Rethinking data collection for quantifying end uses of water: the impact of data temporal resolution

Knowledge of water use behavior at the household level (e.g., the different daily patterns in consumption and the distribution of water use across end uses) is needed to effectively manage and forecast water demand. Availability of high temporal resolution data (i.e., observations recorded with a time interval < 1 minute) can help increase our understanding of residential water use. Most water meters operating today are not capable of collecting and storing this type of data. Hence, studies collecting high resolution data have relied on datalogger devices that operate on top of existing meters and temporally aggregate data in an effort to extend battery and storage capacity of such devices. In the past, data have been collected at different temporal resolutions (e.g., 4 s or 10 s) without assessing the impact of the temporal aggregation interval on the identification and quantification of end use event features (e.g., volume, flowrate, duration) due to the unavailability of data at a sufficient temporal resolution to enable such analyses. We recorded the time between every magnetic pulse generated by a magnetically-driven residential water meter's measurement element (full pulse resolution) using a new, open-source datalogging device and collected data for two residential homes in Utah, USA. We then examined water use events without temporally aggregating data and compared to the same data aggregated at different time intervals to evaluate how temporal resolution of the data affects our ability to identify end use events, calculate features of individual events, and classify events by end use. Our results show how collecting full pulse resolution data can provide more accurate estimates of event occurrence, timing, and features along with producing more discriminative event features that can only be estimated from full pulse resolution data to make event classification easier and more accurate - all without significantly increasing the volume of generated data.