Statistical aspects of the identification of groundwater pollution trends, and aggregation of monitoring results for the EU Water Framework Directive (WFD/DDG)

Report by Drafting Group no.5 of the Expert Advisory Forum for the

Water Framework Directive of the European Union

Bilthoven/ Brussels, 20020308

rapporteur

C.R. Meinardi (Nl),

RIVM

P.O. Box 1

3720 BA Bilthoven

tel 31(0)302743367

Contents

1Introduction......

2Outline of the statistical tool made under the lead of Austria......

3Practical experience with the statistical tool......

4Remarks from the meeting in Brussels on 20020301......

5Discussion and conclusions......

1Introduction

The European Union (EU) issued the Water Framework Directive (Directive 2000/60/EU, shortly WFD) in the year 2000. The WFD provides general rules on water policies, including groundwater, in the EU. The assessment of groundwater quality and trends in that quality are described in general terms, notably in Articles 4 and 17, which are aiming at measurement programmes to prevent and to control groundwater pollution. Article 17 requires the development of criteria (a method) to evaluate the good status of groundwater and criteria to identify the presence of significant and sustained upward trends in pollutant concentrations. The trend analysis should include the detection of possible reversals in trend and the starting point of such reversals. The starting point for a trend reversal should not exceed 75% of the limit values for groundwater to be developed within the WFD elaboration or prescribed in EU legislation, e.g. the Nitrates Directive. The above criteria should be established within five years after publication of the WFD. Annex 2 and 5 provide information on characterisation of groundwater bodies, criteria for (good) status assessment, monitoring and reporting. The WFD contains relatively rough statements that have to be elaborated in further detail in a WFD Daughter Directive on Groundwater (WFD/DDG).

The European Commission (EC) has initiated a number of groups to elaborate specific subjects and it installed also the Expert Advisory Forum (EAF). EAF has established five drafting groups on groundwater protection strategies to further elaborate some issues of the DDG. However, the recommendations of other Common Implementation Strategy (CIS) working groups will also be important. The groups belonging to EAF are dealing with:

  1. Unpolluted groundwater;
  2. Polluted groundwater- diffuse sources;
  3. Polluted groundwater- point sources;
  4. Interaction surface water-groundwater
  5. Statistic case studies;

These five groups should report before 8 March 2002, so that their results can be discussed in a plenary session of the EAF on 25-26 March.

The present report contains the findings of Drafting Group no.5 on Statistic Case Studies. The Austrian Federal Ministry of Agriculture, Forestry, Environment and Water Management, with the help of project partners from other EU countries, laid the basis for the statistical studies in a project financed for roughly one third by the EC and for two thirds by Austria. The project partners provided data from their respective countries and commented on the study. A subcontractor, the firm “quo data” from Germany, elaborated on statistical aspects. This project started in 1999 and it was reported in 2001 in a Final Report, entitled:

The EU Water Framework Directive: Statistical aspects of the identification of groundwater pollution trends, and aggregation of monitoring results.

The statistical tool (GWStat), described in the final report, was demonstrated in a workshop in Vienna on 25 January 2002. The participating European countries received a computer programme containing the algorithms on statistics and they were asked to apply the tool to groundwater data in their country. Drafting Group no.5 on Statistic Case Studies should collect these experiences and report them to EAF, together with the outcome of a Drafting Group meeting on 20020301. The following report of Drafting Group no.5 contains an outline of the developed method and its merits. Experiences of the participating countries will be represented and, finally, conclusions will be drawn.

2Outline of the statistical tool made under the lead of Austria

The objective of the project on the statistical tool was as follows: The main goal of the project was to establish methods for the calculation of representative mean concentrations, for data aggregation and trend (reversal) assessment at the groundwater body level, respectively for groups of groundwater bodies. The methods had to be suitable for Europe-wide application and implementation based on provisions of the Water Framework Directive taking into account influences originating from diffuse and/or point sources.

