WFD-GW Trend
Minutes, Workshop 8–9 March 2001, Vienna
DraftMinutes of the 2nd Workshop
8–9 March 2001, Vienna
The EU Water Framework Directive: Statistical Aspects of the identification of groundwater pollution trends, and aggregation of monitoring results
Participants: 9 from 11 partners in the project including the project leader; the subcontractor, representatives of the contracting parties as there are the European Commission DG Environment and the Federal Ministry for Agriculture, Forestry, Environment and Water Management Austria; and an expert from Finland as observer. The partners from Ireland and Belgium were not able to attend the meeting.
All slides presented at the workshop by the project leader and the subcontractor and a detailed list of participants can be found in the CIRCA library .
Agenda: see Annex A
1 / Welcome / Participants were welcomed to the 2nd Workshop by the project leader. The provisional agenda was presented and accepted by the participants. (A–Intro-Progress)2 / Introduction
and progress made so far / The project leader informed briefly on different activities carried out so far. A work report was delivered to the Fed. Min. for Agriculture, Forestry, Environment and Water Management Austria in September 2000, in October 2000 a project information sheet was drafted and distributed to the partners and finally in January 2001 the first Interim Report was delivered to the contracting parties.
Project presentations were given to the contracting parties (unfortunately the representative of DG Env. was not able to attend the meeting), to a Delegation of the French Ministry for the Environment (both in Dec. 2000) and at the "Bridging the Gap Workshop" (Paris Feb. 2001).
Information requests were received from Portugal (during the Portuguese EU presidency), Norway and Finland.
Then progress of the project was described with reference to the initial work plan (discussed at the 1st workshop) which is still valid (A–Intro-Progress). Up to now all milestones were delivered/ elaborated in accordance with this work plan. As the provisions for data aggregation in the final version of the WFD differ slightly from the version which was the basis for the project proposal the need arose to develop and test a greater number of candidate methods.
3 / 1st Interim
Report / The 1st Interim Report was elaborated and delivered to the contracting parties by the end of Jan. 2001. The project leader briefly presented the content of the report. (B–Interim Report)
4 / Data availability / An overview of the available test data sets was given by presenting some key figures on the size of the GW-bodies, the types of sampling sites and the type of sampling, the number of parameters per GW-body and the amount of values below the LOD (exemplary for ammonium) (C–Data availability).
5 / Requirements of the WFD and different candidate data aggregation methods were presented and characterised with regard to applicability, interpretability and some statistical properties. The discussion focused on the treatment of unevenly distributed monitoring sites and the treatment of measurements below the limit of quantification and the limit of detection (LOQ/LOD).(D–Aggregation methods)
All calculations were based on annually aggregated data. The term regularised in the data sheet refers to this temporal aggregation.
With regard to the term “hot spots” mentioned in the overheads it was agreed that this refers to “uneven distribution of pollution caused by local or diffuse sources, observed at some points in a groundwater body which show higher concentrations than the rest of the groundwater body”.
Treatment of LOQ/LOD values: For the assessment carried out so far a replacement method based on minimum and maximum calculations was applied. The discussion showed that there is an urgent need to provide sufficient information on both LOQ and LOD. This will have to be considered when defining monitoring requirements and analytical procedures (which is not within the frame of this project).
It was demonstrated by examples and theoretical considerations that the Maximum Likelihood (ML) approach based on the log-normal distribution may produce artificial outliers if the data are not log-normally distributed. Problems appear especially if more than 80% of the values are below the LOQ and in case of a high variation of the numeric values. As this was frequently the case with test data sets it was decided that this methodology will no longer be taken into account.
The fact was raised that in several countries small GW-bodies exist, which, due to different hydrogeological conditions, cannot be grouped. Therefore, a data aggregation method has to be provided for GW-bodies with less than 10 sampling stations. For that reason the methodologies with percentiles were excluded from further discussion (would require at least 10 sites for statistically sound results). As data sets on small GW-bodies have so far not been available for testing Ana Rita Lopes (PT) offered to provide typical examples.
Finally, the kriging method and its advantages was demonstrated on several examples. Especially the effect of down-weighting the impact of clusters of stations and giving higher weights to solitary stations was shown. It was pointed out that there is a close relationship between the arithmetic mean and the kriging mean in case of evenly distributed sites or in absence of spatial auto-correlation. Some concerns were raised that due to different land use etc. a spatial modelling of the concentration levels by Kriging might not be appropriate since land use and hydrogeological conditions should also be taken into account.
It was stressed that the monitoring network should fulfil some minimum requirements. After a short discussion the following minimum requirements on monitoring to allow for sound statistical assessment in accordance with the requirements of the WFD were enumerated:
-Homogeneity of the monitoring network (with regard to the spatial distribution of stations and the types of stations)
-In case of changes of monitoring stations the monitoring network manager should assure that these changes do not affect the outcome of the assessment.
