2009 Portland GSA Annual Meeting (18-21 October 2009)

Paper No. 14
Presentation Time: 9:00 AM-6:00 PM

A MULTIPLE-SENSOR/MULTIVARIATE SIGNAL PROCESSING ARCHITECTURE FOR IN-SITU WATER CHEMICAL ANALYSIS


MUELLER, Amy V., Civil and Environmental Engineering, Massachusetts Institute of Technology, 15 Vassar St. 48-212, Cambridge, MA 02139 and HEMOND, H.F., Civil and Environmental Engineering, Massachusetts Institute of Technology, 15 Vassar St. 48-425, Cambridge, MA 02139, amym@mit.edu

The capability for comprehensive, real-time, in-situ characterization of the chemical constituents of natural waters is a powerful tool for the advancement of the ecological and geochemical sciences, e.g. by facilitating rapid high-resolution adaptive sampling campaigns and avoiding the potential errors and high costs related to traditional grab sample collection, transportation and analysis. Such capacity would additionally find application to problems of environmental remediation and monitoring of industrial waste waters.

An instrument for in-situ measurement of all ions contributing to the charge makeup of natural fresh water is thus pursued via a combined multi-sensor/multivariate signal processing architecture. Based primarily on existing commercial electrochemical sensor hardware, the system employs a novel architecture of multivariate signal processing to extract accurate information from in-situ data streams via an "unmixing" process that accounts for non-linear sensor cross-reactivities. Along with ionic sensor signals, known chemical properties of the water (e.g. conductivity and charge neutrality) can be used as additional mathematical constraints, while inclusion of temperature and pH probes provides added information about the chemical state of the system.

Initial work demonstrates the effectiveness of this methodology at predicting inorganic cations (Na+, NH4+, H+, Ca+2, and K+) in a controlled system containing only a single anion (Cl-) in addition to hydroxide, thus allowing charge neutrality to be explicitly invoked. Calibration of every probe relative to each of the five cations present is undertaken, and resulting curves are used to create a representative environmental data set based on USGS data for New England waters. Signal processing methodologies, specifically artificial neural networks (ANNs), are tuned to optimize performance of the algorithm in predicting actual concentrations from these simulated signals. Results are compared to use of component probes as stand-alone sensors.

Future extension of this instrument for multiple anions will ultimately provide rapid, accurate field measurements at high resolution, improving sampling abilities while reducing costs and errors related to transport and analysis of grab samples.