30, 45, 60, and 75 degrees over a square plastic reflector within an enclosed dark chamber, the researchers established that an angle of 60 degrees maximized reflected signal capture while suppressing direct emitter-to-detector interference. This optimal configuration defined a path length of 2 centimeters between the components and the reflector, along with an emitter-detector separation distance of 2.4 centimeters.

Sensor calibration was conducted using thirty experimental samples systematically organized into six distinct categories to account for different fertilizer matrices. The base soil comprised a black agricultural soil harvested from a depth of zero to twenty centimeters from a farm in the Auvergne-Rhone-Alpes region of France. The collected material was air-dried, finely pulverized with a mechanical blender, sieved through a two-millimeter mesh to eliminate coarse debris, and thoroughly homogenized. The soil was subsequently amended with three distinct fertilizer sources: a composite NPK fertilizer containing six percent nitrogen, three percent phosphorus, and twelve percent potassium, an ammonium nitrate fertilizer containing thirty percent nitrogen, and an industrial urea fertilizer containing forty-six percent nitrogen. To benchmark these formulations against natural conditions, an unamended humus soil sample, designated Sample 10, was preserved as an unfertilized reference standard.

Because the prepared sample categories featured elevated nutrient concentrations relative to average arable soils in order to establish controlled calibration lines, the exact elemental contents were verified via analytical laboratory methods at the Institute of Analytical Sciences at Université Claude Bernard Lyon 1. Nitrogen content was quantified using high-temperature combustion elemental analysis at 1050 degrees Celsius followed by thermal conductivity detection, while phosphorus and potassium concentrations were determined via laser-induced breakdown spectroscopy calibrated against certified reference materials. During optical testing, each prepared powder was securely mounted in a transparent plastic holder inside an enclosed dark box. The sample holder was rotated across four orientations to capture spatial variance, and the recorded photodiode voltages were averaged over three independent replicate readings per configuration.

The experimental trials demonstrated that reflected light intensity increased systematically as nutrient concentrations grew, confirming that higher nutrient loading in the soil-fertilizer matrix enhances reflectance at the designated wavelengths. Mathematical models were subsequently developed using both linear and non-linear regression techniques to map output voltage against chemical concentration. Model evaluation was guided by four key statistical criteria: the coefficient of determination, root mean square error, predicted residual error sum of squares, and standard error. Across all sample matrices and individual nutrients, second-degree polynomial regression equations consistently outperformed linear, exponential, and logarithmic functions, providing superior fitting accuracy and minimized residual error values across the calibration curves.

In specific nitrogen calibration trials, second-degree polynomial models demonstrated marked improvements over alternative curve-fitting equations. For instance, in sample type five, the polynomial equation achieved a coefficient of determination of 0.8205 and a root mean square error of 4.7698 volts, whereas the linear model yielded a lower coefficient of 0.7589 and a larger root mean square error of 5.1676 volts. Phosphorus models exhibited exceptionally strong predictive fitting, with sample type two reaching a coefficient of determination of 0.9764 and a root mean square error of 0.2474 volts under polynomial regression, compared to 0.9432 and 0.3596 volts for linear regression. Potassium calibration exhibited an identical trend, where sample type five registered a coefficient of determination of 0.9245 and an error of 0.8072 volts using a polynomial fit, substantially improving upon linear and logarithmic alternatives.

Comprehensive validation of the sensor system further evaluated stability across four core analytical dimensions: repeatability, reproducibility, sensitivity, and linearity. Repeatability trials comprising five successive cycles under identical laboratory conditions revealed minimal signal fluctuation, recording low average standard errors of 0.00048 volts for nitrogen, 0.00232 volts for phosphorus, and 0.00512 volts for potassium. Reproducibility evaluations conducted over three separate days with ambient temperature variations of 3 degrees, 7 degrees, and 10 degrees Celsius yielded average standard errors of 0.00061 volts, 0.00103 volts, and 0.00111 volts. Sensitivity analyses revealed an asymmetrical response profile, where the sensor achieved an average sensitivity of 9.4445 percent for phosphorus, 7.5912 percent for potassium, and 5.0027 percent for nitrogen, indicating that the system is most responsive to phosphorus fluctuations while nitrogen detection requires greater signal amplification.

Linearity assessments evaluating plotted relationships between measured and predicted concentrations revealed strong agreement, maintaining coefficients of determination consistently above 0.8 across the target nutrients. However, the researchers observed that sample type six, which incorporated a combined mixture of black soil, NPK compound fertilizer, ammonium nitrate, and urea, exhibited greater spectral variability, registering a lower coefficient of determination of 0.4975 for nitrogen. The authors attributed this deviation to intricate optical interactions among overlapping fertilizer constituents, where complex reflectance and scattering behaviors introduce non-linear signal interferences that standard single-variable polynomial regressions cannot fully resolve.

Benchmarking against prior literature confirmed measurable performance advantages for the optimized free-space prototype. Whereas earlier optical reflectance systems had recorded nitrogen detection errors of 0.0421 volts or root mean square errors reaching 1.923 volts, the current configuration achieved residual errors as low as 0.00094 volts under comparable settings. The study authors emphasized that this increased precision stems directly from the geometrical optimization of the optical path, which effectively suppressed direct emitter radiation and stabilized diffuse signal reception. Furthermore, the inclusion of predicted residual error sum of squares and standard error metrics provided a broader statistical validation framework than is typically reported in similar proof-of-concept hardware studies.

Notwithstanding these favorable laboratory findings, the authors documented several clear boundaries regarding the scope and generalizability of their results. The reported coefficients of determination and error metrics reflect sensor-level accuracy in measuring voltage signals rather than absolute nutrient prediction accuracy in uncalibrated field environments. Because the experimental library relied on thirty composite samples featuring elevated fertilizer ratios and controlled moisture levels, the calibration curves do not yet capture the operational complexities of dynamic open fields. Variations in soil moisture content, natural soil roughness, textural diversity, and ambient illumination present recognized confounding factors. Consequently, future developments will require integrating advanced machine learning architectures, such as support vector regression, random forest algorithms, and artificial neural networks, to resolve multivariate interference across diverse soil classifications before practical field deployment can occur.

Subject of Research: Optical visible-near infrared reflectance sensor characterization for rapid soil macronutrient monitoring

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

Article References: Jaja, M. E., Nkemeni, V., Tsafack, P., Brosselard, P., Nicole, G., & Patrick, J. (2026). Optical sensor characterization for rapid detection of nitrogen phosphorus and potassium concentrations in soil. Discover Sensors, 2(1), Article 17. https://doi.org/10.1007/s44397-026-00048-y

Image Credits: AI Generated

DOI: 10.1007/s44397-026-00048-y

Keywords: Soil Science, Visible-Near Infrared Spectroscopy, Optical Sensors, Soil Macronutrients, Precision Agriculture, Nitrogen Detection, Phosphorus Detection, Potassium Detection, Sensor Calibration, Agricultural Technology, Diffuse Reflectance, Soil Nutrient Monitoring, Free-Space Optics, Photodiode Detection, Smart Farming, Polynomial Regression