Limited data coverage presents a central challenge to hydrologic monitoring. Optimal networks use the minimum number of measurements to characterize hydraulic processes of interest. This requires informed selection of sensor types and locations to reduce redundancy while capturing critical spatial and temporal patterns in the measured properties.
We consider an example problem: how to choose the optimal number and locations of water content sensors to quantify the change in the field-average water content in the upper meter with time during precipitation followed by drainage and ET. We assume that the sand/silt/clay (SSC) distribution is known.
We consider three approaches. We first examine random sensor placement. Second, we use the SSC distribution to map soil texture and then use soil type to reduce redundant sampling and to assign representative areas to each measurement. Third, we conduct process-specific response clustering, which groups SSC values based on the expected similarities of their hydrologic responses to precipitation followed by drainage and ET. We use the response clusters to reduce redundant sampling and to assign representative areas. For each approach, we quantify the uncertainty of the inferred average water content time series as a function of the number of measurements collected. We also examine the use of different sensors, which are sensitive within different depth intervals. We show that textural classification reduces the uncertainty for any given number of sensors. Response clustering further reduces the uncertainty, especially if loamy soils are present. Further, clustering can be tailored to different numbers and types of sensors, so it was especially useful for sensors that measured over a depth interval that differed from the depth interval of interest. We discuss how response clustering can be applied to a wide range of hydrologic monitoring objectives.
See more of: Topical Sessions