Policy context and targets
Context description
There are a number of EU directives that aim to improve water quality, and reduce the loads and impacts of organic matter. First, the Water Framework Directive requires the achievement of good ecological status or good ecological potential of rivers across the EU by 2015 and repeals, step-by-step, several older water related directives. Alongside this, the following directives stay in place: the Nitrates Directive (91/676/EEC), aimed at reducing nitrate and organic matter pollution from agricultural land; the Urban Waste Water Treatment Directive (91/271/EEC), aimed at reducing pollution from sewage treatment works and certain industries (see also CSI24 Urban waste water treatment); and the Integrated Pollution Prevention and Control Directive (96/61/EEC), aimed at controlling and preventing the pollution of water by industry.
Targets
The environmental quality of surface waters with respect to organic pollution and ammonium, and the reduction of the loads and impacts of these pollutants are, however, objectives of several directives. These include the Surface Water for Drinking Directive (75/440/EEC), which sets standards for the BOD value of surface water intended for the abstraction of drinking water. The level should not exceed 3 mg O2/l if the water is intended for abstraction with simple physical treatment and disinfection only, e.g. rapid filtration and disinfection.
A subsequent Drinking Water Directive 98/83/EC does not set guideline or mandatory values of BOD but defines indicator limits for ammonium: remedial actions for improving drinking water quality should be taken if an indicator value of 500 µg NH4/l (i.e. 388 µg N/l) is exceeded.
The guide values of BOD and ammonium are also set in the Fish Directive (2006/44/EC) and focus on supporting fish life, while also aiming at improving and maintaining freshwater quality from an ecological and economic point of view. BOD values should not exceed 3 mg O2/l in salmonid waters and 6 mg O2/l in cyprinid waters, while ammonium concentrations should not exceed 40 µg NH4/l (i.e. 31 µg N/l) in salmonid waters and 200 µg NH4/l (i.e. 155 µg N/l) in cyprinid waters.
The UN Sustainable Development Goals also include targets related to river oxygenation conditions. The most relevant targets are 6.3 on improving water quality, specifically addressing wastewater treatment, and 6.6 on protecting and restoring water-related ecosystems, but also 6.1 on safe drinking water and several other targets with a more indirect coupling to organic pollution.
Related policy documents
Methodology
Methodology for indicator calculation
Data source
The data on water quality on rivers in Waterbase are collected annually through the WISE SoE - Water Quality (WISE-6) data collection process. It includes data on nutrients, organic matter, hazardous substances and general physico-chemical parameters, as well as biological data for rivers and lakes. The WISE-6 data flow was new as of 2019, and has from 2020 replaced WISE-4. This reporting obligation is an EIONET core data flow. A request is sent to National Focal Points and National Reference Centres every year with reference to templates to use and guidelines. As of 2015, WISE SoE - Water Quality (WISE-6) supersedes Eurowaternet reporting on river quality (EWN-1).
The data in Waterbase are a sub-sample of national data assembled for the purpose of providing comparable indicators of the pressures, state and impact of waters on a Europe-wide scale. The data sets are not intended for assessing compliance with any European directive or any other legal instrument. Information on the sub-national scales should be sought from other sources.
Monitoring sites selection
Data from all reported monitoring sites are extracted for the indicator assessment. Some data are excluded following the QC process (see QC below). The time series is based on complete time series only (see Inter/extrapolation and consistent time series below).
Determinants
The determinands selected for the indicator and extracted from Waterbase are BOD5, BOD7 and ammonium.
Most countries monitor BOD5. Finland monitors BOD7. Lithuania monitored BOD5 up to 1995 and started monitoring BOD7 in 1996. Latvia monitored BOD7 from 1996 to 2001. Estonia has monitored BOD5 since 2010, replacing BOD7 which was monitored until 2009. 'BOD' is commonly used for BOD5. To be comparable the data of BOD7 have been converted to BOD5 (BOD7=1.16 BOD5).
