Lab Validation of ISE-based Soft Sensors for Environmental Applications

Chemical sensing of natural waters at a spatial resolution relevant to understanding and managing natural-built coupled systems demands development of sensors that (1) are relatively inexpensive to support deployment of large numbers in highly heterogeneous systems, (2) capture holistic characteristics of variable chemical background, and (3) are able to detect target pollutants which may represent a minority fraction of the charge balance. While the cost, size, and power demands of ion selective electrodes (ISEs) are compelling, the extent to which they can be realistically operationalized as salinity increases or water chemistry background varies must be realistically assessed. Miniaturization efforts unlock an ability to design multi-sensor arrays at reasonable cost and size (i.e., 10s to 100s of individual sensors per array), however achieving best possible detection limit and accuracies requires the design of an array of complementary membrane chemistries be conducted in parallel with design of sensor data fusion approaches that integrate knowledge of the sensor physics as well as the chemical properties of the target waters. This talk will overview the integrated approach, where machine learning methods are used to fuse sensor data in real time while also embedding chemical and physical knowledge through the design of the training dataset, alongside data from an initial lab validation. Samples are generated using a custom sequential injection analysis (SIA) system for scalability and represent the transect from fresh to coastal waters via variable salinity estuarine waters. Primary target for the study is nitrate, although the approach can be flexibly adapted to quantify any constituent of the samples. The work is a step in the direction of expanding utility of ISE arrays for application in environmental systems.