The project included an inventory of groundwater bodies (GWB) recognised by participants in the working group, resulting in the description of 21 GWB in 9 countries with data collected on hydrogeology and groundwater quality. The project stated that the monitoring network should fulfil a number of minimum requirements. Homogeneity of the network (reflecting spatial representativeness) was a prerequisite and should be assured. A representativeness index was developed to determine the homogeneity of the sampling sites within the monitoring network. If the GWB is hydrogeologically heterogeneous and if a spatially homogeneous monitoring network is not feasible or sensible, the monitoring network should first be developed to be at least hydrogeologically representative.

Conclusion: The starting point for application of the developed statistical tool is that results are only valid if data from a suitable monitoring network are used.

Given the above starting points, the project developed a practical method as became clear from the Final Report and the demonstration at the Vienna workshop on 25 January 2002. Drafting Group no.5 agrees generally with the choices made by the project. The tool is pragmatic in the sense that complex techniques are avoided in cases where they do not entail an extra benefit. A somewhat more sophisticated tool, such as Kriging, is applied in cases where the result would yield a better representativeness. The presented computer program, being freely available, appears to be consumer-friendly. Also a practical hydrogeologist may apply it. It contains some graphical assets such that the results can be illustrated. The methods used for the trend analysis again represent comprehensible and easily applicable ways to indicate trends. Yet, the methods used are in all cases scientifically sound.

One major objective of the method is in our opinion that much attention is given to the two-dimensional representativeness of the monitoring network, elaborated in a representativeness index, but that the true spatial representativeness is less adequately addressed. The GWStat solution is to repeat the same procedure for other layers in the subsoil, or for average values over depth. However, this will only give a partial remedy. The groundwater quality in an aquifer will often show gradients with depth, for example in cases of salt-water intrusion. This feature is even the more important when recent anthropogenic influences are present. The diffuse pollution by fertiliser and biocides has strongly increased since World War II. These pollutants penetrate from land surface towards deeper layers. The depth of penetration depends on the travel times of groundwater in the soil. The most recent pollutants only reached the upper meters of the groundwater body concerned, deeper layers remaining unattained up to now. Moreover, the Netherlands experience shows that a small variation in boundary conditions (e.g. land cover) or aquifer structure may lead to significant changes in travel time such that nitrate concentrations at one and the same depth will strongly vary. The analysis of trends in a recently increased pollution should take groundwater travel times in the soil into account.

3Practical experience with the statistical tool

Comments were received in February 2002 from Austria (D. Müller, J. Grath, K. Schwaiger), Belgium (F. Delloye), Estonia (A. Marandi), France (A-P. Duboulet),

Germany (J. Grimm-Strele), the Netherlands (L. Boumans) and the UK (R. Ward).

The German remarks can be summarised as follows:

1.Status and boundaries of groundwater bodies

  1. Use of CLAM, resp. CLwAM or CLKM appears justified for the description of the average state as long as it is applied to an appropriately selected data set. The requirements are described below.
  2. Prime interest lies on the possibly influenced part of the three-dimensional groundwater body, that usually is the upper part of the groundwater in the aquifer (like the uppermost 20m to 30m). Where two distinct groundwater bodies lay over each other, the lower groundwater body has to be treated separately.
  3. The boundary of the groundwater body should reasonably coincide with the surface water body. This is often achieved, if the body has an area of approximately 1.500 km² to 5.000 km².
  4. In order to have a sufficient degree of uniformity, which is required when using arithmetic means and confidence limits, sub bodies have to be defined within the groundwater body because of the given variability of the groundwater status.
  5. In order to define a sub body, all available experience about the spatial distribution of the following factors influencing groundwater quality have to be considered:
    Land use: Generally a division into a few categories is sufficient like:
    Arable land (pastures); Urban (industrial, dense traffic areas); Forests; Special agriculture like vine yards, professional gardening.
    Hydrogeology: A further division in large entities is desirable, subdivision into four to five units seems to be sufficient in most cases.
    Additional factors: Protection capacity of the unsaturated zone, groundwater renewal rates, groundwater age, hydraulic conditions, interaction with surface waters have to be considered too. By superimposing these layers (perhaps with the help of a GIS), areas should be defined, where it seems reasonable to expect a relatively homogeneous groundwater quality.
  6. Such sub bodies have – as an order of magnitude and for the conditions given in Germany – an area of approx. 50 km² to 500 km²; in cases where only very diffuse pressures like atmospheric deposition are relevant, even up to approx. 5.000 km².
  7. For the sub bodies it may be sufficient to determine the status from the data from three monitoring sites only, but this poses strict requirements on the selection and quality of the monitoring stations themselves.

