Data-driven statistical optimization of a groundwater monitoring network

Abstract

We propose a comparative study of three different methods aimed at optimizing existing groundwater monitoring networks. Monitoring piezometric heads in subsurface porous formations is crucial at regional scales to properly characterize the relevant subsurface hydrology and to assess water resources management and protection. Here, the basic idea to optimize the efficiency of existing gauging networks is to identify correlated timeseries to guide the removal of redundant measurement sites. Three data-driven statistical methods are compared: Oscillation correlation (OC) hierarchical (HC) and timeseries clustering (TSC). These methods are applied to a hydrogeologically complex groundwater system within the Bacchiglione basin (IT). Results suggest that: (i) the OC method returns well-gathered correlation clusters while being fast and easy to apply; (ii) HC underpins more spread clusters but it is useful when considering multiple groundwater characteristics; and (iii) TSC proves the best performing method for the study area, at the cost of being the most complex to implement. The latter identified 30 out of 59 existing timeseries as redundant, i.e., where sensors might be relocated elsewhere thus gaining in information quality (or else simply saving money if dismissed). We also suggest that the microscale of a random piezometric head field is a suitable measure to extract from data the monitoring frequency of manual measures in dismissed locations.