All countries reported total ammonium until 2006. In 2007, Greece and Liechtenstein started reporting ammonium instead of total ammonium. Instead of total ammonium, Cyprus, Lichtenstein and Slovenia began reporting ammonium in 2008, as did Austria and the Netherlands in 2009, Bulgaria and Latvia in 2010, and Estonia, Norway and Poland in 2011. Besides total ammonium, Slovakia also started to report ammonium for some sites in 2008. Belgium, Germany, Italy, Luxembourg, Slovakia and the United Kingdom report either ammonium or total ammonium for an individual site in a selected year from 2008. Data for either of the two determinands were included in the assessment. For those sites in Slovakia where both were reported, total ammonium data were included in the assessment.
Mean
Annual mean concentrations are used as a basis in the present concentration and indicator analyses. Unless the country reports aggregated data, the aggregation to annual mean concentrations is done by the EEA. Countries are asked to substitute any sample results below the limit of quantification (LOQ) by a value equivalent to half of the LOQ before calculating the site annual mean values. The same principle is applied by the EEA.
The annual data in most cases represent the whole year, but data are used also if they represent shorter periods. Up until 2012, data could be reported at different temporal aggregation levels. Here, annual data have been selected, but if these were not available, seasonal data were selected according to a specific order of preference.
Quality control (QC)
An automatic QC procedure is applied when data are reported, including checking that the values are within a certain range defined for each determinand. Automatic outlier tests based on z-scores are also applied, both to the disaggregated and aggregated data, excluding data failing the tests from further analysis. In addition, a semi-manual procedure is applied, to identify issues which are not identified in the automatic outlier tests. The focus is particularly on suspicious values having a major impact on the country time series and on the most recently reported data. This comprises e.g.:
- values not so different from values in other parts of the time series, but deviating strongly from the values closer in time;
- consecutive values deviating strongly from the rest of the time series (including step changes);
- whole time series deviating strongly in level compared with other time series for that country and determinand;
- where values for a specific year are consistently far higher or lower than the remaining values for that country and determinand.
Such values are removed from the analysis (both time series/trend and present state analysis) and checked with the countries. Depending on the response from the countries, the values are corrected, flagged as outliers or flagged as confirmed valid. Any response affecting the indicator analysis is corrected in the next update of the indicator.
Inter/extrapolation and consistent time series
For time series analyses, only series that are complete after inter/extrapolation (i.e. no missing values in the monitoring site data series) are used. This is to ensure that the aggregated time series are consistent, i.e. including the same monitoring sites throughout the time series. In this way, assessments are based on actual changes in concentration, and not changes in the number of monitoring sites. For the trend analysis, it is essential that the same time period is used for the different monitoring sites, so that the results are comparable. However, the statistical approach chosen can handle gaps in the data series, so inter/extrapolation is not applied here.
Changes in methodology: Monitoring site selection and inter/extrapolation
Until 2006, only complete time series (values for all years from 1992 to 2004) were included in the assessment. However, a large proportion of the monitoring sites were excluded by this criterion. To allow the use of a considerably larger share of the available data, it was decided in 2007 (i.e. when analysing data up until 2005) to include all time series with at least 7 years of data. This was a trade-off between the need for statistical rigidity and the need to include as many data as possible in the assessment. However, the shorter series included might represent different parts of the whole time interval and the overall picture may therefore not be reliable.
In 2009, it was decided to inter/extrapolate all gaps of missing values of 1-2 years for each monitoring site. At the beginning or end of the data series, one missing value was replaced by the first or last value of the original data series, respectively. In the middle of the data series, missing values were replaced by the values next to them for gaps of 2 years and by the average of the two neighboring values for gaps of 1 year.
In 2010, this approach was modified, allowing for gaps of up to three years, both at the ends and in the middle of the data series. At the beginning or end of the data series up to three years of missing values were replaced by the first or last value of the original data series, respectively. In the middle of the data series, missing values were replaced by the values next to them, except for gaps of one year and for the middle year in gaps of three years, where missing values were replaced by the average of the two neighboring values.
