Combination of Multivariate Standard Addition Technique and Deep Kernel Learning Model for Determining Multi-Ion in...

Area of Science:

• Agricultural Engineering
• Analytical Chemistry
• Machine Learning Applications

Background:

• Ion-selective electrodes (ISEs) are crucial for monitoring hydroponic systems but suffer from signal drift, interferences, and high ionic strength issues.
• Accurate nutrient monitoring is essential for optimizing plant growth and resource efficiency in controlled agricultural environments.

Purpose of the Study:

• To overcome the limitations of traditional ISEs in hydroponic applications.
• To develop an enhanced sensing structure combining multivariate standard addition (MSAM) and deep kernel learning (DKL) for accurate ion prediction.
• To improve the prediction accuracy and precision for eight essential ions (NO3-, NH4+, K+, Ca2+, Na+, Cl-, H2PO4-, Mg2+) in hydroponic solutions.

Main Methods:

• Employed a six-ISEs array integrated with the multivariate standard addition (MSAM) sampling technique.
• Utilized feature enrichment (FE) from MSAM to enhance data features for the deep kernel learning (DKL) model.
• Applied the combined MSAM-FE-DKL model to analyze ten real hydroponic samples.

Main Results:

• Achieved low root mean square errors (RMSE) below 63.8 mg·L-1 and coefficients of variation (CVs) below 8% for predicting six major ions (nitrate, ammonium, potassium, calcium, sodium, chloride).
• Successfully predicted phosphate (H2PO4-) and magnesium (Mg2+) with RMSEs of 29.6 and 8.7 mg·L-1, respectively, addressing limitations in detecting these ions.
• Demonstrated improved prediction reliability for existing ISE ions and enhanced detection capabilities for previously unavailable ions.

Conclusions:

• The proposed MSAM-FE-DKL sensing structure significantly improves the accuracy and precision of ISEs in hydroponic systems.
• This approach effectively mitigates inherent ISE shortcomings, making them more feasible for real-time monitoring in closed hydroponic environments.
• The study validates the successful application of advanced machine learning techniques for robust chemical sensing in agriculture.