Introduction

Environmental problems related to arable land have a long history in Europe, and they have accelerated in recent decades associated with changes to landscapes, vegetation and animal communities. This has mainly resulted in increased adoption of external inputs such as fertilizers and pesticides (Stoate et al. 2001). Strebel et al. (1989) highlighted nitrate pollution in Western Europe, the main source of nitrogen being agricultural leaching caused by excess inputs of fertilizers and manure, this being especially evident in north-western Europe (Iversen et al. 1998).

Agrochemicals that pose the greatest threat to human health are nitrate fertilizers and pesticides (Bouman et al. 2002; Gardner and Vogel 2005). The high nitrate concentrations in some groundwaters used for human consumption are causing increasing concern (Rajagopal and Tobin 1989). Diffuse (non-point) nitrate contamination originating from agriculture is a worldwide challenge which has been widely documented throughout the world (e.g. Mitchell et al. 2003; Mohamed et al. 2003; Thorburn et al. 2003; Oren et al. 2004; Liu et al. 2005). Many studies have shown the relation between agricultural practices and diffuse contamination of groundwater (e.g. Hudak 2000; Spalding et al. 2001; Harter et al. 2002; Johnsson et al. 2002; Lake et al. 2003). In contrast, point sources of nitrate contamination are more related to urbanized areas and septic tanks (Hantzsche and Finnemore 1992; Aravena et al. 1993; Aravena and Robertson 1998).

As a response to this threat, the European Union (EU) adopted in 1991 the Nitrate Directive 91/676/EEC (EU 1991) requiring that Member States take measures to minimize agricultural nitrate sources in nitrate-contaminated zones. More recently, the “Water Framework Directive 2000/60/EC” (EU 2000) was published, stating that a “good” status of groundwater is required for all EU Member States and specific measures have to be adopted to prevent and control pollution of groundwater. In order to achieve the environmental objectives for groundwater, the Water Framework Directive 2000/60/EC requires Member States to identify and reverse any significant upward trend in the concentration of pollutants and to achieve a good groundwater status by the end of 2015.

For EU Member States, a series of recommendations were made for groundwater quality trend assessment and data aggregation (Grath et al. 2001), proposing parametric (linear regression) and non-parametric (Mann-Kendall) methods, owing to their capability to detect different types of patterns of change and their robustness. Using very detailed nitrate datasets, an efficient methodology for trend analysis derived from the general guidelines of the EU Water Framework Directive is tested and evaluated in the Hesbaye chalk aquifer in the Geer watershed basin in Belgium. This is a representative northwest European chalky aquifer, from the point of view of its geology and spread of nitrate contamination (Downing et al. 1993). This analysis allows one to define the minimal requirements in terms of field data acquisition and processing. It also shows the limitations of the regulations in containing the problem and the fact that the nitrate problem is probably more acute than expected in the Geer basin. The study also suggests that many other cultivated groundwater basins in Europe have a similar problem because of the expected delay between changes in manure application at the soil surface and changes in groundwater quality and so reversal of nitrate trends.

Methodology for statistical trend analysis

The methodology used in this research mostly follows the work of Grath et al. (2001) who proposed particular algorithms and techniques for the identification of pollutant trends in groundwater. As suggested by Hirsch et al. (1991), a three-step procedure is considered (Fig. 1): (1) normality test of the dataset; (2) trend detection; and (3) trend estimation. This three-step procedure is described here after. Results of the three-step procedure are presented in Table 1.

Fig. 1
figure 1

A three-step procedure is adopted for nitrate trend analysis: (1) normal/non-normal distribution data; (2) trend detection; (3) trend estimation

Full size image
Table 1 Trend test results for the observation points (groundwater) used in the statistical analysis using the Shapiro-Wilk (SW), Shapiro-Francia (SF) and D’Agostino’s (D) tests
Full size table

Normality of the dataset

As will be discussed further in this report, the application of a normality test to the datasets is a priori necessary, in order to select the trend detection method to be applied (parametric or non-parametric). For this purpose, the Shapiro-Wilk test (Shapiro and Wilk 1965), hereafter called SW-test, or the Shapiro-Francia test (Shapiro and Francia 1972), hereafter called SF-test, was used depending on the number of records of the dataset (Conover 1980; Helsel and Hirsch 1995). Both tests have been widely used to test the normality of environmental datasets (Kumagai et al. 1997; Bonett and Seier 2002; Henderson 2005; Zhang and Wu 2005). They are generally considered as the most powerful tests of normality (Stephens 1974; Gan and Koehler 1990; Bonett and Seier 2002; Mudholkar et al. 2002).

