Water quality control through spatial and temporal analysis of water quality monitoring systems

With the advancement of sensor technology for drinking water distribution systems and attention on surface water quality at a watershed scale, managers of our water resources must be prepared to quickly manage and analyze large spatial data sets to obtain understanding and control of our drinking and recreational water. Integrating geographic information systems (GIS) with real-time water quality data and system models will provide the capability of advanced spatial and temporal data storage and analysis, assisting managers in disaster response preparedness, improving system control, and determining a priority list for watershed restoration. This work evaluated, using GIS and distribution system modeling, the effects of spatial and temporal variability in drinking and recreational water on quality control practices.;Optimal sensor placements in a water distribution system were first evaluated to determine if they were congruent across different optimization criteria and attack scenarios. Sensor networks were designed based on optimization of time to detection, population affected, volume of consumed contaminated water, and detection likelihood. A fifth optimization criteria case considered an equal weighting of the four criteria as a multiple-objective function. GIS was used to discover spatial relationships among the 40 different sensor networks in a large (12,000+ node) network. Frequency, average nearest neighbor, and spatial autocorrelation analyses indicated that sensor placements corresponding to the different intrusion scenarios and optimization cases are likely to overlap and cluster. In addition, the different scenarios and cases tend to place the sensors in similar locations and in the same order, particularly for the first few sensors placed. Thus, sensor networks based on different intrusion scenarios will provide similar protection to this water distribution system. The dependence of sensor location on the reachability and reachable average demand of each network node was also analyzed using GIS and a Chi-Square analysis. The analysis illustrated that sensor locations defined by optimizing volume of consumed contaminated water and population affected are likely to be dependent on network nodes with a high reachable average demand and high reachability. Sensor locations defined by optimizing detection likelihood are likely to be dependent on network nodes with a low reachable average demand and low reachability.;Secondly, the potential for a boost-response system to provide substantial protection against an intrusion event to allow for uninterrupted service even during a contamination incident was evaluated. Random contamination events were simulated in a model water distribution system with an optimized sensor network. A disinfection boost was simulated to begin the instant contamination reached a sensor, and a range of decay coefficients were applied to the contaminant to simulate reaction with the disinfectant. Cumulative distribution curves of the volume of consumed contaminated water for various response levels were prepared to analyze how each response affected the vulnerability of the system. This analysis illustrated that a boost-response system could be effective in significantly reducing the volume of consumed contaminated water, but only in very specific circumstances. Most importantly, the booster must be located at a node with high reachability. Further, if the disinfectant cannot rapidly inactivate the contaminant, the effectiveness of a boost-response system is much reduced.;Finally, the effect of temporal variability on stream classification based on indicator data was evaluated. Fecal coliform and E. coli samples were taken weekly during the prized into wet vs. dry days, and upper vs. lower watershed, and the geometric means and geometric standard deviations of various 5-sample data sets were analyzed to determine if selecting particular sets of sampling data would cause an appreciable difference in regulatory decisions. Results indicate if the bacterial indicator samples are near the regulated limits for water contact recreational use, temporal bias could sway impairment classification decisions. To reduce the temporal bias, sampling data submitted for stream classification should include several sampling groups within the recreational season, particularly for sites near point sources of pollution and with indications of low fecal contamination.