Application of near-infrared spectroscopy to predict chemical properties in clay rich soil: A review

• Published in: European journal of agronomy Vol. 159; p. 127228
• Main Authors: Park, Sihyun, Jeon, Sangho, Kwon, Na- Hyun, Kwon, Muhyeok, Shin, Jae-Ho, Kim, Won-Chan, Lee, Jeong Gu
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2024
• Subjects: absorption; adsorption; agronomy; calcium; chemical analysis; Clay soil calibration; clay soils; cost effectiveness; near-infrared spectroscopy; Non-destructive soil analysis; nutrient uptake; prediction; Proximal sensing; soil organic carbon; Soil organic matter prediction; total nitrogen; water content; CEC; CCD; SOC; ELM; AI; Clay soil calibration; UAV; Soil organic matter prediction; IICO; Non-destructive soil analysis; RF; FOD; Proximal sensing; GADF; AOTF; ML
• ISSN: 1161-0301
• DOI: 10.1016/j.eja.2024.127228

Proximal soil sensing is a highly advanced and rapidly evolving technique for predicting soil chemical properties. NIR spectroscopy is expected to offer an easier and more cost-effective alternative to traditional soil chemical analysis methods for broader agricultural applications. This review aims to address the challenges in high-clay content soils by 1) understanding the relationship between NIR spectroscopy and its predictive capabilities of chemical properties, and 2) proposing the application of advanced technologies to improve their practical applications. The high-clay content soils increase the adsorption of various cations, including Al3+, Ca2+, Mg2+, K+, Na+, and NH4+, and affect crop root density, absorption area, and nutrient uptake efficiency. However, predicting soil characteristics using NIR spectra can be difficult due to the complex nature of soil metrics. Under clay-rich soils, the high moisture content leads to a significant presence of O-H vibrations (1395 and 1415 nm), which overlaps with crucial function groups. Recent advancements using deep learning models have significantly improved prediction accuracy and efficiency for soil properties. The results showed that the R² values for predicting soil organic C, total N, and pH were 0.94, 0.89, and 0.87 for S-AlexNet; 0.95, 0.94, and 0.90 for GADF-Swin Transformer; and 0.88, 0.84, and 0.87 for PLSR, respectively. Additionally, integrating hyperspectral and multispectral sensors further improves prediction efficiency by providing higher spatial and spectral resolutions. In conclusion, this review suggests that adopting cutting-edge technologies can improve the predictions accuracy of chemical properties in the clay-rich soils. Nevertheless, overcoming the limitations of advanced technologies, such as high computational resource requirements, is necessary for practical application in agriculture. [Display omitted] •Challenges of model prediction using NIR in soils with high clay content.•Impact of overtones in hydroxyl (O-H) functional groups on NIR spectral interpretation.•Advancements in model accuracy by noise reduction in NIR analysis.•Developing novel models for improved soil chemical property prediction.•Utilizing UAV and AI technologies for enhanced soil analysis.