In order to corroborate the results obtained using the SW and SF-tests, the D’Agostino’s test (D’Agostino 1970a, b), hereafter called the D-test, has also been applied. It is also a very popular and commonly used test (Mudholkar et al. 2002). Generally, the D-test results have corroborated those obtained with the SW-test or the SF-test.

Trend detection

For normally distributed datasets, a simple linear regression is applicable as a tool for trend detection. Examples of application of this technique in groundwater studies include the works of Hanson (2002), Valverde Ramírez et al. (2005) and Zhang et al. (2005). Trend detection is based on the calculation of the correlation coefficient r, also called Pearson’s r, which is a quantitative measure of correlation between time t and concentration C:

$$ r = \frac{{{\sum\limits_{{\text{i}} = 1}^n {{\left( {t_{{\text{i}}} - \overline{t} } \right)}{\left( {C_{{\text{i}}} - \overline{C} } \right)}} }}} {{{\sqrt {{\sum\limits_{{\text{i}} = 1}^n {{\left( {t_{{\text{i}}} - \overline{t} } \right)}^{2} {\sum\limits_{{\text{i}} = 1}^n {{\left( {C_{{\text{i}}} - \overline{C} } \right)}^{2} } }} }} }}} $$
(1)

In accordance with Carr (1995), three ranges of correlation degrees have been considered:

  • Strong correlation for r values ranging between 0.8 and 1 (or −0.8 and −1)

  • Moderate correlation for r values ranging between 0.5 and 0.8 (or −0.5 and −0.8)

  • Weak correlation for r values ranging between 0.1 and 0.5 (or −0.1 and −0.5)

  • No correlation for r values ranging between −0.1 and 0.1

For non-normally distributed datasets, trend detection is performed using the non-parametric Mann-Kendall test (Mann 1945; Kendall 1975), hereafter called MK-test. Examples of use of the MK-test for detecting trends in hydrological and hydrogeological time series data include the works of, e.g. Hirsch et al. (1982), Taylor and Loftis (1989), van Belle and Hughes (1984), Yu et al. (1993), Lee and Lee (2003), Kahya and Kalayci (2004), Zhang et al. (2005).

The MK-test has the advantage that it does not assume any distribution for the data and it has similar power as parametric methods (Serrano et al. 1999). The MK-test determines whether a trend is present or not with an indicator (T) based on the calculation of differences between pairs of successive data:

$$ T = {\sum\limits_{{\text{i}} = 1}^{n - 1} {{\sum\limits_{{\text{j}} = {\text{i}} + 1}^n {\operatorname{sgn} {\left( {C_{{\text{j}}} - C_{{\text{i}}} } \right)}} }} }, $$
(2)

where:

$$\operatorname{sgn} {\left( {C_{{\text{j}}} - C_{{\text{i}}} } \right)} = {\left\{ {\begin{array}{*{20}c} {1} & {{{\text{if}}}} & {{C_{{\text{j}}} - C_{{\text{i}}} > 0}} \\ {0} & {{{\text{if}}}} & {{C_{{\text{j}}} - C_{{\text{i}}} = 0}} \\ {{ - 1}} & {{{\text{if}}}} & {{C_{{\text{j}}} - C_{{\text{i}}} < 0}} \\ \end{array} } \right\}},$$
(3)

where C j and C i are concentration data at different time i and j, with j>i and n is the size of dataset.

A 95% significance level has been used for the trend detection test (Helsel and Hirsch 1995), corresponding to a threshold value of T thresh=1.65. It is thus considered that a trend is present for values of T>T thresh. The MK-test being non-parametric, it is not possible to define various degrees of trend robustness, as performed with linear regression.

Trend estimation

The trend magnitude is expressed in units of increment of nitrate concentration per year (mg/L \( {{NO}}^{ - }_{3} \) per year).

