Monitoring Soil Quality Using MIR and NIR Spectral Models: An Interview with Felipe Bachion de Santana

Soil quality is the bedrock of agriculture, serving as the essential foundation for crop growth and food production. It directly impacts the health and productivity of plants by providing crucial nutrients, water retention, and root support. Healthy soil teeming with diverse microorganisms ensures efficient nutrient cycling and pest control, reducing the need for chemical inputs. Moreover, soil quality plays a pivotal role in carbon sequestration, contributing to climate change mitigation. Therefore, soil quality is the cornerstone of ensuring a prosperous agricultural future that meets the growing demands.

Felipe Bachion de Santana of Teagasc in Wexford, Ireland, is exploring new ways to monitor soil quality through using spectroscopic techniques. His research interests are focused on the development of a fast, robust, precise and accurate methodologies based on the use of mid-IR (MIR), near-IR (NIR), and X-ray fluorescence (XRF) spectroscopy in tandem with chemometrics and machine learning regression and classification algorithms (random forest, cubistic, support vector machine [SVM], artificial neural network [ANN], partial least squares [PLS], principal component analysis [PCA], and PLS-discriminant analysis [PLS-DA]) to determine soil attributes. Bachion de Santana, working under the guidance of Karen Daly of Teagasc, recently published a study that assessed the impact of different soil particle sizes on the accuracy of infrared spectra acquisition, particularly in the MIR and NIR ranges, using particles smaller than 0.100 mm (ball-milled) versus particles smaller than 2 mm (1).

Spectroscopy spoke to Felipe Bachion de Santana about his team’s work in monitoring the quality of soil to improve agricultural efficiency.

Felipe Bachion de Santana of Teagasc | Image Credit: © Felipe Bachion de Santana

Would you be able to set the stage for our readership? The introduction to your study discusses how the United Nations (UN) estimated that our global population increase over the next several decades, which will require increased agricultural efficiency. What are the biggest challenges that the agriculture industry faces in meeting this increased production efficiency?

Improving agricultural production efficiency is a complex task involving interdisciplinary areas. In addition, there are particular cases involving the weather, the amount of water available, the type of agriculture, the soil parent material, and the region. In general, the main challenges are:

• Soil degradation: Refers to the deterioration of chemical attributes, physical structure, and biological activity. Soil degradation and loss of fertility are long-term challenges that require sustainable practices such as crop rotation, cover cropping, and organic matter management.
• Resource Scarcity: There are limitations about the land that we can grow crops, which can be mainly attributed to soil degradation, water scarcity, limited arable land, and weather. In general, extending the agricultural areas is not an option.
• Climate Change and Global Warming: The increase in the earth’s temperature and extreme weather events can result in low production for many crops.
• Pest and Disease: Pests and diseases can significantly reduce crop yields.
• Knowledge and Access to Technology: It is recommended that farmers hire an advisor to help them with land management. This information, combined with modern machinery, can minimize fertiliser wastes and water pollution and increase crop yields. However, because of the high initial costs, access to expert information and modern machinery is unavailable for several farmers.
• Supply Chain Issues: Efficient transportation and storage are the minimal requirements to reduce post-harvest losses. This problem affects mainly developing countries.
• Economic Viability: Farming needs to be economically viable for farmers to continue their operations. The commodities prices can vary significantly, so farmers face financial challenges and may require access to special credit lines.

Therefore, improving agriculture efficiency is a multidisciplinary task, and all these issues need to be addressed in an environmental and sustainable way, which increases the challenge.

Your study tackles the issue of global agricultural production by monitoring soil quality. What are the most important soil attributes that need to be measured to ensure optimal soil for agriculture?

Generally, for farmers that already know their lands, the minimum soil analyses that should be performed to obtain a good crop yield are pH (for lime requirement), macro and micro-nutrients analysis (available nitrogen, phosphorus and potassium, calcium, magnesium, boron, copper, iron, manganese, zinc), carbon sequestration, and organic matter, among others. This can vary according to the region and the crops. In addition to these parameters, the Soil Monitoring Law from the European Union (EU) suggested some indicators for monitoring soil health and quality. Among these parameters we can highlight include salinization, soil erosion, loss of organic carbon, soil compaction, excess of nutrient (P and N), contamination of heavy metals, water capacity retention, loss of soil biodiversity, soil particle size, bulk density, total nitrogen—these are among several other parameters.

What makes MIR and NIR spectroscopy the techniques of choice for determining a range of soil attributes?

There are several parameters that the EU recommends analyzing to monitor soil quality and health. However, these analyses are expensive and time-intensive, demanding significant investments from farmers or the government to build regional, national, or semi-national maps (large-scale) of high-resolution soil attributes (that is, <5 km for national surveys). As a consequence, the soil attribute maps are created using low resolution and do not represent the real situation of the region well.

MIR and NIR spectra contain many chemical information, such as C, H, O, S, and minerals. Consequently, they can supplement or replace the use of traditional analytical wet methods, which are currently used to determine several of these parameters, such as pH, organic matter, soil particle size, bulk density, and total nitrogen, among other parameters. Using spectroscopy techniques, all these parameters can be determined in a few minutes with good accuracy and reduced costs in an environmentally friendly way for soil monitoring proposes.

Can you discuss what chemometrics and machine learning models were used in your study, and why these models were better suited for your study?

Initially, we wanted to define that machine learning algorithms applied to chemical data can be considered a chemometric tool. Our paper tested the most common chemometric algorithms commonly used in spectroscopy data: the PLS, SVM, and Cubist. For our data, we noticed that PLS works better when the data does not present high variability. SVM was demonstrated to deal better with data with high variability, and Cubist was the best choice for large variability and a high number of samples. Of all these algorithms, only PLS has well-established methods to identify outlier samples during the calibration modeling and prediction steps. The other two calibration modeling algorithms must use other approaches, such as PCA, to identify outliers in the prediction step.

Can you summarize the results of your study?

In our study, we noticed that MIR spectroscopy can provide more accurate results compared to NIR spectroscopy for bulk density, carbon, nitrogen, pH, soil particle size, and other attributes. In addition, we observed that drying and sieving the soils are enough to analyze the soils in MIR or NIR. This was a novelty because it was common sense in the academic community that soil samples should be ball-milled for MIR analysis. To prove that hypothesis, we tested three algorithms and the calibration and validation set were randomly selected 29 times before the median accuracy parameters were compared. In general, the accuracy was the same for ball-milled and <2 mm soil samples. Besides that, the variables that are important in projection (chemical information used to build the regression models) were the same for both particle sizes.

What are the next steps in this work?

The following steps are to investigate strategies that can enable the prediction of soil quality and health parameters using MIR and NIR handheld devices on the field. We will also explore the limitations and accuracy of handheld devices used in laboratory and field conditions.

Reference

(1) Bachion de Santana, F.; Daly, K. A comparative study of MIR and NIR spectral models using ball-milled and sieved soil for the prediction of a range soil physical and chemical parameters.Spectrochimica Acta Part A: Mol. Biomol. Spectrosc. 2022, 279, 121441. DOI: 10.1016/j.saa.2022.121441