2.Trend, trend reversal

If the data set for the status assessment is selected appropriately, the procedure for the determination of trends and of trend reversals gives plausible results.

It would be desirable, to provide a method, to use data with missing values for some years also, because otherwise the data pool will easily become restricted to very few monitoring stations.

The Netherlands experience with the analysis of monitoring results.

The subsurface of large regions in the sandy areas in the Netherlands would by most classical hydrogeological definitions be classified as a homogenous GWB. The National Institute of Public Health and Environment, RIVM, has installed monitoring networks in the area with observation wells provided with screens at depths from groundwater table to 1m below that level (ls-1m) at a level of 10m (ls-10m) and at a level of 25m below land surface (ls-25m). These three levels clearly indicate different averages for parameter values indicating (recent) pollution. The conclusion is that data from one level do not fully represent the groundwater status of the aquifer. Hence, they should be considered as separate subsets, which in itself might be homogeneous. However, even within the subsets great differences in concentrations occur between different categories of land cover near the observation wells. This would lead to an even further subdivision in regions with the same land cover by assuming that recharge will mostly occur close to the observation wells. Nevertheless, the number of subsets becomes large and the distinction between different subsets becomes rather arbitrary.

The Netherlands monitoring networks were installed after 1982. Analysis of results was done since 1984. The interpretation of monitoring data from sandy areas in the Netherlands, obtained at a uniform depth of, for example, ls-10m, was difficult, even for a detailed subdivision. The concentrations of nitrate in observation wells below grassland would represent a case where results might be expected. The load at land surface increased dramatically after roughly 1970, when intensive husbandry started to produce large volumes of manure. The determination of an arithmetic mean for nitrate was accompanied by large confidence limits and the data did only show weakly significant trends not complying with the requirements as formulated in the Final Report. The most probable explanation is that both differences in travel times and denitrification capacities in the soil will widely vary in an almost unpredictable way. The general conclusion is that the presence of a homogeneous data set within a homogeneous groundwater body for elaboration with the statistical tool is only to be found in Utopia and probably not in the Netherlands.

The Belgian experience on application of GWStat

GWStat was applied to a chalk aquifer where monitoring stations consisted of drinking water wells. Note that by using pumping wells with long screens, much of the vertical variation in the aquifer will be averaged (dampened). The main conclusion is that the tool yielded satisfactory results with regard of the quality status of the aquifer. Mean values, established in various ways, and confidence limits were determined which were according to expectations. The representativeness index RI was too small, partly because the monitoring network had to be based on drinking water wells not evenly distributed over the area. The Belgian remark is, consequently, that the network should be pressure-based, or receptor-based (which is not the same but which has the same implications). The French delegate in the Vienna Workshop also stated that land cover (=pressure) is one of the determining factors used in France for the delineation of groundwater bodies. When detecting trends, it became clear that groundwater in the investigated aquifer showed strong seasonal fluctuations and maybe also long-year variations, although GWStat did not detect such effects. The conclusion is that causal factors like variations in net precipitation or the level of the groundwater table should be incorporated as a determining parameter in the statistical tool. However, an additional remark is that such factors will probably have a variable effect on different European aquifers.

The general conclusion is that GWStat is applicable in the Belgian situation. There is a need to incorporate causal factors like water-table fluctuations. The WFD/DDG should also take into consideration a minimum site density based on pressures on groundwater quality.

UK remarks

The purpose of GWStat was to produce something that could be widely used to support the implementation of the WFD. The tool was developed without prior knowledge of the outcomes of discussions on the WFD/DDG. The WFD in Annex 5 specifies criteria for interpretation of chemical status and trends (Sections 2.4.4 and 2.4.5), however, the status assessment is much more than this. In Annex 5, section 2.3.2, Status is defined. This definition refers not only to concentrations and standards but in part also to an achievement of environmental objectives (Article 4) for associated surface waters and associated ecosystems. The criteria set out in 2.4.4 and 2.4.5 only go some way to addressing this and hence the tool too. We must therefore be careful of how the tool is used and realise its limitations. The tool is a technical tool that provides a method of analysing data in a consistent way and which is generally applicable. However, because of its necessarily simple approach it makes a number of compromises and assumptions that are not always valid in all situations. Hydrogeology is heterogeneous and groundwater impacts are heterogeneous. The use of the tool must always be accompanied by a professional appraisal and interpretation along with the use of other more appropriate tools, where necessary, to support the implementation of the WFD and enable the objectives to be met.