For normally distributed datasets, the magnitude of the trend is estimated using the slope of the linear regression equation (USEPA 2000):

$$ C = a + bt, $$
(4)

where a is the y-intercept (t=0) of the line and b the slope.

For non-normally distributed datasets, the trend magnitude is based on calculation of the Sen’s slope estimator (Hirsch et al. 1991), which is the nonparametric alternative to the linear regression slope. The Sen’s slope is obtained as follows. First, one computes the slopes (b ij) for all pairs of successive data:

$$ {\mathop b\nolimits_{{\text{ij}}} } = \frac{{C_{{\text{j}}} - C_{{\text{i}}} }} {{t_{{\text{j}}} - t_{{\text{i}}} }}, $$
(5)

where C i and C j are nitrate concentrations at time t i and t j, respectively.

Finally, the value of the Sen’s slope estimator (H) is the median of those slopes (USEPA 2000):

$$b = {\text{median}}{\left( {b_{{{\text{ij}}}} } \right)}$$
(6)

Trend analysis as applied to the Geer Basin

The Hesbaye aquifer, located in the Senonian chalk formations of the Geer basin, is an important groundwater resource for drinking water supply for the city of Liège and its suburbs, where around 600,000 people live and consume about 30 million m3 of water per year (Brouyère et al. 2004a). The land use is dominated by agriculture, covering about 65% of the catchment area, the remaining space being divided between 15% of pasture, 13% of housing and 7% of forests (Broers et al. 2005). As shown further on, a very detailed nitrate dataset is available for this aquifer. These elements make it a very representative example of groundwater resources at risk in the sense of the EU Water Framework Directive.

Geographical, geological and hydrogeological context

The Geer River is a tributary of the Meuse River downstream of the city of Liège. The basin, located in the northern part of the Walloon Region in Belgium (Fig. 2), extends over about 480 km2, with altitudes ranging between 80 m in the northeast and 206 m in the southwest, with a relatively flat topography. The Geer basin corresponds mostly to the unconfined part of the Hesbaye aquifer. However, the chalk formation dips northward and continues out of the basin. Because of that, the chalk aquifer extends also outside of the hydrological basin. Most of the aquifer is thus unconfined except in the northern part of the basin, where semi-confined conditions prevail on the north bank of the Geer River and outside of the Geer basin.

Fig. 2
figure 2

a Geer basin location (A–A′: geological cross section, see Fig. 3); b Location of sampling points used in the trend analysis

Full size image

The geology is made of a succession of Mesozoic and Cenozoic layers that dip northward, with a slope gradient between 1 and 1.5%. From top to bottom, the geology consists of (Fig. 3):

  • A Quaternary loess of variable thickness, up to 20 m

  • A maximum of 10 m of flint conglomerate, a highly heterogeneous geological formation made of dissolved chalk residues (flints, sand, clay and locally phosphate)

  • Locally, several metres of Tertiary sand deposits, mostly in the north of the basin, where they take the place of the flint conglomerate

  • Cretaceous chalks forming the main groundwater reservoir, showing depths ranging from a few metres in the south up to 100 m in the north-eastern part of the basin. In most of the area, this layer is divided into two main units by a thin layer of hardened chalk called the “Hardground”. This low permeability layer is not continuous, and where there is discontinuity, the hydraulic connectivity between the two main parts of the chalk aquifer can be enhanced.

  • Several metres of smectite clay of low hydraulic conductivity, considered as the base of the aquifer

Fig. 3
figure 3

Geological cross-section in the west part of the Hesbaye aquifer (see Fig. 2 for location of the cross section in the Geer basin; modified from Brouyère et al. 2004b)

Full size image

The mean hydraulic gradient in the aquifer is north-oriented (Fig. 4), ranging from 0.01 in the south to 0.003 in the north, close to the Geer River (Dassargues and Monjoie 1993). Groundwater levels are at depths ranging from 10 m to more than 60 m below the land surface.

