Introduction
Background and objectives
Similarity-based approaches in groundwater: state of the art
-
Hydrogeological classification schemes are widespread, but usually at a very low level of formalization, mainly descriptive and only applicable in specific regional contexts.
-
With the exception of hydrochemistry, classification has generally received little attention in hydrogeology as a scientific concept.
-
Classification in groundwater hydrology is usually based on static properties and conditions, and rarely includes similarities in the dynamic responses of groundwater systems. Similarity analysis and classification of groundwater time series are hardly ever used.
Study area and data
Locations and general description of the data
-
Relatively shallow, with only a few deeper than 200 m and, in Fennoscandia, with only a few deeper than 10 m
-
Mainly located in productive aquifers (gravel and sand, karstic limestone, fractured sedimentary rocks), with most in Quaternary alluvial aquifers
-
Often clustered in river valleys
-
Often located relatively close to human settlements
-
Differed greatly in terms of length and regularity of measurement intervals, and total length of observation period
-
Very often contained gaps, outliers, and (sometimes quite peculiar) irregularities
Anthropogenic influence
Preprocessing and working dataset
Examples of groundwater time series
-
Time series show very different types of dynamics.
-
Similarity occurs between groundwater time series from very different regions (e.g. southern Germany and Sweden), i.e. it is not only the result of spatial proximity of observations.
Methods
Time series similarity
Groundwater systems similarity
Dependency between groundwater systems and groundwater dynamics
Results
Time series similarity
Visual classification
-
Results are generally intuitive.
-
Visual classification is rather tolerant with respect to irregularities and gaps in time series, varying measurement intervals, time series length and monitoring periods.
-
It requires relatively little data preprocessing.
-
The human eye is very sensitive to differences in visual appearance.
-
Results remain subjective and generally not reproducible.
-
Preprocessing layout of time series presentation (plotting) plays a decisive role.
-
The approach is time-consuming.
-
Carrying out classification of new data and/or different plot types requires repetition of the entire process.
Direct comparison of original time series
-
Once implemented, the analysis can be carried out and repeated on very large numbers of time series.
-
Results (degree of similarity) can be expressed and analysed using standardized, validated methods.
-
It can detect similarity of features which are not discernible to the human eye.
-
Approaches require times series of equal length and equal spacing of measurements, gaps are not allowed.
-
Irregularities in time series may strongly influence results.
-
Results are not immediately intuitive, which makes it difficult to offer conceptual explanations.
-
Dependencies between the classifications found and groundwater systems properties are hard to establish (see sections ‘Groundwater systems similarity’ and ‘Dependency between groundwater systems characteristics and groundwater hydrograph characteristics’).
Classification based on indices (feature extraction)
-
Time series do not have to cover the same time interval.
-
The large variety of indices, describing different characteristics and emphasizing different features, allows for a better targeted, more transparent, time series characterization and statistical analysis, e.g. through cluster and regression analysis.
-
The individual indices can be conceptually related to physical properties and boundary conditions which makes the approach more intuitive and more useful for gaining improved understanding of groundwater systems.
-
Certain indices are rather sensitive to noise and irregularities.
-
The algorithms to calculate certain indices contain parameters that need to be adjusted for a specific dataset and are thus not immediately transferable.
Groundwater systems similarity
Dependency between groundwater systems characteristics and groundwater hydrograph characteristics
Semiquantitative dependency analysis
-
A quite clear relationship appears to exist between the two “deep aquifer” groundwater system types, where the deep unconfined locations mostly have a dynamic type of group 2, while the deep confined locations reflect a behaviour associated with group 1 (compare with Fig. 2).
-
In the groundwater system type “shallow unconfined (2)”, a clear separation into two types of dynamics (two groups: 7 and 9) can be observed. This implies that there is a crucial hydrogeological difference that distinguishes the two which was not taken into account in the expert-based grouping. The aquifers in this groundwater system type are shallow peat aquifers, not all of which are artificially drained (ditches), creating very distinct types of dynamics.
-
Groundwater dynamics for limestone aquifers seem not to show specific characteristics—limestone alone is not a well-defined enough type of groundwater system. Unfortunately, there are too few limestone records in the dataset to allow subdivisions into deep-shallow, confined-unconfined, and fractured-karstic.
Formal quantitative dependency analysis
-
G1 contains wells screened in deep unconfined sand and gravel aquifers with a depth to groundwater level between 25 and 40 m.
-
G2 contains wells screened in deep confined gravel aquifers. The depth to groundwater level is 30–50 m.
-
G3 comprises wells located in shallow unconfined aquifers.
Applications
-
Using index-based time series characterization and clustering for drought analysis (Heudorfer 2019).
-
Case study-based evaluation of index-based time series characterization and clustering to describe groundwater flows in different shallow aquifer settings (Giese et al. 2020).
-
Identification of anthropogenic influence on groundwater hydrographs (unpublished).
Discussion and conclusions
Promising results and potential benefits
Main challenges, open questions and suggestions for further research
-
It would be highly beneficial to apply approaches to characterization of groundwater dynamics, for example the index approach as presented in (Heudorfer et al. 2019) to a wider range of different hydrogeological, climatic and geographic contexts. The tools required for this are available on request from the authors.
-
A promising path for future research is cross-validation of the approach with other models, namely numerical models. This could be achieved by comparing the results of existing models with classifications of the time series data used for calibration of those models, or by setting up numerical models to try to reproduce the dynamic behaviour of typical hydrogeological settings. Studies of this nature have been carried out (e.g. Hellwig et al. 2020; Stoll et al. 2011) albeit with a different scope.
-
Establishing consistent, systematic approaches to groundwater system characterization has proven to be one of the challenges that needs to be overcome to realise the proposed concept successfully. However, consistent, systematic approaches to groundwater system characterization can be of great value even beyond the context classification and similarity approaches, in terms of resource and vulnerability assessment, and management of ground resources. Many hydrogeological classification systems have been developed, often at a local level, and often dedicated to a specific purpose (e.g. hydrochemical characterization). They almost never involve groundwater dynamics. The classification and similarity approach suggested here provides a great toolbox for unifying the currently separate characterization and classification schemes into a widely applicable groundwater systems typology.