In 2018, this approach was slightly modified using linear interpolation for gap filling in the middle of the time series. Moreover, if data were available from 1989-1991 these were applied in the gap filling procedure, making it possible to interpolate instead of extrapolating at the beginning of the time series.
Only time series with no missing years for the whole period from 1992 or 2000 after such inter/extrapolation are included in the assessment. Even if the gap filling is not applied in the trend analyses, the same time series are used, for easier comparison of the time series and trend analysis results. This gap filling procedure increases the number of monitoring sites that can be included in the time series/trend analysis. Using also the shorter time period starting in 2000 allows the inclusion of more monitoring sites, and hence increases the representativeness. Still, the number of monitoring sites in the time series/trend analysis is markedly lower compared to the analysis of the present state, where all available data can be used.
Aggregation of time series
The selected time series (see above) are aggregated to country and European level by averaging across all monitoring sites for each year.
Trend analyses
Trends are analysed by the Mann-Kendall method (Jassby et al., 2020) in the free software R (R Core Team 2020), using the wq package. This is a non-parametric test suggested by Mann (1945) and has been extensively used for environmental time series (Hipel and McLeod, 2005). Mann-Kendall is a test for monotonic trend in a time series y(x), which in this analysis is nutrient concentration (y) as a function of year (x). The test is based on Kendall's rank correlation, which measures the strength of monotonic association between the vectors x and y. In the case of no ties in the x and y variables, Kendall's rank correlation coefficient, tau, may be expressed as tau=S/D where S = sum_{i<j} (sign(x[j]-x[i])*sign(y[j]-y[i])) and D = n(n-1)/2. S is called the score and D, the denominator, is the maximum possible value of S. The p-value of tau is computed by an algorithm given by Best and Gipps (1974). The tests reported here are two-sided (testing for both increasing and decreasing trends). Data series with p-value < 0.05 are reported as significantly increasing or decreasing, while data series with p-value >= 0.05 and <0.10 are reported as marginally increasing or decreasing. The results are summarised by calculating the percentage of monitoring sites within each category relative to all monitoring sites within the specific aggregation (Europe or country). For the relative Sen slope (%), the slope joining each pair of observations is divided by the first of the pair before the overall median is taken and multiplied by 100. Again, this is summarised for Europe or individual countries by averaging across sites. The Sen slope was introduced for this indicator in 2013.
The size of the change is estimated by calculating the Sen slope (or the Theil or Theil-Sen slope) (Theil 1950; Sen 1968) using the R software. The Sen slope is a non-parametric method where the slope m is determined as the median of all slopes (yj−yi)/(xj−xi) when joining all pairs of observations(xi, yi). Here the slope is calculated as the change per year for each monitoring site. This is summarised by calculating the average slope (regardless of the significance of the trend) for all monitoring sites in Europe or a country. The relative change per year (Sen slope %) is calculated as the Sen slope relative to the time series average. Again, this is summarised for Europe or individual countries by averaging across monitoring sites. The Sen slope was introduced for this indicator in 2013.
The Mann-Kendall method or the Sen slope will only reveal monotonic trends, and will not identify changes in the direction of the time series over time. Hence a combination of approaches is used to describe the time series: a visual inspection of the time series, describing whether the general impression is a monotonic trend, no apparent trend, clear shifts in direction of the trend or high variability with no clear direction; an evaluation of significant versus non-significant and decreasing versus increasing monotonic trends using the Mann-Kendall results; an evaluation of the average size of the monotonic trends using the Sen slope results.
Current concentration distributions
For analysis of the present state, average concentrations are calculated across the last 3 years. Outliers and suspicious values are removed before averaging. In this case all monitoring sites can be used, which is a far higher number than those that have complete time series after inter/extrapolation. The 3-year average is used to remove some inter-annual variability. Also, since data are not available for all monitoring sites each year, selecting data from 3 years will give more sites. The average value thus represents 1, 2 or 3 years.