Fig. 4
figure 4

Piezometric map of the Geer basin, based on a groundwater survey campaign in 1984 on the Hesbaye aquifer (in metres asl)

Full size image

The aquifer is exploited by two subsurface galleries and pumping wells owned by water companies, local industries and agricultural settlements. The northern gallery was dug in the chalk at a mean depth of 60 m. The southern gallery was dug at a mean depth of 30 m. These 40 km of galleries play a key role in the shape of the piezometric surface, acting in most part as a drain.

As highlighted by various groundwater studies in the basin, the chalk formation shows clear dual-porosity characteristics (Biver 1993; Hallet 1998; Brouyère 2001; Brouyère et al. 2004b). The large total porosity of the chalk (30–50%) allows for an important water storage capacity and the fissure porosity (about 1%) drains groundwater stored in the chalk matrix and provides the fastest pathways for transport.

Description of the nitrate dataset

The nitrate dataset used in this study comes mainly from the Nitrate Survey Network (NSN) established by the Walloon Region water authorities. In this network, existing boreholes, springs, galleries and traditional wells have been selected as monitoring points, where sampling and water analyses are carried out regularly, providing a spatial distribution of nitrate concentrations in the aquifer; in the best cases, data are available from 1957. Collected data are stored in a database of the Walloon Region water authorities. Each new dataset from the NSN is periodically imported into a hydrogeological database (Gogu et al. 2001; Wojda et al. 2005) developed and managed by the Hydrogeology Group of the University of Liège (HGULg).

The official NSN dataset has been extended using complementary points coming from other sources such as private owners or the VMW (Vlaamse Maatschappij voor Watervoorziening), the Flemish water supply company managing pumping wells in the north western part of the Geer basin and in the confined extension of the chalk aquifer in the Flemish region.

Finally, the network in the Geer basin nitrate study consists of 57 groundwater sampling points, among which 24 have been considered as containing suitable records for trend analysis (i.e. a minimum of 10 nitrate records over time). Importantly, however, there is a lack of knowledge on the position of the screens in most of the wells. Because of that, it was assumed that the deepest wells (generally the public water supply wells) were most probably screened in the totality of the aquifer. For the non-public water-supply wells, a similar criterion was used for the cases where knowledge on the screen position was not available. The remaining 33 points could not be used for trend analysis because of either a too small number of records or a lack of nitrate detections (mostly in the north-western part of the basin corresponding to the confined zone). However, they were considered as useful for estimating the present distribution of nitrate in the basin.

Figure 5 shows the location of selected sampling points, with some representative time-nitrate concentration plots. Sample points H-19 and H-20 correspond to water reservoir tanks where groundwater drained by the galleries is collected for further distribution by gravity flow. Even if they are reservoir tanks and not sampling points in the aquifer, these two points have been considered in the statistical trend analysis because they constitute amalgamation points, representative of most of the groundwater in the Hesbaye aquifer, at least of the large portion of the aquifer drained by the subsurface galleries.

Fig. 5
figure 5

Sampling locations in the unconfined part of the Hesbaye aquifer (inside the Geer basin) and the confined part of the Hesbaye aquifer (northwest of the Geer basin), and nitrate concentration as a function of time at selected locations

Full size image

In the north-west, outside of the Geer watershed basin, nitrate is often below detection limit in groundwater, the time-concentration plot for HF-1 being illustrative of the confined zone of the Hesbaye aquifer. The absence of nitrate in this part of the basin may have two explanations: denitrification processes in this part of the aquifer where prevailing confined conditions are favourable for anoxic reducing conditions, or the occurrence of very old, uncontaminated groundwater. This point will be discussed later.

The first columns of Table 1 summarize time periods of nitrate records for the 24 sampling points selected for trend analysis. In many locations, concentrations are approaching the drinking limit of 50 mg/L (e.g. H-12, H-18 and H-20), sometimes above (e.g. H-10 and H-15) (Fig. 5). As indicated by Hallet (1998), the unconfined part of the aquifer is characterized by nitrate concentrations frequently over 45 mg/L. EPIC (Erosion-Productivity Impact Calculator) simulations (Dautrebande and Sohier 2004) suggest that 88% of the groundwater nitrate contamination originates from diffuse agricultural sources, while domestic, point-sources are responsible for the remaining 12%. The EPIC simulations take into account statistical data on agricultural land use and spreading of domestic waste from the sewage system.