It was noted that it was not within the scope of the project to check whether these requirements are fulfilled and that this was left to the entire discretion of the Member States.
For data aggregation it was concluded that:
-Kriging reflects the spatial structure of the GW-quality data and also to a certain extent the impact of factors affecting the concentration level within the area such as land use and hydrogeological conditions, if these are spatially correlated.
-In case of no spatial correlation of these factors, results of kriging will be similar to those of arithmetic mean calculations.
-In general the difference between the kriging mean and the arithmeticmean decreases if the monitoring network is homogeneously designed and representative with regard to human impacts and environmental conditions.
-Compliance with good groundwater chemical status at a given level of confidence can be demonstrated with a statistical test for the null hypothesis
H0: „gw body is not in good status, i.e. true mean level above the limit value“
and the alternative hypothesis
H1: „gw body is in good status, i.e. true mean level below the limit value“
H1 may be considered statistically proven at significance level alpha (e.g. 5%), if the corresponding upper confidence limit (CL) at confidence level (1-alpha/2) is below the limit. Alpha denotes the probability of making a wrong decision for a good status (although the true, unknown mean exceeds the limit value); alpha could vary for different parameters.
The upper confidence limit (CL) depends on the variability of the concentration level within the GW-body and on the number of stations. CL decreases with an increasing number of stations within the GW-body. To some extent it is therefore in the hands of the monitoring manager whether the CL will be below or above the limit value. The use of the CL allows to reduce the number of stations in GW-bodies with levels far below the limit value, and enforces a higher number of stations in GW-bodies with levels close to the limit value, thus allowing an effective allocation of analytical resources.
With regard to the further procedure of the project it was decided:
-Representatives of the contracting parties emphasised that the outcome of the project should be a proposal of one data aggregation method suitable for small and large as well as for groups of GW-bodies.
Hence, it was decided to further explore and compare the properties of the arithmetic mean and kriging mean with special focus on the variogram. In order to guarantee the required level of confidence for GW-bodies with only a few stations it was decided to use the upper confidence limit of arithmetic mean and kriging mean instead of the mean values itself. For the same reason percentiles were not considered appropriate for the assessment of the chemical status.
6 / Candidate methods for both trend and trend reversal assessment were presented (E–Trends).
Based on the provisions of the WFD and the following requirements, methods for trend and trend reversal assessment were selected:
For the assessment of monotonic trends the test of Mann-Kendall and a generalized linear regression test (ANOVA test) based on the LOESS smoother were taken into consideration. For the assessment of trend reversals the following methods were considered:
- test of Mann-Kendall applied to subsequent differences
- test of the quadratic trend component based on a quadratic regression model
- two-sections test (based on a linear model in two sections)
For the selection of the methods the following criteria were taken into consideration:
- applicability for all types of parameters
- extensibility to potential adjustment factors
- sufficient power for the detection of trend/reversal
- robustness was considered less important than power and extensibility (data validation will be responsibility of MS)
The goal of the project is not to provide tools for the calculation of cause-effect relationships but leave it to the MS. The applied trend calculation model should be extendable to an adjustment of factors such as precipitation, water level, nitrogen application, etc.
With regard to extensibility and power the linear methods (based on a linear model) outperform non-parametric methods based on the test of Mann-Kendall, and therefore the decision was in favour of the linear methods. The consequence was a decision for the generalized linear regression test (ANOVA test) for the assessment of monotonic trends. For the assessment of a trend reversal, the consequence was a decision for the two sections model, due to its simple interpretability.
It was noted that for many GW-bodies and parameters the detectable trend (a hypothetical trend that can be statistically detected with 90% probability) exceeds 50% relative increase between starting point and end point of the time series. This implies a considerable risk of exceeding the limit value without detecting a trend, if the trend analysis is starting at a concentration level of 75% of the limit value. This implies further that the trend detection would loose its "early warning function". To avoid that risk it was considered necessary to start the trend analysis at a level clearly below that percentage (eg. at 50% of the limit value).
In case of pesticides it might occur, that the LOQ is equal to the limit value of 0.1 µg/l. In this case there is a risk that trend detection is not appropriate any more. It was considered necessary to establish criteria for the decision whether or not a trend analysis is appropriate. One criterion to be investigated more closely is the relation between the minimum and the maximum mean (with LOD=0 and LOD =LOD, respectively). In any case sensitive analytical methods should be applied, respectively developed.
Within the frame of the project it should be tested if the developed and proposed methods are applicable to various types of GW-bodies, parameters etc. It is not the goal to make validated calculations of trends or trend reversals with the provided test data sets.
