Since 1979 the Dutch National Institute of Public Health and Environmental Protection (RIVM) has been developing the Dutch Groundwater Quality Monitoring Network (LMG). This network presently consists of about 350 monitoring sites. At each site, well screens are placed at two depths: 10 and 25 m below surface level. Samples are collected every year and are analyzed for all macrochemical parameters and some trace elements. Tritium contents were measured in the first sampling round. The geochemistry of Dutch groundwater is complex, due to the different sources (seawater, surface water and rainwater), complicated hydrogeology, and human impact on flow systems and pollution. Structuring or data analysis is required for the interpretation of the large number of hydrogeochemical data from such a monitoring network. An exploratory approach is to look within the data set for homogeneous groups, each with a typical (macro)chemistry. The selection criteria for the location of the monitoring sites of the LMG are mainly based on soil type and land use, and to some extent on the hydrogeological situation. However, a classification based on the two most reliable criteria, soil type and land use, does not result in chemically distinguishable homogeneous groups or water types. Fuzzy c means clustering was successfully used to discern structure and natural groups in the LMG data for 1 year. A seven‐cluster model was adopted. The number of clusters was decided heuristically with the aid of nonlinear mapping, on the basis of the geographic distribution, the hydrogeochemical interpretability, and the unimodality of the distribution of the parameters per cluster. The consistency of the model is illustrated by the reproducibility of the clusters in different years. The clusters are related to geochemical processes, natural sources, and anthropogenic input and are designated as follows: (1) “seawater” in coastal areas, (2) “desalinization” in organic‐rich Holocene marine and peat deposits, (3) “surface water” for downward seeping river water or surface water near the main rivers, (4) “carbonate/reduction” in peat areas or old groundwater in seepage zones, (5) “carbonate” for carbonate‐dissolving “precipitation,” (6) acid “precipitation” water in sandy topographic highs, and (7) a “polluted” cluster characterized by agricultural contaminants. Although the influence of soil type and land use is noticeable in some of the clusters, the geochemical controls, which characterize the different clusters, appear to outweigh their anticipated influence on the hydrogeochemistry on the scale of the national groundwater network. Consequently, the homogeneous groups, obtained through the cluster analysis, present a better base for further statistical and hydrogeochemical evaluation than, e.g., a stratification of the data based on soil type and land use.