Some of the datasets exhibit clear periodic variations in nitrate concentrations (e.g. H-15 in Fig. 5). As discussed by Brouyère et al. (2004b), such periodic variations are explained by groundwater level fluctuations in the variably saturated dual-porosity chalk. In principle, nitrate spread over the land surface progressively infiltrates through the soil and the unsaturated zone and migrates slowly downward through the unsaturated chalk matrix. Under low groundwater level conditions, the nitrate contamination front in the unsaturated chalk can be disconnected from the saturated zone and nitrate concentrations in the aquifer tend to diminish because of dispersion and mixing processes. When groundwater levels rise, the contamination front is quickly reached and leached. The contamination source is then re-activated and nitrate concentrations are likely to increase rapidly in the saturated zone. Generally, the periodic variations of nitrate concentrations are better observed at observation wells screened in the shallowest part of the chalk aquifer, which confirms the explanation given here above.

In the subsequent statistical analysis, the seasonal effect of such periodic variations has not been considered explicitly because of the difficulty in defining a clear periodicity which is related to pluri-annual variations in the precipitation regime. Neglecting the seasonality (or pluri-annual periodicity) is not a problem in trend detection because the datasets integrate several periods and the global trend emerges. Considering only a reduced period of time (about 1 year) would of course lead to dubious conclusions because the general trend is less likely to be observed during a shorter observation window.

Point-by-point trend estimation

Table 1 summarizes the results obtained in terms of statistical trend analysis for the 24 sampling locations selected in the Geer basin. A total of 17 sampling points are characterized by an upward trend (71% of the points), the annual increase of nitrate concentration ranging between 0.3 and 0.8 mg/L. The remaining points (7 out of 24) that do not show any evidence of upward or downward trend, generally correspond to sampling points with limited nitrate records, irregularly distributed in time. One can notice that four of these datasets (H-6, H-11, H-9 and HF-18) consist of between 9 and 14 records, distributed over 8 years. These results imply that in the Geer basin and similar systems, a minimum of 20 samples (the more the better), distributed over a period of 10 years (frequency: two samples/year) is required in order to perform effective statistical trend analysis. Three sample points (H-4, H-15 and HF-21) from those seven do not show evidence of a trend, although they have a sufficient number of records. In the case of the sample point H-15, for example, where a high number of records exist, one could think that the absence of trend detection is masked by the very marked pluri-annual periodicity in nitrate concentrations.

Concerning the points where results were contradictory in terms of normality tests, both the parametric linear regression and the non-parametric Sen’s slopes were used to estimate the trend magnitude (see Table 1). Figure 6 shows that there is a very good agreement between slopes calculated using both techniques. Practically speaking, this means that the trend analysis methodology could be simplified by using the non-parametric method (Mann Kendall and Sen’s slope estimator) for both normally and non-normally distributed datasets, without performing first the normality test (Hirsch et al. 1991). On the other hand, the SW/SF-test indicates that all datasets are non-normally distributed except in three cases: H-7, H-16 and HF-18 (see Table 1). Among these, only one (H-7) shows a trend for which the trend slope estimation is similar using parametric and non-parametric methods. Based on this example, one could thus conclude that the linear regression estimator is also good for non-normally distributed data. However, there are not enough data (just H-7) on which to validate such a conclusion.

Fig. 6
figure 6

Correlation between simple linear regression slopes and Sen’s slope values

Full size image

Figure 7 gives a regional overview of nitrate trends in the Geer basin. Results indicate a general upward trend in the whole basin. However, as mentioned earlier, the unconfined part of the chalk aquifer, presenting high nitrate concentrations, has to be distinguished from the confined part of the aquifer (north of the watershed basin), where nitrate concentrations have been either below detection limit or very low. Figure 7 also indicates that the most significant upward trends are mainly located in the centre of the basin, where agriculture is more developed (Broers et al. 2005). One could think that subsurface galleries may act as a barrier to the migration of nitrate towards the north, explaining, in this way, the absence of nitrate in the confined zone. This is not likely to be the case because, even if the galleries do modify groundwater flows in the aquifer, they do not produce a reversal of groundwater flow at their vicinity. Furthermore, several points in the basin located to the north of the northern gallery (i.e. H-10 and HF-17) show high concentrations in nitrate and important upward trends (see Fig. 5 and Table 1), which confirms that the aquifer is also contaminated in this region.

