Assessment of Ten Standardization-Based Calibration Transfer Techniques for Gas Sensors in a Small Data Context

The calibration transfer techniques aim to standardize secondary instruments to a primary instrument, enabling the utilization of the primary instrument’s calibration model with minimal additional experiments. While widely employed in spectroscopic datasets, their application in environmental sensors and sensor arrays is less common despite the pressing need due to issues, such as sensor batch variability and time drift. This study assesses ten calibration transfer techniques on three experimental gas sensor datasets that have a small number of transfer standards (TSs). Model-based [linear regression (LR)] and model-free [K-nearest neighbor (KNN)] calibration methods are also compared. Results unexpectedly show that, despite the reduced data availability, direct recalibration of the secondary sensor (e.g., without transfer) may be sufficient for the secondary sensor to reach the same performance as the primary one. In the other cases, partial-least-square standardization and direct standardization are the most robust transfer methods across the use cases, and they can outperform direct recalibration. Other methods, such as single sensor standardization (SSS), mean correction, principal component (PC) standardization, and procrustes transform, have consistently lower performance in this small data context.