Advancing Soil Phosphorus Management with Vis-NIR and AI

A recent study explored how analytical spectroscopy can be used to monitor and measure the phosphorus content in soil. This study, published in Soil and Tillage Research, brought together scientists from Isfahan University of Technology, The Ohio State University, the University of Nebraska-Lincoln, and the University of Valencia (1). In their study, they demonstrated the utility of visible-near-infrared (Vis-NIR) spectroscopy combined with artificial intelligence (AI) to improve phosphorus management in agricultural soils (1).

Phosphorus is an important ingredient in soil. It is considered a macronutrient that is essential for living organisms (2). Ultimately, the soil’s ability to maintain high levels of phosphorus are critical to the agriculture industry delivering excellent yield on their crops (2). The problem the agriculture industry faces is that using traditional methods requires extensive laboratory analysis, which means it is an expensive and time-consuming processes (1).

Expert hand of farmer checking soil health before growth a seed of vegetable or plant seedling. Gardening technical, agriculture concept. Image Credit: © piyaset - stock.adobe.com

Soil phosphorus testing measures the sorption parameters to better understand how fertilizers are interacting with the soil and how much phosphorus is lost through leaching (1). In this collaborative study, the research team explored whether Vis-NIR spectroscopy could provide accurate estimations of phosphorus soil parameters (PSPs) (1). Vis-NIR is a rapid and non-destructive technique that measures light reflectance in the 350–2500 nm range that analyzes the chemical or biological composition of any object (3).

As part of their procedure, the researchers collected 100 soil samples from western Iran and applied partial least squares regression (PLSR) and artificial neural networks (ANN), which are two of the most common advanced modeling techniques. The results revealed that PSP values varied significantly across sampling sites because of the high diversity of soil properties such as clay content, soil organic carbon (SOC), and cation exchange capacity (CEC), all of which strongly influence phosphorus sorption (1). This variability made it challenging for traditional PLSR models to achieve accurate predictions because linear models struggle to capture the complex interactions between spectral data and multiple soil properties (1).

As a result, the ANN model was better suited in predicting PSPs. Specifically, ANN improved the prediction accuracy of MBC by 39.25%, PBC by 50%, Q_max by 19.28%, SBC by 39.41%, and SPR by 59.32% (1). The best coefficient of determination (R²) values in the validation data set ranged from 0.65 to 0.85, indicating that the ANN models could provide reliable estimates suitable for practical application by farmers and policymakers (1).

The researchers recommended in their study that ANN-based Vis-NIR spectroscopy can potentially serve as a viable tool for large-scale soil phosphorus assessments. The researchers further stated that this approach could be especially beneficial in regions where conventional soil fertility data is lacking, enabling farmers and decision-makers to optimize phosphorus fertilization strategies while reducing costs and the environmental impact (1).

PSPs are complex by design. As a result, the researchers suggest that ANN is not a panacea to all the issues outlined earlier when testing PSPs in soil. Although ANN contributed to improved prediction accuracy, further refinements continue to be needed (1). The authors recommend that future studies can work on integrating mid-infrared (MIR) spectral data and other environmental variables into these machine learning (ML) models (1).

By leveraging AI and spectroscopy, this research presents a new method that can help advance agriculture. With further advancements, these methods could become essential tools for optimizing phosphorus use, ensuring better crop productivity, and reducing agricultural runoff pollution (1).

References

1. Saidi, S.; Ayoubi, S.; Shirvani, M.; et al. Use of Vis-NIR Reflectance Spectroscopy for Estimating Soil Phosphorus Sorption Parameters at the Watershed Scale. Soil Till. Res. 2025, 248, 106460. DOI: 10.1016/j.still.2025.106460
2. Dunne, K. S.; Holden, N. M.; O’Rourke, S. M..; et al. Prediction of Phosphorus Sorption Indices and Isotherm Parameters in Agricultural Soils Using Mid-Infrared Spectroscopy. Geoderma 2020, 358, 113981. DOI: 10.1016/j.geoderma.2019.113981
3. Eurofins, Ultraviolet/Visible/Near Infrared Spectroscopy (UV/VIS/NIR). EAG.com. Available at: https://www.eag.com/techniques/spectroscopy/uv-vis-spectroscopy/#:~:text=UV/VIS/NIR%20spectroscopy%20is,and%20enters%20the%20integrating%20sphere. (accessed 2025-02-04).