Comparison of two methods for optimizing existing groundwater monitoring networks: application to the Bacchiglione Basin, Italy

In the presence of a groundwater monitoring network (GMN) of sensors aimed at measuring the hydraulic head in a given domain, the statistical analysis of time series not only provides insight into the general aquifer behaviour, but it can also return parameters useful to optimize and enhance the GMN’s efficiency.

Several methods to design new GMNs are available, but few of them are useful for optimizing existing networks. This study compares two methods in order to define pros and cons of their applicability and effectiveness.

They are carried out for the case study of the alluvial basin of the Bacchiglione river, near Vicenza (Veneto, Italy). The existing network comprises 92 groundwater data-loggers, installed in wells screening mostly the unconfined aquifer.

The first simple method, here proposed, is based on the Pearson correlation coefficient and the microscale parameter, which shows the time interval in which data are perfectly correlated. The coefficients were calculated between detrended time series. Firstly, based on the correlation coefficient threshold of 0.95, areas of intercorrelated couples are defined. They are characterized by similar hydrological behaviour, therefore it is sufficient to constantly monitor only one location in each area, while other interesting correlated points can be measured manually at longer sampling time. The microscale can be used to estimate this sampling time in order to see the water table trend (between 7 and 78 days in this domain), even if shorter oscillations are obviously missed and some peaks could remain unseen. This way, extra sensors can be moved to other critical areas, in order to improve the system knowledge.

The second method defines the seasonal Mann Kendall (sMK) test for detecting monotonic trends, that are used into Principal Component Analysis (PCA). Finally, a Hierarchical Clustering Analysis is carried out to group sensors with similar factors of the PCA. This method is more articulated than the previous one and entails some informed choices to be made about the distance measure and the clustering algorithm. Thanks to the sMK test and the PCA, a high insight of the system is achieved, however the clustering result may strongly variate depending on the expert’s knowledge and expectation.

The two proposed statistical analyses of hydrogeological data provide integrative decision support to improve representativeness and effectiveness of monitoring networks aimed at both qualitative and quantitative groundwater control.