The sites are assigned to different concentration classes and summarised per country (count of sites per concentration class). The classes defining values are based on the range of concentrations found in Waterbase and only give an indication of the relative distribution of values of BOD and ammonium in each country.
Methodology for gap filling
The methodology for gap filling is described under inter/extrapolation and consistent time series.
Methodology references
-
Best, D., and Gipps, P. (1974).
Algorithm AS 71: The Upper Tail Probabilities of Kendall's Tau. Journal of the Royal Statistical Society. Series C (Applied Statistics), 23(1), 98-100.
-
Jassby, A.D., Cloern, J.E. and Stachelek, J. (2017).
wql: Exploring Water Quality Monitoring Data. R package version 0.4.9.
-
Mann, H.B. (1945).
Nonparametric tests against trend, Econometrica, 13, 245-259.
-
R Core Team (2020).
R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
-
Sen, P. K. (1968).
"Estimates of the regression coefficient based on Kendall's tau", Journal of the American Statistical Association 63: 1379–1389
-
Theil, H. (1950).
"A rank-invariant method of linear and polynomial regression analysis. I, II, III I, II, III". Proceedings van de Koninklijke Nederlandse Akademie van Wetenschappen 53: 386–392, 521–525, 1397–1412.
Uncertainties
Methodology uncertainty
The methodologies used for aggregating and testing trends in concentrations illustrate the overall European trends. Organic and oxygen conditions vary throughout the year, depending especially on flow conditions, affected by weather events etc. Hence, the annual average concentrations should ideally be based on samples collected as often as possible. Using annual averages representing only part of the year introduces some uncertainty, but it also makes it possible to include more river sites, which reduces the uncertainty in spatial coverage. Moreover, the majority of the annual averages represent the whole year.
Data sets uncertainty
This indicator is meant to give a representative overview of oxygenation availability in European rivers. This means it should reflect the variability in conditions over space and time. Countries are asked to provide data on rivers according to specified criteria.
The datasets for rivers include almost all countries within the EEA, but the time coverage varies from country to country, both through the analysed period and within the year for which the aggregated mean value is provided. It is assumed that the data from each country represents the variability in space in their country. Likewise, it is assumed that the sampling frequency is sufficiently high to reflect variability in time. In practice, the representativeness will vary between countries.
Each annual update of the indicator is based on the updated set of monitoring sites. This also means that due to changes in the database, including changes in the QC procedure that excludes or re-includes individual sites or samples and retroactive reporting of data for the past periods, which may re-introduce lost time series that were not used in the recent indicator assessments, the derived results of the assessment vary in comparison to previous assessments. As an example, the 2016-2018 assessment of current concentrations of ammonium in rivers is based on 7797 sites, compared to 6577 sites in the assessment of 2015-2017 (i.e. the last year assessment). However, some sites available in the previous assessment are not part of this year’s assessment and vice versa.
Waterbase contains a large amount of data collected throughout many years. Ensuring the quality of the data has always been a high priority. A revision of Waterbase reporting and the database-composition process took place in the period 2016–2018. This included restructuring of the data model and corresponding reporting templates; transformation of the legacy data (i.e. data reported in the past, for the period up to and including 2012); re-definition of specific data fields, such as aggregation period defining the length of aggregation in a year; update of the datasets according to correspondence with national reporters; re-codification of monitoring site codes across Eionet dataflows; and connection of the legacy data time series with the newly-reported data in restructured reporting templates. Still, suspicious values or time series are sometimes detected and the automatic QA/QC routines exclude some of the data. Through the communication with the reporting countries, the quality of the database can be further improved.
Rationale uncertainty
Biochemical oxygen demand and total ammonium show oxygen consumption and are thus well suited to illustrating water pollution. However, using annual average values may not fully illustrate the severity of low oxygen conditions.