Fig. 7
figure 7

Spatial distribution of nitrate trends in the Geer basin groundwater, with nitrate upward trend values in mg/L \( {{NO}}^{ - }_{3} \) per year

Full size image

Trend extrapolation

Presently, nitrate concentrations in groundwater are generally still lower than the drinking water limit of 50 mg/L, but from time to time, this threshold may already be exceeded. This is particularly the case in the southern subsurface gallery, which is at a shallow depth and thus more vulnerable to nitrate contamination. When exceedance occurs, the local water company mixes groundwater coming from the southern gallery with less contaminated groundwater coming from different points of the northern gallery in order to reduce the nitrate concentration of the supplied water to below the drinking water limit. Based on trend results obtained in the previous section, it is likely that this mixing procedure will be progressively less of an option in the future.

As mentioned by Grath et al. (2001), statistical techniques by themselves do not allow identification of trend reversal because they do not consider possible changes in manure application at the soil surface. However, in a context such as in the Geer basin, i.e. where there is a thick unsaturated zone, there is a long delay between changes at the land surface (i.e. reduction in manure-spreading) and the observed impacts in the aquifer (i.e. trend reversal). At first glance, this might lead to the conclusion that time is still available for taking adequate measures to effect a trend reversal. However, this oversimplified judgment assumes that measures taken to reduce the quantity of nitrate leaching to groundwater would have an immediate effect on groundwater quality. In fact, before reaching groundwater, nitrate must travel through the thick unsaturated formations (from 10 to 70 m) over a long period. Brouyère et al. (2004b) have estimated the mean vertical nitrate velocity across the unsaturated zone in the Geer basin to be approximately 1 m/year. Practically speaking, this means that measures taken today will have an observable effect in the groundwater with a delay of 10–70 years. Compared to that, the travel time of nitrate in the saturated zone to the pumping wells and galleries is extremely small, as it is dominated by fast preferential migration through the intense fracture network in the chalk (Hallet 1998; Brouyère 2001). Because changes in agricultural practices have started recently, one can anticipate that the upward trends observed in groundwater presently are not likely to reverse until many years into the future. Based on this assumption, a “simple” trend extrapolation is relevant in order to estimate the time remaining before groundwater is unusable for public water supply.

A rough estimation of the time remaining before the threshold concentration of 50 mg/L would be reached in various parts of the chalk aquifer, has been calculated based on a point-by-point extrapolation of current nitrate contamination levels using nitrate trend estimates as obtained in this research. To do so, the present contamination level has been considered at 36 of the 57 available sampling points (excluding the area to the northwest of the basin, where nitrate concentrations are mainly below detection limit) based on a point-by-point calculation of mean nitrate concentration measured over the period 1999–2003, a period for which the nitrate dataset is well furnished. The resulting nitrate distribution is considered to be representative of year 2001, as the mean year of the period 1999–2003. Then, using the calculated slope value at the nearest available point, an estimation of the time remaining to reach the drinking water limit was performed. The extrapolation exercise was applied to the points where a trend was detected but also to observation points where datasets were not sufficient to perform a trend analysis and to points where no trend was detected in the analysis. In the latter case, one can expect that the absence of a trend at these points is most likely related to the restricted size of the corresponding datasets but nitrate is generally present at high concentrations. One exception is well H-15 where, as mentioned before, the dataset is very detailed and does not show any trend. Anyway, this point has already reached the drinking water limit of 50 mg/L.

Table 2 lists all the points included in this analysis and associated extrapolation results (number of years before reaching the threshold concentration of 50 mg/L \( {{NO}}^{ - }_{3} \) and year of occurrence). Figure 8 shows the results of this calculation classified in four categories. Black circles correspond to locations where nitrate concentrations already exceed 50 mg/L \( {{NO}}^{ - }_{3} \). Red, orange and yellow circles correspond to locations where the limit will be reached within 30 years; between 30 and 60 years, and in more than 60 years, respectively. This confirms that the situation is more critical in the central part of the basin where for almost 80% of points, the drinking limit is expected to be reached within 30 years. Sampling points H-19 and H-20 are not represented because they correspond to water reservoirs tanks, not observation points in the aquifer.

