The use of NIR sensor technology for soil test-based decision making in agriculture

Fertilisers are a major cost item for almost all farmers all over the world. The vast majority of farmers apply fertilisers, but only a fraction have access to information about the crop needs in relation to their soils’ fertility. This problem is particularly evident in Sub-Saharan Africa, where farmers are often barred from soil testing services because of financial constraints (too expensive), logistic constraints (too far away), or simply because it is unavailable. In developed countries, a transition is occurring towards circularity and precision farming, where insights into soil fertility variations in space and time are required. In both cases, the use of sensor technology may be relevant because of its mobility, affordability and lack of consumables.

Near Infrared (NIR) spectroscopy has been known to soil scientists for years, especially in the area of soil carbon monitoring (e.g. Viscarra Rossel and Bouma, 2016). This is because NIR directly measures chemical bonds in organic matter. To use NIR spectroscopy in the field to additionally measure nutrients in moist soils several building blocks are required:

• A high-quality sensor in a fool proof device;
• A clean, and sufficiently large database;
• Advanced statistical models to convert spectral readings into desired parameters;
• Quality checks and corrections, e.g. for moisture content;
• Translation of results into reports based on user needs.

In practice, a number of constraints and limitations have been found to exist in the deployment of this technology, which are still in the process of being overcome. AgroCares has developed a NIR based operational system called the SoilCares Scanner, which was introduced to the market in early 2017, and which overcomes several of these limitations. The sensor and lightsource generate a spectral signal which is transmitted to an application on a smartphone. This application sends the signal to the AgroCares cloud where it is converted into soil characteristics using the calibration database and deep learning models. Key aspects of the AgroCares system will be described.