A low-cost AI-based sensing approach to quantify ammonia volatilization as a driver of indirect greenhouse gas emissions

This study presents the development of a low-cost, portable, and AI-enhanced electronic nose (e-nose) system for quantifying ammonia (NH₃) volatilization from fertilized agricultural soils, with a specific emphasis on its implications for indirect greenhouse gas concentrations. Although NH₃ is not a greenhouse gas itself, its volatilization contributes significantly to indirect nitrous oxide (N₂O) emissions, one of the most potent GHGs regulated under IPCC guidelines. The proposed system integrates a MICS-6814 metal oxide sensor, ESP32 microcontroller, cloud-based data transfer, and machine learning algorithms to provide real-time monitoring and predictive analysis of NH₃ losses. Time-series sensor data were normalized, converted into area-under-the-curve (AUC) metrics, and modeled using eight machine learning algorithms. After preprocessing and hyperparameter tuning, Gradient Boosting achieved the highest performance (R² = 0.84; MAE=0.86). Laboratory evaluations demonstrated strong correlations between AUC values and NH₃-N measurements obtained through classical boric acid trapping, validating the system’s accuracy. The findings confirm that rapid detection of NH₃ volatilization can support digital nitrogen management strategies, reduce fertilizer-derived nitrogen losses, and ultimately help mitigate indirect N₂O emissions by minimizing surplus reactive nitrogen in agricultural fields. By enabling real-time emission monitoring through a low-cost digital platform, this research contributes to emerging precision agriculture solutions aimed at reducing the environmental footprint of nitrogen fertilization.