Sampling Design Methods for Water Distribution Model Calibration

Abstract

Field sampling is sometimes performed to support modeling activities—specifically, to estimate the parameters of a mathematical model or, more accurately, to calibrate the model. In this case, a relevant question for field samplings design is “how to maximize the confidence in estimated parameter values, given a level of sampling effort?” We approach this question using established ideas in parameter estimation and sampling design theory and propose general sensitivity-based methods to rank the locations and types of measurements for estimating the parameters of a water distribution network model. The proposed methods are suboptimal, yet practical, and are applied to select good tracer and pressure measurement locations for estimating pipe roughness coefficients. These particular results suggest that, when compared to pressure measurements, tracer measurements can be informative for calibrating network hydraulic parameters but one must take more care in selecting their location. Using the proposed methods, a selection of both tracer and pressure measurements improves estimation confidence by a factor of 2, over that obtained using tracer or pressure measurements alone.