French comment on GWStat

The comments on the application of GWStat to French data series contains two aspects. The first one is a wish to apply a further weighing of data to take into account a distinction between different types of observation wells. Moreover, the French experience indicates that it would be useful to add pre-processing and post-processing tools to GWStat.

Estonian experience

In general, the Estonian feeling is that GWB and groundwater status should be clearly defined before GWStat can be applied. GWStat is in itself a good instrument. The Representativeness Index poses difficulties insofar that the RI >80% requirement is not easily met. Furthermore, it would be useful to improve the importation of GIS data.

Austrian remarks

The Austrian delegates stressed the fact that the algorithms developed within EC-Austrian project by Working Group 2.8 should be considered separately from its application in GWStat. The experience obtained with GWStat is a mixture of evaluating the algorithms in itself and their application to groundwater problems. Both aspects should be separately evaluated. The Austrian delegates also remarked that point sources need a specific approach.

4Remarks from the meeting in Brussels on 20020301

Drafting Group no.5 decided on the following remarks in a Brussels meeting on 2002-03-01:

  • The GWStat requirement of a representativeness index RI >80% is too severe in many cases, thereby implying non-consideration of valuable data. Hence, RI >80% should not be included as obligatory in the WFD. Nevertheless, the statistical description of monitoring data should yet discuss the network representativeness.
  • Drafting Group no.5 agrees with the suggestion not to focus on a rigid definition of a groundwater body and on the status of it. The approach to assess groundwater danger zones is a more fruitful approach. Yet, the consistency of an adapted approach of groundwater status with the WFD document should be taken into consideration.
  • Pollution by line and point sources needs a specific monitoring system and, consequently, the elaboration of monitoring data will also need a specific statistical approach, probably deviating from GWStat. The same arguments are valid for systems on the monitoring of ground- and surface water interactions.
  • All aspects of a monitoring system (such as design of monitoring wells, way of sampling, sample treatment, chemical analyses) need to be precisely described in the WFD/DDG and, if possible, follow relevant European CEN Standards.
  • The monitoring network should in most cases focus on the upper strata of a groundwater body to detect pollutants arriving from land surface. Deviations of this rule may be necessary in specific cases (e.g. salt-water intrusion). When the most upper strata are monitored, the two-dimensional approach of GWStat will yield a correct representation of the groundwater situation in those strata and give an indication of groundwater threats to the underlying groundwater body.
  • Important extensions of GWStat might consist of:
  1. Pre-processing tools for input data and post-processing tools for results;
  2. Elaboration of the GIS link;
  3. Treatment and inclusion of series with missing values;
  4. Inclusion of causal parameters in the statistical technique (WARNING: this should not lead to an incorrect representation of results, being affected by causal factors);
  5. Development of a statistical tool for early warning systems.
  • The necessity to remove artefacts is important in some cases and the use of percentiles (discouraged by GWStat) might be helpful.
  • The monitoring network manager should give due attention to the suitability of proposed individual wells for the purpose of monitoring.
  • Monitoring frequency should be more flexible than indicated in GWStat. Notably karst aquifers might need more frequent monitoring than four times per year.

5Discussion and conclusions

The statistical algorithms were developed in a joint Austrian-EC project, being assisted by various European experts. That tool was further elaborated in GWStat, describing the status of groundwater bodies and quality trends. Evaluation of GWStat covers various aspects, such as prerequisites before application, merits of the tool in itself and practical applications. The tool assumes that a number of prerequisites has been fulfilled and, notably, the presence of a suitable data set. Two factors play a mayor role, these are the design of the monitoring network determining the input data and representativeness of the obtained results.