Table 2 List of sampling points used for the calculated remaining time before exceeding the 50 mg/L \( {{NO}}^{ - }_{3} \) drinking water standard. ExcC currently exceeds the drinking water standard
Full size table
Fig. 8
figure 8

Time (years) foreseen before the 50 mg/L \( {{NO}}^{ - }_{3} \) threshold will be exceeded in groundwater

Full size image

Conclusions and perspectives

From a theoretical point of view, the statistical trend analysis, as performed in this research, seems to be robust. This is a result of the two-step procedure, with trend detection first and subsequent trend quantification. Results indicate that the preliminary normality test does not seem to be strictly necessary because the non-parametric MK-test works well with both normally and non-normally distributed datasets and the calculation of the trend amplitude using the Sen’s slope estimator consistently gives results that are comparable to the linear regression estimator. Indeed, it has been shown that in cases where parametric and non-parametric methods can be applied, results are similar for both methods, indicating that non-parametric methods (Mann-Kendall test in this case) are applicable for normally distributed datasets, making this test suitable for the Water Framework Directive application.

However, any statistical trend extrapolation relies on the strong and restrictive assumption that driving factors such as land use, agricultural practices, climate, and the dynamics of solute transport do not change with time. Thus, statistical trend extrapolation must be considered just as a warning procedure and not a forecasting tool. In order to go further in forecasting the evolution of nitrate concentrations in groundwater and in assessing or optimizing mitigation measures, more detailed and elaborated approaches are required such as transfer functions directly relating land-use and agricultural practices to the evolution of nitrate concentrations in groundwater; or even better, integrated mathematical modelling approaches based on observed physical processes of nitrate transport in the unsaturated and saturated zones.

The analysis presented here indicates that the problem associated with the evolution of nitrate concentrations in groundwater of the Geer basin is acute. Indeed, nitrate concentrations are already relatively high, sometimes locally above the drinking water limit; moreover the statistical trend analysis confirms that a general upward trend is observed throughout the basin. Furthermore, no downward trend has been identified, which means that there is no indication of trend reversal, despite the fact that first measures (EU 2000) have been introduced in relation to application of the EU nitrate directive, like control of manure spreading, and more recently the EU Water Framework Directive (Hall 1992). Extrapolation of the results, considering the “worst case scenario” of no significant changes in agricultural practices and land-use, indicates that the concentration threshold of 50 mg/L will be reached in 10–70 years in most of the unconfined aquifer. Measures should thus be adopted as soon as possible, having in mind the fact that, because of the long delay related to nitrate migration in the thick unsaturated zone, they will produce an observable effect within 10–70 years.

Further investigations have been performed in order to improve knowledge and understanding on the dynamics of nitrate in groundwater of the Geer basin. As pointed out by various authors (e.g. Broers 2004; Broers and van der Grift 2004) a strong relation is often observed between groundwater quality and age. In order to accommodate this factor, a groundwater dating campaign is ongoing in the Geer basin, using tritium and CFC-SF6 (chlorofluorocarbons and sulphur hexafluoride, respectively) as environmental tracers. However, the interpretation of the results is difficult and ambiguous because of the heterogeneity and dual-porosity of the chalk (Weissmann et al. 2002; Cook et al. 2005; LaBolle et al. 2006). Such interpretation should however contribute to a better explanation of the absence of nitrate in the semi-confined area north west of the Geer basin.

Finally, the establishment of a good network to monitor diffuse nitrate groundwater contamination is a difficult challenge in many regions (Smith and Ritzi 1993; Nunes et al. 2004). Further investigations and geostatistical analyses will be performed in the Geer basin in order to propose an optimized monitoring network.

Beside these new field investigations, a regional groundwater model is being developed in the scope of a FP6-IP project (AquaTerra 2004). This model will be used for nitrate trend analysis and forecasting in the Geer basin. For this purpose, nitrate trend results obtained in the present study will be aggregated and used as calibration and validation datasets for the groundwater flow and transport model.