Further procedure:
For the selection of the base year and starting point of trend calculations a proposal will be elaborated and distributed to partners for comments.
7 / GIS Information / Most important available GIS information (within the project) was presented - e.g. GW-body information (boundaries, name and size, hydrogeologic information) sampling sites (type of sampling sites and referring quality data), land use information (CORINE Landcover)
(F–GIS-CORINE).
Since it is explicitly expressed in the contract that the final report has to be presented in the internet, the question was raised if there are any restrictions/objections against providing available information in the WWW (G–GIS Info). The maximum scale for maps will be 1:200 000.
It was concluded that there are no restrictions by the partners. A concept of the presentation in the web will be developed by the project leader and a password protected test version will be made accessible to the partners for comments before allowing general access.
8 / Reporting /
Interim Report
It was discussed whether the 1st Interim Report should be further distributed or be treated as an internal document. Within the discussion it was emphasised that other groups dealing with the implementation of WFD or MS not directly involved in the project will be highly interested in the results of this project. For that reason the interim report should be made available on request. This view was shared by all participants – project partners as well representatives of contracting parties.The project leader will provide the 1st Interim Report to project partners and Reetta Waris (FI observer) on CD-ROM.
Final Report
It was agreed that the final report will be made available as printed version in limited distribution to contracting parties and partners. This printed version shall contain the main findings of the project (H–Final report)Furthermore a website will be installed. On this website the above mentioned summary of findings and all further results and products of the project (tables, figures and maps) shall be presented. On the website links to all partners will be placed. Annika Nilsson stated that information will also be made available on the EC CIRCA server.
9 / 3rd Workshop / The 2nd workshop was held previously to the envisaged date which was due to the reason that already several results had been available for discussion and decisions of the group were necessary as basis for the further procedure.
During the discussion it became evident that some minor decisions, which will be needed for the finalisation of the project, will have to be taken before the summer break. Otherwise finalisation of the project in due time will be endangered. It was agreed that proposals will be distributed to partners for comments to be made within four weeks at the latest (unless stated otherwise) .
Topics left for discussion at a 3rd workshop around the middle of September would mainly deal with contents of the final report, layout and contents of the website and the conclusions of the project. It was agreed that such a meeting would be very helpful. Therefore two possible dates were proposed: 17/18. Sept. 2001 or 20/21. Sept. 2001. Partners will check possibilities and reply to the project leader within a few days.
10 / AOB / The project leader will keep the partners informed on the progress of work and on further activities via the CIRCA server.
The project leader thanked all partners and the subcontractor for their participation and contributions to the workshop.
Annika Nilsson expressed her appreciation of the project work and emphasised the importance of the project. It was stressed that results from this project will be used as a recommendation for the further implementation of the WFD and results will also be guidance for other WFD-related projects.
Contracting parties and partners emphasised that a method for data aggregation and trend (-reversal) assessment developed by the project team will be used as recommendation for further implementation of the WFD.
Participants expressed their thanks to the host organisation.
page 1 of 6
WFD-GW Trend
Minutes, Workshop 8–9 March 2001, ViennaAnnex A: Agenda and Timetable
Annex A: Agenda and Timetable
2nd Workshop, 8–9 March 2001
(Federal Environment Agency ltd. – A-1090 Vienna, Spittelauer Lände 5)
Thursday, 8 March 2001
10:00–10:30 / Welcome address10:30–10:45 / Progress made so far
10:45–11:45 / 1st Interim Report / Presentation of the first Interim Report (contents, relevant provisions of the WFD with particular emphasis of the final version, description of GW-bodies, groundwater quality data, statistics applied in Member States, statistical methods discussed within the project)
11:45–12:15 / Data availability / (see Interim Report) GW-bodies and parameters provided, length of time series, type of sampling (mixed, upper, medium, deep), sub-bodies, point sources,... monitoring network design, ...
12:15–13:30 / Break for lunch
13:30–16:00 / Data aggregation / Requirements of the WFD, statistical methods for data aggregation, examples, discussion, summary of findings
16:00–16:15 / Coffee break
16:15–16:45 / GIS / GIS-information - provision of information (e.g. scale, type of information) Discussion
Friday, 9 March 2001
9:00–11:30 / Assessment of trends and trend reversal / Requirements of the WFD, statistical methods for trend calculations, examples, discussion, summary of findings11:30–12:30 / Reporting / Final report - printed version and Internet
12:30–13:15 / Break for lunch
13:15–13:30 / Discussion of possibilities for a 3rd workshop / Topics left for further discussion => 3rd Workshop in September?
13:30–14:00 / AOB
page 1 of 6