Monitoring the total organic carbon concentrations in a lake with the integrated data fusion and machine-learning (IDFM) technique

Abstract

The concentration of total organic carbon (TOC) in surface waters is subject to seasonal variation, as well as abrupt changes in concentration due to events. In drinking water treatment, TOC is a precursor to disinfection byproducts such as total trihalomethanes (TTHM). With the aid of an early warning system for the detection of TOC concentrations, water treatment operators could make more informed decisions and adjust the treatment processes to minimize the generation of disinfection byproducts. In this paper, a near real-time monitoring system is explored using the Integrated Data Fusion and Machine-learning (IDFM) technique to predict the spatial distribution of TOC in a lake based upon surface reflectance data from satellite imagery. Landsat 5 TM and MODIS Terra satellite imagery can be acquired free of charge, yet MODIS has coarse spatial resolution, while Landsat has a lengthy 16 day revisit time. This difficulty is solved using data fusion algorithms to fuse the fine spatial resolution of Landsat with the daily revisit time of MODIS to generate a synthetic image with both high spatial and temporal resolution. To demonstrate the capabilities of IDFM, this case study uses the fused surface reflectance band data and applied machine-learning techniques to reconstruct the spatiotemporal distribution of TOC in Harsha Lake, which serves as the source water intake for the McEwen Water Treatment Plant in Ohio. The results of this application of IDFM were analyzed using 4 statistical indices, which indicated that the Artificial Neural Network model is capable of reconstructing TOC concentrations throughout the lake.